(Focus on Biotechnology 3C) S. R. Weijers (auth.), Spiros N. Agathos, Walter Reineke (eds.) - Biotechnology for the Environment_ Wastewater Treatment and Modeling, Waste Gas Handling-Springer Netherla

BIOTECHNOLOGY FOR THE ENVIRONMENT:
WASTEWATER TREATMENT AND MODELING, WASTE GAS HANDLING

VOLUME3C
FOCUS ON BIOTECHNOLOGY
Volume3C
Series Editors
MARCEL HOFMAN
Centre for Veterinary and Agrochemical Research, Tervuren, Belgium
JOZEFANNE
Rega Institute, University of Leuven, Belgium
Volume Editors
SPIROS N. AGATHOS
Universite Catholique de Louvain,
Louvain-Ia-Neuve, Belgium
WALTER REINEKE
Bergische Universitiit,
Wuppertal, Germany
COLOPHON
Focus on Biotechnology is an open-ended series of reference volumes produced by
Kluwer Academic Publishers BV in co-operation with the Branche BeIge de la Societe
de Chimie Industrielle a.s.b.l.
The initiative has been taken in conjunction with the Ninth European Congress on
Biotechnology. ECB9 has been supported by the Commission of the European
Communities, the General Directorate for Technology, Research and Energy of the
Wallonia Region, Belgium and J. Chabert, Minister for Economy of the Brussels Capital
Region.
Biotechnology for the Environment:
Wastewater Treatment and Modeling,
Waste Gas Handling '-
Volume 3C
Edited by
SPIROS N. AGATHOS
UlIil'ersite Catho/ique de LOlll'Gill.
LOlll'Gill-Ia-Nelll'e. Belgium
and
WALTER REINEKE
Bergische UlIil'ersitiit.
Wuppertal. Germany
SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

A CI. P. Cataloglle record for this book is available from the Library of Congress.
ISBN 978-90-481-6224-6 ISBN 978-94-017-0932-3 (eBook)

DOI 10.1007/978-94-017-0932-3
Printed on acid-free papa
AlI Rights Reserved

© 2003 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 2003
Soficover reprint ofthe hardcover 1st edition 2003
No part of this work may be reprodllced, stored in a retrieval system, or transmitted
in any fonn or by any means, electronic, mechanical, photocopying, microfilming, recording
or otherwise, WithOllt written permission from ilie Publisher, with the exception
of any material supplied specificalIy for ilie purpose of being entered
and exeCll ted on a computer system, for exclusive use by the purchaser of the work.
EDITORS PREFACE
At the dawn of the 21st century we are witnessing an expanding human population
in quest of survival and continued well-being in harmony with the environment. Many
segments of society are increasingly preoccupied with the battle against both diffuse
and concentrated pollution, the remediation of contaminated sites, the restoration of
dainaged areas due to anthropogenic activities and the re-establishment of functioning
biogeochemical cycles in vulnerable ecosystems. There is an enhanced awareness of the
value of pollution prevention and waste minimization in industrial, urban and
agricultural activities, as well as an increased emphasis on recycling. Faced with these
major contemporary challenges, biotechnology is emerging as a key enabling
technology, and, frequently, as the best available technology for sustainable
environmental protection and stewardship.
Although the activities of microorganisms and their subcellular agents have been
recognized, studied and harnessed already for many years in the environmental arena,
there is a new dynamism in the in-depth understanding of the molecular mechanisms
underlying the functioning of microorganisms and their communal interactions in
natural and polluted ecosystems, as well as an undeniable expansion of practical
applications in the form of the new industry of bioremediation. A number of distinct but
increasingly overlapping disciplines, including molecular genetics, microbial
physiology, microbial ecology, biochemistry, enzymology, physical and analytical
chemistry, toxicology, civil, chemical and bioprocess engineering, are contributing to
major insights into fundamental problems and are being translated into practical
environmental solutions and novel economic opportunities.
The book set <<Biotechnology for the Environment», based on a compilation of
some of the outstanding presentations made at the 9th European Congress on
Biotechnology (Brussels, Belgium, July 11-15, 1999) and enriched with newly updated
thematic chapters, captures the vitality and promise of current advances in the field of
environmental biotechnology and is charting emerging developments in the beginning
of the new millennium. This third volume, subtitled 'Waste Water Treatment and
Modeling, Waste Gas Handling' presents current technological applications of
microorganisms in wastewater treatment and in the control of waste gas emissions,
illustrating the importance of multidisciplinary methodologies for years to come.
In the first section of the book special emphasis is placed on the use of rigorous
mathematical and conceptual models for an in-depth understanding of the complex
biology and engineering aspects underlying the operation of modem wastewater
treatment installations. Biological treatment plants today are without a doubt the major
man-driven microbiological process in terms of sheer volume and societal impact.
Although the design of wastewater treatment installations has evolved over many
decades, it is only recently that detailed biological knowledge has started being
translated into reliable dynamic models ensuring the predictable operation of such
plants and their robustness in the face of external shocks. A first chapter addresses the
theory behind the reduction of complex mathematical models such as the one describing
I
the activated sludge process, to make them more suitable for process control. This is
followed by three contributions demonstrating, respectively, the determination of
parameter sensitivity in the dynamic model of a relatively simple wastewater treatment
plant, the development of a general protocol to assess the performance of activated
sludge processes from a respirometry-based control perspective and, finally, the
comparative merits of model-, fuzzy logic- or artificial neural network-based
approaches for diagnostic purposes in wastewater treatment plants, including those
employing anaerobic treatment. The next chapter offers an extensive and structured
state-of-the-art review of numerous aspects underlying the evolving versions of
comprehensive activated sludge models, aiming at model calibration for any scale or
practical situation. This represents a major contribution in this field, since, until now,
the requisite information for model calibration and process characterization was
scattered in a vast amount of literature. Given the enormous growth in recent years of
biological nitrogen removal from wastewater, a final chapter in this section critically
reviews emerging developments in process design, operational optimization and control
of nitrogen removal.
The second section of the book addresses waste gas biofiltration, an expanding
biotechnological application of microbial metabolism for air quality assurance through
processes ranging from the abatement of hazardous volatile pollutants to the elimination
of nuisance odors. The opening chapter here provides an extensive up-to-the-minute
review of the diversity of biological waste gas treatment methodologies and applications
with a balanced attention on both the microbiological fundamentals and the engineering
principles of bioreactor design. In the next chapter an illustration of biofiltration
technology covers current developments in the design and installation of prototype and
full-scale bioreactors for the destruction of toxic or otherwise undesirable volatile
organic chemicals in industry. Finally, the power of an experimental biofilter in
efficiently biodegrading low-concentrations of organic volatile contaminants in indoor
air is presented in view of space travel applications.
The Editors hope that the integration of the depth of scientific fundamentals with
the breadth of current and future environmental applications of biotechnology so
evident in these selected contributions will be of value to microbiologists, chemists,
toxicologists, environmental scientists and engineers who are involved in the
development, evaluation or implementation of biological treatment systems. Ultimately,
a new generation of environmental scientists should take these lessons to heart so that
new catalysts inspired from the biosphere can be designed for safe, eco- and energy-
efficient manufacturing and environmental protection.
Spiros N. Agathos Walter Reineke
2
TABLE OF CONTENTS
Colophon ..................................................................................................................... II
EDITORS PREFACE ................................................................................................... 1
Table of contents .......................................................................................................... 3
PART 1 Wastewater treatment ..................................................................................... 9
Model reduction of activated sludge model no. 1 and bioprocess models for
identification and control ........................................................................................ 11
S.R. Weijers ............................................................................................................ 11
Summary ............................................................................................................. 11
1. Introduction .................................................................................................... 11
1.1. Need for reduction of rigorous, mechanistic models ............................... 11
1.2. Problem statement and methodology ....................................................... 13
1.3. Reduction approaches for process engineering systems .......................... 15
1.3.1. New model building from 'scratch' .................................................. 15
1.3.2. Simplifying assumptions ................................................................... 15
1.3.3. Neglect of dynamics by quasi steady state assumptions and singular
perturbation ................................................................................................. 16
1.3.4. Order reduction methods .................................................................. 16
1.3.5. Black-box identification ................................................................... 17
2. Reduction approaches for ASM1 .................................................................... 17
2.1. New model building from 'scratch' ......................................................... 17
2.2. Simplifying assumptions ......................................................................... 18
2.2.1. Simplification with respect to components ....................................... 18
2.2.2. Simplification due to separation of aerobic and anoxic conditions .. 19
2.2.3. Simplifying assumptions with respect to kinetics ............................. 19
2.2.4. Simplification with respect to dynamics ........................................... 19
2.2.5 Reduced order models ....................................................................... 19
2.3. Dynamics of variables ............................................................................. 23
2.4. Order reduction methods ......................................................................... 25
2.5. Black-box identification .......................................................................... 25
2.6. Discussion and conclusions ..................................................................... 25
3. Singular perturbation of bioprocess systems: theory and review .................... 26
3.1. Singular perturbation theory .................................................................... 26
3.2. Review of model reduction of (bio )process systems using singular
perturbational approach .................................................................................. 28
3.2.1. General rule for order reduction ....................................................... 28
3.2.2. Other reduction approaches .............................................................. 30
3.3. scaling in model reduction ...................................................................... .32
3.3.1. A method for state partitioning based on scaling .............................. 32
3.3.2. Relationship of scaling with regime analysis and dimensional
analysis ....................................................................................................... 36
3.4. Other methods for timescale analysis ...................................................... 36
3.5. Conclusions ............................................................................................. 37
3
4. Scaling for singular perturbation in a simple bioprocess system ................... .37
4.1. Introduction and methodology ................................................................ .37
4.1.1. Procedure 1: Direct scaling ............................................................... 38
4.1.2. Procedure 2: Timescale estimation for variables .............................. 38
4.1.3. Procedure 3: Analytical scaling procedure ...................................... .42
4.2. Model system: chemostat with biomass and substrate ............................ .42
4.2.1. Model system definitions and analysis ............................................ .42
4.2.2. Cases 1 and 2: Ratio SIX very low, 'C* very large ............................. .46
4.2.3. Case 3: 'C* intermediate, Mo«l ...................................................... .51
4.2.4 Case 4: S and X comparable, 'C' moderate, Mo "" 1........................... .53
4.2.5. Case 5: 'C* close to critical, Mo« 1, 'Ce "" Mo .................................. .54
4.2.6. Case 6: 'C* close to critical, Mo« 1, 'Ce « Mo ............................... .54
4.3. Other results ............................................................................................ .57
4.4. Conclusions ............................................................................................ .57
5. Conclusions and perspectives ......................................................................... 58
References .......................................................................................................... 60
Parametric sensitivity of a dynamic model for a pilot single-stage wastewater
treatment plant ........................................................................................................ 65
Igor Plazl, Goran Pipus, and Tine Koloini .............................................................. 65
Summary ............................................................................................................. 65
1. Introduction .................................................................................................... 65
2. Materials and methods .................................................................................... 66
3. Mathematical model ....................................................................................... 67
4. Results and discussion .................................................................................... 69
5. Conclusions .................................................................................................... 72
Acknowledgment ................................................................................................ 72
References .......................................................................................................... 72
Applicability of a simulation benchmark to respirometry-based control strategies73
John B. Copp and Henri Spanjers ........................................................................... 73
Abstract. .............................................................................................................. 73
1. Introduction .................................................................................................... 73
2. General benchmark description ...................................................................... 74
3. Plant layout ..................................................................................................... 75
4. Process models ............................................................................................... 75
5. Test protocol ................................................................................................... 76
6. Performance assessment ................................................................................. 77
7. Application to respirometry-based control strategies ..................................... 78
7.1. Joyce et al. (1974) .................................................................................... 79
7.2. Stenstrom and Andrews (1979) ............................................................... 79
7.3. SlIlrensen (1980) ....................................................................................... 79
7.4. Sekine et al. (1988) .................................................................................. 80
7.5. Kim et al. (1996) ..................................................................................... 80
7.6. Larose et al. (1997) .................................................................................. 80
8. Overview and discussion of results ................................................................ 81
9. Conclusions .................................................................................................... 84
Acknowledgements ............................................................................................ 84
4
References .......................................................................................................... 84
Fault detection and isolation in wastewater treatment plants ................................. 87
Jean-Philippe Steyer and Jerome Harmand ............................................................ 87
Abstract. .............................................................................................................. 87
1. Introduction .................................................................................................... 87
2. Examples of fault detection and isolation methods for anaerobic digestion
processes ............................................................................................................. 89
2.1. The experimental process ........................................................................ 89
2.1.1. Characteristics of the wastewater ...................................................... 89
2.1.2. The anaerobic digestion process ....................................................... 89
2.2. The model-based FDI approach ............................................................... 91
2.3. The ANN-based FDI approach ................................................................ 92
3. Fuzzy supervision of an industrial equalization process ................................. 94
3.1. The equalization process .......................................................................... 94
3.2. The fuzzy supervisor ................................................................................ 95
4. Conclusions .................................................................................................... 98
References .......................................................................................................... 99
Calibration of activated sludge models: a critical review of experimental designs
.............................................................................................................................. 101
B. Petersen, K. Gemaey, M.Henze, P.A. Vanrolleghem ................................. 101
Abstract ............................................................................................................. l 0 1
1. Introduction .................................................................................................. 101
2. Description of the state-of-the-art activated sludge models .......................... 102
2.1. Activated Sludge Model N°.1 (ASMl) .................................................. 102
2.1.1. COD components in ASMl ............................................................ 102
2.1.2. Nitrogen components in ASMl ...................................................... 103
2.1.3. Processes in ASMl ......................................................................... 106
2.1.4. Restrictions of ASMI ..................................................................... 108
2.2. Activated Sludge Model N°. 3 (ASM3) ................................................. 109
2.2.1. COD components in ASM3 ............................................................ 111
2.2.2. Nitrogen components in ASM3 ...................................................... 111
2.2.3. Processes in ASM3 ......................................................................... 114
2.2.4. Restrictions of ASM3 ..................................................................... 115
3. Model Calibration ......................................................................................... 115
3.1. Information set for model calibration .................................................... 117
3.2. Model calibration levels ........................................................................ 119
3.2.1. Steady state model calibration ........................................................ 119
3.2.2. Dynamic model calibration ............................................................. 120
4. Characterisation of wastewater and sludge kinetics ..................................... 124
4.1. Wastewater characterisation .................................................................. 124
4.1.1. Physical-chemical characterisation ................................................. 124
4.1.2. Summary and discussion of physical-chemical wastewater
characterisation ......................................................................................... 128
4.1.3. Biological characterisation ............................................................. 129
4.1.4. Summary and discussion of biological wastewater characterisation
.................................................................................................................. 143
5
4.1.5. Discussion on physical-chemical vs. biological wastewater
4.2. Characterisation of sludge composition ................................................. 148
4.3. Characterisation of stoichiometric and kinetic parameters .................... 150
4.3.1. Respirometry .................................................................................. 150
4.3.2. Nitrate utilisation rates .................................................................... 160
4.3.3. Titrimetry ........................................................................................ 161
4.3.4. Summary and discussion of biological characterisation of
stoichiometric and kinetic parameters ...................................................... 161
4.4. Is characterisation via lab-scale experiments relevant? ......................... 163
4.4.1. Kinetic and stoichiometric parameters ............................................ 166
4.4.2. Relevant kinetic and stoichiometric parameters for lab-scale
4.4.3. Relevant wastewater components for lab-scale characterisation .... 171
5. Biological experimental constraints .............................................................. 172
5.1. Transferability between model concepts: an example ........................... 173
5.2. Review and discussion of S(O)/X(O) ratio .............................................. 174
5.2.1. Effect of S(O)/X(O) on stoichiometry .............................................. 175
5.2.2. Effect of S(O)/X(O) on kinetics ....................................................... 178
5.2.3. Discussion on S(O)/X(O) ratio ......................................................... 179
6. Summary ....................................................................................................... 180
Acknowledgement ............................................................................................ 181
References ........................................................................................................ 181
Optimization and control of nitrogen removal activated sludge processes: a review
of recent developments ......................................................................................... 187
Zhiguo Yuan, Jiirg Keller and Paul Lant .............................................................. 187
Abstract. ............................................................................................................ 187
1. Introduction .................................................................................................. 187
2. Elementary analysis of biological nitrogen removal systems ....................... 189
2.1. System analysis ...................................................................................... 189
2.1.1. SRT and volume requirement ......................................................... 189
2.1.2. Anoxic fraction and volume requirement ....................................... 190
2.1.3. Anoxic fraction and nitrification capacity ...................................... 190
2.1.4. Anoxic fraction and influent COD utilization efficiency for nitrate
reduction ................................................................................................... 191
2.1.5. Alkalinity and pH ........................................................................... 193
2.2. Optimization opportunities .................................................................... 193
2.2.1. Optimisation by on-line process control.. ....................................... 193
2.2.2. Optimisation by improved process design ...................................... 194
2.3. Conclusions ........................................................................................... 195
3. Aeration control ............................................................................................ 195
3.1. Aeration phase length control based on ORP and pH measurement... ... 195
3.1.1. Aeration control based on absolute values ofORP and pH ............ 197
3.1.2. Aeration control based on bending points ofORP and pH ............. 198
3.2. Aeration control based on respirometry ................................................. 199
3.3. Aeration control based on ammonia and nitrate measurement .............. 200
6
3.3.1. Objective functions ......................................................................... 200
3.3.2. Aeration control of alternating systems .......................................... 201
3.3.3. Aeration control of pre-denitrification systems .............................. 202
3.3.4. Model-based control ....................................................................... 203
3.4. Conclusions ........................................................................................... 204
4. External COD dosage optimisation and control ........................................... 204
4.1. External carbon sources ......................................................................... 205
4.2.Utilization efficiency of external COD ................................................... 206
4.3. External carbon dosage control systems ................................................ 207
4.3.1. Control of external carbon dosage to recirculating biological nitrogen
removal systems ....................................................................................... 208
4.3.2. Control of external carbon dosage to alternating biodenipho®
systems ..................................................................................................... 210
4.3.3. Control of external carbon to two-sludge post-denitrification systems
.................................................................................................................. 210
4.4. Conclusions ........................................................................................... 210
5. SRT optimisation and controL ..................................................................... 211
5.1. Impact of SRT on a biological nitrogen removal plant... ....................... 211
5.2. SRT minimization via surplus sludge waste flow control ..................... 212
5.3. Conclusions ........................................................................................... 213
6. Side-stream nitrifier supplies ........................................................................ 213
6.1. Shortening the RTs of inert solids via sludge storage ............................ 214
6.2. Side stream nitrification of reject water ................................................. 215
6.3. Conclusions ........................................................................................... 216
7. Novel multi-sludge systems .......................................................................... 217
7.1. Attached growth processes .................................................................... 217
7.1.1. SRT and volume requirement ......................................................... 218
7.1.2. Treatment capacity .......................................................................... 218
7.1.3. Influent COD utilization efficiency for nitrate reduction ............... 218
7.1.4. Comparison with other multi-sludge systems ................................. 219
7.1.5. Aeration in an attached growth system ........................................... 219
7.2. COD preservation for denitrification ..................................................... 219
7.3. Conclusions ........................................................................................... 221
8. Conclusions .................................................................................................. 221
Acknowledgment .............................................................................................. 222
References ........................................................................................................ 222
PART 2 Waste gas biofiltration ............................................................................... 229
Performance and characterisation of a membrane biological air filter for space
applications ........................................................................................................... 231
Jaap Van Der Waarde, Arjan Van Der Werf, Maurice Henssen, Bert Geurkink,
Klaas Van Der Marel, Piet Paul and Marc Gent.. ................................................. 231
Summary ........................................................................................................... 231
1. Introduction .................................................................................................. 231
2. Materials and methods .................................................................................. 232
3. Results and discussion .................................................................................. 233
3.1. Physiological characterisation ............................................................... 233
7
3.2. Performance of the biological air filter (BAF) ....................................... 235
3.3. Molecular characterisation ..................................................................... 235
3.4. Performance of the BAF at extremely low concentrations of volatile
contaminants ................................................................................................. 236
4. Conclusion .................................................................................................... 236
References ........................................................................................................ 237
Biofiltration for waste gas handling ..................................................................... 239
Frederic Thalasso, Maria C. Veiga and Christian Kennes ............................. 239
I. Introduction .................................................................................................. 239
2. Bioscrubbing ................................................................................................. 239
3. Trickling filters ............................................................................................. 240
4. Biofiltration with organic packing materials ................................................ 242
5. Applications .................................................................................................. 246
5.1. General aspects ...................................................................................... 246
5.2. Industrial scale applications ................................................................... 248
6. Other biological gas treatment technologies ................................................. 250
7. Conclusions .................................................................................................. 251
Acknowledgements .......................................................................................... 251
References ........................................................................................................ 252
Bioreactors for the treatment of industrial waste gases containing formaldehyde
and other aliphatic compounds ............................................................................. 259
6scar J. Prado, Marta Eiroa, Maria C. Veiga and Christian Kennes .................... 259
1. Introduction .................................................................................................. 259
1.1. Gas phase bioreactors ............................................................................ 259
1.2. Formaldehyde as industrial air pollutant... ............................................. 260
2. Biodegradation of formaldehyde .................................................................. 261
3. Case studies on biological abatement of formaldehyde in industrial waste
gases ................................................................................................................. 264
3.1. Background ............................................................................................ 265
3.2. Optimization of the removal of formaldehyde in biofilters and
biotrickling filters ......................................................................................... 270
Acknowledgements .......................................................................................... 272
References ........................................................................................................ 273
INDEX ...................................................................................................................... 275
8
PARTl
WASTEWATER TREATMENT
MODEL REDUCTION OF ACTIVATED SLUDGE MODEL NO.1 AND
BIOPROCESS MODELS FOR IDENTIFICATION AND CONTROL
S.R. WEIJERS
Systems and Control Group, Faculty ofApplied Physics
Eindhoven University of Technology, Eindhoven, The Netherlands
email: S.R.Weijers@tue.nl
Summary
This paper treats model reduction of the important Activated Sludge Model No.1
(ASMl) for dynamic modelling of wastewater treatment plants. Section 1 motivates the
need for model reduction and gives an overview of reduction approaches in the
literature. Section 2 reviews model reduction of activated sludge models, especially
ASMI. The review shows that model reduction is often performed quite heuristically
and that more insight in the timescale properties of ASMI is desired. Singular
perturbation is a systematic technique for order reduction. Section 3 reviews application
of singular perturbation to bioprocess systems to reveal if methods exist to obtain or
detect the so-called standard form, which is the difficult part of reduction by singular
perturbation. Because existing methods appear insufficiently straightforward, Section 4
develops a method for this task. Three procedures are proposed and tested on a simple
continuous general bioprocess model, which yields valuable insight into properties of
bioprocess models. A proposed timescale estimation procedure appears a helpful tool in
model reduction through timescale separation and provides a good basis for further
reduction studies.
1. Introduction
1.1. NEED FOR REDUCTION OF RIGOROUS, MECHANISTIC MODELS
Dynamic simulation based on rigorous, physics-based mechanistic modelling has

become a standard tool in many engineering fields. Examples are finite element models
in structural dynamics, computational fluid dynamics, the SPICE program for circuit
11
S.N. Agathos and W. Reineke (eds.), Biotechnology for the Environm
Wastewater Treatment and Modeling, Waste Gas Handling, 11-63.
© 2003 Kluwer Academic Publishers.
S.R. Weijers
analysis in electrical engineering and dynamic flowsheeting in chemical engineering. In

biotechnology, process modelling is well established (Roels, 1983).
Rigorous models are applied for a variety of tasks. They are useful in system design,
especially to check system behaviour under extreme dynamic loading conditions. They
are very helpful in understanding system behaviour in science and engineering. As they
typically contain much prior knowledge from the relevant application domain, their
prediction accuracy and range of validity can be high in principle, if the model building
and model tuning tasks are designed to achieve this accuracy. However, the costs of
rigorous models are generally high because of the effort required to construct them.
For engineering tasks described above, typically high-dimensional models result.
Despite the usefulness of rigorous, large models for the tasks described above, for the
following important tasks simple models are better suitable:
• Process control: Many control theory concepts are only applicable to low-
order models. The high dimensionality of large models results in enormous
computational requirements, ill-conditioned problems and often stiff numerical
problems due to interaction of slow and fast dynamics (Kokotovic et al., 1986).
Relatively low-order reduced models are therefore required for controller
design and as internal models in model-based control.
• Model identification: Rigorous mechanistic models typically require high
investments in model tuning and validation. Moreover, problems that are more
fundamental exist because large models typically exhibit lack of parameter
identifiability. In addition, mechanistic models contain internal states whose
behaviour is difficult to verify (or falsify), which is referred to as verifiability
(Jeppsson, 1996). The need for simple, well-identifiable models holds true for
off-line identification and even stronger for on-line identification for process
monitoring and adaptive control.
• Understanding of model behaviour: While rigorous models are helpful in
system understanding, at the same time this understanding is hampered by their
complexity. Much understanding can be acquired from reduced models
describing only the most important phenomena.
• System design through rigorous optimisation: Many design problems might be
solved more straightforwardly by applying an analytic design procedure, using
(mathematical) multi-criteria optimisation. Due to their size, however, rigorous
models are not suitable for direct system design using optimisation; rather,
they are most often used to check designs. Straightforward, systematic system
design employing rigorous multi-objective optimisation would be facilitated if
simple models containing the most important phenomena would be available.
Concluding, one may state that model reduction is required for several important tasks.
In wastewater engineering, rigorous modelling and dynamic simulation have become
well accepted during the past decade since the publication of Activated Sludge Model
No.1 (ASM1, Henze et aI., 1987). This model is now becoming routinely used in
wastewater engineering. Model tuning of ASM1 has become an important task for
process analysis, process optimisation and process control. Furthermore, advanced,
model based control of wastewater treatment plants is expected to be important in
12
Model reduction of activated sludge model no. 1 and bioprocess models for identification and control
meeting more stringent treatment requirements that are imposed on wastewater

treatment, especially with respect to nitrogen removal, in an economical way.
Mechanistic models based on ASMI are however not directly suitable for control. If
these mechanistic models accurately describe the process dynamics, however, reduced
models can be extracted for controller design thus re-using the knowledge in the
mechanistic models. Consequently, model reduction of ASMI has received
considerable attention in the literature.
In bioprocess engineering, reduced models may be fruitfully applied in process
optimisation and process control, for example for dynamic optimisation of individual
(fed-) batch processes, batch process scheduling and model based process monitoring.
1.2. PROBLEM STATEMENT AND METHODOLOGY
This paper wants to contribute to advancing the field of model reduction of ASMI and
bioprocess models, with emphasis on application in (model based) control and
identification. The ultimate goal is to provide a methodology to derive nonlinear
reduced models for different timescales in a straightforward, systematic manner. In this
paper, the goal is to develop a method that can provide the starting point for such a
straightforward nonlinear reduction procedure. This section outlines the methodology
that is followed in the sequence to develop such a method.
To select a suitable reduction method, it is useful to first list desired properties of a
candidate reduction method. The following properties are considered as desired in this
paper.
Nonlinear reduction methods are preferred, because nonlinear models can have a
larger validity range than linear models.
The stiffness ofthe activated sludge process requires the development of models that
are suitable for different timescales. For example, control of dissolved oxygen (DO)
requires a different timescale than control of ammonia (and nitrate) or control of sludge.
Models on these different timescales are therefore required for each of these control
tasks, either for controller design or as internal model. Moreover, this fits well in a
hierarchical control approach. Therefore, especially model reduction based on timescale
separation is a logical approach.
Moreover, it is desired that the reduction method is systematic and straightforward
in order to avoid time-consuming trial-and-error and iterations, and to be independent of
(too much) application-domain dependent knowledge. Ideally, the reduction method
would also supply an estimate of the error induced by the reduction.
Another desired property of reduction methods is that the states of the model retain
their physical interpretation after reduction. This will enable a more direct interpretation
of the controller design results and of the control actions. Moreover, if adaptive control
is applied, identified parameters have a direct physical interpretation.
Finally, as this paper focuses on the pre-denitrification plants and carrousels, the
method should be applicable to derive reduced methods for these systems. Several
reduced models have been derived for pre-denitrification plants, or, more exactly, for
aerobic and anoxic conditions separately. However, reduced models for systems where
simultaneous denitrification takes place, such as typically is the case in carrousels, are
lacking. Methods that are helpful in reduction for these systems are therefore desired.
13
S.R. Weijers
The need for reduced models in wastewater engineering has resulted in a variety of
reduction approaches and reduced order models of ASM1. Section 1.3 discusses the
most important reduction approaches in the literature. Section 2 gives a state-of-art
overview of ASMI model reductions that are classified along the discussed reduction
approaches. Aim of this overview is to obtain clarity into the limitations and
possibilities of different reduced models and reduction methods. This should support
selection of a suitable reduced model or reduction method.
The review in Section 2 shows that reduction through timescale separation is applied
by several authors, but often in a heuristic fashion. Moreover, the timescale properties
of ASMI are not well understood.
The technique of singular perturbation is therefore studied in depth to better
understand the timescale properties of ASMI and thus provide a basis for developing a
reduction methodology. This technique has the desirable properties of a candidate
reduction technique listed above. It provides the mathematical basis for reduction by
timescale separation, provides an error estimate, it is a systematic technique, and may
therefore lead to a more straightforward reduction procedure. Moreover, it is applicable
to nonlinear systems and, under certain conditions, the physical interpretation of states
is retained.
In section 3.1, the theory of singular perturbations (as used by Kokotovic et at.,
1986) is summarised. For application of the singular perturbation technique for order
reduction, the model must be in the so-called standard form. This means that the states
of the model can be partitioned into 'fast' and 'slow' scales. (Also partitioning into
more timescales is possible (e.g., fast, medium and slow». In that case, the reduction
boils down to either eliminating the fast states to obtain the slow model or to
eliminating the slow states to obtain the fast model. Upon reduction, the physical
interpretation of states is then preserved.
If the model is not in standard form, a state transformation may be applied to bring it
into standard form. In that case, the physical interpretation of the states is not retained.
In fact, the fast and slow modes cannot be assigned to disjunct sets of states. This
situation is not studied here.
It also happens that the model can be partitioned into slow and fast states, but that it
is not easy to recognise this from the model equations. In that case, writing the model
equations in a different way, e.g. by scaling, may show immediately from the scaled
equations that the partitioning is possible indeed. Finding such a scaling is difficult
however. In fact, recognising whether the model is in standard form and detection of the
fast and slow timescales is the difficult part of the reduction technique.
Therefore, the study in this paper focuses on recognising the state partitioning into fast
and slow states and on recognising or obtaining the standard form.
One application of singular perturbation theory to an activated sludge plant model
has been reported, which is discussed in section 2.3. The authors applied a method that
is based on eigenvalues for the state partitioning. The reduction was only partly
successful. section 3.2 reviews application of singular perturbation to reduction of -
more general and closely related - bioprocess systems into more detail to obtain a more
fundamental insight into timescale properties of bioprocess models in general and of
ASMI in particular. The review shows that singular perturbation of bioprocess models
is not sufficiently well understood and it does not yield a method or rule for the state
14
partitioning. Therefore, a method for this task is developed. A scaling procedure

summarised in section 3.3 provides the starting point for this method.
Three procedures are proposed in section 4 as candidates for the state-partitioning task.
They are tested on a simple bioprocess model in different operating conditions.
1.3. REDUCTION APPROACHES FOR PROCESS ENGINEERING SYSTEMS
Starting point in model reduction is the definition of the goal for which the model is
intended. It can be argued that, for biological systems even stronger than in other basic
sciences, the goal of the model determines its formulation (Vansteenkiste and Spriet,
1982). The goal determines selection of model inputs, including possible reference
trajectories and disturbances, selection of model outputs, and determines which model
accuracy is required over which time horizon.
For the actual model reduction in this review, several approaches are distinguished
and briefly explained below. The models obtained with these approaches differ in their
degree of 'greyness'. Models that are obtained via systematic reduction of a white
model whilst preserving the physical interpretation of the system states can be
considered light grey. Simple, mechanistic input-output models are considered grey
(Carstensen et al., 1995). The last category is that of "black box models", such as
polynomial models or artificial neural networks. It is noted that also mixed forms can be
applied, e.g. models contain a mechanistic part and an artificial neural network (e.g. van
Can, 1997). Alternatively, artificial neural networks can be configured to contain prior
knowledge, e.g. of model structure.
The models are expected to allow further extrapolation beyond the experimental
domain with increasing 'lightness'.
1.3.1. New model building from 'scratch'

One way to construct simple models is to disregard existing rigorous models and to
build simple models from scratch. Of course, strictly speaking this is no model
reduction. Nevertheless, the prior knowledge contained in rigorous models is often
either implicitly or explicitly used in this approach.
1.3.2. Simplifying assumptions

Application of simplifying assumptions is a very frequently applied approach in model
reduction. Here, we distinguish simplifying assumptions with respect to:
• Components; e.g. aggregation of variables (for example COD and BOD as

total measures of pollutant concentrations);
• Processes; e.g. aggregation of reactions (for example modelling nitrification as
a one-step process whilst it is a two-step process);
• Lumping of space distribution: neglect or simplification of gradients in one or
more directions;
• Kinetics, e.g. simplification of complicated kinetic schemes;
• Dynamics; these are treated in the next paragraph.
15
S.R. Weijers
Often, simplifying assumptions are applied in a rather heuristic fashion, especially with
respect to components. For reduction of distributed systems, several systematic methods
can be applied (Dochain, 1994), which are outside the scope of this paper.
1.3.3. Neglect of dynamics by quasi steady state assumptions and singular perturbation
The neglect of dynamics that are fast or slow compared to the timescale of interest is
discussed in its own right here, as it is a central reduction approach. For example, it
provides the basis for application of hierarchical control. In hierarchical control, a
layered control approach is applied to control large, composite systems, for example in
plant wide control. The control problem is decomposed into a hierarchical set of several
levels of smaller sub-problems. On each level, dynamics of lower levels are assumed
very fast and considered to be in (pseudo) steady state and dynamics of higher levels are
assumed very slow and considered as constant.
In many cases, neglect of dynamics is performed heuristically. From field specific,
physical understanding and process knowledge, fast states are omitted or slow states
considered constant, without firm motivation. Although heuristically reduced models
may (seem to) perform satisfactory, there is a need for a more thorough understanding
of criteria for performing this reduction, preferably supported by formal proof or error
analysis.
A well-known heuristic reduction approach! is the quasi-steady state approximation
(QSSA) for reactive intermediate species. This approach, first introduced by Bodenstein
and Lutkemeyer (1924) (Bowen et aI., 1962), has been extensively used in kinetic
modelling. Criteria for its application were originally that concentration and timescale
of the intermediate species are small. Later, the nature and consequence of the QSSA
have been studied more thoroughly.
Singular perturbation is a mathematical technique to analyse timescale multiplicity
and to perform a systematic order reduction and error analysis. It is the appropriate tool
to provide the mathematical basis for application of quasi steady state assumptions. Its
application to the QSSA in Michaelis-Menten kinetics is well studied, as will be
summarised in section 3.3.
1.3.4. Order reduction methods

In modem control engineering, order reduction of linear models is a very important task
in control system design. Models for control are often obtained from linearization of
high-dimensional rigorous models, resulting in very high dimensional models. Modem
controller design methods, especially robust control design methods, yield even higher
dimensional controllers. Order reduction of model or controller has become a necessity
if they are to be implemented, as high-order controllers are usually not accepted in
industry. Moreover, they may exhibit extremely poor robustness against controller
parameter errors.
Consequently, order reduction has rapidly progressed and since the milestone
publication by Moore (1981), a variety of reduction methods have been developed and
are relatively well understood. Examples of reduction methods are Hankel norm
reduction, balanced reduction and modal reduction. More recently, closed loop model
! With 'model reduction', we will also denote 'model order reduction'.
16
reduction has received attention; see for example Wortelboer (1994), who also gives a
thorough discussion on linear reduction methods.
For nonlinear system, balanced reduction methods have been developed (see
Scherpen, 1994). With these methods, however, it may be more difficult to preserve the
physical interpretation of states, as a state transformation is typically involved.
Moreover, these methods do not provide a timescale separation. Instead, they eliminate
states that are poorly jointly observable and controllable and thus contribute least to
input-output behaviour. Because we want to preserve the physical interpretation of
states and focus on timescale separation, these methods are not considered further.
1.3.5. Black-box identification

Reduction through black box identification is an approach that requires little prior
knowledge. Linear black box identification can be applied for weakly nonlinear systems
within a limited domain. To construct reduced models of nonlinear systems with
validity over a larger domain, nonlinear black box modelling techniques may be
applied, especially artificial neural networks (ANN's). The design of test signals in
reduction through identification is very important. By selecting appropriate frequency
ranges for the excitation signals, black box models may be obtained for different
timescales.
2. Reduction approaches for ASMI
This section reviews approaches applied to reduce ASM1. The emphasis is put on
systems with enhanced nitrogen removal. In most of the cases presented, the purpose of
model reduction is application for identification or control. The subsections follow the
arrangement of reduction approaches as presented in section 1.3. For each case
discussed, the treatment system, goal for model reduction, motivation for selected
approach will be indicated together with (a description of) the reduced models. For a
description of ASMI please consult Henze et al. (1987) or other sources (e.g. Dold and
Marais, 1986).
2.1. NEW MODEL BUILDING FROM 'SCRATCH'
Marsili-Libelli (1989) developed a low-order model for conventional activated sludge

systems with BOD removal and nitrification. Motivation was that literature models are
not suitable for control, due to their complexity and poor identifiability. The model was
developed to describe a) biodegradation carbonaceous COD b) nitrification c) DO
utilisation d) sludge sedimentation. As a growth model a predator/prey modified
Volterra-Leslie logistic equation was used instead of the usually applied, poorly
identifiable Monod model. Theoretical and practical identifiability of the model were
investigated and an observer was constructed based on this model.
Isaacs (1996) formulated several simple models for control of the Biodenipho® system,
which is an alternating, sequential semi-batch type system with nitrogen (and
phosphorus) removal. Four (or six) phases are applied in each cycle. The control and
model horizon is one phase, with a timescale in the order of several minutes. Three
model-based control strategies were tested, external carbon addition control (ECAC),
17
S.R. Weijers
dissolved oxygen setpoint control (DOSPC) and cycle length control (CLC). All
controllers employ a relational model and a predictive model. The predictive model is
used to compute required denitrification rate, nitrification rate or cycle length during
one phase. The predictive model was roughly the same in all control strategies,
assuming zero-order kinetics both in nitrification and denitrification phase. Changes in
biomass amount and composition need not be predicted, as actually measured
denitrification and denitrification rates from the last preceding cycle are used. Different
relational models were applied in the different strategies, as explained below. The
relational model in CLC is trivial.
In ECAC, the prediction model computed the required denitrification rate rd; the
required external carbon addition rate qCOD to achieve this rd is computed from the
following relational model with the constants rd, b and rd, max:
(2.1)
In the model, the half-rate constant KCOD was fixed, so the model is linear in parameters.
The model parameters were estimated on-line from measured carbon addition and
denitrification rates in preceding cycles using recursive least squares.
In DOSPC, the required nitrification rate rn is computed with the prediction model.
The required DO setpoint to achieve this nitrification rate is computed from the
relational model and is held constant during one phase (using DO control).
(2.2)
This relational model is estimated from measured DO and nitrification rates in

preceding cycles using recursive least squares.
2.2. SIMPLIFYING ASSUMPTIONS
Several simplifying assumptions are applied in ASMI reduction, which can be

classified as follows (for an explanation of the symbols, see the symbols list at the end):
2.2.1. Simplification with respect to components
COD components:
• No hydrolysis, no distinction between soluble and suspended biodegradable
COD
• Do not consider inert products as a separate component (XI includes Xp)
• Do not consider inert components at all (no Sj, XI, Xp)
N components:
18
Model reduction of activated sludge model no. I and bioprocess models for identification and control
• No suspended organic bound entrapped N (No XND )

• No ammonia production from organic bound nitrogen (No SND)
Alkalinity:
• Do not consider alkalinity
2.2.2. Simplification due to separation of aerobic and anoxic conditions
• No denitrification under aerobic conditions

• No nitrification under anoxic conditions
2.2.3. Simplifying assumptions with respect to kinetics
• Replace terms with Monod kinetics by fITst-order or zero-order kinetics (the

last corresponds to neglect of dependency on one or more limiting substrates)
2.2.4. Simplification with respect to dynamics
• Neglect DO dynamics
• Neglect dynamics of solute components
• Neglect dynamics of suspended components
2.2.5 Reduced order models

If one considers the huge number of possible combinations of simplifying assumptions,
it is clear that a large number of reduced order models can be formulated. Up till now, a
limited number of reduced order models have been published, which are presented in
this section.
A reduced and modified ASMI model to be used in adaptive control was formulated
by van Impe et al. (1991) and by Vanrolleghem (1994) . This is not discussed further, as
it was developed for conventional activated sludge systems.
Carstensen et al. (1995) formulated grey models to have operational dynamic
models capable of giving on-line information on the present state of the wastewater
treatment plant. The system was the Biodenipho process, which has an alternating,
sequential semi-batch type of operation, where nitrification and denitrification are
separated in time. Different models are applied for both conditions. No biomass growth
is modelled.
Nitrification:
• The rate equation for ammonia removal contains both the maximal autotrophic
growth rate and yield, which are not identifiable if the rate equation is not
coupled to growth model for autotrophic biomass. Therefore, one overall
maximal nitrification rate parameter rnit,max is used
19
S.R. Weijers
• No method for measuring the concentration of the autotrophic biomass exists,

so the measured total solids concentration Xss is used. This gives the following
rate expressions during one batch phase.
(2.3)
(2.4)
Denitrification:
• One overall maximal denitrification rate rdenit,max is used.

• No method for measuring the concentration of heterotrophic biomass exists, so
the measured total solids concentration is used.
• The ammonium load to the plant is used as this proved a correlated measure of
the readily biodegradable substrate concentration, similar to Isaacs (1996) (Eq.
2.1).
dS NO S NO rload X (2.5)
- d t = -rdenit,max . KNO +SNO K SS'
load + rload
Lukasse et al. (1997a,b, 1998a,b, 1999) developed reduced models for control of
nitrogen removal in an activated sludge process consisting of one CSTR; clarifier
dynamics were neglected. Two levels of model reduction were applied:
• For nonlinear (open-loop) optimisation (optimisation horizon 1 day, Lukasse et

al., 1998a). To investigate optimal aeration strategies, a nonlinear model was
developed. Dynamics of all suspended components, including biomass, were
neglected. With respect to solute components, inert COD and alkalinity were
not considered; DO dynamics were neglected by considering DO as an input
instead of a state variable (assuming tight control). Ammonia production due
to hydrolysis and ammonification was simplified by using an (constant)
influent ammonia correction factor (in fact, this last simplification implies a
quasi-steady state assumption on hydrolysis and ammonification). These
assumptions result in a reduction of ASMI to a reduced third-order nonlinear
model with the state variables Ss, SNH, SNO'
• In receding horizon optimal control (Lukasse et al., 1997a,b, 1998a,b, 1999).
To circumvent the disadvantages associated with the open-loop strategy,
especially associated with uncertainty of the affinity constants, feedback was
introduced by applying RHOC (Receding Horizon Optimal Control), a Model
Predictive Control law. Here, the requirement was to obtain an identifiable
model, suitable for on-line optimisation of an alternating process, with a
20
timescale of interest of hours. Assumptions were stated in Lukasse et al.

(1997a and 1998a,b) (slightly different models were used, however): the
system was forced to be either aerobic or anoxic by restricting So E {O, 3},
transients between aerobic and anoxic phase were neglected, Monod kinetics
was simplified to Blackman kinetics, arguing that the corresponding affinity
constants were small and the substrate dependency only serves as a switching
function. Additional assumptions were 1) If SNH=O and the system is aerobic,
then the nitrification rate is such that all ammonia entering the reactor is
nitrified. 2) ammonia and nitrate concentrations in the sludge recycle are equal
to the concentrations in the reactor. This leads to the following equations:
(2.6)
(2.7a)
(2.7b)
with UE {O, 1 }. (1 for aerobic, 0 for anoxic) and Din is the dilution rate. For use in
adaptive RHOC, SNH,in, rNH,rnax and rNO,rnax were estimated based on measurement of
SNH, SNO and So·
Jeppsson (1996) developed a reduced model for ASMI for application in control
including adaptive control and requiring identifiability and 'verifiability' of the
parameters and states (Jeppsson and Olsson, 1993). This last property expresses the
specification that the states of the model are verifiable with information obtained from
measurements. Reactors are assumed to be either aerobic or anoxic (separated in space)
and ideally mixed (pre-denitrification system). The model has been used in studies by
other authors (Ayesa et aI, 1995) and is presented here into more detail.
The following assumptions and simplifications were made:
• Do not consider alkalinity

• Do not consider inert components (biologically not relevant, XI is slow)
• No ammonia production from organic bound nitrogen (No XND , SND)
• No hydrolysis, no distinction between soluble and suspended biodegradable
COD (hydrolysis is not well understood; difficult to measure Ss and Xs;
neglect fast dynamics of Ss).
• Neglect DO dynamics: assumed to be controlled
21
S.R. Weijers
• Approximate Monod expressions by Blackman kinetics (it is assumed that

substrate concentration is low).
Under these assumptions, five model components remain, namely X BR, X BA, SNR, SNO
and XCOD ' Two models were derived, one for anoxic and one for aerobic conditions.
Under anoxic conditions, there is no nitrification. Three processes take place:
growth of heterotrophs, decay of heterotrophs and decay of autotrophs. The resulting
model contains four parameters: rR, YR, bR and bA.
(2.8)
(2.9)
l-YH
--""'-rH . XCODX BH (2.10)
2.86YH
(2.11)
(2.12)
Under aerobic conditions, there is no denitrification. Four processes take place: growth
of heterotrophs, decay of heterotrophs, growth of autotrophs and decay of autotrophs.
The model contains six parameters: rR, YR, bR, rA, YAand bA.
(2.13)
(2.14)
22
Model reduction of activated sludge model no. 1 and hioprocess models for identification and control
(2.15)
rx BH =rH ,XCODXBH -bHX BH (2.16)
(2.17)
Further reduction can be achieved by assuming all decay constants equal. Motivations
for this reduction are that mainly net growth rate of importance and that the decay
parameter is difficult to estimate.
Julien et al. (1998) developed reduced-order models for identification and control
for aerobic and anoxic conditions, with ammonia, nitrate and DO as state variables.
DO dynamics are often decoupled from the other ASM 1 equations, for use in
(adaptive) control of DO. An example is the following (Lindberg, 1997). The decoupled
balance equation for DO over a CSTR by assuming quasi-steady-state for the other
components is written as
(2.18)
The conversion term, rso , (which is the Oxygen Uptake Rate, OUR), is a time-varying
parameter into which other states and parameters are lumped. It is modelled with a
simple, discrete-time black-box mode1 2:
1
rs (k) = OUR(k) = 1 1 e(k) (2.19)
o (1- fq - )(1- q - )
2.3. DYNAMICS OF VARIABLES
It is generally known that activated sludge systems exhibit stiff dynamics, with
timescales ranging from seconds to weeks. This also holds true for activated sludge
models based on ASMI. This section first summarises published material on timescale
properties of ASMI and subsequently summarises reductions based on simplifying
assumptions with respect to dynamics.
IAWQ Report No. 1 (Henze et aI., 1987) distinguishes three groups of variables:
dissolved oxygen So, dissolved components and particulate components. This was
concluded from a timescale estimation employing only the output (no input) transport
2 There may however be a danger in doing so, as the OUR is also affected by So itself
and by Ss, which can both vary relatively quickly.
23
S.R. Weijers
terms and consumption (no production) terms of the balance equations. This yielded a
time constant for So in the order of 1 s., for Ss, SND SNH SALK in the order of 1 min, for
XBH, XBA, X s, X p, and XND in the order of 10 min. These were estimates for time steps
in Euler integration, and then this poses no problems because the resulting estimate is
rather conservative. This analysis however may give misleading results as it may
grossly underestimate the time constants.
Weijers et al. (1995) carried out an analysis based on the Jacobian matrix of a pre-
denitrification system, consisting of one anoxic and one aerobic reactor. The time
constants, computed in a steady-state under typical operating conditions, ranged form
30 s., associated to a very low DO concentration (0.0114 mgll) in the anoxic reactor, up
to 13 days, associated to inert particulate COD. It is noted that the fastest time constant
of 30 s. is much larger than 1 s., reported in (Henze et al., 1987). It was observed that at
elevated DO concentrations, DO dynamics were much slower (approximately 10 times
as slow) than at the very low concentration of 0.0014 mgll. This was observed both in
the aerobic reactor (So =1.2 mgll) and in the anoxic reactor (if for computation of the
Jacobian matrix So was artificially elevated to 2 mgll to be in the same order of
magnitude as the other solute components). These results indicate that stiffness of DO
dynamics is introduced if the DO concentration is much lower than the concentrations
of other components that are involved in the same reactions.
Based on qualitative reasoning, Olsson and Jeppsson (1994) classified cause-effect
relationships between available manipulated variables and measurable variables into
different timescales. The motivation for classification was to achieve decomposition
based on timescales for plant-wide control. Jeppsson (1996) summarised these in an
incidence matrix, distinguishing fast (minutes), medium (hours), and slow (> days)
dynamic influence. The matrix displayed that most outputs are effected by several
inputs, where different variables may act on different timescales, which is caused by the
strong internal couplings in the system. Interactions occur also within a timescale.
Consequently, the authors expect that control of the activated sludge process requires a
multivariable approach.
In a quantitative study of dynamics of ASM1, Steffens et al. (1997) used a
procedure developed by Robertson and Cameron (1997a,b; see section 3.4), to make a
state partitioning. This required association of states to eigenvalues. They classified as
fast Ss, X s, SNH, SND and XND , as medium Sr. SALK and SNO and as slow X BA, XBH, XI
andXp
Subsequently, they applied singular perturbation to obtain a systematic reduction of
ASMI by removing fast and slow states. In addition to the eigenvalues, to select
reducible states they applied additional criteria on the relative error introduced by the
reduction (Robertson and Cameron, 1997a). For fast mode reduction, Ss, SNH, SND and
XND were obtained using a relative error of 5% and with a lower bound on the timescale
of interest of 12 h. The fast mode reduced model without these state variables showed a
large error (100%) in SNH however as a result of nonlinearity. For slow mode reduction,
only XBA was obtained as reducible state using a (large) relative error of 10% and with
an upper bound on the timescale of interest of 20 minutes. The slow mode reduced
model (constant XB~ however showed a large error of 40% in SNH after one day.
Consequently, slow mode reduction with XBA in this case was possible only for
relatively short time horizons of approximately 6 h.
24
2.4. ORDER REDUCTION METHODS
Van Schagen et al. (1995) applied algebraic reduction techniques to reduce a linearized
ASM1 model of a carrousel system for application in LQG control of ammonia and
nitrate with aeration in a carrousel system. No details on the method applied are given
however.
2.5. BLACK-BOX IDENTIFICATION
Lindberg (1997) applied subspace state space identification to identify a fourth-order

linear model from a nonlinear model including 5 reactors based on ASM1 and a settler
model. The linear model was used in multivariable setpoint control of ammonia and
nitrate in a pre-denitrification plant, with external carbon dosage, internal recirculation
flow rate and the DO setpoint as manipulated variables. The study was to illustrate
application of black-box identification, the nonlinear model being used to generate
artificial data, rather than to perform model reduction. Nevertheless, it also illustrates
how identification may be used to reduce a calibrated ASM1 model when this is
available.
Some case studies of neural network application to model activated sludge systems
have been reported (Su et aI., 1992; Cote et ai., 1995). Neural networks may also be
used to obtain nonlinear reduced models. By incorporating prior knowledge into the
structure of neural net, a larger validity range may be obtained. Identification of ANN's
for modelling of a phosphorous removal activated sludge plant model was unsuccessful,
even for interpolation only (Vanrolleghem, personal communication). No other cases
for wastewater treatment plant modelling however are known to the author.
2.6. DISCUSSION AND CONCLUSIONS
The literature review shows that several approaches are applied in ASM1 model
reduction. In almost all reported cases, the reduction was applied to lumped systems
without concentration gradients, assuming the system to be either aerobic or anoxic.
Thus, the reduction concentrated on reduction of the reaction kinetics, rather than
reduction of transport and mixing in the reactor.
Of the approaches applied to model reduction of ASM1, simplifying assumptions
have been applied most frequently. These assumptions lead to a variety of reduced
models, depending upon the goals and system for which the model is intended.
Proposed models range from simple black-box zero-order kinetic models via grey-box
models with different Monod terms both neglecting biomass dynamics, to the more
complicated model of Jeppsson, which includes biomass growth and decay. The first
category is only valid over a short time horizon and within the domain of experimental
conditions. The latter category may be used for process optimisation over a longer time
horizon, and might be to some extent valid beyond the domain of experimental
conditions.
The selection of simplifying assumptions is generally guided by the type of the
treatment system, the input/output relations to be modelled and the time horizon of the
model prediction, in a heuristic fashion. In several cases, the accuracy of the reduced
models is not tested. From the set of reduced models, no single best reduced model can
25
S.R. Weijers
be selected. A more systematic investigation of the validity and implications of the

various simplifying assumptions under different conditions (system, goals,
input/outputs) would facilitate a more straightforward selection of simplifying
assumptions and resulting reduced models.
Assumptions with respect to dynamics are usually applied to dynamically decouple
(some of) the system equations, especially for application in hierarchical control. A
thorough understanding of the timescale properties of the model is required to value
validity and implications of reduced models based on these assumptions. Still however,
little systematic efforts have been done to analyse and understand these properties. The
biokinetic model itself is complicated, describing several biological and chemical
processes that are strongly coupled. Furthermore, it is used to describe relatively
complicated process configurations. Consequently, the dynamics of the model are not
fully understood.
A disadvantage of the order reduction techniques and linear identification techniques
is that the validity of the linear model thus obtained is limited. It is concluded that there
is a need for more systematic order reduction methodology for activated sludge models.
Moreover, more insight in dynamics of ASMI is desired. In section 3, for these reasons,
singular perturbation is studied in further detail.
3. Singular perturbation of bioprocess systems: theory and review
Singular perturbation application to bioprocess systems is reviewed in this section. This

method is a promising candidate to be part of a systematic methodology to derive
reduced nonlinear models for different timescales, e.g. for application in hierarchical
control, as was discussed in section 1.2.
In 3.1, that part of the theory is presented that is required to understand the material
in sections 3 and 4. Section 3.2 reviews application of singular perturbation to
bioprocess models. Section 3.3 presents a procedure to obtain the standard form that is
required for application of singular perturbation for timescale separation. This
procedure provides the basis for developing the procedures for state partitioning in
section 4. Section 3.4 discusses some methods that can be considered complementary to
scaling in detecting fast and slow states.
3.1. SINGULAR PERTURB AnON THEORY
Let the system be described by n+m equations in state-space notation
(3.1)
£1: = g(x, Z, u, t, £), z(to) = zo,z E R rn • (3.2)
with £>0 a small scalar. Then for £~O the order reduces to n, because substituting a root
zi = <Pi (X, 11, t) of the equation 0 =g(x, Z, u, t,O) into (3.1) yields a reduced model:
26
~ == f(X,c1>i (X, U, t), U, t,O) == r(X, U, t), x(tO) == xO' (3.3)
which describes the slow dynamics of the system, also referred to as the outer system or
outer layer, or quasi steady state (x refers to the quasi-steady-state). Model (3.1), (3.2)
is said to be in the so-called standard form if and only if the following crucial condition
is satisfied.
Condition 3.1:
In a domain of interest, the equation 0 == g(x, z, u, t,O) has ~1 distinct real roots
zi == c1>i (X, u, t), i == 1,2, .. , k.
The quasi steady state x(t) can be prescribed to start from Xo and thus be a uniform
approximation of x(t), that is, x(t) ==x(t) + 0(£) holds for all tE [to, 1:,,], including to. The
quasi steady state z(t) however is not free to start from a prescribed initial condition,
and the approximation z(t) ==z(t) + 0(£) can be expected to hold only on an interval
excluding to, that is, for tE [5,1:,,], with 5>to. During an initial interval [to,5] (the so-called
boundary layer), the original variable z approaches Z. The substitution tf == tI £
("stretching" the initial time) converts (3.1), (3.2) to a set of equations describing the
fast dynamics of the system, the so-called boundary layer or inner system or inner layer
(3.4).
dZ == g(xo, Z(tf + z(to),O, to), z(O) == Zo - z(t o ) (3.4)

dtf
The solution to this problem provides a boundary layer correction term z == z - z which
is used in a possible approximation z= z(t) + Z(tf) +0(£), valid for tE [to, 1:,,], The
important Tikhonov's theorem with respect to the boundary layer system states that
(3.3) is a valid approximation of (3.1), (3.2) for all tE [to, 1:,,] if the following two strong
stability conditions on the boundary layer system are satisfied.
Condition 3.2: The equilibrium z(tf) == 0 of the boundary system (3.4) IS
asymptotically stable uniformly in Xo and to, and z(O) == Zo - z(to) belongs to its
domain of attraction, so z(tf) exists for tr-0.
Condition 3.3: The eigenvalues of og/oz evaluated, for £ == 0, along x(t), z(t) , have
strictly negative real parts, i.e. Re A.{ ogloz}~ c <0.
Thus, singular perturbation theory allows us to treat slow and fast dynamics
separately. Equations (3.3), (3.4) provide a zero-order approximation of the system
behaviour in the slow and fast timescales respectively, which is exact for £=0. Higher
order approximations are required to perform a formal error analysis and to obtain more
accurate reduced models as DO. They are obtained through series expansion of the state
variables in powers of the perturbation parameter. Two possible procedures are to apply
the matching procedure (Kokotovic et al., 1986) or the boundary function method
27
S.R. Weijers
(Vasil' eva et al., 1995). A further treatment of approximations is beyond the scope of
this paper.
3.2. REVIEW OF MODEL REDUCTION OF (BIO)PROCESS SYSTEMS USING

SINGULAR PERTURBATIONAL APPROACH
In this section, fIrst a rule proposed by Bastin and Dochain (1990) is discussed. It is
shown that this rule is not generally valid. Other references are therefore investigated to
fInd a general rule for reduction of bioprocess systems through singular perturbation.
3.2.1. General rule for order reduction
Bastin and Dochain (1990) propose a simple, general rule for order reduction in their
book on estimation and control of bioreactors. Given the balance for component Si:
(3.5)
(with kij the stoichiometric coefficients, Cl'j the reaction rates, D the dilution rate, Fi the
inflow and Qi the gaseous outflow (if applicable) of component i) for a continuous,
ideally mixed bioreactor, the simplifIcation is achieved by setting Si and dS/dt to zero
which yields the algebraic equation:
I (±)kijCl'j = Qi - Fi . (3.6)
Jt'i
This rule is not general however for several reasons. Firstly, it is not indicated in a
general sense in which cases the dynamics of a component can be neglected, that is,
which component i can be assumed fast, and when and why ~i can be assumed zero. The
general rule is motivated with two specifIc situations only, which are briefly discussed
below, namely 1. neglect of product dynamics for volatile products with low solubility,
and 2. neglect of substrate dynamics in a model with biomass and substrate. These
examples do however not cover nor explain all possible situations where multiple
timescales allow reduction.
Secondly, the rule is defIned for a single, isolated equation. In the more general case
however, several equations are coupled, e.g. via the reaction network, in which case the
proposed general rule cannot be applied in its simple form. In those cases, the reduction
procedure is much more involved, as will be discussed in section 3.2.2.
Singular perturbation technique for products. This case considers product formation in
the following reaction scheme involving one substrate and one product:
S --> P
28
When the product is volatile and has low solubility, its concentration remains relatively
low. In this case, the saturation constant can be chosen as the perturbation parameter.
Writing the product concentration P as llPsab O::;;n::;;l, with Psat the saturation constant,
with E=Psat> gives the standard form:
dS
- = <p-DS+DSin (3.7)
dt
ill
E-= k<p-EDn-Q (3.8)
dt
with Q the rate of mass removal in gaseous form. For E~O, a reduced model is obtained
by substituting the resulting algebraic equation, k<p=Q into (3.7):
(3.9)
In this case, the rule holds (for Psat sufficiently small).
Singular perturbation technique for substrate. For the following simple microbial
(autocatalytic) growth process (See also section 4.2):
the component balance equations are written as:
dX
-= IJX -DX (3.10)
dt
(3.11)
Multiplying by V and using Vasa perturbation parameter gives (with VD Sin= FSin and
X~VX):
(3.12)
29
S.R. Weijers
dS
E- = - kJ flXT - ED S + FS. (3.13)
dt ill
which for E~O, with kJ flXT = Fs.ill , reduces to:
(3.14)
In this derivation, the authors state that it should be understood that the volume is not
assumed to be zero, but small enough to neglect E(dS/dt) and EDS. Therefore, it is
considered legitimate to divide the reduced equation by V again to obtain:
dX 1
-=-DX+ -DS· . (3.15)
dt kl III
However, it can be argued the volume is not the adequate perturbation parameter in this
case because:
• It is not the volume that makes the term EDS go to zero; in fact, the dilution
rate D=FyN (with Fy the volumetric flow rate) becomes very large for small E
and the term reads FyS.
• Without any additional information, there is no reason to select the substrate as
the fast state; in fact, the same argumentation can be applied to neglect the
biomass dynamics.
• The physical reasoning is not sound, as smallness of the volume is not the
cause of multiple timescales. Consequently, the derivation is mathematically
not consistent.
In section 4, it will be shown that in some cases the ratio of the biomass and substrate
concentration can be used as a perturbation parameter; procedures that are more
systematic are developed there.
3.2.2. Other reduction approaches

Van Breusegem and Bastin (1991) investigated model reduction of reaction systems
with more general reaction networks, in the case of one (ideally mixed) reactor. The
reduction problem was addressed under the assumption that some reactions are faster
than others. As a perturbation parameter, the average of the fast rate constants was used.
In natural coordinates, the problem is not always in standard form however, as
Condition 3.1, existence of distinct roots, is not guaranteed. A change of coordinates
was suggested to bring the problem into standard form. As the state transformation is
non-singular, the reduced model can be transformed back into natural coordinates. In
natural coordinates, however, there is generally no direct partition in fast and slow
variables and the singular perturbation cannot be applied to the original, untransformed
system, which we however like to preserve the interpretation of states. The procedure
30
Model reduction of activated sludge model no. 1 and hioprocess models for identification and control
was successfully tested on a two-step enzyme reaction (Van Breusegem and Bastin,
1991) and on a two-step model of a methanogenesis process (Van Breusegem and
Bastin, 1992)
Vasil' eva et al. (1995) discern as the critical case that class of applied problems in
which the condition of isolated, distinct roots of the reduced equation
0= g(x, z, u, t,O) is not satisfied. In fact, in many problems in chemical kinetics the
condition is not satisfied. A procedure is presented using the boundary function method
for constructing approximate solutions based on asymptotic expansions for the standard
form. A modification of this procedure is presented to deal with this critical case and
applied to chemical reaction networks (in batch reaction), where small and fast reactions
are involved. As a perturbation parameter, the smallest of the large rate constants was
selected. It is interesting to note that this modified procedure does not start from nor
produces the standard form. It is noted that this case again shows that a separation in
fast and slow reactions does not correspond to a clear partitioning into slow and fast
variables. Instead, it leads to a set of simplified systems for the fast and slow timescales.
Weiss and Preisig (1997) studied simplifying assumptions in the process of
modelling composite process systems. Very large transfer coefficients between
subsystems or very small capacities of subsystems for example can lead to lumping of
the transferred intensive variables in the concerning subsystems. The difficulties and
accuracy in this type of simplifications were studied on a relatively simple yet generic
example of n tanks with various levels and temperatures, assuming that some of the
transfers are very quick. The modified procedure of Vasil' eva et at. was applied (in fact,
the systems studied are very similar, with fast and slow transfers instead of fast and
slow reactions) to reduce the system and to estimate the resulting error. Reduction of
heat balances to describe the temperatures in the tanks was much more involved and
less accurate than reduction of total mass balances to describe the heights. The singular
perturbation parameter in this case is the inverse magnitude of the fast transfers relative
to the slow transfers.
The references reviewed above all apply to lumped systems. Dochain (1994) applied
singular perturbation to reduce an (infinite-dimensional) distributed parameter
bioprocess system model to a (finite-dimensional) lumped parameter model. The
general distributed parameter system studied is a plug-flow reactor with dispersion. For
the reduction, the system was rewritten using dimensionless numbers, employing as
perturbation parameter the mass Peclet number, which expresses the ratio of the
residence time and characteristic time for dispersion. A first -order approximation using
a series expansion of the solution and using matching to determine initial conditions of
inner and outer solutions yielded a lumped reduced model.
To apply singular perturbation for order reduction into fast or slow states of a given
model, the model must be in standard form and a state partitioning must be made.
A general rule for reduction of continuous bioprocess systems was given by Bastin
and Dochain (1990). However, this rule is not generally valid, and is not helpful in
general in deciding which parameters can be considered fast (only products with very
low solubility).
Other procedures have been derived for reduction of models in non-standard form.
For reduction of models in natural (original) coordinates, a procedure exists for the so-
called critical case (no distinct roots of the quasi-steady-state). However, the discussed
31
S.R. Weijers
references do not help to recognise or obtain the standard form. A procedure that does
this is summarised below.
3.3. SCALING IN MODEL REDUCTION
3.3.1. A method for state partitioning based on scaling

Aim of this section is to illustrate the usefulness of a systematic scaling procedure for
state partitioning, to recognise or obtain the standard form. Frequently, the model to be
reduced using is in non-standard form and must be brought into the standard form. One
way to do so is to select or fmd a suitable, small perturbation parameter. This is often
nontrivial and requires considerable physical knowledge; smallness of a parameter
alone is not a sufficient condition. Selection of a perturbation parameter is associated
with the issue of scaling, as smallness is a relative notion, and scaling is helpful to bring
the model in the standard form and to select a suitable perturbation parameter.
Kokotovic et al. (1986) presents several approaches in scaling, including the use of
dimensionless parameters, parameter scaling and state scaling. However, no systematic,
straightforward procedure for scaling is given.
Heineken et al. (1967) applied singular perturbation theory to provide a mathemati-
cally sound analysis of the quasi-steady-state assumption (QSSA) in derivation of
Michaelis-Menten enzyme kinetics. Segel and Slemrod (1989) used a systematic
approach based on scaling to obtain nondimensionalised equations in which it is easier
to reveal the relative magnitude of parameters. With the applied scaling procedure, they
were able to refine the conditions for application of the QSSA. The procedure they
applied is outlined as follows and how they applied it to Michaelis-Menten kinetics is
summarised in section 3.3.2:
• Estimate slow and fast timescales 'tf and 'ts •

• Test on the necessary condition for the QSSA on the timescales.
• Test on the necessary condition on the smallness of the error in the slow state
during the pre-steady-state.
• For both timescales, choose state scaling and subsequently derive scaled
equations; this step should yield the perturbation parameter.
• Reduce the model.
• Test the reduced model, e.g. through numerical simulations.
The tests on the necessary conditions may also provide perturbation parameter
candidates. In their paper, in addition to these steps a formal analysis was performed
using approximate solutions of the scaled equations. Both the derivations of the
approximate solutions and a formal error analysis are beyond the scope of this paper.
Application of an analytic scaling procedure to Michaelis-Menten kinetics. In the

reaction model employed in derivation of Michaelis-Menten kinetics, an enzyme E
combines with substrate to form an ES complex, which dissociates to either E and S or
to E and product P:
32
With the initial conditions E(O)=Eo, S(O)=So, C(O)=O, P(O)=O, and using the fact that
E(t)+C(t)=Eo, the following basic mathematical problem results:
dS
- = -kl (Eo -C)s+L1C, S(O)=So, (3.16)
dt
dC
- = k1(Eo -C)s-(Ll +k 2 )C ,C(O)=O. (3.17)
dt
Application of the QSSA to simplify this system is standard and can be found in any
elementary textbook on biochemistry. The QSSA implies that after a short pre-steady-
state, the complex concentration C is approximately constant and the decrease of the
concentration S due to complex formation can be neglected as Co is small. The
derivation is not repeated here. Only the main points are indicated.
We now focus on the procedure of Segel and Slemrod (1989). The steps to apply the
procedure to Michaelis-Menten kinetics were as follows.
(a) Estimation of timescales:
• The fast timescale is associated with the complex formation. To estimate the
fast timescale 'tf, the approximate analytical solution of (3.17) for the fast
timescale is used by supposing S is slow and can be approximated by So:
C(t) = C[l-exp(-kt)], (3.18)
with:
(3.19)
and we obtain
'tf =k - 1 . (3.20)
The slow timescale is associated with the substrate after the pre-steady-state and is
estimated using the following characterisation of a timescale (Here, Segel and
Slemrod (1989) refer to Segel (1984), p.56):
33
S.R. Weijers
Here, the maximum and minimum concentration of S are Smax=SO and Smin=O. The
maximum rate after the pre-steady-state is estimated by substituting S=So and (3.19)
into (3.16), which gives:
(3.21)
• Test on the necessary condition for the QSSA on the timescales.

A necessary condition for the QSSA to hold is that the duration of the pre-steady
state is much shorter than the characteristic time for substrate change, 1"f «1"s '
which gives:
11 « (1 + K)(l + oi , (3.22)
with the dimensionless parameters
S E k
0" = _0_ 11 = _0_ K = ----=L (3.23)
- Km' - Km' - k2 .
• Test on the necessary condition on the smallness of the error in the slow state
during the pre-steady-state.
For the change in the substrate concentration to be negligible, the relative
concentration change ~S 1So must be very small during the pre-steady state, which
is estimated by:
~SI ""~1 IdSI

I~ dt max ·'tf
(3.24)
Using (3.16) with C=O to determine IdSI dtlmax yields the following additional
condition in dimensional variables, which is stronger than (3.22):
11« (1 +0"). (3.25)
The result also indicates a perturbation parameter candidate, namely 17 1(1 + a") .
• For both timescales, derive scaled equations.
In the pre-steady-state, the time is scaled by 'tf . The substrate concentration is scaled
by So and the complex concentration C by the maximal complex concentration
34
c. With the introduction of the following dimensionless variables into (3.16) and
(3.17),
set
s=-, c==,tf = - (3.26)
So C 'tf
we obtain the scaled equations for the fast timescale:
-ds= c[ -s+--cs+
a K(K+ 1)-1 c ,s(O)=1 1 (3.27)
dtf a+l a+l
dc a 1
- = s - --cs - - - c ,c(O)=O. (3.28)
dtf a+l a+l
After the pre-steady-state, the time is scaled by 't8 • The same state scaling then
yields:
~
-=(K+l)(a+l) -s+--cs+
ilis a+l
[a ~K+D~ c1
a+l
(3.29)
dc
c-=(K+l)(a+l)
dt s a+l
[a
s - - - c s - -1
- c] .
a+l
(3.30)
Thus, we now have the problem in standard form, and have obtained a perturbation
parameter:
(3.31)
This result is more accurate than the traditional perturbation parameter, EofKm.
Segel and Slemrod (1989) obtained this more accurate result through applying this
systematic physical scaling procedure. Besides physical scaling, Segel (1972) also
applied mathematical scaling procedures with the aim to test whether this result can
also be obtained with less prior knowledge. Through one such method, minimal
simplification, the same results were obtained; it is not clear however whether this
method will always work and the method is not further studied here.
• Reduce model and test the reduced model
From the standard form, the model was reduced. In numerical simulations, the
conditions on the timescales and errors were verified and confirmed to be correct.
35
S.R. Weijers
3.3.2. Relationship of scaling with regime analysis and dimensional analysis

In (bio)process engineering, regime-analysis is a tool to detect rate-limiting mechanisms
by comparing the timescales for different transport and reaction processes. For example,
Oosterhuis (1983) applied regime analysis to the scale-up of bioreactors. Evidently,
there is a relationship with the detection of occurrence of multiple timescales as
described above. An important difference is that regime analysis compares the relative
size of the individual terms in the balance equations, while the timescale estimation, as
applied by Segel and Slernrod above, considers the total rate expression of extensive
variables. Both methods can be used in a complementary fashion. Dimensionless
groups, obtained by the ratios of characteristic scales of different phenomena, also
provide insight into the system. The insight obtained by using scaled equations is a
strong argument for the use of scaling to convert the equations to dimensionless form,
as advocated by Segel (1972), even if this scaling requires some physical reasoning.
3.4. OTHER METHODS FOR TIMESCALE ANALYSIS
In section 3.3, a procedure was described for detection of multiple timescales in models
and transformation of the problem to the standard reducible form. Robertson and
Cameron (1997a) presented another approach to detect timescale multiplicity and for
state partitioning for use with singUlar perturbation reduction. Their procedure might be
suitable for large systems if the scaling procedure described above is too laborious, or
could be used as a complementary technique. The essential issues of the procedure are
briefly summarised here.
Starting from the definition of a timescale of interest, slow and fast modes are
detected using a linearized model. The timescale of interest is determined by the
intended application, e.g. control of ammonia and nitrate in activated sludge plants or
control of the sludge concentration, which require different timescales. The fIrst
requires medium timescales, the latter slow timescales. Fast mode reduction is
performed based on an eigenvalue-to-state association. This is done using a so-called
homotopy parameter. The procedure starts with a system in which the eigenvalue-to-
state association is known (in their study, a matrix with only the diagonal of the
Jacobian matrix was used). The eigenvalues are traced when going from this system to
the system for which the eigenvalue-to-state association must be determined (the full
Jacobian matrix). A so-called homotopy parameter is used to gradually change from the
known system (homotopy parameter is zero) to the unknown system (homotopy
parameter equal to one).
If a group of fast time constants exists, then the states associated with these time
constants are candidates for reduction. Slow mode reduction proceeds via Taylor
expansion of the free response. From linear systems, randomly generated by Monte-
Carlo sampling, empirical relationships for the maximum relative error both for fast and
slow mode reducibility were derived. After detection of slow and fast states, these are
removed from the nonlinear model. The procedure provides no guarantee that the
system is in the standard form and no formal error analysis is given. Tests on an
evaporator system and compressor start-up yielded good results; results on ASMI were
less successful as discussed in section 2.3.
36
We indicate two disadvantages of the slow mode detection method. Error estimation can
be too optimistic if in reality a forced response is dominant, because only the free
response is considered in the selection criterion. Furthermore, relatively small changes
of a slow mode variable can be associated with relatively large changes in other
variables, especially if these have small absolute value (such as dissolved oxygen or
ammonia in the case of ASM1). Consequently, on the other hand, the method can be too
conservative.
Wasynczuk and Decarlo (1981) suggested an approach for reduction of composite
models. This method is useful for order reduction of large-scale systems such as in
plant-wide control, and is therefore briefly described in here. The procedure indicates
whether model reduction on a component level is possible or not. This must be checked,
because interaction may introduce dynamics in the timescale of subsystem dynamics. It
this is the case, then the model reduction cannot be performed simply by reduction of
the individual subsystems only.
In the analysis, the so-called component connection model is employed, which
separately defmes the subsystem equations and the interconnection matrix. The
procedure consists of eigenvalue tracking applying a homotopy parameter, which is
multiplied by the interconnection matrix, and is used to vary from completely decoupled
subsystems to the fully interconnected system. This indicates whether reduction on a
subsystem level is sufficient to reduce the complete system.
3.5. CONCLUSIONS
Application of singular perturbation for reduction of (bio)process models was reviewed.

Bastin and Dochain (1990) suggested a simple rule for order reduction, which is not
generally applicable, however. In reduction of a system with biomass and substrate, the
suggested selection of the perturbation parameter was not satisfactory. Segel and
Slemrod (1989) applied a methodology based on scaling to formally derive Michaelis-
Menten kinetics. The methodology can support detection of multiple timescales, which
are necessary for applying quasi-steady state assumptions and to order reduction with
singular perturbation. Furthermore, it may help to bring the model in the standard form
and to select a suitable perturbation parameter that is required for application of the
singular perturbation method. The methodology forms the basis to develop a method to
recognise or obtain the standard form and for state partitioning. This is applied to a
simple yet basic and important bioprocess with biomass and substrate, which is similar
to the model studied by Bastin and Dochain (1990), in the next section.
4. Scaling for singular perturbation in a simple bioprocess system
4.1. INTRODUCTION AND METHODOLOGY
This section investigates application of singular perturbation to reduction of bioprocess

models. The following observations motivate this further investigation. Section 2 made
clear that more insight into the timescale properties of ASM1 is required. Section 3
showed that only a few occasions have been described in which there is a clear
37
S.R. Weijers
perturbation parameter and associated timescale separation, namely the case of low
solubility of volatile products and the occurrence of fast and slow reactions. In ASMI
however, these are not the predominating causes of multiple timescales, as no volatile
products are applied in the model and the reaction rates, although different, do not differ
orders of magnitude. Therefore, we want to find additional causes of multiple
timescales and associated perturbation parameters. Moreover, these findings are also of
relevance for bioprocess systems in general.
The following strategy is adopted. Three procedures for state partitioning are
proposed. They are applied to a very simple but basic and important model system,
closely resembling the system used by Bastin and Dochain (1990) (Section 3.2). This
should reveal to reveal timescale multiplicity and the physical conditions that cause
multiple timescales. These conditions are related to a singular perturbation parameter, if
possible, and it is checked whether the model can be brought into standard form.
The three procedures to obtain or recognise the standard form are evaluated with
respect to straightforwardness and required prior knowledge. The following procedures
are applied:
• Direct scaling
• Timescale estimation for variables
• Analytical scaling procedure
These procedures are now described and are applied to the model system described in
section 4.2.
4.1.1. Procedure 1: Direct scaling

This procedure attempts to bring the model into standard form directly, without a
thorough scaling procedure and without introducing dimensionless parameters. The
background to apply this procedure here is the conjecture that stiffness is associated
with large concentration ratios. For example, in analysis of DO dynamics, it appeared
that a fast timescale occurred only if the DO concentration was very low. Moreover, in
wastewater treatment systems the suspended components are usually regarded as slow,
and (most of) these variables have a much larger concentration than the suspended
components. From these observations, it was hoped that a simple yet physically
meaningful rule for order reduction could be derived.
4.1.2. Procedure 2: Timescale estimation/or variables

Procedure 2 is a systematic procedure for detection of timescale multiplicity through
estimation of timescales of variables. This procedure is proposed as an alternative and
complementary procedure where Procedure 3 fails.
In the example in section 3.3, it was already known that multiple timescales
occurred, and to which variables the fast respectively the slow timescales were
associated. Generally, this is not the case; in fact, the eigenvalue to state association
procedure presented in section 3.4 was developed for this purpose.
Here we will propose a different procedure to estimate timescales of variables. This
procedure is a modification and extension of the procedure of Segel and Slemrod (1989)
in section 3.3, and does not require analytical timescale or error estimates. Instead, a
38
mixed numerical I analytic approach is applied to detect timescale multiplicity and to

estimate the error in slow states during the initial layer, thus checking whether the quasi
steady state approximation applies.
We start with the system (4.1)
(4.1)
and try to find a partitioning into fast and slow variables C T =[Ci C:] so that we can
write (4.1) in standard form as
(4.2)
(4.3)
The procedure consists of the following steps that are subsequently discussed below:
• Step 1. Estimate initial timescales of all variables and select the fast variables.
• Step 2. Estimate timescales of slow variables in the outer layer (in quasi steady
state).
• Step 3. Check the multiplicity of timescales.

• Step 4. Estimate the error of the slow variables.
• Step 5. Reduce model if preceding steps indicate that the QSSA applies.
Step 1: The initial timescale 'efi for variable i in Step 1. is estimated with:
(4.4)
with C? the quasi steady state value of variable i with the other variables C j' j:;t:i at their
initial values CjO and IdC/dtlmax,in the maximal rate during the inner layer, which in this
paper is evaluated with all state variables at their initial conditions. We would like to
avoid the need to compute C? Therefore, one can try to rewrite (4.3) to eliminate
I C i (0) - Cia I as follows. In the quasi steady state for C? we have:
39
S.R. Weijers
(4.5)
Subtracting equation (4.5) from the rate equation for C i (this is allowed, because the
quasi steady state equation is zero) gives (4.6) (Note: 1.1 denotes the absolute value).
(4.6)
We can try to write this as a product with a term (C iO - Cp) :
to obtain:
(4.7)
In some cases, it is thus possible to eliminate Cia completely and to obtain an

expression for the time constant as a function of the parameters and initial conditions
only. The variables that have a (much) faster associated timescale than other variables
are selected as the fast variables, and thus the partitioning is made.
Step 2. The slow timescale 't si for the slow variables i can be estimated with:
(4.8)
with Cio the initial value of variable i, C i its steady state value and IdC/dtlmax,out the
maximal rate during the outer layer. To avoid the necessity of computing Ci , it is
assumed that the difference between initial value and steady state are in the same order
of magnitude as the initial value and that (4.8) can be approximated by (4.9).
40
't . = ICiOl (4.9)

s, IdC.1
dt' max,out
The maximal rate during the outer layer is evaluated with the slow state variables at
their initial values CjO and the fast states at their quasi steady state values.
Step 3. is straightforward, unless the difference between fast and slow timescales is not
very large. Here, we will pragmatically consider a factor of about 10 between largest
fast timescale and smallest slow timescale sufficient for timescale separation.
Step 4. The error in the slow variables during the initial, inner layer is estimated to
check whether the error is small enough to qualify the slow variables as slow. If this is
not the case, this does not mean that there are no multiple timescales, but rather that
they cannot be assigned to disjunct fast and slow variables, i.e. the problem is not in
standard form. A ftrst, conservative approximation employs the maximal rate during the
inner layer, with the advantage of computational simplicity as all required quantities are
known already. The condition then is:
(4.10)
with'tf the largest of the small time constants and IdC/dtlmax,in the maximal rate during
the inner layer. The estimation is more accurate and less conservative if an average rate
is applied:
Co "" CoI IdCI

I
LlCI
dt aV,in . 'tf
(4.11)
In the sequel, the average rate computed as the mean of the initial rate and the rate after
the inner layer at t= 'tf will be used to compute the error, unless stated otherwise.
Step 5. Reduce model if preceding steps indicate that the QSSA applies. If the
conditions checked in steps 3. and 4. are satisfied, then this indicates that the quasi
steady state approximation applies and the model may be reduced. One can try to
formulate scaled equations and select a perturbation parameter and thus bring the
problem into standard form. If this is not successful, then we can proceed as follows.
From the partitioning into slow and fast variables, it can be simply concluded that the
rate for the fast variable is much higher than the rate for the slow variable and therefore
can be written as:
41
S.R. Weijers
where hf(CpCf,u,t,E) is III the same order of magnitude as fs(CpCf,u,t,E),

because frCCs,Cf,u,t,E) is much higher than fs(Cs'Cf,u, t,E). Therefore, (4.1) can
be written as (4.2), (4.12):
(4.2)
(4.12)
and it is seen that the system is almost in standard form. Although we do not have E in
analytic form, we know it is small enough (through the preceding steps) and we may
apply the QSSA to obtain the quasi steady state value for Cf by equation (4.5). An error
estimate is also provided through step 4.
4.1.3. Procedure 3: Analytical scaling procedure

In Procedure 3, the methodology proposed by Segel and Slemrod (1989) (Section 3.3) is
applied. This procedure proceeds via scaling and selection of a perturbation parameter.
For a complete scaling procedure, analytic timescale estimates are required for scaling
the fast and slow timescale and analytic error estimates are derived.
4.2. MODEL SYSTEM: CHEMOS TAT WITH BIOMASS AND SUBSTRATE
4.2.1. Model system definitions and analysis

A simple chemostat with one biomass and one substrate will be used to test the
proposed procedure. In this subsection, the system is described, some analytical
relationships are presented and the cases that are discussed in the next subsections are
indicated.
In a completely stirred tank reactor, biomass X grows on a single substrate S (the
arrow in the formula denotes an autocatalytic reaction). Motivation for choosing this
system is that this is a very simple yet basic and important continuous biotechnological
process that can exhibit multiple timescales. This simple system is well understood and
is a good study object as better understanding its timescale behaviour is helpful to
understand systems that are more complicated as well.
s -C'x
The reactor is schematically shown in figure 1.
42
F'S~ V,S,X I
Fig. 1: Chemostat with biomass growth on one substrate
For this system, assuming Monod kinetics, model equations (4.13), (4.14) can be
written.
S
X= ~--X -DX (4.13)
K+S
. S
S= -kl I I - - X -DS+DS· (4.14)
,..., K+S III
with S the substrate concentration" Sin the substrate concentration in the feed, X the
biomass concentration, D the dilution rate =VIF, kl the reciprocal of the yield, ~ the
maximum specific growth rate, K the Monod constant, The dimensionless parameters
defined for this system are the dimensionless residence time and the Monod number:
(4.15)
K
Mo=- (4.16)
Sin
For the system to be viable, no washout must occur so there is a lower limit to the
dimensionless residence time:
't:m =-~-=l+Mo (4.17)

Dmax
The steady state concentrations are given by (4.18) and (4.19).
1 DK 1 K
Xoo = -{Sin - } = -{Sin - - * - } (4.18)
kl (~-D) kl 't -1
43
S.R. Weijers
S = OK =~ (4.19)
00 (1-1- 0 ) 'C*-1
Upon linearization in a state x={X, S}with u=O, the following linearized system is
obtained:
-!J.S
--0 I-1KX
[ K+S (K+S)2
(4.20)
x- -kl~ -kl !J.KX
K+S (K+S)2
from which the eigenvalues are computed:
2. _S(~S_+_K~)-_k~lX_K_
"1.2 =-0, -0+1-1 (4.21)
. (S+K)2
Starting point in the subsequent timescale analysis is the conjecture that occurrence of
multiple timescales is associated with a large concentration difference between the
states. The ratio SIX in steady state is written as a function of the dimensionless
parameters 'C. and Mo:
Soo =_1 { Mo } (4.22)

Xoo kl 'C*-I-Mo
From (4.22), we see that SIX is low when Mo is very small or 'C. very large and six
cases are distinguished as indicated in table 1 and briefly explained below.
Case 1 and 2: When 'C. is very large (high residence time), substrate conversion is
almost complete and biomass produced is only slowly withdrawn. At high feed
substrate concentration Sin (small Monod, Case 1), the concentration difference is larger
at a given residence time than at lower Sin (Monod ~ 1), because the biomass
concentration is higher whilst the substrate concentration remains unchanged (Case 2).
This is easily seen from (4.19).
44
Table 1: Cases distinguished in timescale analysis; for explanation, see text
Case Ratio S.JX,. Condition for 'C* Other condition Section
1 S«X 'C*»1 Mo« 1 4.2.2
2 S«X 'C*»1 Mo""l 4.2.2
3 S«X 'C* > 0(2)1 Mo« 1 4.2.3
4 S""X 'C* > 0(3)1 Mo""l 4.2.4
5 S""X 'C* 'Ce "" Mo, Mo «1 4.2.5

=1+Mo+'Ce
6 S»X 'C* 'Ce « Mo 4.2.6
=1+MO+'Ce
1: Uppercase o denotes "order of magnitude".
Table 2: Ratio S.IX~ and ratio of eigenvalues for different cases
Case 'C* Mo S.JX,. A./A2 A.l ~
1 20 0.04 0.0032 0.0022 0.2 90

2 20 0.4 0.0323 0.0226 0.2 8.8
3 3 0.04 0.0306 0.0306 1.33 43.5

4 3 0.4 0.375 0.375 1.33 3.56
5 1.244 0.04 0.294 1 3.21 3.21
6 1.044 0.04 15 236 3.83 0.016
Case 3: At moderate dilution rate, but far from washout, which is expressed by the
condition 'C* > 0(2), the ratio SIX can still be small when the feed concentration is very
high (this is the case when the Monod number is very small).
Case 4: If Sin is moderate at moderate dilution rate, then the ratio SIX will be 0(1).
45
S.R. Weijers
Case 5: At high dilution rate, relatively close to washout, we write 'to as 'to =1+Mo+'tE.
This is an interesting situation, as biomass productivity is optimal close to washout,
namelyfor'tE=~Mo+M02 .
Case 6: At still higher dilution rates, when 'tE« Mo, the situation is so close to washout
that substrate conversion is almost zero and high ratio SIX results.
For typical values representing the different cases, the ratio SJX", and the ratio of the
eigenvalues are given in table 2. Equations (4.18), (4.19) and (4.21) were used, with the
following parameter values as default parameters: 1l=4, K=20, k j =1.5.
The results indicate that the supposed relationship between the ratio SJX", and the ratio
of eigenvalues holds indeed. In the subsequent subsections, the cases are analysed more
thoroughly according to table 1.
4.2.2. Cases 1 and 2: Ratio SIX very low, r* very large
Direct. With the direct approach, we introduce scaled variables directly. In this case,
from the supposed association of multiple timescales with a large concentration
difference between the states, both states are scaled with their steady state, assuming
that SJX", is very small.
(4.23)
(4.24)
In dimensionless variables we obtain with k=KlS oc :
. s
X= Il--x -Dx (4.25)
k+s
(4.26)
With E = SJX", (4.26) becomes:
. k s Soc Sin S
E S= - j Il-- X -E D s + D -- -- E = - kj Il-- X -E D s + fi (4.27)
k+s Xoc Soc k+s
46
Now the problem is in standard form, and (4.25), (4.27) are the equations for the slow
timescale. With the substitution tf= et the fast time equations are obtained:
dx s
-=e{Il--x -Dx} (4.28)
dtf k+s
ds s
- = -kl Il--x -eDs+fin (4.29)
dtf k +s
Note: To arrive at (4.27), it must be shown that fin is not very small, O(e), but that it is
of the same order of magnitude as the other non-negligible terms and thus cannot be
neglected. The result is given here as is without proof, but is can be shown that this
condition holds.
With the direct approach, we arrived at the equations in standard form. If the ratio
S.JX., is very small, this ratio can be used as a perturbation parameter to obtain the
problem in standard form. In that case, biomass is the slow variable, substrate the fast
variable.
The required physical knowledge to obtain the standard form in this case is the
association of low S.JX., with timescale multiplicity. However, the standard form
obtained presupposes that S.JX~ be very small, but does not indicate whether this
assumption indeed is valid. In the next two subsections, an alternative formulation of
the standard form will be derived, which states the condition for timescale multiplicity
directly in terms of (dimensionless) system parameters.
Timescale estimation for variables. The procedure described in Section 4.2 is applied to
system (4.13), (4.14).
Step 1. Estimate initial timescales and select fast variables.

For estimation of the initial timescale for biomass X, the quasi steady state equation
(4.30) for X with S=So
(4.30)
and the estimate (4.4) are used to obtain an approximation in the form of (4.7):
1 So
-=Il--- D . (4.31)
'tfX K+So
For substrate S, we similarly obtain the estimate for the initial timescale:
47
S.R. Weijers
(4.32)
with SO the positive root of the quadratic quasi steady state equation of S:
-0
S -0
o= - kl fl Xo - DS + DS in (4.33)
K+So
It is noted that in this form the timescales depend upon the initial state and thus have to
be evaluated considering the initial condition. Consequently, here the timescales will be
estimated numerically. This is a result of the nonlinearity of the model. Results of fast
timescale estimation for representative cases are given in table 3. The default
parameters given in section 4.2.1 were used. For the initial states, for both states a value
of half the steady state value was chosen
Table 3: Fast timescalesfor Case 1 and 2 (x(O)=O.5-x(oo))
Case 't
• Mo
1 20 0.04 0.0022 0.0022 lO.3 0.023
2 20 0.4 0.0226 0.0213 lO.3 0.22
Table 3 shows that the timescale estimation indicates multiplicity of timescales. S is

selected as the fast state. The results obtained with the timescale estimation correspond
well with the eigenvalue results. The advantage with timescale estimation is that the
timescales are directly associated with variables.
Step 2. Estimate timescales of slow variables in the outer layer (quasi steady state).
As S is selected as the fast variable, X is the slow variable. Applying (4.9), the slow
timescale for X is estimated with by substituting SO obtained from (4.33).
(4.34)
Table 4: Slow timescales and error for Case 1 and 2 (x(O)=O.5-x(oo))
Case 't
• Mo
1 20 0.04 0.0042 5.0 0.021 0.0012
2 20 0.4 0.038 5.2 0.20 O.OlO
48
Table 4 shows lhe results of lhe slow timescale estimates for Case 1 and Case 2.
Step 3. Check lhe multiplicity of timescales.

From table 4, it is seen that for bolh Case 1 and Case 2lhe condition 'ttsI'C.x«1 holds.
Step 4. Estimate lhe error of lhe slow variables.

Applying (4.11), lhe error in lhe slow state X during lhe initial layer is estimated by
(4.35) (Xo can be eliminated).
I-L\XI
Xo
"" -1(SO
2
Jl
K+So
+Jl SO) -D ''Cfs
K+S o
(4.35)
The errors are given in table 4 and indeed lhe condition LlXoIXo<<1 holds. In Case 2,
lhe error is approximately 1 %.It is concluded lhat timescale separation can be applied.
Step 5. Reduce model if preceding steps indicate lhat lhe QSSA applies.
From lhe preceding steps, we know lhat lhe QSSA applies because timescale
multiplicity occurs and lhere is a separation into fast and slow states (see Step 5 in
section 4.1.2). Therefore, lhe system can be slow-mode reduced straightforwardly to:
SO
X= Jl oX-DX (4.36)
K+S
wilh SO lhe positive root of lhe quadratic quasi steady state equation for S (4.33).
Analytical scaling procedure. In this subsection, scaling is performed employing
estimates based on lhe analytical solutions of lhe eigenvalues of lhe system, which leads
to lhe standard form.
Estimate timescales
In Case 1 and 2, lhe large eigenvalue of (4.21) can be approximated as follows, because
X»S, S«K, !l»D (because 'C*»1) and X""Sulkl:
(4.37)
As lhe time constant is reciprocal to lhe real part of lhe eigenvalue, lhe fast time and
slow timescales can be scaled as (4.38) and (4.39) respectively.
(4.38) ; 1s=t·D (4.39)
Test for QSSA necessary conditions

For lhe QSSA to be valid, 'Ct«'Cs must hold
49
S.R. Weijers
Mo/f.!« lID or lIMo» 1 (4.40)
which is the same condition as was found before and which holds under the
assumptions made.
Perform error analysis on initial condition for slow state.

For the fast timescale we have 'tf=Mo/f.!. Assuming that So and Xo are in the same
order of magnitude as S~ and Xx, respectively and applying the (conservative)
estimate (4.10) this is approximated by the estimate (E is a small number O(Moh)):
(4.41)
For timescales, choose a state scaling and subsequently derive scaled equations and (try
to) find a perturbation parameter.
The states are scaled with their steady state values, but instead of symbols S~ and X~,
now expressions (4.18) and (4.19) are used to find the approximate steady-state values
Xx,,,,Sinlk l and S~"'KI't*. Substitution of scaled variables x=K-kI/S in and s=S·'t*/K and
for the slow timescale t.= t·D yields:
sK/'t *
f.! *x -Dx (4.42)
K+sKI't
sKI't * Sin K
-kIf.! * x--Ds-* + DS in (4.43)
K + sKI't ki 't
Dividing (4.42) and (4.43) by D, multiplying numerator and denominator of the

Monod term by K, using Mo=KlS in and introducing 1I't* as a perturbation parameter
yield the standard form with the outer equations (4.44) and (4.45).
dx s
= --x -x (4.44)
dt s 1 +cs
ds s 1 1
c-= - - - - x - - c s + - (4.45)
dt s l+cs* Mo Mo
The substitution tf= t·f..lIMo gives the inner equations (4.46) and (4.47).
50
dx s
= EMo(--x -x) (4.46)
dtf 1 + ES
ds
--_s-x-EMo s+l (4.47)
dtf 1 +ES *
Thus the timescale estimation and knowledge of scaling of the variables enables a
scaling of the variables which in turn has led to successful selection of a perturbation
parameter lI't·. This selection is in accordance with the cases studies, as here it is
assumed that 't*»1, so that indeed E«1. Compared to the dimensionless equations
obtained with the direct approach, equations (4.44)-(4.47) have the advantage of
directly showing the physical prerequisite under which the model reduction with
singular perturbation is allowed.
4.2.3. Case 3: l intermediate, Mo«l
Direct. The result obtained in 4.2.2.1 applies.
Timescale estimation for variables

Step 1. Estimate initial timescales and select fast variables.
The timescale estimation results (Table 5) indicate multiplicity of timescales. S is
selected as the fast state.
Table 5: Fast timescalesfor Case 3 (x(O)=O.5-x(oo))
Case 't
• Mo
3 3 0.04 0.036 0.034 1.88 0.063
Step 2. Estimate timescales of slow variables in the outer layer (quasi steady state).
X is the slow variable; its timescale is estimated with (4.33). Table 6 shows the results
of the slow timescale estimates.
Table 6: Slow timescales and errorfor Case 3 (x(O)=O.5·x(oo))
3 3 0.04 0.076 0.83 0.063 0.021
Step 3. Check the multiplicity of timescales.

From table 6, it is seen that for Case 3 the condition 'tfsl'tsx«l holds.
51
S.R. Weijers
Step 4. Estimate the error of the slow variables.

The condition AXofXo<<l holds as the error is approximately 2 % (Table 6). It is
concluded that timescale separation can be applied.
Step 5. Reduce model if preceding steps indicate that the QSSA applies.
The preceding results indicate that the QSSA is valid. The same comments as III
subsection 4.2.2.2 apply.
Analytical scalinfl procedure. In this subsection, scaling is performed employing

estimates based on the analytical solutions of the eigenvalues of the system, which leads
to the standard form.
Estimate timescales
In Case 3, S«K no longer holds, because S= O(K). The approximation proceeds as
follows. S is in the order of magnitude as S~, and with S=KJ('t*-I), the term (S+K) is
written as a·K, with a='C* /('C*-I). Then, with X»S and X""Sinlkl we obtain (4.48).
(4.48)
which is in the same order of magnitude as the estimate (4.38) for 'C* moderate ('C*>2).
Perform error analysis on initial condition for slow state
The same results as in subsection 4.2.2.3 apply.
For timescales, choose state scaling and subsequently derive scaled equations and (try
to) find perturbation parameter.
For S, now another scaling is used as S in O(K). Substitution of scaled variables
x=X·k1/S in and s=SIK and for the slow timescale t8= t·D yields transformed equations.
Dividing the transformed equations by D, multiplying numerator and denominator of
the Monod term by K, using Mo=KJS in yield the standard form with the outer equations
(4.49 and (4.50).
dx * s
'C --x-x (4.49)
dt s l+s
Mo~= -'C*_s-x-Mos+l (4.50)

dts 1+ s
52
with Mo' as a perturbation parameter. The substitution tf= q.tlMo gives the inner
equations (4.51) and (4.52).
dx s 1
--= Mo(-x--x) (4.51)
l+s 't*
ds = -_s-x-Mo s+l (4.52)

dtf 1 + s*
Again, scaling has led to straightforward, successful selection of a perturbation

parameter and reformulation of the problem in standard form. This selection is in
accordance with the case studies, as here it is assumed that Mo<<1, so that indeed £«1.
4.2.4 Case 4: S and X comparable, r* moderate, Mo =1

In the subsequent cases, only the relevant steps of the respective procedures will be
discussed. In Case 4, both the timescales estimated by the eigenvalues and the estimated
timescales according to Procedure 2 indicate that the timescales are relatively close and
no timescale multiplicity occurs (Tables 7 and 8). This was also confIrmed in
simulation. Thus, the conjecture that timescale multiplicity is associated with large
concentration differences is confirmed.
Table 7: Fast timescalesfor Case 4,5 and 6 (x(O)=O.5 x(ac))
Case 't * Mo ')../)..2 'tfS/'tfX 'tfX 'tfS
4 3 0.4 0.375 0.182 1.88 0.34
5 .244 0.04 1.00 0.123 1.90 0.24
6 1.044 0.04 0.0042 0.040 6.45 0.26

..;,v,.~~<,tf09'N5(=~u,.,... ~s ~ ~ # ~ ss; ~"'ZMl'Q,A2M~=~= M&~0VNiISIl.~<I ";==-~Q;'.'J'i'h";6='9W==W«~
Table 8: Slow timescales and errorfor Case 4,5 and 6 (x(O)=O.5·x(ac))
Case 't
• Mo
4 3 0.4 0.227 1.50 0.34 0.022
5 1.244 0.04 0.118 2.01 .024 0.0036
6 1.044 .04 0.002 129 0.26 0.019
53
S.R. Weijers
4.2.5. Case 5: -I close to critical, Mo« 1, 1"e""Mo

Case 5 is very interesting, as this is the situation of optimal biomass productivity in a
chemostat. The eigenvalue ratio (evaluated in steady state) indicates no difference in
timescales (Table 7). Also the eigenvalue ratio in the initial state was computed, which
was also close to 1 (1.56). This seems to be in agreement with the conjecture, because
the ratio SJX,. is close to 1 (Table 2).
However, the fast timescale estimates obtained with Procedure 2 indicate occurrence of
timescale multiplicity, S being the faster variable (Table 7). This is confirmed by the
ratio 'trsl'tsX in table 8, which is much smaller than 1.
Figure 2 shows a simulation of Case 5, which shows that indeed a boundary layer in S
occurs. Apparently, the eigenvalues in the steady state and in the initial state do not
reveal this timescale multiplicity, whereas the timescale estimation by Procedure 2
correctly indicated timescale multiplicity in this case.
The eigenvalues evaluated in the steady state do not correctly reflect the timescale
behaviour of the nonlinear model. The eigenvalue ratio in the quasi steady state was in
better agreement with the timescale estimation (A.l~=O.088; computation not shown)
than the eigenvalue ratio in steady state (A.l~=I). An eigenvalue trace might thus better
reveal the timescale properties of the model, as was observed by Steffens et al. (1997).
However, this is less straightforward than the timescale estimation procedure and it is
concluded that the proposed timescale estimation procedure is to be preferred.
The conjecture that stiffness is associated with large concentration differences is
falsified in this case, as timescale multiplicity occurs when the ratio SJX,. is close to 1
(see Fig. 2).
From the results the timescale and error estimates by Procedure 2, it is concluded
that the QSSA applies in this case and S is the fast state. The model can thus be
reduced. In the next subsection, for the situation that 't* is close to critical, scaling will
be applied to check whether in this case the model can also be written in standard form
and whether a perturbation parameter can be found.
4.2.6. Case 6: -I close to critical, Mo « 1, 1"e« Mo
Direct. From the conjecture that a low ratio XJS= is associated with timescale
multiplicity, both states are scaled with their steady state, assuming that £=XJS= is very
small. With the scaled variables x=X!X.. and s=S/S=, the scaled equations (4.53) and
(4.54) are obtained.
.
x=~--x
s - Dx (4.53)
k+s
s= -Ek1 ~_s_~ -Ds+Dsin (4.54)

k+sXoo
54
with Sin= Sui Soo. This system is not in standard form however, so the direct scaling is
not successful in this case.
fast scale slow scale

350 350
300 300
250 250
/,.._.------ ~.- .............
__.- - __
,..//
.r; 200 .r; 200 //
.c ./
~ .. .. ... . X /'
150,._......_ . _.._ _· 150, . /
100 100
50 50
OL-----~------------~ OL---------~--------~
o 0.2 0.4 0.6 o 2 3 4
t
300 300
250 250
200 /---,."
200 ,,/--"-------'-'-"'-"'--
"' 150 If!,
'" ,,'
150 J "
CI)
_/"
/
100 // 100 ,! "',_._.-...... _.-
50/' 50(;
0L-----~------------~ OL---------~--------~
o 0.2 0.4 0.6 o 2 3 4
Fig. 2: Timescale multiplicity at optimal I
Timescale estimation for variables. From table 7 and table 8, it is seen that the fast
timescale for S (0.26) is much smaller than the fast timescale for X (6.45). The ratio of
the fast timescale for S and the slow timescale for X (129) is even much lower (0.002).
Also the ratio of the eigenvalues is very low. Clearly, multiple timescales are present.
This is confrrmed by numerical simulation, which shows a very short boundary layer for
S (result not shown). Thus the QSSA holds and the system is in the form (4.2), (4.12)
with e. very small and can be reduced.
Analytical scaling procedure. Finally, the analytical scaling procedure will be

employed.
Estimate timescales
In Case 6, the nontrivial eigenvalue of (4.21) is approximated as follows. We have, with
t*=I+Mo+t£ and t£«I,
55
S.R. Weijers
S = Mo "" l __
't_
e (4.55)
Sin l+Mo+'te -1 Mo
(4.56)
which expresses the fact that conversion is very low and consequently the substrate
concentration is close to the feed concentration Sin. With X«S, and 'te«Mo, after some
manipulation and approximation we obtain (the result is given without proof):
/"'2 =- D + f.l -k1XK + S(S + K) "" -D 'te (4.57)

(S+ K)2
The fast and slow timescales can be scaled as (4.58) and (4.59) respectively.
tpt·D (4.58)
(4.59)

For the QSSA to be valid, 'tt«'ts must hold
lID « 11D·'te, or ·'te « 1 (4.60)
which holds in the case studied.
Perform error analysis on initial condition for slow state: this is omitted here.
For timescales, choose state scaling and subsequently derive scaled equations and (try
to) find perturbation parameter.
Substitution of scaled variables x=X·k[·Mo/Sin'te and S=S/Sin and for the slow timescale
ts= t·D·'te after some manipUlation yields: (4.61), (4.62), which is not in standard form.
dx * s
't e - = 't --x-x (4.61)
dts 1+s
ds * s 1 1
't e - = -'t - - x - - £ s + - (4.62)
dt s l+s Mo Mo
56
From this result and the scaling result in section 4.2.6.1, it is concluded that for low
ratio x.JS~ no simple perturbation parameter can be found. The results obtained with
Procedure 2 clearly indicated timescale multiplicity for low ratio X.JS~, the fast
timescale being associated with the substrate concentration S. From the conjecture,
however, it was expected that for low ratio X.JS~,the fast timescale would be associated
with the lower concentration, namely the biomass concentration X. Consequently, for
low ratio x.JS~, the conjecture is falsified.
4.3. OTHER RESULTS
In sections 4.2.2.1 and 4.2.3.1, direct scaling was applied to the system with Monod
kinetics, which showed that the (small) ratio S~ can be applied as perturbation
parameter. It can be shown that is also possible for zero-order and first-order kinetics.
In addition to the simple system presented in section 4.2, the same system but
extended with biomass retention was investigated, as biomass retention is usually
applied in activated sludge systems. Under typical operating conditions, the ratio S.JX~
is very small, which causes an even more pronounced timescale multiplicity between
substrate and biomass in these systems (results not shown).
In addition to the simple system studied above, a slightly more complicated system
was studied which included dissolved oxygen as an additional state variable. Here too
occurrence of timescale multiplicity depends upon the operating conditions, especially
the dilution rate and the oxygen mass transfer rate. The timescale estimation procedure
was successfully tested to this system under operating conditions causing two
timescales (Weijers and Weiss, 1999).
4.4. CONCLUSIONS
A simple bioreactor model of a chemostat with one biomass species and one substrate
species was studied to obtain insight into timescale properties of bioprocess models and
to test different procedures for model reduction. Starting point in the analysis was the
conjecture that occurrence of multiple timescales is associated with a large
concentration difference between the states. Three procedures to bring the problem into
standard form were tested, namely a direct scaling procedure, a systematic, analytical
scaling procedure and a procedure based on timescale estimation of variables.
At low substratelbiomass ratios, which occur at low dilution rate, the conjecture was
valid. The direct scaling procedure showed that the problem can be brought into
standard form and that the ratio substratelbiomass can use as perturbation parameter in
this case. The analytical scaling procedure enabled a more detailed analysis and showed
that, depending upon the operating conditions, the Monod number or the reciprocal of
the dimensionless residence time are suitable perturbation parameters. The substrate
concentration was the fast variable in this case. The timescale estimation procedure
correctly indicated the validity of the quasi-steady-state assumption for substrate.
For intermediate substratelbiomass ratios, the conjecture was falsified. At dilution
rates with optimal biomass productivity, the ratio of eigenvalues evaluated in steady
state did not reveal timescale multiplicity, whereas the timescale estimation procedure
correctly indicated timescale multiplicity in this case. The substrate concentration was
57
S.R. Weijers
the fast variable in this case, while the substrate concentration and biomass
concentration were in the same order of magnitude. The timescale estimation procedure
correctly indicated the validity of the quasi-steady-state assumption for substrate.
For high substratelbiomass ratios, the conjecture was also falsified. At high dilution
rates close to washout, the ratio of eigenvalues evaluated in steady state and the
timescale estimation procedure correctly indicated timescale multiplicity. The substrate
concentration was the fast variable in this case, while from the conjecture it was
expected the biomass concentration would be the fast variable. Neither the direct scaling
procedure, nor the analytical scaling procedure led to the standard form. The timescale
estimation procedure correctly indicated the validity of the quasi steady state
assumption for substrate, as confIrmed by the clear boundary layer for this variable
observed in simulations.
With respect to the procedures tested, it is concluded that the timescale estimation
procedure is a helpful tool to detect timescale multiplicity and check validity of the
quasi-steady-state approximation. The use of eigenvalues for this purpose can be
misleading in nonlinear systems. The analytical scaling procedure can be helpful to
bring the problem into standard form and to obtain a perturbation parameter, thus
providing insight into the cause of timescale multiplicity. Direct scaling is not a
generally applicable procedure.
An important motivation to derive reduced nonlinear models is to obtain models that
have a larger validity range than linear models. It is observed, however, that also
nonlinear models obtained by reduction based on timescale separation have a limited
validity range, namely for that operational range in state space or parameter space in
which the assumptions for reduction are valid. A change of operating point may require
a different reduced model, as states that were partitioned as slow may become fast or
vice versa. This is probably the cause of the large error induced in the reduction of
ASMI studied by Steffens and Lant (1997), discussed in section 2.3. Thus, together
with the reduced models, also a validity range should be indicated.
5. Conclusions and perspectives
Rigorous models often must be reduced to low-order models as these are better suitable
for control, identifIcation and are helpful to obtain a better understanding. The aim in
this paper was to develop a systematic reduction procedure to obtain nonlinear reduced
models for controller design. First, reported reduction approaches and reduced models
of ASMI were reviewed. Several approaches are applied in ASMI model reduction.
Simplifying assumptions have been applied most frequently. Proposed models range
from simple black-box zero-order kinetic models neglecting biomass dynamics, to the
more complicated model of Jeppsson (1996), which includes biomass growth and decay
and consequently has a larger validity range.
Assumptions with respect to dynamics are also frequently usually applied for model
order reduction. This is logical, as the stiffness of activated sludge process argues to
develop models that are suitable for different timescales. It appears that the reduction
often is done heuristically. Relatively little systematic efforts have been done to analyse
and understand timescale properties of ASM 1.
58
It was concluded that singular perturbation is a promising candidate as a method to

develop a systematic reduction procedure. It is a systematic reduction procedure and it
can give more insight in dynamics of ASMI. Moreover, under certain conditions, it
retains the physical interpretation of the states in the order reduction, it provides an error
estimate and it is suitable for nonlinear model reduction, which is desired as it is an aim
to obtain reduced models.
To apply the method, the model must be in the so-called standard form, which means
that the states can be partitioned into fast and slow states. Obtaining this form or
recognising that the model is in this form is the difficult part in the method. Therefore, it
was decided to concentrate on this task.
First, methods in the literature were reviewed. Bastin and Dochain (1990) suggested
a simple rule for order reduction of (bio)process models through singular perturbation.
This simple rule is not generally applicable, however, and can be applied in specific
cases only, e.g. in the case of very low product solubility. Moreover, it does not yield
the state partitioning or tell when the reduction can be made. Segel and Slemrod (1989)
applied a methodology based on scaling to formally derive Michaelis-Menten kinetics
via singular perturbation. This methodology was successful in detection of multiplicity
of timescales, to bring the model in the standard form and provided a suitable
perturbation parameter that gave insight in the physical conditions that lead to timescale
multiplicity. This procedure was therefore used as a starting point for the state
partitioning (or detection of timescale multiplicity).
Three procedures were proposed, namely a direct scaling procedure, a procedure
based on a mixed numerical/analytic timescale estimation and the analytical scaling
procedure of Segel and Slemrod. The procedures were tested on a simple bioreactor
model of a chemostat with one biomass species and one substrate species. Starting point
in the analysis was the conjecture that occurrence of multiple timescales is associated
with a large concentration difference between the states.
At low substratelbiomass ratios, occurring at low dilution rates, the conjecture is valid.
The ratio substratelbiomass, the Monod number and the reciprocal of the dimensionless
residence time are suitable perturbation parameters. For intermediate substratelbiomass
ratios, the conjecture was falsified. The substrate concentration was the fast variable in
this case, while the substrate concentration and biomass concentration were in the same
order of magnitude. For high substratelbiomass ratios at high dilution rates close to
washout, the conjecture was also falsified. The substrate concentration was the fast
variable in this case, while from the conjecture it was expected the biomass
concentration would be the fast variable.
With respect to the procedures tested, the timescale estimation procedure is a helpful
tool in model reduction, which is concluded from the fact that it in all cases correctly
indicated whether the quasi-steady-state assumption was valid. The use of eigenvalues
evaluated in the steady state for timescale multiplicity however can be misleading in
nonlinear systems. The analytical scaling procedure is helpful to bring the problem into
standard form and to obtain a perturbation parameter in some cases, thus providing
insight into the cause of timescale mUltiplicity. The direct procedure is not generally
applicable.
It is expected that model reduction will remain an active research area in the next
years, for application in control, identification and process optimisation. Singular
59
S.R. Weijers
perturbation provides the formal basis for quasi-steady-state assumptions and timescale
separation and can be an important tool in future reduction studies. The procedure for
timescale estimation of variables is a helpful tool for model reduction, as it provides a
timescale to state association as well as an error estimate of the reduction. Scaling can
provide additional insight into causes of timescale multiplicity.
In almost all reported cases of ASMI reduction, the reduction was applied to
lumped systems without concentration gradients, assuming the system to be either
aerobic or anoxic. Thus reduction concentrated on reduction of the reaction kinetics,
rather than reduction of transport and mixing in the reactor. However, many activated
sludge systems are of the plug-flow type and exhibit gradients especially with respect to
dissolved oxygen. This also holds true for full-scale bioreactors. Therefore, reduction of
distributed systems may become an important topic in reduction of ASM 1 and
bioprocess models.
References
Ayesa E., Oyarbide G., Larrea L and 1. L Garcia-Heras (1995) Observability of reduced order models -
application to a modelfor control of alpha process, Wat Sci. Tech. 31 (2),161-170.
Bastin, G. and D. Dochain (1990) On-line estimation and adaptive control of bioreactors, Process
measurement and control, I, Elsevier, Amsterdam
Bodenstein, M and R Lutkemeyer (1924) Z Phys. Chem 114208.
Bowen, J. R. , A Acrivos and A K Oppenheim in (1962) Singular perturbation refinement to quasi-steady
state approximation in chemical kinetics, Chem Eng. Sci. 18, 177-188.
C6te, M , B. P. A Grandjean, P. Lessard and J. Thibault (1995) Dynamic modeling of activated sludge
processes: improving prediction using neural networks, Wat Res. 29,995-1004.
Carstensen,1. , P. Harremoes and R Madsen (1995) Statistical identification of Monod kinetic parameters
from on-line measurements, Wat Sci. Tech. 31 (2), 125-133.
Dochain, D. (1994) Contribution to the analysis and control of distributed parameter systems with application
to (bio)chemical processes and robotics, Aggregate thesis, Universite Catholique de Louvain, Belgium
Dold, P. L and G. v. R. Maris (1986) Evaluation of the general activated sludge model proposed by the
lAWPRC Task group, Wat Sci Tech 18 (6), 63-89.
Heineken, F. G. ,R M. Tsuchiya and R. Aris (1967) On the mathematical status of the pseudo-steady state
assumption of biochemical kinetics, Mathematical biosciences I, 95-113
Henze M. et al. (1987). Activated Sludge Model No. I, Scientific and technical reports No. I, lAWPRC,
London.
Isaacs, S. (1996) Short horizon control strategies for automating activated sludge process. Wat Sci. Tech. 34
(1-2), 203-212.
Isaacs, S. and M Kummel (1993) Dissolved oxygen setpoint control of an alternating activated sludge
process, Proc. 7th Forum for Applied Biotechnology, Med. Fac. Landbouww. Univ. Gent, 2083-2086.
Isaacs, S. R , M. Henze and M. Kummel (1995) An adaptive algorithm for external carbon addition to
alternating activated sludge process for nutrient removal from wastewater, Chem Engng. Sci. 50 (4),
617-629.
Jeppsson, U. and G. Olsson (1993) Reduced order models for on-line parameter identification of the activated
sludge process, Wat Sci Tech. 28 (11),173-183.
Jeppsson, U. (1996) Modelling Aspects of Wastewater Treatment Plants, PhD Thesis, Lund Institute of
Technology.
Julien, S. , 1. P. Babary and P. Lessard (1998) Theoretical and practical identifiability of a reduced order
model in an activated sludge process doing nitrification and denitrification, Wat Sci. Tech. 37 (12), 309-
316.
Kokotovic, P, R Khalil and 1. O'Reilly (1986) Singular perturbation methods in control: Analysis and design,
Academic Press, London.
Lin, C. C. and L A Segel (1974) Mathematics applied to deterministic problems in the natural sciences,
McMillan Publishing Co. Inc. , New York
60
lindberg, C. -F. (1997) Control and estimation strategies applied to the activated sludge process, PhD Thesis,
Uppsala University.
Lukasse, L 1. S. (1999) Control and Identification of Activated Sludge Processes, PhD Thesis, Wageningen
Agricultural University, Wageningen, The Netherlands.
Lukasse, L 1. S. , K 1. Keesman, A Klapwijk and G. van Straten (1997a) Adaptive receding horizon optimal
control of N-removing activated sludge processes, Proc. lllh Forum for Applied Biotechnology, Med
Fac. Landbouww. Univ. Gent 62 (4B), 1665-1672.
Lukasse, L 1. S. , K J. Keesruan, A. Klapwijk and G. van Straten (1997b) Identification for models predictive
control of biotechnological processes, case study: nitrogen removal in an activated sludge process, Proc.
IIIh IFAC symp. on Syst. Ident. 3, Fukuoka, Japan, 1525-1530.
Lukasse, L 1. S. , K 1. Keesman, A Klapwijk and G. van Straten (1998a) Optimal control of N-removal in
ASPs, Wat. Sci. Tech. 38 (3), 255-262.
Lukasse, L 1. S. , K 1. Keesman and G. van Straten (1998b) A recursively identified model for short term
predictions of Nf4 and N03 concentrations in alteruating activated sludge processes, Journal of Process
Control 9, 87 -100.
Lukasse, L 1. S. , K J. Keesman and G. van Straten (1999) A comparison of NHJN03 control strategies for
alternating activated sludge processes, Wat Sci. Tech. 39 (4), 93-102.
Marsili-libelli, S. (1989) Modeling, identification and control ofthe activated sludge process, In: A Fiechter
(Ed ), Advances in Biochemical EngineeringlBiotechnology 38, Springer Verlag, Berlin, Heidelberg,
89-148.
Moore, B. C. (1981) Principal component analysis in linear systems: controllability, observability and model
reduction, IEEE Transactions on Automatic Control, AC-26, 17-32.
Oosterhnis, N. M. G. (1984) Scale-up of bioreactors, a scale-down approach, PhD Thesis, Delft University of
Technology.
Olsson, G. and U. Jeppsson (1994) Establish.ing cause-effect relationships in activated sludge plants - What
can be controlled, Proc. 8th FAB, Med Fac. Landbouww. Univ. Gent. 59(4a), 2057-2071, Gent.
Robertson, G.A and Cameron, I.T. (1997) Analysis of dynamic process models for structural insight and
model reduction-Part 1: Structural identification measures, Computers chern. Engug. 21 (5), 455-473.
Robertson, G.A. and Cameron, I.T. (1997) Analysis of dynamic process models for structural insight and
model reduction-Part 2: A multi--stage compressor shut-down case study, Computers Chem Engug. 21
(5),475-488.
Roels,1. A (1983) Energetics and Kinetics in Biotechnology, Elsevier Science Publishers, Amsterdam
Scherpen,1. M. A (1994) Balancing for nonlinear systems, PhD Thesis, University of Twente, Enschede, The
Netherlands.
Segel, LA (1972) Simplification and scaling, SlAM Review 14 (4), 547-571.
Segel, L A and M. Slemrod (1989) The quasi-steady state assumption: a case study in perturbation. SlAM
Review. 31 (3), 446-477.
Segel, L A (1984) Dynamic phenomena in Molecular and Cellular Biology, Cambridge University Press,
Cambridge.
Steffens, M. A , P. A Lant and R. B. Newell (1997) A Systematic Approach for reducing biological
wastewater treatment models, Wat. Res. 31 (3),590-606.
STOWA (1996). Methods for influent characterization: Inventory and guidelines, Report STOWA 96-8,
Utrecht, The Netherlands (in Dutch).
Su, H -T. , T. 1. Mc. Avoy and P. Werbos (1992) Long-Term Predictions of Chemical Processes Using
Recurrent Neural Networks: A Parallel Training Approach. Ind Eng. Chem Res. 31 (5), 1338-1352.
Van Breusegem, V. and G. Bastin (1991) Reduced order dynamic modelling of reaction systems: a singular
perturbations approach, Proc. 301h conference on decision and control, Brighton, England, 1049-1054.
Van Breusegem, V. and G. Bastin (1992) Order reduction in bioprocess modelling: a singular perturbation
solution, Proc. IFAC modeling and control ofbiotecbnical processes, Colorado, U. S. A, 347-350.
van Can, H 1. L (1997) Efficient mathematical modelling for bioprocesses based on macroscopic balances
and neural networks, PhD. Thesis, Delft University of Technology.
Van Jmpe, 1. F. , P. A Vanrolleghem, W. Verstraete, B. De Moor and 1. Vandewalle (1991) Model based
monitoring and control of activated sludge wastewater treatment processes, Part II: nonlinear control of
the biotransformation and sedimentation process, Proc. of the 1991 European Simulation Symposium,
Nov. 6-8, Gent (Belgium), 221-226.
Vanrolleghem, P. (1994) On-line modelling of activated sludge processes: development of an adaptive sensor,
PhD Thesis, University of Gent.
61
S.R. Weijers
van Schagen, K M. , Veersma, AM. J. ,Meinema, K and Roest, H F. van der (1995) Multivariable: the new
generation?, H 20 28 (16),480-483 (in Dutch).
Vansteenkiste, O. C. and 1. A Spriet (1982) Modelling ill-defined systems, In: Progress in Modelling and
Simulation, Ed. Cellier F. E. , Academic Press, London, 11-38.
Vasil'eva, A B. , V. F. Butuzov and L V. Kalachev (1995) The boundary function method for singular
perturbation problems, SlAM Studies in Applied Mathematics, Vol. 14.
Wasynczuk, O.and Decarlo, R.O. (1981) The components connection model and structure preserving model
reduction, Automatica 17 (4), 619-626.
Weijers, S. R. ,1. J. Kok and H. A Preisig (1995) Analysis of dissolved oxygen dynamics in the IAWQ
Model No.1, Proc. 9th Forum for Applied Biotechnology, Med. Fac. Landbouww. Univ. Gent 60(4B),
2455-2458.
Weijers, S. R. and M. Weiss (1999) Quasi-steady state assumptions in the modeling of bioreactors: model
reduction and validation, Proc. American Control Conference, 1999.
Weiss, M. and H A Preisig (1997) Simplifying hypotheses in computer-aided modelling: a singular
perturbation approach, Computers Chem Engng, 21, Suppl., S721-726.
Wortelboer, P. (1994) Frequency-weighted Balanced Reduction of Closed-loop Mechanical Servo-systems:
Theory and Tools, PhD Thesis, Delft University of Technology.
Symbols
ASMI symbols
Stoichiometric parameters:
Y H :: Fleterotrophic yield (-)
Y A :: Autotrophic yield (-)
fp :: Fraction biomass yielding inert (-)
products
ixb :: Fraction N in biomass (kg N/kg COD)
iXD :: Fraction N in inert products (kg N/kg COD)
Kinetic parameters:
!LH :: Fleterotrophic growth rate (d")
constant
b H :: Fleterotrophic decay rate constant (d-')
Ks :: Affinity constant for Ss (kg m· j )
KoH:: Fleterotrophic affinity constant (kg m· j )
for So
KNHH:: Fleterotrophic affinity constant (kg m· j )
for SNH
KALKH :: Fleterotrophic affinity constant (mol m· j )
for SALK
ll g :: Correction factor for anoxic (-)
growth
!LA :: Autotrophic growth rate constant (d")
b A :: Autotrophic decay rate constant (d")
KoA:: Autotrophic affinity constant for (kg m- j
)
So
KNHA:: Autotrophic affinity constant for (kg m'J)
SNH
KNo:: Affinity constant for SNO (kg m'j)
KALKA :: Autotrophic affinity constant for (Mol m- J)
SALK
62
Symbols
kt. .... Hydrolysis rate (d- 1)
Kx:: Hydrolysis affinity constant (bm-~)
ka :: Ammonification rate (d- 1)

'T\h :: Correction factor for anoxic (-)
hydrolysis
Components:
Ss :: Readily biodegradable COD (k1l; m-~)
SI :: Soluble inert COD (kg m-3 )
SNH :: Ammonia and ammonium (bNm-3 )
SNO :: Nitrite and nitrate (kg N m-3 )
SND :: Soluble biodegradable or1l;anic N (k1l;Nm-3)
SALK:: Alkalinity (Mol m-3 )
So :: Dissolved oXY1l;en (k1l; m-~)
X BH :: Active heterotrophic biomass (kg m-3 )
X BA :: Active autotrophic biomass (k1l; m-~)
Xs:: Slowly biodegradable COD (kg m-3 )
XI :: Particulate inert COD (k}!; m-~)
X ND :: Particulate biodegradable org. N (kg N m-3)
Xp:: Particulate COD from decay (kg m-~)
63
PARAMETRIC SENSITIVITY OF A DYNAMIC MODEL FOR A PILOT
SINGLE-STAGE WASTEWATER TREATMENT PLANT
IGOR PLAZL, GORAN PIPU8, AND TINE KOLOINI

Department of Chemical Engineering. University of Ljubljana
Askerceva 5, 1001 Ljubljana, Slovenia
igor.plazl@Uni-Lj.si
Summary
The application of models in most treatment plants is limited due to a lack of advanced
input parameter values required by the models. Although the numbers of parameters are
presented in the literature, the range of some parameters is too wide considering the
parametric sensitivity of the dynamic model. On the other hand, some parameters
depend on the nature of a specific wastewater treatment plant. The present paper is
concerned with a dynamic model of the activated sludge process in a single stage
wastewater treatment plant with parametric sensitivity and evaluation. Model
calibration was successfully experimentally confirmed for the steady-state operational
conditions.
1. Introduction
The model parameters and state estimation associated with modern control studies
based on the available noisy process measurements. An alternative parameter approach
is via laboratory analysis [1]. These procedures of the IAWPRC Activated Sludge
Model I [2] are presented in Ekama et al. [3,4]. However, an important aspect of
instrumentation theory is that measurements are never deterministic variables, since
they always involve random noise as well as experimental random errors [5]. On the
other hand, many authors discuss various nonlinear numerical estimation methods
(Bayesian estimation method - [6]; Maximum Likelihood methods - [7]). Kabouris and
Georgakakos [1] presented a continuous process model formulation in a state-space
form. They presented the measurement model, including the relationship between the
measured quantities and model state variables, followed by the development of the
Linearized Maximum Likelihood (LML) algorithm. In this study attempts were made to
present a dynamic model for activated sludge process. Efforts were made to simplify the
parameter evaluation procedure. The calibration of the dynamic model was successfully
experimentally confirmed.
65
s.N. Agathos and W. Reineke (eds.), Biotechnologyfor the Environment:
Wastewater Treatment and Modeling, Waste Gas Handling, 65-72,
Igor Plazl, Goran Pipus, and Tine Koloini
2. Materials and methods
The laboratory at Vodovod-Kanalizacija Ltd., Ljubljana, monitors daily the activated

sludge process of a pilot single-stage wastewater treatment plant with reactor volume,
VI = 1.585 m3, and settler volume, V2 = 0.709 m3 (Fig. 1). For the purpose ofthis work,
some additional analyses were made. The concentrations of autotrophic nitrifying
biomass, X A, heterotrophic biomass, X H, activated sludge concentration, inlet and outlet
ammonium, A, dissolved oxygen, So, and the concentration of inlet and outlet substrate
(in g COD m- 3), SCOD, were determined by standard laboratory methods. The average
measured values for the period from 4 to 20 July 1998 are presented in table 1.
2, Q.
VI
Fig. I. Sketch of the single-stage wastewater treatment plant. ( 1 - reactor, 2 - settler)
Table 1. The average measured values of process parameters for the period from 4 to 20
July 1998 (source: JP Vodovod Kanalizacija Ud., Ljubljana).
Qo (inlet flow) = 0.263 m3 h- I

So = 3.6 g 02m-3
SCOD,O (inlet) = 416 g COD m- 3
Ao (inlet) = 16.1 g N-NH/ m- 3
SCOD.2 = SCOD,I (outlet) =42gCODm-3
A2 = Al (outlet) =0.5 g N-NH/ m-3
X H•1 = 3104 g m-3
X A•I = 96 g m- 3
Activated sludge concentration =4920 g m- 3
Biomass fraction =65%
66
Parametric sensitivity of a dynamic model for a pilot single-stage wastewater treatment plant
3. Mathematical model
The dynamic model is obtained by mass balances of substrate, ammonium, autotrophic

and heterotrophic biomass for the single-stage wastewater treatment plant, separately
for the reactor and the settler [1,2,8], considering aerobic growth of heterotrophs and
autotrophs and equations of time dependent flow rates. The set of linear differential
equations can be solved by an appropriate numerical method, using literature data on
the kinetic and stoichiometric parameters. However, in order to get an acceptable
agreement between the measured (Table 1) and predicted sludge process, numerical
estimation of some parameters is necessary. To simplify this procedure, considering that
the experiments were performed at steady-state conditions, the dynamic model was
rebuilt for the case of steady-state operation. In this way, the set of differential
equations becomes a set of ordinary equations and the equations for flow rates become
time independent. On the basis of the experimental values of sludge process variables
(Table 1) and by using some of the parameters from literature (YH, YK heterotrophic
and autotrophic yield coefficient; Ko,H, Ko,K oxygen half saturation for heterotrophs
and autotrophs; KA- ammonia half saturation; Ks- substrate half saturation) the
estimation of other parameters (~ax,H' ~ax,A- maximum growth rate for heterotrophs
and autotrophs; bH, bA- decay rate for heterotrophs and nitrifiers; ix-N content of
biomass) can be obtained by steady-state model calibration, using an appropriate
numerical method. However, it was found again that the parametric sensitivity of the
simplified model is still very high and we are unable to simply calibrate the parameters
with ordinary numerical methods, built in a powerful mathematical package
(Mathematica 3.0). The following iterative procedures were eventually found to be the
most effective and successful method for model calibration:
Step a - Some parameters (YH, YA, Ko,H, Ko,A, K~ were adopted from literature.
Step b - The assumed value for Ks can be found in the literature as the initial value
for calibration. The study of parametric sensitivity of the dynamic model has shown
that the influence of Ks (in range 4-60 gCOD m-3) on system stability is negligible
in our case. On the other hand, the linear dependence of Ks on evaluation of the
maximum growth rate of heterotrophic organisms can be obtained (Eq. 2:
~ax,H[Ks=4 g COD m-3] = 0.0146 h-'; ~ax,H [Ks=60 g COD m-3] = 0.0324 h-').
Step c - The assumed value for ix can be found in the literature (0.07 g N (g CODY
') as the initial value for calibration.
Step d - The assumed value for bAcan also be found in the literature (0.0063-0.002 h-')
as the initial value for calibration (0.002 h-').
Step e - The value for heterotrophic decay rate, bH, is calculated from the following
equation (Eq. 1):
67
Igor P1azl, Goran Pipu~, and Tine Koloini
(1)
which can be obtained by equalising equations of mass balances of substrate and

heterotrophic biomass arranged for steady-state, considering a substrate concentration
equality, SCOD,1=SCOD,2, and the condition XH,2=2XH ,l.
Step f - The value for the maximum growth rate of heterotrophic organisms, ~,H' can
now be calculated from the equation (Eq. 2):
JlJD1JX,H = ( Qo (S COD,O -.S COD,l )YH )( So + K O,H J( S COD,l +Ks ) (2)

X H,l V1 So SCOD,l
which is derived from the mass balance of heterotrophic organisms arranged for steady-
state operation, considering again the substrate concentration equality, SCOD,l=SCOD,2,
and the condition XH,2=2XH,1. Finally, equation 2 is obtained after simple mathematical
procedures considering the expressions for ~ (Eq. 1).
Step g - The maximum value of N content of biomass, ix,JD1JX' is determined from the
condition of positive growth rate at defined operational conditions (see Eq. 5):
(3)
If the initial value of ix is bigger than ix,JD1JX' the procedure is then repeated from step c)
until reaching a satisfying system solution (Eq. 6). The study of parametric sensitivity of
the dynamic model has shown that the influence of ix on system stability is very high.
After a number of iterations the final value, ix = 0.001 g N (g CODr1, was found for our
experimental conditions at which the system error is negligible.
Step h - The value for the maximum growth rate of autotrophic organisms, /lmax,A, can
be calculated from the equation (Eq. 4):
(4)
which is derived from the mass balance of ammonium arranged for steady-state
operation, considering the ammonium concentration equality, A1=A2' and condition
XA,2=2XA,1. After some simple arrangements, RA can be expressed:
RA = Qo(Ao -A1)YA ixRHVlYA

(5)
Vl (l+i x YA ) Vl (l+i x YA )"
68
Step i-The system error considering the initial value of b A can be determined from the
following equation (Eq. 6):
(6)
which is derived from the mass balance of autotrophic organisms arranged for steady-
state operation. From Equation 6 the new value for bA can be obtained. The procedure is
then repeated from step d) to i) until reaching a system solution.
4. Results and discussion
On the basis of the described method the calibrated parameters defined for the steady-
state activated sludge process taking place in a pilot single-stage wastewater treatment
plant (Table 1) can now be used for successful process simulation based on the dynamic
model. The calibrated parameters used in the dynamic model are presented in table 2.
Table 2. Estimated and typical literature values for some kinetic and stoichiometric
coefficients [8J
Parameter Steady-state model Literature data Units

calibration
bH 0.0079 0.008-0.016 h- l
bA 0.0018 0.002-0.006 h- l
ix 0.001 (0.066) 0.07 gN (gCOD)-l
J..lmax,H 0.0146 (0.0324) 0.125-0.25 h- l
~ax.A 0.0218 0.015-0.042 h- l
Substantial deviation can be seen for ix and J..lmax,H when the system calibration error is
practically zero. However, for practical purposes, which still allowed stability of the
system, the value 0.066 g N (g CODr l for ix and the value 0.0324 h- l for J..lmax,H[Ks=60
g COD m-3] were calculated. An example of dynamic model experimental confirmation
for the single-stage wastewater treatment plant (JP Vodovod Kanalizacija Ltd.,
Ljubljana) is presented in figure 2.
69
400
.-. Q. = 0.263 m3 iii
'?
E ~ = 0.256 m3 h- I - prediction
~ 300 3 -I __
0 ~= 0.004 m h - predictIOn
~
'-"
200
8
VJU __ - - - SCOD.'- inlet
- - SCOD.I= SCOD,2
100
42
00
50 100 150 200 250 300
t (h)
Fig_ 2_ Dynamic model prediction of substrate concentration for the steady-state activated
sludge process taking place in a single-stage wastewater treatment plant_
~O~==~====~====~====~--~-'--~I
QO(t) = 0.263 + /l30Exp(-0.2 t)Jf8.1fl I
- - " - reactor
- - A: - settler
I
O.OO!;-~-~50::--~--:1-!:-OO::--~--::"15:;;O:--~--:2~O""O-~---=2:'=50::--~-=!300
t (It)
Fig_ 3_ Dynamic response of ammonium concentration in the reactor and the settler during
the simulation of "real" two-day disturbance of inlet wastewater flow-
70
gJ(t) = 0.3 - 0.04 Sin(xtl24)

• - - !i:OD,l- reactor
--- ScoD,2- settler
o~--~----~--~----~----~--~----~--~
o 50 100 150
t(h)
Fig. 4. Dynamic response of substrate concentration in the reactor and the settler during the
simulation of sinuous fluctuation of inlet wastewater flow.
Dynamic simulation of the activated sludge process is presented in figure 3 during the
simulation of ''real'' disturbances of inlet wastewater flow. Dynamic responses of
substrate and heterotrophs concentration during the sinuous simulation of inlet
wastewater flow are presented in figures 4 and 5.
7000
I I
gJ(t) = 0.3 - 0.04 Sin(x 1124)
6000 ---------------------------------
3000
- - Xa.t -reactor
2000 I -- ~ - settler
o 50 100 150
t(h)
Fig. 5. Dynamic response of heterotrophs concentration in the reactor and the settler during
the simulation of sinuous fluctuation of inlet wastewater flow.
71
5. Conclusions
A mathematical model of the activated sludge process taking place in a pilot single-
stage wastewater treatment plant was built. A relatively simple procedure is proposed
for the evaluation of the kinetic and stoichiometric coefficients. The dynamic model
calibration was successfully confirmed experimentally for steady-state operational
conditions.
Acknowledgment
The authors thank the representatives of JP Vodovod-Kanalizacija Ltd., Ljubljana, in

particular Maja Drolka, Jurij Kus and Alojz Hojs for their encouragement, cooperation,
and financial support. We also express our appreciation to Mitja Lakner for useful
mathematical discussions.
References
[1] Kabouris, J.C., Georgakakos, A.P.: Wat. Res. 30 (1996), 2853-2865.

[2] Henze, M., Grady, C.P.L., Gujer, W., Marais, G.R., Matsuo, T.: Activated Sludge Model No.1, IAWQ
Scientific and Technical Reports No.1, IAWPRC, London, ISSN 1010-707X, 1987.
[3] Ekama, G.A., Marais, G.V., Pitman, A.R., Keay, G.F.P., Buchan, L., Gerber, A., Smollen, M.: South
African Water Research Commission, Pretoria, South Africa, 1984.
[4] Ekama, G.A., Dolt, P.L., Marais, G.R.: Wat. Sci. Tech., 18 (1986), 91-114.
[5] Willard, H.H., Merritt, Jr.L.L., Dean, J.A., Seattle, Jr.F.A.: Instrumental Methods of Analysis; Wadsworth,
Belmont, Calif., 7'" edition, 1988.
[6] Beck, M.B.: System Simulation in Water Resouces; North-Holland Publishing Co., Amsterdam, 1976.
[7] Mybeck, P.S.: Stochastic Models, Estimation and Control; Volume 2, Academic Press, New York, 1982
[8] Henze, M., Grady, C.P.L., Gujer, W., Marais, G.R., Matsuo, T.: Activated Sludge Model No.2, IAWQ
Scientific and Technical Reports No.3, Preprlnt for IAWQ Specialised Seminar on Modelling and
Contrul of Activated Sludge Processes, Copenhagen, (1994),22-24.
72
APPLICABILITY OF A SIMULATION BENCHMARK TO RESPIROMETRY-
BASED CONTROL STRATEGIES
JOHN B. COPP AND HENRI SPANJERS

AEST, Environmental Technology, Wageningen University and Research
Centre, Wageningen, The Netherlands. Fax: +31317 482108;
e-mail: John.Copp@algemeen.mt.wau.nl
Abstract
A simulation benchmark is applied to a number of respirometry-based control

strategies. The benchmark is a platform-independent simulation environment defining a
plant layout, a simulation model, influent loads, test procedures and evaluation criteria.
The purpose for defining a benchmark is to create a general simulation environment for
evaluating and comparing different control strategies. In the paper we describe the
definition of a general benchmark for the evaluation of control strategies in wastewater
treatment plants and identify the features of the benchmark which need adjustment to
make it applicable to a particular family of control strategies, namely those based on the
measurement of respiration rate.
1. Introduction
The activated sludge process aims to achieve, at lllllllmum costs, a sufficiently low
concentration of biodegradable matter in the effluent together with minimal sludge
production. To do this, the process has to be controlled. Many control strategies have
been proposed in the literature. However, the literature does not provide a clear basis for
comparison because of the many confounded influences, which have an impact on the
system. Many of these influences are easily recognised. For instance, the influence of a
control strategy on process performance is expected to vary with different disturbances,
thus the disturbances used to test the control strategy become important. Additionally,
the objective of reported strategies is not always consistent which may result in the
omission of data necessary to make a fair and unbiased comparison. As well, physical
characteristics of the process can have an impact on process performance, which makes
the comparison of strategies applied to different reactor layouts difficult. Also
complicating the evaluation is the lack of standard evaluation criteria. That is, effluent
requirements and treatment costs (i.e. labour costs) are often location specific. This
73
S.N Agathos and W Reineke (eds.), Biotechnologyfbr the Environment:
© 2003 KhMer Academic Publishers.
John B. Copp and Henri Spanjers
makes it difficult to judge the particular influence of an applied control strategy from a
reported performance increase.
Two common aerobic control strategies involve the maintenance of biomass levels
and dissolved oxygen concentrations in the aeration tanks by manipulating waste sludge
flow, return sludge flow or aeration capacity. Such strategies are based on
measurements of mixed liquor suspended solids and dissolved oxygen; however, alone
these measurements give limited information as to process stability and performance.
Alternatively, the respiration rate of activated sludge provides valuable information for
modelling and control of the activated sludge process because respiration is linked to
two important biochemical processes in a wastewater treatment plant: biomass growth
and substrate consumption.
The respiration rate is an important variable in the activated sludge process as it
provides information on activity and concentration of the biomass, influent waste
concentration and composition, toxicity and concentration of biodegradable matter in
the effluent. Highly sophisticated and reliable respirometers have been developed in
recent years, but the application of these instruments in control of full-scale activated
sludge plants lags behind the recognised potential of respirometry. This is due, at least
in part, to a lack of understanding with respect to the information content of
respirometric data and confusion arising from inconsistency in implementation of
control strategy methods.
The literature related to respirometry is substantial and numerous control strategies
involving respirometry have been proposed and tested. However, from a practical
standpoint, it is not reasonable to experimentally test and verify the effectiveness of all
reported control strategies. Alternatively, given a standardised procedure, it is possible
to efficiently evaluate numerous strategies through computer simulations. Validation of
the computer simulations is difficult without supporting experimental or full-scale data,
but the value of the work is enhanced through the use of accepted activated sludge
models. Because appropriate simulation tools for the activated sludge process are
available this approach has numerous advantages, but still there is a need for a
standardised evaluation procedure.
In an attempt to standardise control strategy evaluation by simulation, a defined
methodology, called a benchmark, has been developed. The simulation benchmark is a
comprehensive simulation environment, which defines a plant layout, a simulation
model, influent loads, test procedures and evaluation criteria. In this instance, the
evaluation criteria relate the model output to performance indicators such as costs and
effluent qUality. This paper briefly describes the features of a generalised simulation
benchmark for evaluating and comparing different control strategies and then
specifically identifies features of that benchmark that need adjustment to make it
applicable to respirometry-based control strategies.
2. General benchmark description
The assessment of activated sludge control strategies is confounded in most cases by the
multifaceted nature of the process under study. That is, a true comparison of reported
strategies is difficult because of the many variables that have an impact on process
performance. To combat these problems and decrease the influence of these many
74
Applicability of a simulation benchmark to respirometry-based control strategies
variables, there has been a recent effort to develop a standardised testing procedure - a
benchmark.
Once defined, the benchmark forms the building block from which the various control
strategies are evaluated. Several benchmarks have been published or are under
development [1,8,11]. Spanjers et al. (1998) defined a simulation benchmark and used it
for the evaluation of a respirometry-based control strategy. The European Co-operation
in the field of Scientific and Technical Research (COST) 624 Action is currently
working on the development of a general benchmark. The COST benchmark (in this
paper referred to as the general benchmark), although based on the work of Spanjers et
at., is an attempt to create a generally accepted benchmark for the evaluation of all types
of control strategies. Cross-platform testing of the COST benchmark has been
successfully demonstrated such that similar results can be attained using GPS-XTM
(Hydromantis Inc), Simba™ (Ifak System GmbH), WESr ™ (Hemmis n.v.) and a
user defined Fortran code [1,8]. A brief description of the COST generalised benchmark
is outlined below.
3. Plant layout
The COST benchmark plant was designed to treat an average flow of 20000 m3 d- I with
an average influent biodegradable COD concentration of 300 g m- 3. The flant was
designed for pre-denitrification using a 5 tanks-in-series layout (5999 m) with a
secondary settler (6000 m\ The first two tanks are fully mixed, but unaerated and the
last three are fully mixed and aerated. For modelling purposes, the secondary settler is
comprised of 10 equal volume layers. The plant has two internal recycles including one
from the underflow of the settler to the head of the plant, and a second recycle directly
from the fifth to the frrst tank. figure 1 shows a schematic representation of the
proposed configuration.
Fig. 1. General benchmark plant layout
4. Process models
The IAWQ Activated Sludge Model No 1 [4] was chosen to simulate the biological
processes. The double-exponential settling velocity model proposed by Takacs et at.
[14] was selected to describe the behaviour of the settler. Within the framework of the
general benchmark, several manipulated variables are defined. In particular, oxygen
supply, the internal and external recycle flow rates, and the waste flow rate are defined
as potential manipulated variables.
75
5. Test protocol
The test protocol begins with a series of procedures meant to verify the simulation
software being used. The first procedure involves simulating to steady state (or, if
preferred, for lOx the sludge age) the described layout with all process variables
constant including all flows and influent constituents. The characteristics of the
prescribed influent are defined and represent the flow-weighted average concentrations
of a 14-day dry weather influent data file [2,15]. Following verification of the steady
state output data, the dynamic response of the simulation is checked by first using a
dynamic dry weather influent file as input to the system.
35000 1000
900
30000
- 800 .r
;..;:
,:;"" 25000 .. 700
..,"=
!~ 20000
600
~
500
~
1:: 15000 400
j
;S 10000 300
- 200
5000
100
o ........................;.........................
i ...................... L ..................... .I ...•.•.•................ 0
o 0.2 0.4 0.6 0.8 1
Time (days)
Fig. 2. A typical dry weather day as depicted in the dry weather data file.
The dry weather file is one of three weather files (each two-weeks in duration)
generated to test the dynamic response of the system [2,15]. This dry weather file
exhibits characteristic diurnal variations in flow and component concentrations. Also
incorporated in the file is a substantial (20%) decrease in flow and load during the
'weekend'. Figure 2 shows a sampling from a typical day of data. The second and third
files are based on the dry weather data with an added rain event during the second week.
The first of these rain files has, during the second week, a simulated rain event that
results in a constant increase in influent flow and lasts for approximately 2 days. In
particular, this file depicts a constant hydraulic load increase without any increase in
COD or nitrogen load as compared to the dry weather file. The second of these rain files
has two storm events during the second week. These storm events are shorter in
76
duration than the rain event, but are more intense. Also, in addition to increasing the
hydraulic load, these storm events have an associated increase in particulate COD load
as compared to the dry weather data.
Verification of the dynamic response is done using all three of the weather files
following simulation with three weeks of dry weather data. That is, from the steady
state results the system is simulated using the dry weather file as input and the state of
the entire system is saved. Then, in turn, each of the three weather files is used from that
saved state. As the first week of each file is the same, this provides three weeks of dry
weather data response followed by one week of alternative weather. Analysis of the
system response is calculated based on the output data from the fourth week. At each
stage, the output data are verified with published data given in the benchmark definition
[3]. It is assumed that the model implementation is correct once there is agreement
between the published data and the user's output data.
To further validate the user's software, a basic control strategy is proposed. That is,
prior to defining and testing a new control strategy users must validate their software by
implementing a defined control strategy. Again, the generated output can be compared
to a standardised output as defined in the benchmark description. The basic control
strategy has two parts and consists of:
• dissolved oxygen (DO) control in the 5th tank (set-point 2 mg L-') by

manifulation of the aeration rate via the oxygen transfer coefficient, KLa (in
the 5 tank).
• nitrate control in the 2nd unaerated tank (set-point of 1 mg L-') by manipulation
of the internal recycle flow rate (5 th 7 1st tank).
When performing the control test, the system is again simulated to steady state (or, for
lOx the sludge age if preferred), but this time with active controllers. Following this, the
dry weather file is applied and then each of the weather files is used as described
previously, also with active controllers. The results are then compared to the available
data. By following these steps and validating the results, it is assumed that the simulator
is correctly tuned for further control strategy evaluations.
6. Performance assessment
Performance assessment of the controlled system is made at two levels. The first level
concerns the local control loops and is assessed by:
• IAE (Integral of the Absolute Error)

• ISE (Integral of the Squared Error)
• maximum deviation from set-points
• error standard deviation
Basically, this serves as an indication that the proposed control strategy has been
applied properly.
77
The second level quantifies the effect of the control strategy on plant performance and it
can be divided into three sub-levels:
• effluent quality
• operational costs
• potential increase in capacity
Within the general benchmark, effluent quality is considered through an effluent quality
index, which is meant to quantify into a single term the effluent pollution load to a
receiving water body. Constraints with respect to the effluent quality are defined and the
percentage of time that the constraints are not met is reported. As well, the number of
violations also is reported based on a 15-min sampling time. The operational costs are
considered through three terms: sludge treatment/disposal, pumping and aeration
energy, and an empirical estimation of the wear and tear on the pumps due to the control
strategy. Finally, a potential increase in capacity (PIC) is calculated for the controlled
case and is defined as the potential increase in influent flow that could be obtained using
the proposed control strategy. The PIC is an empirical attempt to quantify the savings
resulting from a potential increase in capacity of the plant as opposed to a plant
expansion.
By defining the benchmark in these specific terms and setting up the simulations in
this way, the user ensures a consistent and unbiased methodology for the evaluation of
future control strategies.
7. Application to respirometry-based control strategies
Control strategies can be evaluated by model simulation and respirometry-based control

strategies are no exception. However, as already stated, the methodology used in the
evaluation is a critical step and is the most efficient way to ensure unbiased
comparisons. This implies that to make unbiased comparisons, each control strategy
must be evaluated under the same conditions. It also implies that the effect of the
control strategy must be compared to a fully defined and suitable reference output. Only
then is it possible to truly evaluate a control strategy and compare it with another
strategy.
In an attempt to standardise control strategy evaluation, the general benchmark
simulation concept of the COST 624 Working Group on Operation [8] was evaluated as
a tool for evaluating a number of respirometry-based control strategies. Although the
general benchmark is based on a benchmark developed for the evaluation of
respirometry-based strategies [11], it has since undergone several changes. Hence, it
was of interest to investigate the applicability of the general benchmark to this specific
family of control strategies.
The main characteristic of respirometry-based control strategies is that the measured
variable, and thus the controller input, is either a respiration rate or a variable deduced
from a respirometric experiment. Because of this characteristic, respirometry-based
control strategies are a particular family of control strategies that need a dedicated
simulation tool for their evaluation and comparison. In short, the general benchmark
provides an excellent starting point, but it does not provide sufficient flexibility for the
78
evaluation of respirometry-based control strategies. The following section outlines a

number of respirometry-based strategies and the changes in the benchmark that are
needed for a complete evaluation.
7.1. JOYCE ET AL. (1974)
• Strategic objective - to decrease the variability in the effluent (as defined by

the BODs) and to improve the effluent quality
• Measured variables - respiration rate, DO, 5 min settling volume, influent total
organic carbon (TOC), return sludge suspended organic carbon (SOC)
• Controlled variables - FIM, DO, respiration rate and 5 min settling volume
• Manipulated variables - influent flow distribution (i.e. step-feed),
waste sludge, recycle sludge, and air flow rates
Joyce et al. devised a complicated algorithm to control a full-scale fully aerobic plant
receiving a significant fraction of its influent flow (up to 75%) from several food
processing plants. Necessary changes to the general benchmark include aeration to all
tanks, TOC, SOC and sludge settling test measurements as well as step-feed capability
for the influent flow.
7.2. STENSTROM AND ANDREWS (1979)
• Strategic objective - to maintain a constant biological growth rate in a reactor

subjected to a diurnal variation in influent load
• Measured variables - respiration rate, mixed liquor volatile suspended solids
(MLSS), influent flow rate
• Controlled variables - SRT, specific respiration rate
• Manipulated variables - waste and recycle sludge flow rates, influent
flow distribution (Le. step-feed)
To evaluate the Stenstrom and Andrews control strategies using the benchmark several
changes would be needed including a fully aerobic nitrifying plant layout, with step-
feed capability for the influent flow. Another variation relates to the evaluation of the
effectiveness of the control strategy which Stenstrom and Andrews related to the
variance in the controlled variable, specific respiration rate.
7.3. S0RENSEN (1980)
• Strategic objective - to decrease the variability in the effluent quality (that is,
the author's aim was to devise a strategy which decreased the variance in the
effluent quality, and not necessarily improve the effluent quality)
• Measured variable - blower speed (to maintain constant dissolved oxygen, DO)
in the last tank of a fully aerobic tanks-in series plant layout
• Controlled variable - (derived) respiration rate in the last tank
• Manipulated variable - influent flow distribution (Le. step-feed)
79
S0rensen outlined a control strategy based on a derived oxygen utilisation rate in the
last tank of a fully aerobic three-tanks in series layout. As applied to the general
benchmark, several changes are required including a flow proportional recycle flow
rate, aeration in all tanks, and step-feed capability for the influent flow.
7.4. SEKINE ET AL. (1988)
• Strategic objective - to maintain optimal conditions for microbial activity (i.e.

to diminish DO limitations in the aeration tank) by controlling the DO set-
point at a adequate level
• Measured variables - influent flow rate, endogenous and aeration tank outlet
respiration rate, DO, mixed liquor suspended solids (MLSS)
• Controlled variables - DO, critical specific respiration rate
• Manipulated variable - air flow rate
Sekine et al. developed a hybrid DO control strategy that used respiration rate
measurements to optimally determine an appropriate DO set-point. The system studied
was different from the general benchmark in several ways. The system was fully
aerobic and nitrifying. It was operated at a long (15 day) SRT and had a feed-forward
control loop based on influent flow measurements for DO in the aeration tanks. Unlike
the general benchmark, waste sludge was not continuously removed from the system,
but rather was removed once per day.
7.5. KIM ET AL. (1996)
• Strategic objective - to maintain a constant organic loading rate

• Measured variable - respiration rate
• Controlled variable - loading rate
• Manipulated variable - influent flow rate
Kim et al. linearly related measured mixed liquor respiration rate to organic loading rate
and devised a control strategy for maintaining a constant loading rate through the
manipulation of the influent flow rate. However, to evaluate such a strategy using the
benchmark, the benchmark would have to allow for manipulation of the influent flow
what it doesn't. Further, the system studied by Kim et al. was fully aerobic and
nitrifying with a solids retention time of 17 days.
7.6. LAROSE ET AL. (1997)
• Strategic objective - to optimise the cycle time in a biological phosphorus

removing sequencing batch reactor
• Measured variable - respiration rate
• Controlled variable - anaerobic rapidly biodegradable COD concentration
• Manipulated variable - anaerobic cycle time
80
Larose et al. applied a respirometric technique to a sample of sludge from the anaerobic
stage of a cyclical SBR being used for biological phosphorus removal. Several problems
become immediately apparent when an attempt is made to apply the general benchmark
to this strategy. For instance, biological phosphorus removal is not included in the
biological process model ASMI. Also, the SBR has intermittent feed and this strategy
has potential variable cycle lengths (or variable volumes, if translated) that are not
defined in the general benchmark description [3].
8. Overview and discussion of results
From the vast literature of respirometry-based control [12], a selection of six

respirometry-based control strategies was examined for the applicability of the general
benchmark as developed by COST. Table 1 lists some of the necessary changes needed
to make the general benchmark applicable. The general benchmark can be used to
evaluate respirometry-based control strategies. However, for each strategy a number of
benchmark features must be modified especially those related to the controller and plant
layout. In particular, several features, not included in the general benchmark, are
commonly associated with respirometry-based strategies. For instance, often these
strategies are applied to fully aerobic systems including short SRT, carbon removing
only plants and longer SRT, nitrifying systems. Also, manipulation of the influent flow
(i.e. step-feed) is a popular mechanism.
Table 1. Overview of necessary changes to the general COST benchmark for the evaluation
of a selection of respirometry-based control strategies.
Strategy Fully Need for Step-feed External Variable Need for

aerobic process and respirometer influent evaluation
process flow criteria
model changes
changes
Joyceetal. ./ ./ ./
Stenstrom and ./ ./ ./
Andrews
SS!lrensen ./ ./ ./
Sekine et al. ./ ./ ./
Kimetal. ./ ./ ./
Larose et al. ./ ./ ./
In an attempt to make the general benchmark applicable to this family of control

strategies, several necessary additions are suggested. Figure 3 shows a schematic
representation of the slightly modified layouts. In this case, two new fully aerobic
81
designs were developed including one for carbon removal and one for nitrification.
Also, step-feed capability was added to all three designs. The adopted layouts are all 5
tanks-in-series designs, but of course not all proposed strategies are applied to this type
of configuration. So, for those strategies, assumptions need to be made to fit the
strategies to the 5 tanks-in-series benchmark layout and may include approximating the
source of samples. Changes to the general benchmark should be limited to those that are
required, and those that have a direct and significant impact on the control strategy
performance.
Performance assessment remains the most important part of any strategy evaluation.
The general benchmark lays out a performance assessment procedure for evaluating
control strategies, but the applicability of such a procedure is highly dependent on the
strategy objective. The general benchmark assessment assumes that the objective of all
strategies is to improve effluent quality or decrease the cost of treatment. As shown in
this paper, that is not always the case. Table 1 shows that the general benchmark
evaluation criteria are not totally applicable to respirometry-based strategies. That is, in
many respirometry-based strategies the objective is to decrease the variability in a given
parameter and not necessarily improve it [6,10,13]. In these instances, the performance
assessment defined in the general benchmark is unsuitable because it does not provide a
means to evaluate whether or not the strategy was successful in achieving its stated
objective. Hence, the evaluation assessment must be modified to be applicable in these
cases. Further work to develop a comprehensive and applicable assessment technique
clearly is needed. In particular, an evaluation criteria for variability assessment is
required.
Fig. 3. Schematic representation of the altered plant layouts. Layout A is consistent with the
carbon-only and nitrifying layouts showing 5 completely mixed and aerated tanks in series.
Layout B is a schematic of the denitrifying layout with tanks 1 and 2 mixed and unaerated,
and tanks 3, 4 and 5 aerated.
Strategies based on respirometry include the need to measure a respiration rate. In many
instances, as in the Larose et at. strategy outlined above, there may be a need to measure
respiration rate in a separated mini-reactor, i.e. a respirometer. These measurements
often are complicated and need to be accounted for in any evaluation procedure because
82
they may have a direct impact on the performance of the control strategy. For instance,
time delays, which would be inherent in any measurement of respiration rate, need to be
taken into account. Further, respirometric measurements are often obtained through
sophisticated experiments, which may combine substrate and biomass from various
sources within the process. As such, these experiments often need to occur outside the
process itself, in a separated environment. Also, when variables are deduced from
respiration rate data, the practical methodology of deducing the variables needs to be
considered.
Take for example a strategy, which aims to maintain a constant active biomass
concentration. As it is not possible to measure active biomass in practice, an alternative
measuring approach must be devised. For instance, it might be proposed that the
endogenous respiration rate be used as a measurable indicator of the active biomass. In
practice, the endogenous respiration rate may be measured by sampling sludge from a
completely mixed reactor during low loading conditions or from the end of a plug-flow
reactor. Alternatively, the active biomass concentration may be estimated from the
maximum respiration rate measured in an external apparatus using an excess of a
defined readily biodegradable substrate. Such a procedure would require a separate
experimental set-up. There are several ways to measure indicators of active biomass,
but for benchmarking, the problem remains as to how practice is represented in the
modelling environment. That is, any model simulation study which aims to evaluate a
given control strategy must mimic (as closely as possible) what can be done in practice.
In a simulation environment, using ASMI, it is possible to simply calculate the
concentration of active biomass from the state variables; however, as already stated, in
practice this is not possible. So, to more closely approximate what can be measured in
reality, the endogenous respiration rate might be proposed as a modelled indicator of the
active biomass. Unfortunately, in ASMI, the endogenous respiration rate is not well
defined; and therefore, a direct correlation between respiration rate and active biomass
concentration may not be an accurate assumption in the modelling environment. An
alternative approach for simulating the measurement of the endogenous respiration rate
is to model an external un-fed recirculating plug-flow reactor to approximate a practical
method. The maximum respiration rate might also be used. In either of these cases, the
respirometry experiment would have to be modelled because the procedure can have a
direct impact on the effectiveness of the control strategy. However, such capabilities are
not included in the general benchmark.
The benchmark is a comprehensive description of a simulation environment and
evaluation procedure including the definition of a plant layout, a simulation model,
influent loads, test procedures and evaluation criteria. A benchmark, by definition, must
be independent of the simulation software being used. That is, the simulation software
should have no effect on the modelling output such that different simulators modelling
the same system should give the same result. Because of the many simulator specific
options, this is not always the case, nor is it a trivial task to ensure similar results using
different simulators. A substantial effort has gone into this aspect of the benchmark
development. However, by stipulating specific model equations, modelling procedures
and simulator specific options, similar results can be achieved. It should be noted that
synchronising the output is not limited to commercially available simulation software.
That is, similar results should be attainable irrespective of the wastewater simulation
83
tool that is used including user defined computer code. Demonstrating similar results in
this way is the first step in the evaluation procedure, and ensures that the simulators are
tuned in the same way and also ensures the consistent comparison of control strategies.
Four different simulation platforms have been used (BioWin™, GPS-XTM, Simba™ and
WESTfTM) in order to verify that the results were independent of the software.
9. Conclusions
In the paper we describe a number of respirometry-based control strategies and identify

the features of the COST general benchmark, which need adjustment in order to make it
applicable to these strategies. The results of this investigation clearly demonstrate that
development of a universally applicable benchmark is not possible because of the many
variables and permutations that can occur in the activated sludge process. However, the
general benchmark does provide a reasonable starting point for evaluations. That is, the
procedures described in the benchmark definition for tuning the simulation software are
universally applicable. These procedures are valid and provide a standardised method
for software/model verification even if several changes to the benchmark are required
for the evaluation of a particular control strategy.
Acknowledgements
The authors are pleased to acknowledge the paid corporate sponsors who have made
this work possible: Aquafin (Belgium), Biotim n.v. (Belgium), Dow Chemical (USA),
Eco Process and Equipment IntI. Inc. (Canada), ENEA (Italy), Hemrnis n.v. (Belgium),
Hydromantis Inc. (Canada), International Water Association (United Kingdom),
CIRSEE - Lyonnaise des Eaux-Durnez (France), Severn Trent Water (United
Kingdom), STOWA - Dutch Foundation for Applied Water Research (The
Netherlands), WRc (United Kingdom).
References
[1] Alex J., Beteau J.F., Copp J.B., Hellinga C., Jeppsson U., Marsili-Libelli S., Pons M.N., Spanjers H. and.
Vanhooren H. (1999) Benchmark for evaluating control strategies in wastewater treatment plants.
Proceedings of the European Control Conference, ECC '99, Karlsruhe, Germany, August 31-September
3,1999.
[2] Copp J. B. (1999) Development of standardised influent files for the evaluation of activated sludge control
strategies. IAWQ Scientific and Technical Report Task Group: Respirometry in Control of the Activated
Sludge Process - internal report.
[3] COST 624 Action website (http://www.ensic.u-nancy.fr/COSTWWTP) - The European Co-operation in
the field of Scientific and Technical Research.
[4] Henze M., Grady Jr c.P.L., Gujer W., Marais G.v.R. and Matsuo T. (1986) Activated sludge model No.1,
IAWQ Scientific and Technical Report No.1, IAWQ, London.
[5] Joyce, R.J., Ortman, C. and Zickefoose, C. (1974) How to optimise an activated sludge plant. Wat.
Sewage Wks 121, 96-99 .
[6] Kim C.W., Choi E.H., Kim B.G., Lee T.H. and Park T.J. (1996) Stable loading control in a field-scale
activated sludge plant using on-line respiration meter. Proceedings of the IAWQ's 18th Biennial
International Conference and Exposition - 'Abstracts - Poster Presentations', June 23-28, 1996.
84
Applicability of a simulation benclunark to respirometry-based control strategies
[7] Larose, A., Perrier, M. and Comeau, Y. (1997) Respirometric control of the anaerobic period duration of
an SBR Bio-P process. Wat. Sci. Technol. 36 (5), 292-300.
[8] Pons M. N., Spanjers H. and Jeppsson U. (1999) Towards a benclunalX for evaluating control strategies in
wastewater treatment plants by simulation. Proceedings of 9th European Symposium on Computer Aided
Process Engineering, Budapest, Hungary, May 31-June 2,1999.
[9] Sekine, T., Sato, S., Furuya, N. and Sunuhara, H. (1988) Supervision and control of the activated sludge
process utilising the respiration rate activity. Environ. Technol. Lett. 9, 1317-1326.
[10] Sj/lrensen, P.E. (1980) Evaluation of operational benefits to the activated sludge process using step feed
control strategies. Prog. Wat. Technol. 12 (6),109-125.
[11] Spanjers H., Vanrolleghem P., Nguyen K. Vanhooren H. and Patry G.G. (1998) Towards a benclunalX
for evaluating control strategies in wastewater treatment plants by simulation. Wat. Sci. Technol., 37(12),
219-226.
[12] Spanjers H., Vanrolleghem P., Olsson G. and Dold P.L. (1998) ) Respirometry in Control of the
Activated Sludge Process: Principles, lAWQ Scientific and Technical Report No.7, lAWQ, London.
[13] Stenstrom, M.K. and Andrews, J.F. (1979) Real-time control of the activated sludge process. J.
Environ.Engng 105, 245-260.
[14] Takacs I., Patry G.G. and Nolasco D. (1991) A dynamic model of the clarification thickening process,
Wat. Res., 25,10,1263-1271.
[15] Vanhooren H. and Nguyen K (1996) Development of a simulation protocol for evaluation of
respirometry-based control strategies, Technical Report, University of Gent, Gent, Belgium.
85
FAULT DETECTION AND ISOLATION IN WASTEWATER TREATMENT
PLANTS
Comparison of different approaches and experimental results
JEAN-PHILIPPE STEYER AND JEROME HARMAND

Laboratoire de Biotechnologie de l'Environnement - INRA, Avenue des
etangs, 11100 Narbonne, France email: steyer@ensam.inra.jr
Abstract
This paper presents the problem of detecting and isolating faults in biological
wastewater treatment plants. Depending on the specific problems to be solved and on
the knowledge available on the process, model-based, fuzzy logic-based and artificial
neural network-based approaches are investigated and tested on different processes.
Finally, some conclusions and perspectives are drawn.
1. Introduction
Due to the increasing complexity and necessity of safety in industrial processes,

efficient diagnosis systems are becoming more and more important. Indeed, even in
normal operation conditions, several types of disturbances can be present and they can
largely affect the operating conditions of the process. A clear need for advanced control
and diagnosis systems is thus expressed in order to keep the system performances as
close as possible to the optimal conditions. This fact is particularly true in wastewater
treatment processes (WWTPs) where the normative constraints are becoming more and
more stringent.
Anaerobic digestion processes are commonly used for the treatment of industrial
wastewaters and their use presents a number of advantages. In particular, anaerobic
sludge production is much lower compared to aerobic activated sludge processes. The
organic matter is also anaerobically degraded into a gas mixture of carbon dioxide and
methane, thus producing valuable energy. In addition, this process is able to operate
under severe conditions : high strength effluents and short hydraulic retention time. For
all these reasons, anaerobic digestion processes are particularly well adapted for
concentrated wastes such as agricultural (e.g. plant residues, animal wastes) and food
industry wastewater. However, biological processes in general - and anaerobic
87
S.N. Agathos and W. Reineke (eds.), Biotechnology for the Environment:
Jean-Philippe Steyer and Jerome Harmand
digestion processes in particular - exhibit some very specific behaviours and the control
needed is not the same as it would be usually understood by a control engineer: see for
instance the survey paper by Andrews (1994) in which it is stressed that, in biological
processes, things seem to work fairly well until some failures or faults occur. Hence,
there is an increasing need for Fault Detection and Isolation systems (also known as
FDI systems).
Automatic FDI systems use analytical or heuristic knowledge in order to detect as
early as possible deviations from the normal operation that tend to degrade the overall
system performance. Such malfunctions may occur in the sensors, the actuators, the
components of the process (e.g., inhibition of the biomass in a biological reactor) or in
the control system, if the process is running in closed loop.
Model-based FDI methods have been first initiated by Willsky (1976). These
methods can be divided into two categories. The first method is based on state
estimation. It includes detection filters presented by White and Speyer (1987), parity
space approaches developed by Chow et at. (1986), Patton and Chen (1991), and
Delmaire et al. (1994), as well as works of Frank (1993) on diagnostic observer based
methods. Parameter estimation techniques, proposed in particular by Isermann (1984)
belong to the second category.
In parallel to these model-based methods, there has been rapidly a growing interest
for using techniques such as artificial neural networks or fuzzy logic for fault detection
purposes (e.g., Frank, 1994; Isermann and Balle, 1997) and few demonstrations of the
advantages of fuzzy-based FDI systems can be found in the literature (see for example
Ulieru and Isermann, 1993; Montmain and Genti11993; Kiupel and Frank 1996; Giraud
and Aubrun, 1996; Genovesi et al. 2000). Other recent studies use this technique for
sensor validation (Boudaoud and Masson, 1998) or combine it with artificial neural
networks (Steyer et al. 1997).
However, a good FDI system for a biological process also requires inclusion of
specific biological knowledge. Artificial Intelligence (AI) methods such as knowledge-
based systems or fuzzy logic allow one to bring new insights into biological process
control and to introduce a "biological dimension" (See for example Konstantinov and
Yoshida, 1992; Aynsley et al. 1993; Steyer et at. 1993; Siimes et al. 1995, Roca et al.
1996; Steyer et al. 1996).
This paper presents the advantages and the drawbacks relevant to three design
methods respectively based on explicit mathematical models, Fuzzy Logic and Artificial
Neural Networks (ANNs). In section 2, the model-based and ANN-based approaches
are applied to an anaerobic biological process. Then, in an other context, a fuzzy-based
approach is used to adapt functioning conditions for the supervision of an industrial
equalization process. Experimental results are presented and analysed before some
conclusions are given.
88
Fault detection and isolation in wastewater treatment plants
2. Examples of fault detection and isolation methods for anaerobic digestion

processes
2.1. THE EXPERIMENTAL PROCESS
2.1.1. Characteristics o/the wastewater

The wastewater to be treated is an industrial wine distillery wastewater. The wine
industries have changing operating conditions characterised in particular by an increase
of activity in spring and autumn months while decreasing the rest of the year. Their
rejected wastewaters contain the following elements: sugars, alcohol, esters, organic
acids, yeast and bacteria. Their pH is about 4 and, to improve the treatment efficiency,
these wastewaters have to be neutralised before entering the reactor. In addition, their
Chemical Oxygen Demand (COD) is about 10 to 45 times more important than in urban
wastewaters. The general characteristics are summarised in table 1.
Table 1: Characteristics of wine distillery wastewaters
Component Range Mean value
Volatile Fatty Acids (gil) [5.00 - 6.00] 5.50
Total Organic Carbon (gil) [2.50 - 6.00] 4.25
Phenol (mgll) [90.0 - 275] 182.5
COD (gil) [9.00 - 17.4] 13.2
Soluble COD (gil) [7.60 - 16.0] 11.8
Total Suspended Solids (gil) [2.40 - 5.00] 3.7
Volatile Suspended Solids (gil) [1.20 - 2.70] 1.95
Alkalinity (meqll) [30.8 - 62.4] 46.6
pH [4.00 - 5.40] 4.7
2.1.2. The anaerobic digestion process

The experiments were performed using an anaerobic up-flow 120 litre fluidised bed
reactor at the "Laboratoire de Biotechnologie de IEnvironnement" of INRA in
Narbonne, France. The process is associated with a 520 litres circular settler. The
feeding tank consists in 3 tanks (27 m3 each) that are connected to an intermediate
20-litre buffer tank by pipes of about 0.4 m3• Figure 1 describes the pilot plant.
The reactor is made of a 3.65 meter high circular column. Support for bacteria
attachment is made of small volcanic particles of 300 ~ mean diameter. Fluidisation
of the bed is ensured by a high liquid recirculation flow rate (Le., 630 IJh using a
centrifugal pump). The 20 litre buffer system has a floater in order to regulate the
volume and a degassing system which eliminates the gas contained in the liquid phase
89
of the effluent to be treated. A remote controllable peristaltic pump ensures the

connection between this tank and the process.
flujdjzed Bed RWtor
(120 liters)
. .- .... Recirculation pump

(J500I/h)
.
DisliUm Vinqsses 0----~ Feeding Pump
(0-50 I/h)
Storage Tank (27 m 3)
Liquid Recirculation
(0 -J500I/h)
Fig. 1__ Synthetic synoptic view of the 120 litre anaerobic digestion fluidised bed reactor
The input liquid is mixed into the recycled flow just before entering a heat exchanger.
This heat exchanger is used to regulate the temperature of the liquid phase into the
reactor (nominal set-point is 35°C). The heated liquid is then introduced at the bottom
of the reactor. The gas is collected at the top of both the column and the settler. The
liquid from the top of the reactor is collected by overflow and rejected in the public
sewage system.
Four different local control loops are present within this process:
• the pH of the influent is regulated at 6.3 in the 20 litre buffer tank using a local
PID controller,
• the temperature of the liquid phase is regulated in the recycled loop by a local
PID controller at the set-point 35°C,
• the recirculation flow rate is controlled through a controllable valve using a
third local PID controller,
• finally, the output gas flow rate is regulated around 110 1Jh by adjustment of
the input liquid flow rate. The control input is saturated at 30 1Jh for security
reasons.
The sensors and actuators are connected to an input/output device that allows the
acquisition, the treatment and the storage of data on a PC. For this purpose the "Modular
SPC" software developed by the French company S.E.R.I. Environnement is used. This
software performs advanced control law calculations as well as process supervision.
90
2.2. THE MODEL-BASED FDI APPROACH
The ftrst example is concerned with the closed-loop detection and isolation of an
actuator problem; see Harmand (1997) and Harmand and Steyer (1998) for further
details. This study was in fact initiated since the experiments run for several years
reveal either continuous or abrupt deviations between the expected values and the
measurements of the Input Liquid Flow Rate (ILFR). This problem is very important
since it can lead to wrong actions that can degrade nominal performances of the process
(i.e., under or over feeding of the reactor according to the sign of the disturbance).
Several reasons can explain these observed biases:
• the practical experience has clearly established that, most of the time, slow-
acting deviations are the consequence of a clogging of the pipes used to feed
the reactor with the wastewater.
• the abrupt deviations can be explained, depending on the magnitude of the
bias, either by a leak in the input pipes or by the appearance of a gas bubble
into the sensor.
In fact, the present FDI model-based approach is integrated with a control approach
called "Disturbance Accommodating Control" or DAC (Johnson 1976). Mathematically
speaking, the DAC is the association of the "Disturbance Modelling Principle" (DMP)
with a classical regulator, the DMP assuming that the disturbance signal (here the bias
on the ILFR) is a linear combination of four different basic functions: constants, ramps,
sine or polynoms. When the disturbances are modelled by these mathematical functions,
the model of the process is augmented by the model of the disturbances. Notice
however that the higher the degree of the disturbance model, the higher the degree of
the controller. Then, under appropriate observability hypothesis, the DAC allows us to
efficiently control the process while the use of a state estimator (e.g., a Kalman ftlter)
allows us to on-line estimate the process state together with the expected disturbances
without actually measuring them. If disturbances vary with time, this strategy holds if
the dynamics of the estimator are faster than those of the disturbances.
There are a number of advantages to use this approach:
• this FDI method can be used without the DAC, that is either in open-loop or in
closed-loop conftgurations,
• in case of a closed-loop conftguration, the method makes explicit use of the
structure of the system to estimate and attenuate the negative effects of a
disturbance on the controlled variables.
A number of experiments have been performed in order to test this integrated

FDIIcontrol methodology in closed-loop. In particular, the accommodation of an
actuator constant bias has been tested while controlling the output gas flow rate at a set-
point of 110 l/h using the input liquid flow rate (See ftgure 2). At t = 100 h, an artiftcial
actuator bias of -2 l/h was applied to the ILFR until t = 165 h. To this end, the
calibration parameter of the input pump controlling the ILFR was voluntarily and
91
manually biased. It means that when the output of the controller is u (what is called the
free-disturbance-expected-value ofthe input), then the real input (called the measured or
real value of the input) of the process equals (u-2) instead of u (see figure 2.a). In such a
case, the controller reacts almost immediately in estimating the bias (see the artificially
applied bias and its estimation in figure 3) and corrects the ILFR signal in order to
maintain the output gas flow rate around the functioning point (see the accommodation
results in figure 2.b). Finally, it is to be noticed that the exact same FDIIcontrol
integrated approach could be used for other WWTPs (for an application to a nitrification
process, see Harmand et al. 1996 and 1997).
30 120
~ ~
*'"
25 ----- ~-----+- ----4 ----- !l
e>.: 20
I
'"
e>.: liS
;l
..'"
0 0
Ii: IS Ii: 110
.":J
.",
a"
10
":;
B-
IDS
E0. 0'"
oS 0
115 13S ISS I7S
9S lIS 135 155 175
Fig. 2.a: Free-disturbance-expected-value (black- Fig. 2.b: Regulation of the OGFR at 110 Vh in the
line) and the real value (grey line) of the ILFR (in presence of an actuator bias of - 2 Vh magnitude
Vh)
Fig. 2: Accommodation of the control law and OGFR regulation
4 .0
__ ____ LI __ ____ JI _ _____ I _____ _
2.0 , , ,
~ 0 .0
~ -2.0
:E
....
0
\ii
-4.0 -- ----,------,- -----,------
, I
B -6.0 --- ---+------~---- -- ~ ----- -

«
CJ
I I I
-8.0 ______ 1 __ ____ J ______ 2 _____ _
I , I
- 10.0
I Time h
95 115 13S 155 175
Fig. 3: Actuator bias (bold line) and its estimation (thin line) in /Ih
2.3. THE ANN-BASED FDI APPROACH
The previous method is perfectly adapted when the process is run in closed-loop and
when the identified model is valid. However, if these conditions are not fulfilled, the
DMP principle cannot be applied and other techniques are to be used.
92
In the following, a hybrid approach that uses both fuzzy logic and artificial neural
networks is presented for on-line detection and analysis of problems occurring when the
process is run either in open or closed-loop (for further details, see Steyer et al. 1997).
~ 1.0
(a)
%l 0.8
~
~
0.6
't:l 0.4
"g. 0.2
;::s Time (h)
15 0.00
.s- 30 40 50 60 70
~1.0~
~ 0.8 (b)
I
i:;~;~.)
} 0.0 0 10 20 30 40 50 60 70
1.0
(e)
en" 0.8
fJ 0.6
is 0.4
;>,
~
&:! 0.2 Time (h)
0.0 0 10 20 30 40 50 60 70
1.0
(d)
" 0.8
~ 0.6
is 0.4
~ 0.2
-< Time(h
70
Fig. 4: Comparison offault detection using fu?2Y qualification and artificial neural network
when the organic loading rate is changed (measurements are centred between 0 and 1).
("Fu?2Y Diag_Sin" is the signal indicating the presence of a fault on the input of the process
using the fu?2Y logic approach and "ANN Diag_Sin" has the same meaning but it is
generated from an artificial neural network. When these signals are close to 0, there is no
fault and their increase indicates the presence of a fault)
The raw data available on the process (i.e., pH, temperature, recirculation flow rate,
input flow rate and gas flow rate) and their signal characteristics (i.e., mean value,
variance, etc.) are pre-processed using fuzzy logic to build a vector of features (i.e., a
pattern vector). This feature vector is classified into a pre-specified category (i.e., a
class) that is a state of the system, according to discrimination fuzzy rules. An artificial
neural network is then used to classify the process states and to identify the faulty or
dangerous ones: if there is no fault, the signal provided by the FDI system is equal to
93
Jean-Philippe Steyer and Jerome Hannand
zero. On the other hand, if a dangerous fault is detected and isolated, the signal is set to
100 %. In between these two extremes, the higher the value of the signal, the more
dangerous a fault is.
This approach was developed to handle in real time problems such as, for example,
foam forming, sudden changes in the effluent to be treated (due to a change in
concentration), pipe clogging (due to struvite formation) or bad temperature regulation
(due to improper setting of the control parameters).
In the example presented in figure 4, the influent liquid flow rate (ILFR) was disturbed
on purpose (see Fig. 4.a) without indicating it to the FDI procedure (i.e., the ILFR
measurements provided to the FDI were constant and equal to the ILFR value at t = 0).
Effects of these changes can be seen on the output gas flow rate (Fig. 4.b). Only
analysing the gas flow rate measurements, the fuzzy FDI procedure and the ANN FDI
procedure provided respectively the results in figure 4.c and in figure 4.d. It can be seen
that both approaches are able to detect efficiently the changes of the ILFR with a better
detection though using the ANN FDI procedure.
3. Fuzzy supervision of an industrial equalization process
3.1. THE EQUALIZATION PROCESS
Any industrial WWTP is subject to large magnitude variations in both the influent flow
rate and in the influent pollutant concentration and it is well known that either flow or
concentration shockwaves can have catastrophic consequences on the downstream
biological processes. To deal with this problem, a solution consists in implementing an
equalization system at the primary stage of the WWTP to minimise the influence of the
input disturbances on the downstream biological treatment processes. Such systems
consist in a number of buffer tanks interconnected through pipes, pumps and on/off
valves that can be either manually or automatically controlled.
Equalization systems are thus used:
• to overcome the operational problems caused by flow rate and load variations,
• to improve the performance of the downstream processes,
• to reduce the size and cost of the plant.
The equalization process under interest in the study is represented in figure 5 where
EQOi (i = 1 to 3) represents the three interconnected buffer tanks, Q represents the
liquid flow rates and C represents wastewater concentration measured as Total Organic
Carbon concentration or TOC. The control problem can then be stated as developing
algorithms that automatically operate pumps and eventually on/off valves so that
constant - or nearly constant - flow rates and output concentrations are achieved before
being introduced in the treatment plant itself. In terms of control, this problem is very
challenging since the control system must take into account hard constraints. These
constraints are related to the physical capability of the equalization system to face
increases in the influent flow rate reflected by the total capacity of the interconnected
tanks as well as the pump characteristics.
94
Fig. 5: Configuration of the equalization system
3.2. THE FUZZY SUPERVISOR
Very important problems can arise even though the equalization process is carefully
controlled (see Devisscher et al., 1999, De Clercq et al., 1999 and Harmand et al.,
1999). In fact, the control schemes give good performance but only when the buffer
tanks are neither full nor empty. In these two extreme situations, the control system
misses the necessary degree of freedom to buffer the flow rate and the TOC output
concentration at the same time. In fact, this comes from the inability of the control
algorithms to fulfil the control objectives.
In order to better understand what happens, let us consider the two following
extreme cases. On one hand, if it is assumed that the input flow rate is significantly
higher than its expected mean value for a long period of time, then, without any
adaptive ability of the control system, it is obvious that, after a given time, tanks will
overflow. On the other hand, if it is now assumed that the input flow rate is significantly
smaller than its mean value for a long period of time, then it is quite obvious that the
mean values have to be decreased to avoid emptying the tanks. The "adaptation ability"
of the control system has also to deal with different time constants. In other words, the
"corrections" to be applied to the flow rates computed by the controllers have to deal
with both short and long time disturbances. In order to overcome this difficulty, a fuzzy-
based supervisory system and a long-term filter are added to a classical control strategy.
This supervisor operates both short time and long time corrections on the flow rates
computed by the controllers in order to avoid saturation of the volumes over short and
large periods of time. The supervisor is tested using a data set obtained from the real
process over a two-month period. The simulation results are provided in figure 6. From
these simulations, different statistical characteristics were obtained and summarised in
table 2. The general structure of the overall supervisor system is presented in figure 7.
Notice that these statistical results have only a relative value from a control
engineering perspective. Indeed, they only have a real pertinence when concerned with
a classical regulation problem evaluation. However, this is not exactly the case in this
fuzzy model-based approach since the set-point (due to the corrections computed by the
fuzzy system) is adjusted depending on the input disturbances. As a consequence, the
set-point is time-varying and the variance loses its absolute physical meaning.
95
160r-----------------------------------------,
150
140
:e 130
g
~ 120
~IIO
100
90
80~----~----~----~----~----~~----~----~
o 200 400 600 800 1000 1200 1400
Time (b)
Fig.6 a: Input (thin line) and output (bold line)flow rates with the supervisor
11000
10000
9000
8000
] 7000
.'!!J
=>0 6000
u
.SU 5000
4000
3000
2000
1000
0 200 400 600 800 1000 1200 1400
Time (h)
Fig. 6b: Model-based control: input (thin line) and output (bold line) TOe concentrations
with the supervisor
96
x 10 5
12
11
10
;e
l!!J
7
""
.S
aI
.3
200 400 600 800 1000 1200 1400

Time (h)
Fig. 6c: Input (thin line) and output (bold line) loading with the supervisor
1500 , - - - - - - - - - - - - - - - - - - - - - - - - ,
1000
500
o~--~--~--~--~--~--~~-~
o 200 400 600 800 1000 1200 1400
Time (h)
Fig. 6d: Volumes (VI to Vdrom thinner to thicker lines) with the supervisor
97
The following conclusions can be drawn from these results:

• The control objectives are quite well fulfilled. Indeed, Qout is almost constant
and the output waste concentration is very well attenuated compared to the
input.
• At the same time, using the fuzzy adaptation system, it is possible to manage
the overall time period without any saturation in the volumes while variations
of the both input flow rates and input Toe concentration are quite well
attenuated.
Table 2: Statistical evaluation of the equalization peiformance
Variances
Flow rate Waste Concentration Loading
Influent 209 1.5 10 2.8 10
"Effi~~;;-i"""""""""""""""""""" ··55························································i·ij··lOS···············································6·:6·"1"ii9······················· ..················...
Qin]' Cin ]
External
Qin3' Cin3
Disturbance
output of the system
(output to be controlled
+ measurements)
Process
Q,
Controllers I+----{ +
( ~) (~J
CO",
1&2
([J Vi
Vi
Fig. 7: Configuration of the global control system of the equalization process

(Q are the mean values of the flow rates)
4. Conclusions
Basically, the choice for an FDI approach is determined by the problem to be solved
and by the knowledge that is available on the process to be diagnosed. On one hand, if a
98
mathematical model of the process can be obtained, then it can be expected that a
model-based method will give satisfactory results. On the other hand, as it often
happens in the case of a WWTP, if only qualitative expert knowledge is available,
fuzzy-based methods will rather be used. However, another context is common when
dealing with the WWTP: a large amount of data is available but the structure for a
general mathematical model is very difficult to define. In this case, an approach based
on the ANN usually will give efficient and pertinent results for FDI purposes.
References
Andrews 1.F. (1994) Dynamic control of wastewater treatment plants, Environ. Sci. Techno!. 28(9), 434A-
439A
Aynsley M., Hofland AJ., Morris AJ., Montague G.A., Di Massimo C. (1993) Artificial intelligence and the
supervision of bioprocesses (Real-time knowledge-based systems and neural networks) in A Fiechter ed.,
Bioprocess Design and Control, Springer-Verlag, Advances in Biochemical EngineeringlBiotechnology
48,1-27.
Boudaoud, AN. and Masson, M.H. (1998) An adaptive fuzzy diagnosis system for on-line sensor data
validation. 3m lFAC Workshop on On-line Fault Detection and Supervision in the Chemical Process
Industries, Lyon, France.
Chow E.Y., X.c. Lou, G.C. Verghese and AS. Willsky (1986) Redundancy relations and robust failure
detection, in Basseville M. and A. Beneveniste Eds., Springer Verlag, Berlin, Detection of Abrupt
Changes in Signals and Dynamic Systems, Lecture Notes in Control and Information Sciences 77, 275-
294.
De Clercq B., B. Vanderhaegen, 1. Harrnand, and P. A Vanrolleghem (1999) Evaluation of a rule-based
control strategy for an equalization facility with technical/physical constraints, European Control
Conference ECC'99, 8 pages (on CD-ROM), Kalsruhe, Germany, 31 st August-3'd September 1999.
Delmaire G., J.P Cassar and M. Staroswiecki (1994) Identification and parity space techniques for failure
detection in SISO systems including modelling errors, 33'd IEEE CDC Conf., 2279-2285, Lake Buena
Vista (USA).
Devisscher M., J. Harrnand, 1. P. Steyer, B. Vanderhaegen and P. A Vanrolleghem (1999) A combination of
fuzzy and linear control techniques for the equalization of an industrial wastewater treatment plant, 18m
Benelux Meeting on Systems and Control, Houthalen, Belgium, March 3-5.
Frank P.M. (1993) Advances in observer-based fault diagnosis, TOOLDIAG'93 Conference, 3, 817-836,
Toulouse (France).
Frank, P.M. (1994) Application of fuzzy logic to process supervision and fault diagnosis. lFAC Symp. on
Fault Detection Supervision and Safety for Technical Processes, SAFEPROCESS'94, Espoo, Finland, 2,
531-538,13-16 June 1994.
Giraud, D. and Aubrun, C. (1996) A fuzzy fault diagnosis method applied to a steam circuit. IEEE
International Conference on Fuzzy Systems FUZZIEEE'96, 3, 1944-1950, New Orleans, Louisiana, USA
Genovesi, A, Harmand, J. and Steyer, J-P. (2000) Integrated fault detection and isolation: Application to a
winery's wastewater treatment plant. Applied Intelligence Journal (APIN), 13,207-224.
Harrnand J., R.E. Skelton, J-Ph. Steyer (1996) Disturbance accomodation control of a nitrification process
with structural uncertainties, IEEE-IMACS Conference CESA'96, 70-75, Lille, France, 9-12 July 1996.
Harrnand 1., J-Ph. Steyer, R.E. Skelton (1997) On disturbance accommodating control of biological processes,
Forum for Applied Biotechnology FAB'97, Gent, Belgium, 25-26 September 1997.
Harmand J. (1997) Identification et commande avancee de procedes biologiques de depollution, Ph.D. Thesis
(in French), University of Perpignan, France, September, 18m 1997, 192 pages
Harrnand J., J-Ph. Steyer (1998) The disturbance accommodating control of biological processes, lFAC-
EurAgEng International Workshop on "Decision and Control in Waste Bio-Processing", WASTE-
DECISION'98, 7 pages (on CD-ROM), Narbonne, France, 25-27 February 1998.
Harmand J., J. P. Steyer, M. Devisscher, B. De Clercq, B. Vanderhaegen and P. A Vanrolleghem (1999)
Advanced control of an industrial equalization system, European Control Conference ECC'99, 8 pages
(on CD-ROM), Kalsruhe, Germany, 31 st August-3'd September 1999.
Isermann R. (1984) Process fault detection based on modelling and estimation methods - A survey,
Automatica 20,387-404.
99
Isennann, R. and Balle, P. (1997) Trends in the application of model-based fault detection and diagnosis of
technical processes, Contr. Eng. Pract. 5(5), 709-719.
Jobnson C. D. (1976) Theory of disturbance-accommodating controllers, Control and Dynamic Systems :
Advances in Theory and Applications, 18, Editor C. Leondes, London.
Kiupel, N. and Frank, P.M. (1996) Fuzzy supervision for an anaerobic wastewater plant. IEEE-IMACS Int.
Conf. CESA'96, Lille, France, 1,362-367.
Konstantinov, K.B. and Yosbida, T. (1992) The way to adequate control of microbial processes passes via
real-time knowledge-based supervision. J. Biotecbnol. 24, 33-51.
Montmain, J. and Gentil, S. (1993) Decision-maldng in fault detection : A fuzzy approach. Int. Conf.
TOOLDIAG'93, Toulouse, France.
Patton RJ. and J. Chen. (1991) A review of parity space approaches to fault diagnosis, Proc. IFAC/IMACS
Symp. SAFEPROCESS'91, 1,239-255, Baden-Baden (Gennany).
Roca, E., Flores, J., Rodriguez, I., Cameselle, C., Nunez, M.J. and Lema, J.M. (1996) Knowledge-based
control applied to fixed bed pulsed bioreactor, Bioproc. Eng. 14, 113-118.
Siimes, T., Linko, P., von Numers, C., Nakajima, M. and Endo, I. (1995) Real-time fuzzy-knowledge-based
control of baker's yeast production. Biotecbnol. Bioeng., 45, 135-143.
Steyer, J-P., Queinnec, I. and Simoes, D. (1993) Biotech: a real-time application of artificial intelligence for
fennentation processes. Cont. Eng. Pract. 1(2),315-321.
Steyer, J-P., Queinnec, I., Capit, F. and Pourciel, J-B. (1996) Qualitative rules as a way to handle the
biological state of a fennentation process: an industrial application. Journal Europeen des Systemes
Automatises RAlRO-APII, 30(213), 381-398.
Steyer J.Ph., Rolland D., Bouvier J.C. and Moletta R (1997) Hybrid fuzzy neural network for diagnosis :
Application to the anaerobic treatment of wine distillery wastewater in fluidised bed reactor. Wat. Sci.
Tecbnol. 36, 209-217.
Ulieru M. and Isennann R (1993) Design of a fuzzy-logic based diagnostic model for technical processes.
Fuzzy Sets and Systems 58, 249-271.
White J.E. and J.L. Speyer. (1987) Detection filters design: spectral theory and algorithms, IEEE Transaction
on Automatic Control, 32, 593-603.
Willsky A.S. (1976) A survey of design methods for failure detection in dynamic systems, Automatica 12,
601-611.
100
CALm RATION OF ACTIVATED SLUDGE MODELS: A CRITICAL REVIEW
OF EXPERIMENTAL DESIGNS
B. PETERSEN 1,2, K. GERNAEY 1,

M.HENZE 3, P.A. VANROLLEGHEM 1
1 BIOMATH Department, Ghent University, Coupure Links 653,
B-9000 Gent, Belgium. E-mail: Peter.Vanrolieghem@rug.ac.be
2 EPAS n.v., Technologiepark 2, B-905J Zwijnaarde, Belgium.
3 Department of Environmental Science and Engineering, Technical
University of Denmark, Building J IS, DK-2800 Lyngby, Denmark.
Abstract
This review begins with an overview of literature data on methodologies that have been
applied in other studies to calibrate Activated Sludge Model No. 1 (ASMl). An attempt
was made to gather and summarise the information needed to achieve a successful
model calibration, and based on this a general model calibration procedure is proposed.
The main part of the literature review is devoted to the different methods that have been
developed and applied for the characterisation of wastewater and reaction kinetics in
relation to ASMI. The methodologies are critically discussed and it is attempted to
illustrate the power of the different methods for characterisation, all within the frame of
ASMI calibration. Finally, it is discussed which wastewater components and
parameters are most relevant to be characterised via lab-scale experiments. This
discussion also includes the problem of transferability between lab-scale and full-scale
observations, and potentially different model concepts. One of the most discussed
experimental factors determining the experimental response is the ratio between initial
substrate and biomass concentration (S(O)/X(O». A separate section is focusing upon
this factor.
1. Introduction
One of the most widespread biological wastewater treatment techniques is the activated
sludge process. In this process, a bacterial biomass suspension is responsible for the
removal of pollutants. Depending on the design and the specific application, an
activated sludge wastewater treatment plant can achieve biological nitrogen removal
and biological phosphorus removal, besides removal of organic carbon substances. The
increased knowledge about the mechanisms of different biological processes taking
101
Wastewater Treatment and Modeling, Waste Gas Handling, lOl-186.
Petersen B., Gemaey K., Henze M., Vanrolleghem P.A.
place in an activated sludge plant was translated into dynamic models that were
developed to describe the degradation processes in the activated sludge plant. This
review will focus on the Activated Sludge Model No.1 (ASMI) (Henze et al., 1987),
which through the years has been the state-of-the-art model for activated sludge plants
with biological nitrogen removal.
2. Description of the state-of-the-art activated sludge models
In the following the model concepts of ASMI (Henze et al., 1987) and the recent
modifications leading to ASM3 (Gujer et al., 1999) are described. A description of
ASM2/ASM2d (Henze et al., 1995, 1999) is, however, not included since phosphorus
removal is not dealt with in this review.
2.1. ACTIVATED SLUDGE MODEL No.1 (ASMI)
ASMI is presented in a matrix format in table I according to Henze et al. (1987). Many
of the basic concepts of ASMI were adapted from the activated sludge model defined
by Dold (1980). Some of the central concepts (the different model components and
processes) of ASMI are summarised below. For further details the reader is referred to
the IAWQ Task group reports.
2.1.1. COD components in ASMI

COD is selected as the most suitable parameter for defining the carbon substrates as it
provides a link between electron equivalents in the organic substrate, the biomass and
oxygen utilised. In ASMI the COD is subdivided based on (I) solubility, (2)
biodegradability (3) biodegradation rate and (4) viability (biomass):
• The total COD is divided into soluble (S) and particulate (X) components.
• The COD is further subdivided into non-biodegradable organic matter and
biodegradable matter. The non-biodegradable matter is biologically inert and
passes through an activated sludge system in unchanged form. The inert
soluble organic matter (Su leaves the system at the same concentration as it
enters. Inert suspended organic matter in the wastewater influent (Xu or
produced via decay (Xp) becomes enmeshed in the activated sludge and is
removed from the system via the sludge wastage.
• The biodegradable matter is divided into soluble readily biodegradable (Ss)
and slowly biodegradable (Xs) substrate. Already here it should be stressed
that some slowly biodegradable matter may actually be soluble. The readily
biodegradable substrate is assumed to consist of relatively simple molecules
that may be taken in directly by heterotrophic organisms and used for growth
of new biomass. On the contrary, the slowly biodegradable substrate consists
of relatively complex molecules that require enzymatic breakdown prior to
utilisation.
102
• Finally, heterotrophic biomass (XBH) and autotrophic biomass (XBA) are

generated by growth on the readily biodegradable substrate (Ss) or by growth
on ammonia nitrogen (SNH)' The biomass is lost via the decay process where it
is converted to Xp and Xs (death regeneration, see below).
Summarising, the total COD balance of ASMI is defined by Eq. 1 and further
illustrated in figure 1.
(1)
Fig.l. COD components in ASM1 and ASM3 (figure modified from Jeppsson, 1996),
components specifically related to ASM3 are given in bold and the ones only related to
ASM1 are given in italics
2.1.2. Nitrogen components in ASM1

Similar to the organic matter, total nitrogen can be subdivided based on (1) solubility,
(2) biodegradability and (3) biodegradation rate:
• Total nitrogen can be subdivided into soluble (S) and particulate (X)
components.
• The nitrogen is divided into non-biodegradable matter and biodegradable
matter. The non-biodegradable particulate organic nitrogen (XNI) is associated
with the non-biodegradable particulate COD (XI or Xp), whereas the soluble
non-biodegradable organic nitrogen (SNI) is assumed to be negligible and
therefore not incorporated into the model.
103
Table 1. The ASM1 process matrix (Henze et ai., 1987) (cont' on next page)
Component (i)--7 1 2 3 4 5 6 7 8 9 10 11
.J.. Process G) SI Ss XI Xs X BH XBA Xp So SNO SNH SND
1 Aerobic growth of
heterotrophic
1 l-YH
biomass - 1 -iXB
YH YH
2 Anoxic growth of
heterotrophic
1
biomass - 1 -iXB
YH
3 Aerobic growth of
autotrophic
4.57 - YA 1
biomass 1
YA
- Y1 -iXB-
YA
A
4 Decay of
heterotrophic
biomass 1-fp -1 fp
5 Decay of
autotrophic
biomass 1-fp -1 fp
6 Ammonification 0 -1
soluble organic
nitrogen 1
7 Hydrolysis of
slowly
biodegradable 1 -1
substrate
8 Hydrolysis of 1
organic nitrogen
104
Table 1. The ASM1 process matrix (Henze et al., 1987) (cont' Jromprevious page)
12 13 Process rate (Pi)
XND SALK
iXB Ss So
JimaxH . - - - . ·X BH
14 Ks +Ss KOH +So
2
l-YH iXB 17 · Jimax H ._S_s_. KOH
--...!.!..----'-''''-
SNO X
. BH
g
14·2.86·YH 14 Ks +Ss KOH +So KNO +SNO
__
2_~
14YA 14
iXB -fp .i xp
14
-1
105
• The biodegradable nitrogen is subdivided into ammonia nitrogen (Srm), nitrate

+ nitrite nitrogen (SNO), soluble organic nitrogen (SND) and particulate organic
nitrogen (XND). The particulate organic nitrogen is hydrolysed to soluble
organic nitrogen in parallel with hydrolysis of the slowly biodegradable
organic matter (Xs) (either present in the wastewater or produced via the decay
process). The soluble organic nitrogen is converted to ammonia nitrogen via
ammonification. Ammonia nitrogen serves as the nitrogen source for biomass
growth (the parameter iXB indicates the amount of nitrogen incorporated per
COD unit). Finally, the autotrophic conversion of ammonia results in nitrate
nitrogen (SNO), which is considered to be a single step process in ASMl.
Summarising, the total nitrogen balance for the components in ASMI is defined by Eq.
2 and further illustrated in figure 2.
Fig.2. Nitrogen components in ASMI (modified from Jeppsson, 1996); components

specifically related to ASM3 are given in bold and the ones only related to ASM1 in italics
2.1.3. Processes inASM1

Basically there are four different main processes defined in ASMI (Henze et al., 1987):
• Growth of biomass
• Decay of biomass
• Ammonification of organic nitrogen
106
• Hydrolysis of particulate organic matter
The substrate flows in ASMI are illustrated in figure 3.
ASMl ASM3
Growth Endogeno s
res on
x So So So
HYdrOlYSr s
Xs--SS~XSTO~XH~XI
Hydrolysis Storage Growth Endogeno
Ss~x res' on
o Growth
Fig.3. Substrate flows in ASM1 andASM3 (modifiedfrom Gujeretal., 1999)
2.1.3.1. Aerobic growth of heterotrophic biomass Growth takes place by degradation of

soluble readily biodegradable substrate (Ss) under the consumption of oxygen (So).
Ammonia nitrogen (SNH) is incorporated into cell mass, as described above. Both the
concentrations of Ss and So may be rate limiting for the growth process. The Monod
relationship is used to describe the growth of heterotrophic and autotrophic organisms.
2.1.3.2. Anoxic growth of heterotrophic biomass (denitrification) In the absence of

oxygen the heterotrophic organisms are capable of using nitrate as the tenninal electron
acceptor with Ss as substrate resulting in biomass growth and nitrogen gas. The same
Monod kinetics as used for aerobic growth is applied except that the kinetic rate
expression is multiplied by a correction factor ll g (<1). This factor is accounting for the
fact that the anoxic substrate removal rate is slower compared to aerobic conditions.
This can either be caused by a lower maximum growth rate or because only a fraction of
the heterotrophic biomass is able to denitrify. Furthermore, anoxic growth is inhibited
when oxygen is present which is described by the switching function KOH/(KOIP·SO).
The coefficient KOH has the same value as in the expression for aerobic growth. Thus, as
aerobic growth declines, the capacity for anoxic growth increases.
2.1.3.3. Aerobic growth of autotrophic biomass (nitrification) Ammonia nitrogen (S~

is oxidised to nitrate resulting in production of autotrophic biomass. Furthermore, a part
of the SNH is also incorporated in the autotrophic cell mass. As for heterotrophic growth
the concentrations of SNH and So can be rate limiting for the process. Nitrification has a
considerable effect on the alkalinity (SALiJ.
2.1.3.4. Decay of heterotrophic biomass The death regeneration concept of Dold (1980)
is applied to describe the different reactions that take place when organisms die. The
107
traditional endogenous respiration concept describes how a fraction of the organism

mass disappears to provide energy for maintenance. However, in the death regeneration
concept oxygen is not directly associated with microbial decay. Decay is assumed to
result in the release of slowly biodegradable substrate that is recycled back to soluble
substrate and used for more cell growth. Thus, the oxygen utilisation normally
associated directly with decay is calculated as if it occurs indirectly from growth of new
biomass on released substrate. A parallel conversion of organic nitrogen to ammonia
nitrogen occurs. It should be noted that the magnitude of the decay coefficient used in
this approach is different from that of the endogenous respiration. In endogenous
respiration the loss of one unit of biomass COD leads to the utilisation of one unit of
oxygen minus the COD of the inert particulate products that are formed. However, in
the death regeneration model the loss of one biomass COD unit results in the ultimate
formation of one unit of COD due to the formed readily biodegradable substrate minus
the formed inert particulate products. When the readily biodegradable COD is used for
cell synthesis, only a fraction of a unit of oxygen (determined by the yield) will be
required because of the energy incorporated into the cell mass. That cell mass
undergoes in turn decay etc. before the unit of oxygen is finally removed.
Summarising, to give the same amount of oxygen utilisation per time due to the
decay process, the decay rate coefficient must be larger for the death regeneration
concept than if a more traditional endogenous decay process was adopted. This has the
effect that the cell mass turnover rate increases, resulting in a higher microbial growth
rate in the death regeneration model.
2.1.3.5. Decay of autotrophic biomass The decay of autotrophs is described similar to

the heterotrophic decay process.
2.1.3.6. Ammonification of soluble organic nitrogen (SND) Biodegradable soluble

organic nitrogen (SND) is converted to ammonia nitrogen (SNH) in a frrst order process.
Hydrogen ions consumed in this conversion process result in an alkalinity change.
2.1.3.7. Hydrolysis Slowly biodegradable substrate (Xs) enmeshed in the sludge is

broken down producing readily biodegradable substrate (Ss). The degradation of slowly
biodegradable matter has appeared rather important to realistic modelling of activated
sludge systems because it is primarily responsible for realistic electron acceptor profiles
(Dold, 1980). This process is modelled on the basis of surface reaction kinetics and
occurs only under aerobic and anoxic conditions. The hydrolysis rate is reduced under
anoxic conditions in the same way as anoxic growth, by applying a correction factor l1h
«1). The rate is also first order with respect to the heterotrophic biomass concentration
present but saturates, as the amount of entrapped substrate becomes large in proportion
to the biomass.
2.1.4. Restrictions of ASM1

A number of restrictions concerning ASMI are summarised below (Henze et at., 1987):
• The system must operate at constant temperature.
108
• The pH is constant and near neutrality. It is known that the pH has an influence
on many of the parameters, however only limited knowledge is available to be
able to express these possible influences. Consequently, a constant pH has
been assumed. The inclusion of alkalinity in the model, however, does allow
for detection of pH problems.
• No considerations have been given to changes in the nature of the organic
matter within any given wastewater fractions (e.g. the readily biodegradable
substrate). Therefore, the parameters in the rate expressions have been
assumed to have constant values. This means that only concentration changes
of the wastewater components can be handled whereas changes in the
wastewater character can not.
• The effects of nutrient limitations (e.g. Nand P) on the cell growth have not
been considered. It is, however, easy to add limitation terms in the model if
needed.
• The correction factors for denitrification (llg and llh) are fixed and constant for
a given wastewater, even though it is possible that their values are depending
on the system configuration.
• The parameters for nitrification are assumed to be constant and to incorporate
any inhibitory effects that wastewater constituents may have on them.
• The heterotrophic biomass is homogeneous and does not undergo changes in
species diversity with time. This assumption is inherent to the assumption of
constant kinetic parameters. This means that any changes in substrate
concentration gradients, reactor configuration, etc. on sludge settleability are
not considered.
• The entrapment of particulate organic matter in the biomass is assumed to be
instantaneous.
• The hydrolysis of organic matter and organic nitrogen are coupled and occur
simultaneously with equal rates.
• The type of electron acceptor present does not affect the loss of biomass by
decay.
• The type of electron acceptor does not affect the heterotrophic yield
coefficient.
• ASMI is developed for simulation of treatment of municipal wastewater, and it
is therefore not advised to apply the model to systems where industrial
contributions dominate the characteristics of the wastewater.
• ASMI does not include processes that describe behaviours under anaerobic
conditions. Simulations of systems with large fractions of anaerobic reactor
volume may therefore lead to errors.
• ASMI can not deal with elevated nitrite concentrations.
• ASMI is not designed to deal with activated sludge systems with very high
load or small sludge retention time (SRT) (<1 day).
2.2. ACTIVATED SLUDGE MODEL NO.3 (ASM3)
ASM3 is presented in matrix form in table 2. In the development of ASM3 some

limitations of ASMI were evaluated, and combined with the experiences gained with
109
the application of ASM1 the following list of "defects" of ASM1 was defined (Gujer et
al., 1999):
• ASM1 does not include expressions to deal with nitrogen and alkalinity
limitations.
• ASM1 considers biodegradable soluble and particulate organic nitrogen as
model components. These can, however, not easily be measured and may in
most cases unnecessarily complicate the use of ASM1.
• The ammonification kinetics can not be easily quantified, and moreover this
process is typically rather fast and does therefore not affect model predictions
significantly.
• ASM1 differentiates between inert suspended organic matter present in the
influent wastewater and produced within the activated sludge process. In
reality, however, it is impossible to distinguish between these two components.
• Hydrolysis has a rather dominating effect upon the predictions of the oxygen
consumption and denitrification by heterotrophic organisms. In reality this
process includes different coupled processes such as hydrolysis, lysis and
storage of substrates. Therefore, the identification of the kinetic parameters of
this combined process is difficult.
• The death regeneration concept is covering lysis combined with hydrolysis of
released substrate and subsequently growth on this substrate. In reality it is
difficult to determine the decay coefficient related to the death regeneration
concept.
• Elevated concentrations of readily biodegradable organic substrates can lead to
storage of poly-hydroxy-alkanoates, lipids or glycogen. This process is not
included in ASM1.
• ASM1 does not include the possibility to differentiate between decay rates of
nitrifiers under aerobic and anoxic conditions. This may lead to problems with
the predictions of the maximum nitrification rates in cases of high SRT and
high fractions of anoxic reactor volumes.
The main difference between ASM1 and ASM3 is the recognition of the importance of
storage polymers in the heterotrophic conversions in the activated sludge processes in
ASM3. The aerobic storage process in ASM3 describes the storage of the readily
biodegradable substrate (Ss) into a cell internal component (XSTO). This approach
requires that the biomass is modelled with cell internal structure similar to ASM2. The
energy required for this process is obtained via aerobic respiration. This internal
component is then subsequently used for growth. In ASM3 it is assumed that all Ss is
fIrst taken up and stored prior to growth. Thus, a division of the storage and growth
process, allowing growth to take place on external substrate directly, is not considered.
Furthermore, the death regeneration concept is replaced by endogenous respiration,
which is closer to the phenomena observed in reality. Endogenous respiration can
readily be obtained from a simple batch test (see below, section 4.1.3.1). Also, ASM3
allows a differentiation between aerobic and anoxic decay.
110
Figure 3 illustrates the difference in COD flows between ASM1 and ASM3. The first
thing to notice is that the conversion processes of both groups of organisms (autotrophs
and heterotrophs) are clearly separated in ASM3, whereas the decay regeneration cycles
of the autotrophs and heterotrophs are strongly interrelated in ASMI. This change of
decay concept (and introduction of the storage step) means that there exist more "entry"
points for oxygen utilisation resulting in, at some points, easier separation and
characterisation of the processes. Second, there is a shift of emphasis from hydrolysis to
storage of organic matters. This gives a change in how wastewater characterisation
should be defined since the separation between Ss and Xs now should be based on the
storage process rather than on the growth process. Still, the separation remains
somewhat based on biodegradation rates. In ASM3 hydrolysis is obviously of a less
dominating importance for the rates of oxygen consumption since only hydrolysis of Xs
in the influent is considered.
Below the components and processes of AMS3 are summarised focusing on the
differences between ASM1 and ASM3.
2.2.1. COD components in ASM3

The COD components in ASM3 are basically defined in the same way as in ASMI.
Only the separation between inert suspended organic matter in the wastewater influent
(XI) and produced via the decay process (Xp) is no longer maintained, and, second, the
component XSTO is introduced, as described above. The substrate Ss goes through the
storage process but is basically still biodegradable. Thus, the total COD balance is
defined by Eq. 3 and further illustrated in figure 1, where the components specifically
related to ASM3 are given in bold and the ones only related to ASM1 are given in
italics.
CODtot =SI +Ss +XI +Xs +XH +XA +XSTO (3)
2.2.2. Nitrogen components in ASM3

The nitrogen balance in ASM3 is simplified compared to ASM1, since the soluble and
particulate organic nitrogen components are no longer considered. Furthermore, a
nitrogen gas component (SN2) is included allowing for a closed nitrogen mass balance.
The nitrogen incorporated in SJ, Ss, XJ, Xs, and the biomass is defined in ASM3 as a
fraction of these components. This fraction is consumed or produced when the
corresponding COD fraction is formed or degraded respectively. Summarising, the total
nitrogen balance for the components in ASM3 is defined by Eq. 4, and further
illustrated in figure 2. Again, the components specifically related to ASM3 are shown in
bold and the ones related to ASM1 in italics.
111
Table 2A. Rate expressions and stoichiometry ofASM3 (Gujer et al., 1999)
j Process Process rate equation Pi" all Pi ~ 0
k H' XS/XH . XH
1 Hydrolysis
Kx +XS/XH
~eterotrophic organisms, aerobic and denitrifying activity
2 Aerobic storage of Ss ksro' SOl '~'XH

K02 +S02 Ks +Ss
K02 SNOX '~'XH

3 Anoxic storage of SS ksro '17N0x .
K02 +S02 KNOX + SNOX Ks + Ss
S02 SNH4 SALK XSro/XH

4 Aerobic growth PH' 'XH
K02 + SOl KNH4 +S NH4 KALK + SALK Ksro+XSro/XH
K02 SNOX SNH4

PH . I1Nox .
Anoxic growth K02 + SOl KNOX +SNOX KNH4 +SNH4
5
(denitrification) SALK XSro/XH ,X H
KALK + SALK Ksro +XSro/XH
Aerobic endogenous bH,02 . S02 ,X H

6
respiration K02 +S02
Anoxic endogenous b K02 SNOX ,X H

7 H,NOX . K02 + S02 KNOX +SNOX
respiration
Aerobic respiration bSTO,02 . S02 . XSTO

8
ofXsTO K02 + SOl
Anoxic respiration b K02 SNOX .X

9
ofXsro sro,NOX . K02 + SOl KNOX+SNOX sro
Autotrophic organisms, nitrifying activity
~erobic growth of S02 SNH4 SALK ,X A

10 PA'
peA, Nitrification KA,02 +S02 KA,NH4 + SNH4 KA,ALK + SALK
Aerobic endogenous bA,02' S02

11 ,X A
~piration KA,02 + SOl
Anoxic endogenous b KA,02 SNOX ,X A

12 A,NOX' KA,02 +S02 KA,NOX + SNOX
~piration
112
Table 2B. Rate expressions and stoichiometry of ASM3 (Gujer et a/., 1999)
compound i > 1 2 3 4 5 6 7 8 9 10 11 12 13
j Process S02 SI Ss SNH4 SN2 SNOX SAL!{ XI Xs Xu XSTO XA Xss
v expressed as > O2 COD COD N N N Mole COD COD COD COD COD SS
1 Hydrolysis fSI I i 0.001 -I -0.75
Heterotrophic organisms, aerobic and denitrifying activity
Aerobic storage 1-
2 -1 0.03 0.002 0.85 0.51
ofSs YSTO.2
Anoxic storage
3 -1 0.03 0.07 -0.07 0.007 0.80 0.48
ofS s
4 Aerobic growth -0.60 -0.07 -0.005 1 -1.60 -0.06
Anoxic growth
5 -0.07 0.30 -0.30 0.016 1 -1.85 -0.21
(denitrific. )
Aerobic endog.
6 -0.80 0.066 0.005 0.20 -1 -0.75
respiration
Anoxic endog.
7 0.066 0.28 -0.28 0.025 0.20 -1 -0.75
respiration
Aerobic
8 respiration of -1 -1 -0.60
XsTO
Anoxic
9 respiration of 0.35 -0.35 0.025 -1 -0.60
XSTO
Autotrophic organisms, nitrifying activity
Aerobic growth
10 -18.04 -4.24 4.17 -0.600 1 0.90
ofXA
Aerobic endog.
11 -0.80 0.066 0.005 0.20 -1 -0.75
respiration
Anoxic endog.
12 0.066 0.28 -0.28 0.025 0.20 -1 -0.75
respiration
113
2.2.3. Processes in ASM3

In ASM3 there are also four basic processes, however, slightly different from ASM1
(Gujer et aI., 1999):
• Storage of readily biodegradable substrate

• Growth of biomass
• Decay of biomass
• Hydrolysis of particulate organic matter
2.2.3.1. Aerobic storage of readily biodegradable substrate This process describes the
storage of readily biodegradable substrate (Ss) in the form of XSTO with the
consumption of oxygen. As stated above, it is assumed that all Ss first becomes stored
material before use for cell growth. It is realised that this is not in accordance with
reality. However, no model is currently available to predict the separation of Ss into
direct growth and storage. Gujer et al. (1999) therefore suggested to apply a low storage
yield (YSTO) and a higher growth yield (YH) to approximate direct growth.
2.2.3.2. Anoxic storage of readily biodegradable substrate This process is identical to

the aerobic storage, only is nitrate used as terminal electron acceptor instead of oxygen.
Furthermore, a correction factor (llNO) is applied to indicate that only a fraction of the
heterotrophic biomass may be capable of denitrifying.
2.2.3.3. Aerobic growth of heterotrophs Aerobic heterotrophic growth takes place by

degradation of XSTO with the consumption of oxygen (So). Ammonia nitrogen (SNH) is
incorporated into cell mass, as described above for ASMl.
2.2.3.4. Anoxic growth of heterotrophs (denitrification) Anoxic growth is similar to

aerobic growth but respiration is based on denitrification. Again, a correction factor
(llNO) is applied to account for the observation of reduced anoxic respiration rates
compared to aerobic respiration.
2.2.3.5. Aerobic growth of autotrophs (nitrification) This process is described similarly

to ASMl.
2.2.3.6. Aerobic decay of heterotrophs The energy requirements not associated with
growth but including maintenance, lysis, etc. are described by endogenous respiration in
ASM3 according to a simple first order reaction kinetics.
2.2.3.7. Anoxic decay of heterotrophs ASM3 allows for a description of anoxic decay in
a similar way as the aerobic decay process.
2.2.3.8. Aerobic and anoxic decay of autotrophs The decay of autotrophs is described in
the same way as the heterotrophic decay process.
114
2.2.3.9. Aerobic and anoxic respiration of storage products These processes are
analogous to endogenous respiration and ensure that the storage product XSTO decays
together with the biomass.
2.2.3.10. Hydrolysis Just as in ASMI hydrolysis is responsible for the breakdown of

slowly biodegradable substrate (Xs) to readily biodegradable substrate (Ss). However,
in ASM3 hydrolysis is assumed to be electron donor independent, and as stressed above
the hydrolysis does not play the same dominating role as in ASM1.
2.2.4. Restrictions of ASM3

The number of restrictions listed for ASMI above (see 2.1.4) basically still holds for
ASM3, except for the restriction stating that the type of electron acceptor does not affect
the biomass decay.
3.~odeICalibration
In this review model calibration is understood as the adaptation of a model to fit a

certain set of information obtained from the full-scale WWTP under study. This task is
often rather time-consuming, and typically the time needed for a model calibration is
underestimated. Even though more than a decade has passed since the publication of
ASMl, a fully developed model calibration procedure has not been defined yet. We
have not been able to find a complete model calibration report in literature. There may
be many reasons for this. Important to realise is that the purpose of a model being built
is very much determining on how to approach the calibration, making it difficult to
generalise (Henze et al., 1995). Still, considering the wide application of the activated
sludge models there are surprisingly few references that contain details on the applied
model calibration procedure. Most often it is not specified in detail how the model was
calibrated but the focus is more on the applications, e.g. for process scenarios and
optimisations etc. Thus, to obtain information on model calibration procedures one
often has to collect bits and pieces from various sources to obtain an overview.
Before going on with a discussion on how to approach a model calibration of
ASMl, it is relevant to define how parameter estimation is understood in this review
and what the difference is between parameter estimation and model calibration.
Furthermore, the term identifiability will be defined and the problem of identifiability
with respect to ASM in general will be addressed.
Parameter estimation consists of determining the "optimal" values of the parameters
of a given model with the aid of measured data. Here, the numerical techniques for
estimation will not be discussed, but reference is made to the literature (Robinson, 1985;
Vanrolleghem and Dochain, 1998). Only the basic idea behind parameter estimation is
schematised in figure 4. Initially, the model structures, of which selected parameters
need to be estimated, and the experimental data need to be defined. Moreover, first
guesses of the initial conditions, i.e. concentrations, and parameters, have to be given.
The parameter estimation routine then basically consists of minimising an objective
function, which for example can be defined as the weighted sum of squared errors
between the model output and the data. When the objective function reaches a minimum
with a certain given accuracy the optimal parameter values are obtained.
115
Petersen 8., Gemaey K., Henze M., Vanrolleghem P.A.
Thus, parameter estimation is carried out via specific mathematical search algorithms.
However, due to the high complexity caused by the numerous parameters and the
unidentifiable nature of the ASM models, it will be rather cumbersome to apply
mathematical calibration techniques.
Indeed, a major problem encountered in calibration of ASM is the (lack of)
identifiability of the model parameters. Identifiability is the ability to obtain a unique
combination of parameters describing a system behaviour. A distinction should be made
between theoretical and practical identifiability. Theoretical identifiability is a property
of the model structure, and relates to the question whether it is at all possible to obtain
unique parameter values for a given model structure considering certain selected
outputs, and assuming ideal measurements. Practical identifiability, on the other hand,
includes the quality of the data. Thus, theoretically identifiable parameters may be
practically unidentifiable if the data are too noise corrupted (Holmberg, 1982; Jeppsson,
1996). This subject is dealt with in great detail in Petersen (2000).
Definition of model structure
Calculation of objective function Experimental data
>------.t Best estimate of parameters

and initial concentrations
Fig.4. Illustration ofparameter estimation routine (modified from Wanner et a/., 1992)
Here, it should only be stressed that a typical problem related to the model calibration of
ASM is that more than one combination of influent characteristics and model
parameters can give the same good description of the collected data (Dupont and
Sinkjrer, 1994, Kristensen et aI., 1998). Indeed, this indicates identifiability problems of
either theoretical or practical origin.
The model calibration of ASM is typically based on a step-wise procedure, and by
changing just a few of the many parameters instead of applying an automatic
mathematical optimisation routine. Based on the above statements concerning
116
identifiability problems it is, however, obvious that a calibration procedure where the
model parameters are changed by trial and error until a good description of the
measured data is reached is not advisable (Dupont and Sinkjrer, 1994, Kristensen et al.,
1998). Thus, it becomes important to gather as much information as possible that can
help the framing of realistic parameter combinations. In this review it was attempted to
gather and summarise the type of information needed for successful model calibration.
3.1. INFORMATION SET FOR MODEL CALffiRATION
The set of information that should be collected for successful model calibration was
extracted and combined from different sources (Henze et al., 1987; Henze, 1992;
Lesouef et al., 1992; Pedersen and Sinkjrer, 1992; Siegrist and Tschui, 1992; Stokes et
al., 1993; de la Sota et al., 1994; Dupont and Sinkjrer, 1994; Funamizu and Takakuwa,
1994; Weijers et al., 1996; Xu and Hultman, 1996; Coen et al., 1997; Mino et al., 1997;
Kristensen et al., 1998), and is summarised below:
• Design data: reactor volumes, pump flows and aeration capacities.

• Operational data:
~ Flow rates, as averages or dynamic trajectories, of influent, effluent,
recycle and waste flows.
~ pH, aeration and temperatures.
• Characterisation for the hydraulic model, e.g. the results of tracer tests.
• Characterisation for the settler model: e.g. zone settling velocities at different
mixed liquor suspended solids concentrations.
• Characterisation for the biological model, ASM, of:
~ Wastewater concentrations of full-scale WWTP influent and effluent (as
well as some intermediate streams between the WWTP's unit processes),
as averages or as dynamic trajectories: e.g. SS, COD, TKN, NR.-N, NOr
N, P04-P etc.
~ Sludge composition: e.g. SS, VSS, COD, Nand/or P content.
~ Reaction kinetics: e.g. growth and decay rates.
~ Reaction stoichiometry: e.g. biomass yields
The list does not discuss on how the particular information can be collected in practice,
since this will be discussed more in detail in the sections below.
As mentioned above, the required quality and quantity of information will depend
very much on the purpose of the modelling exercise. In case the model is to be used for
educational purposes (e.g. to increase basic understanding of the processes), for
comparison of design alternatives for non-existing plants or in other situations where
qualitative comparisons are sufficient, the default parameter values defmed by Henze et
al. (1987) can be applied. A reasonably good description can most often be obtained
with this default parameter set for typical municipal cases without significant industrial
influences (Henze et al., 1997). However, if the calibrated model is going to be used for
process performance evaluation and optimisation, it may be necessary to have a more
accurate description of the actual processes under study. Some processes may need a
more adequate description than others depending on the purpose of the model
117
Petersen B., Gernaey K., Henze M., Vanrolleghem P.A.
calibration. This may especially apply for models that are supposed to describe the
processes in an industrial or combined municipal and industrial treatment plant (Coen et
at., 1997, 1998). In such cases the wastewater characterisation, and thereby the
activated sludge, may differ significantly from standard municipal wastewater. In
addition, special attention often has to be paid to the characterisation of nitrification
kinetics (e.g. Dupont and Sinkjrer, 1994), since nitrification typically is the determining
process for the process designs. Also, the availability of readily biodegradable carbon
substances is important for the successful achievement of both denitrification and
biological P removal, and may need to be characterised in more detail (Coen et at.,
1997).
In this review the focus will mainly be on the information needed for the biological
model. Although not considered in detail, it should be stressed that the information
listed in the first 4 points is also very essential, and should not be neglected for a
successful model calibration. Major calibration problems can, for example, be related to
rather simple errors in the recording of operational data, e.g. erroneous data of the waste
sludge measurements might result in an incorrect sludge balance (Melcer, 1999).
Moreover good characterisation of hydraulics and settling can be of great importance
since e.g. poor or erroneous hydraulic modelling may result in hydraulic effects being
lumped into the biological parameters of ASMI.
•
PURPOSE
Decision on information needed

(1-5) and calibration levels (6-10)
10. Dynamic calibration of ASM
Fig.5. Schematic overview of the different general steps in an activated sludge model
calibration procedure
The information needed for the characterisation of the biological model can basically be
gathered from three sources:
118
• Default values from literature (e.g. Henze et al., 1987).

• Full-scale plant data
~ Average or dynamic data from grab or timelflow proportional samples.
~ Conventional mass balances of the full-scale data.
~ On-line data.
~ Measurements in reactors to characterise process dynamics (mainly
relevant for SBR's and alternating systems).
Information obtained from different kinds of lab-scale experiments with wastewater and
activated sludge from the full-scale plant under study.
Again, the intended use of the model will determine which information source to choose
for the characterisation of the different biological processes in the model. In addition,
the purpose will decide to which level the model has to be calibrated, since the quality
of the desired model predictions will depend strongly on the quality of the model
calibration. Figure 5 illustrates the different general steps in a model calibration
exercise. It should be stressed that not all steps may have to be taken, depending on the
purpose. This will be discussed further with examples below, and the procedure has
been concretised for a municipal-industrial case study in Petersen (2000).
3.2. MODEL CALm RATION LEVELS
Steps 1-5 in figure 5 indicate the collection of information. Design (1) and operational
(2) data are in general always needed for a model calibration. E.g. the flow and load
variations are important in the design of measuring campaigns for hydraulic, sludge
settling and biological characterisation of the full-scale WWTP. The hydraulics (3) are
typically characterised via tracer tests at the full-scale installation (De Clercq et al.,
1999). The settling properties (4) can be characterised via on-line or lab-scale settling
tests (Vanderhasselt et al., 1999). Finally, the biology can be characterised via different
information sources (see below).
In figure 5 steps 6-10 illustrate different calibration levels. The calibration of the
hydraulic model via tracer test results, and the settler model calibration via results from
sludge settling tests are indicated in steps 6 and 7 respectively. A first ASM calibration
level is typically a simple steady state model calibration (8).
3.2.1. Steady state model calibration

In this step data obtained from the full-scale WWTP are averaged, thereby assuming
that this average represents a steady state, and a simple model not including hydraulic
detail is calibrated to average effluent and sludge waste data. Typically, the calibrations
of the ASM and the settler are linked together, since the aim is most often to describe
the final effluent quality. Moreover, the recycle from the settler has an influence on the
activated sludge system. Thus, at this stage, there may be an interaction between the
steady state calibration and the settler model calibration, indicated in figure 5 with the
double arrow. Finally, the characterisation of wastewater components may be adjusted
according to the calibration of the full-scale model, indicated with the double arrow
between (8) and (5) in figure 5.
119
Petersen B., Gemaey K, Henze M., Vanrolleghem P.A.
The next step in the calibration procedure is a steady state model calibration that
includes the hydraulic model (9). In general, with a steady state model calibration, only
parameters responsible for long-term behaviour of the WWTP can be determined, i.e.
YH , fp, ~ and XI in the influent (Henze et al., 1999; Nowak et at., 1999). These
parameters are correlated to a certain degree, meaning that a modification of one
parameter value can be compensated by a modification of another parameter value. In
the study of Nowak et al. (1999) on mass balances of full-scale data, it was therefore
chosen to flx YH and fp, leaving XI in the influent and bH to be determined from the
steady state data. In the study of Lesouef et al. (1992), two WWTP models were
calibrated via steady state calibration only, and this calibrated model was applied to
simulate dynamic process scenarios.
However, if one relies entirely on a steady state calibration some problems may be
encountered since the real input variations are usually faster than the slow process
dynamics that were focused upon during the steady state calibration. In other words, the
process does not operate in steady state but one still attempts to flt a steady state
simpliflcation of the model to an unsteady situation. A steady state calibration is,
however, very useful for the determination of initial conditions prior to a dynamic
model calibration and for the initiation of flrst parameter iteration (e.g. Pedersen and
Sinlqrer, 1992; Stokes et al., 1993; Dupont and Sinkjrer, 1994; Xu and Hultman, 1996;
Kristensen et al., 1998).
3.2.2. Dynamic model calibration

If it is the aim to describe and predict more short-term and dynamic situations, a model
calibration to dynamic data will be needed since such data contain more information
than steady state data, especially on fast dynamic behaviour. The important point in
model calibration based on dynamic data is to obtain a more reliable estimation of the
maximum specific growth rates ~ and ~axA (Henze et al., 1999), which are the
most important parameters in predicting dynamic situations.
At WWTP's data are most often collected routinely with a daily or weekly sampling
frequency. This sampling frequency may, however, not be high enough, and for more
accurate modelling it may therefore be required to run special measuring campaigns
(e.g. Pedersen and Sinlqrer 1992; Dupont and Sinkjrer, 1994; de la Sota et al., 1994; Xu
and Hultman, 1996). The sampling frequencies should be chosen in relation to the time
constants of the process and influent variations. One of the important time constants of
the process is the hydraulic retention time (HRT). Ideally, one should choose to sample
about flve times faster than the hydraulic retention time and have a test duration of 3-4
times this key time constant (Ljung, 1987). However, since measurements on full-scale
WWTP's are relatively expensive these recommendations may not always be
completely fulfilled.
Furthermore, data from the full-scale installation alone may be insufficient for a
dynamic model calibration since the reaction kinetics can not be readily obtained from
such data, except for speciflc designs like SBR's and alternating systems (Vanrolleghem
and Coen, 1995). For a dynamic model calibration on a full-scale WWTP the modeller
is therefore typically aiming at combining more information rich results derived from
lab-scale experiments (carried out with sludge and wastewater from the full-scale
120
installation) with data obtained from measuring campaigns on the WWTP under study
(Dupont and Sinkjrer, 1994; Xu and Hultman, 1996; Kristensen et al., 1998).
In table 3 an attempt is made to gather and summarise the available literature
examples on model calibrations where detailed information is given on the model
calibration procedures. The table should not be regarded as a complete list of
possibilities but can serve as a starting point. The purpose of the different model
calibrations is given together with the applied calibration strategy. Furthermore, the
information sources for the characterisation of (1) wastewater, (2) sludge, (3) kinetics
and (4) stoichiometry, are listed. Table 3 does not indicate the kind of experiments that
may have been carried out to gather the information, since this will be discussed in one
of the next sections of this review. The model parameters that are not mentioned in table
3 have either been taken from literature or their origin may not have been clearly
indicated in the references. Considering wastewater characterisation it is not always
specified how the wastewater information was converted into the wastewater
components according to ASMI. In these cases only the type of measurement (e.g.
COD, TKN etc.) is listed in the table.
Based on table 3, it is obvious that the choice of information needed for the model
calibration is governed by the purpose. E.g. in the studies of Pedersen and Sinkjrer
(1992) and Dupont and Sinkjrer (1994) the emphasis was to have a description of the
nitrification and denitrification, and the model calibrations therefore focused on
adjustment of the parameters related to these processes. In contrast, other studies aimed
at a description of both COD and N removal, and as a result more parameters had to be
considered for adjustment in the model calibration (Siegrist and Tschui, 1992; de la
Sota et al., 1994; Xu and Hultman, 1996; Kristensen et al., 1998).
The wastewater characterisation has both been carried out via full-scale data
combined with mass balances and via lab-scale experiments, e.g. for the inert
components SI and XI (Lesouef et al., 1992) and the Ss component (Xu and Hultman,
1996; Kristensen et al., 1998). In one study all wastewater components were determined
via calibration on the full-scale data (de la Sota et al., 1994). The determination of the
stoichiometric and kinetic parameters is often carried out via calibration of the model to
the full-scale data only. However, some studies have also included the effort of
characterising some parameters in lab-scale experiments, e.g. for the determination of
the specific growth rate of the autotrophic biomass (e.g. Lesouef et al., 1992; Dupont
and Sinkjrer, 1994) or to collect further information on the half-saturation coefficients
(Kristensen et al., 1998).
In addition, table 3 indicates that if the purpose of the model calibration was more
than 'just" a description of the processes, more emphasis was put on the
characterisation of the relevant parameters via lab-scale experiments. For example in the
study of Dupont and Sinkjrer (1994) the aim was to apply the model for optimisation of
nitrogen removal.
121
Table 3. Information sources for model calibration of ASMl (cant' on right hand page)
Reference Purpose Calibration Characterisation

strategy Wastewater
Pull-scale Model components
data
Mass Lab- Model
balances scale calibration
ST92 Description: Steady state * 3,7,8,9 SI XI
Nitrification, Dynamic
COD removal
L92 Optimisation: Steady state 3,4,8,9 Ss,Xs, XBH S"XI

N-removal
PS92 Description: Steady state 3,4,5,7,8, Ss,SI,XI

N-removal Dynamic 9
DS94 Optimisation: Steady state 3,4,5,7,8, *

N-removal Dynamic 9
S93 Description: Steady state 1,3,5,6,8 *

Nitrification, Dynamic
COD removal
dS94 Optimisation: Steady state 3,5,7,8,9, all

All processes Dynamic 10
XH96 Description: Steady state 3,4,6,8,9 SloSS Ss, XB... XI

COD removal, Dynamic Xs
Nremoval
K98 Description: Steady state 1,2,3,4,7, Ss

COD removal, Dynamic 8,9
Nremoval
*: procedure not described in detail, but probably carried out.
1. SS: Suspended Solids 6. TN: Total Nitrogen

2. VSS: Volatile Suspended Solids 7. TKN: Kjeldahl Nitrogen
3. CODtot: total COD 8. NIi!-N: Ammonium Nitrogen
4. CODsol: soluble COD 9. NO.-N: Nitrate + Nitrite Nitrogen
5. BOD5 : Biological Oxygen Demand (5days) 10. P04-P: Ortho-phosphate
122
Table 3. Information sources for rrwdel calibration ofASMI (cont' from left hand page)
Reference Purpose Calibration Characterisation

strategy Sludge
Lab-scale Kinetic and Lab-scale
analyses stoichiometric experiments
Model
calibration
ST92 Description: Steady state * YH.!lH. Ks. kJ.,

Nitrification, Dynamic Kx. bH, !lA. KmI
COD removal
L92 Optimisation: Steady state 1.2.3,7 Kx. kJ. !lA.bA

N-removal
PS92 Description: Steady state !lA. Ks, 1\g

N-removal Dynamic
DS94 Optimisation: Steady state * Ks,1\g !lA. bA. KmI.

N-removal Dynamic KoA
S93 Description: Steady state 1,2,3 !lH,Ks.!lA

Nitrification. Dynamic
COD removal
dS94 Optimisation: Steady state 1.2.3.7 !lA. bA, !lH. Ks,

All processes Dynamic kJ.. Kx.~o. 1\g.
1\h,
XH96 Description: Steady state 1 !lA. KoH, KmI.
COD removal, Dynamic 1\g
Nremoval
K98 Description: Steady state 1.2 kJ.. Kx. bH, 1\g, bH,!lH,!lA,KoA
COD removal. Dynamic KoH
Nremoval
References
ST92 Siegrist and Tshui (1992) dS94 de la Sota et al. (1994)
L92 Lesouef et al. (1992) S93 Stokes et al. (1993)
K98 Kristensen et al. (1998) DS94 Dupont and Sinlgrer (1994)
XH96 Xu and Hultman (1996) PS92 Pedersen and Sinlgrer (1992)
123
Finally, table 4 aims at summarising the most relevant parameters to adjust in the steady
state and dynamic model calibration. The parameters related to the hydrolysis process
are not included in table 4. This was done on purpose since it was not clear from the
literature whether the parameters of this process are most influential to short- or long-
term treatment plant behaviour.
Table 4. Most relevant parameters in steady state and dynamic model calibration.
Steady state calibration Dynamic calibration

Predictions Long-term Short-term
Main relevant parameters YH, f p, ~, XI.influent ~~,~muA,~g'~h,~,KNH,Ko~KoA
4. Characterisation of wastewater and sludge kinetics
Different methods may be proposed to structure the wealth of methods that have been
developed and applied for the characterisation of wastewater and reaction kinetics in
relation to ASM1. At this point it is assumed that the reader is familiar with the ASMI
terminology. In this review it has been chosen to focus on the methodologies, i.e. what
can be achieved with different methods, their advantages and disadvantages, rather than
focus on the different wastewater components and processes separately. This choice
was motivated by the fact that some methods typically can yield information on more
than one component or process. In the end of the review it is attempted to illustrate the
power of the different methods for wastewater and sludge kinetics characterisation in
the frame of ASM1. Finally, the relevance of characterising the different components
and processes in the frame of ASMI model calibration is critically evaluated.
4.1. WASTEWATER CHARACTERISATION
Wastewater can be characterised either with physical-chemical methods or with

biological methods. In practice one typically ends up with a combined approach to
obtain an estimate of the concentrations of all components. In the following physical-
chemical and biological methods will fIrst be described separately to obtain an overview
of what can be achieved with the different methods. Finally, an overview of what can be
achieved by combining both approaches is illustrated and discussed. In ASMI the
CODtot of the wastewater is considered to consist of inert soluble organic matter (S,),
readily and slowly biodegradable substrate (Ss and Xs respectively) and inert suspended
organic matter (Xi), whereas biomass in the wastewater is considered to be
insignificant:
(5)
4.1.1. Physical-chemical characterisation
124
A wastewater can be separated into different components in a relatively simple manner

via physical-chemical separation methods. The difference in molecular size can give an
indication on biodegradability because small molecules can be taken up directly over
the cell membranes whereas bigger molecules need to be broken down prior to uptake.
Enzymatic hydrolysis is primarily a surface phenomenon, which means that the
hydrolysis rate is directly related to the surface area. Thus, smaller molecules are readily
degraded whereas degradation of larger material can be kinetically limited.
In early studies the wastewater components were separated physically into four size
depending fractions by successive sedimentation, centrifugation, and filtration. The
fractions were classified as settleable, supracolloidal, colloidal, and soluble (Rickert and
Hunter, 1971), and were analysed for chemical oxygen demand (COD). An important
conclusion from these studies was that the particles smaller than 1.0 J.Ull were
approximated to be the true soluble fraction. Moreover, the particles smaller than 1.0
!lm were observed to be more rapidly degradable than particles larger than 1.0 !lm. In a
more recent study Levine et ai. (1985) studied the size distribution of the organic matter
in wastewater and the relationship to different wastewater treatment processes. In this
study it was concluded that separation over a membrane with a pore size of 0.1 !lm was
valid for a differentiation between the true soluble and particulate organic fractions. The
organic particles smaller than 0.1 J.Ull are typically cell fragments, viruses,
macromolecules and miscellaneous debris. The major groups of macromolecules in
wastewater are polysaccharides, proteins, lipids and nucleic acids. The fraction
measured by the standard test for suspended solids (1.2 !lm) includes protozoa, algae,
bacterial flocks and single cells. However some bacterial cells, cell fragments, viruses
and inorganic particles have a size from 0.1 to 1.2 J.Ull and will thus also pass through
the more typically applied filter size of 0.45 J.Ull for separation between soluble and
particulate matter (Levine et al., 1985). The size of colloidal matter is typically in the
range 0.1-50 !lm whereas material with a size larger than 50 !lm usually settles (Levine
et al., 1985).
The ASM models do not differentiate between filtered, colloidal and settleable
wastewater fractions. It is therefore necessary to convert the fractions resulting from a
physical-chemical characterisation to the ASM components. The possibilities and
limitations of physical-chemical methods to accomplish this task are summarised and
discussed below.
4.1.1.1. Inert soluble organic matter S[Soluble inert organic matter SI is present in the
influent, but, importantly, is also produced during the activated sludge process
(Chudoba, 1985; Orhon et ai., 1989, Boero et al., 1991; Germirli et al., 1991; Sollfrank
et al., 1992). Most of the evidence for the production of soluble organics by
microorganisms is collected from experiments with simple known substrates, e.g.
glucose (Chudoba, 1985; Boero et al., 1991). However, the production has also been
proven to take place with wastewater (Orhon et al., 1989; Germirli et al., 1991;
Sollfrank et al., 1992). The SI production seems to depend on the initial substrate
concentration and on cultivation conditions (Chudoba, 1985). A model has been
proposed relating the SI formation to the hydrolysis of non-viable cellular materials in
the system, thereby linking the SI production to the initial substrate concentration and
the decay of the produced biomass (Orhon et al., 1989). This model was verified in a
125
study with different industrial wastewaters and, although the data were not of very high
quality, some evidence was given that the SI production depends very much on the
wastewater type (Germirli et al., 1991). The hypothesis that the SI production originates
from the decay process was, however, contradicted in a study on municipal wastewater
(Sollfrank et at., 1992) where it was concluded that the SI production was related to the
hydrolysis of slowly biodegradable COD of the incoming wastewater.
Thus, although the origin of the SI production may remain unexplained, it seems
clear that it does take place to various extents depending on different factors as
mentioned above, resulting in a SI concentration in the effluent that may be higher than
the influent. Such SI production is, however, not included in the ASM models, where SI
is considered a conservative component. To deal with this discrepancy between model
concept and reality a simplified approach is typically applied by the definition of a
fictive model influent concentration SI that includes the produced SI together with the
real SI influent concentration (Henze, 1992).
It is not possible to measure SI directly and different approximations are therefore
usually applied. Most often SI is determined by the soluble effluent COD, which has
appeared to be a good estimate for SI in case of a low loaded activated sludge process
(Ekama et at., 1986). On the other hand Siegrist and Tschui (1992) suggested that the
influent SI could be estimated as 90% of the effluent COD. These approximations may
hold in most cases, but a more correct approach would be to consider it as the soluble
effluent COD minus the soluble effluent Biochemical Oxygen Demand (BOD)
multiplied with a BOD/COD conversion factor (Henze, 1992). Furthermore, SI can be
determined as the soluble COD remaining after a long-term BOD test with the influent
(Henze et at., 1987; Lesouef et at., 1992). The latter approach is in fact a combination
of physical-chemical and biological methods. However, in case of significant SI
production during the test the influent SI may be overestimated (Sollfrank et at., 1992),
which may lead to an underestimation of influent Ss eventually. Finally, a procedure
was developed to distinguish between SI of the influent wastewater and SI produced
during degradation (Germirli et al., 1991). However, in order to achieve significant
response glucose was added in these tests assuming that the wastewater under study
resembled glucose, an assumption that may not hold in practice.
Summarising, it will be case depending whether it is needed to characterise the
produced SI or whether the model component can be approximated as described above.
4.1.1.2. Readily biodegradable substrate Ss The soluble COD fraction excluding the
soluble inert organic matter (SI) is mostly considered to be the readily biodegradable
substrate Ss. The correctness of this approach does however evidently depend on the
pore size of the filters used for the separation. As described above the "true" soluble
fraction passes through a 0.1 !lm filtration step according to Levine et al. (1985).
However, in practice larger filter sizes are most often used, which may result in an
overestimation of the soluble readily biodegradable substrate concentration, assuming
that the definition of Levine et at. (1985) holds.
Another study confirmed that the fraction passing a 0.1 !lm filter gave a good
representation of the soluble readily biodegradable substrate (Torrijos et at., 1994). It
was confirmed biologically (via respirometry, see below for a detailed description) that
the studied wastewater did not contain any particulate readily biodegradable matter. In
126
contrast with this, Spanjers and Vamolleghem (1995) found, also via respirometry, that
filtered wastewater (0.45 !lm) had a lower biological response than unfiltered
wastewater, indicating that parts of the readily biodegradable wastewater fraction was
retained on the filter. Similarly, for an industrial wastewater it was found that the filtrate
fraction produced via ultrafiltration (pore size < 0.001 !lm) had a lower biodegradability
(13% of CODtot) than the fraction determined with a respirometric characterisation
method (20% of CODtot) (Bortone et al., 1994). Further it was also found that part of
the soluble COD can be slowly biodegradable (Sollfrank and Gujer, 1991).
Finally, a method based on flocculation with Zn(OHh has been developed to
remove colloidal matter of 0.1-10 !lm that normally passes through 0.45 !lm filter
membranes, and was successfully applied to a phosphorus removal activated sludge
system (Mamais et al.,1993). However, the flocculation has appeared to be rather
sensitive to interference and appears highly depending on the pH value during the
flocculation (Haider, 2000).
4.1.1.3. Inert suspended organic matter XI The test proposed for the determination of S],
as the residual soluble COD remaining after a long-term BOD test, by Lesouef et al.
(1992) can also be applied to determine XI. The XI concentration is then determined as
the residual particulate COD, assuming that XI is not produced during the test. This
assumption may, however, be questionable since XI will be produced due to decay
during the long-term BOD test and corrections for this will have to be considered.
4.1.1.4. Slowly biodegradable substrate Xs As mentioned earlier, a physical

characterisation based on different molecular sizes can be used to distinguish between
readily biodegradable substrate Ss and slowly biodegradable substrate Xs. In one study
it has been proposed that Xs may be determined as the colloidal fraction defined by 0.1
- 50 !lm (Torrijos et aI., 1994). However, this hypothesis could not be supported since
the results indicated that the colloids mainly disappeared according to a physical
removal mechanism without any related biological oxidation. In another study of
contact stabilisation, a multiple filtration procedure was used to isolate and monitor the
variation in concentration of the colloidal fraction between 0.03 - 1.5 !lm (Bunch and
Griffin, 1987). Here it was further confirmed that colloidal matter was removed
physically, probably by adsorption. However, the subsequent increase in soluble organic
matter, and corresponding oxygen uptake resulting from breakdown of colloidal
substrate, were not observed. Thus, based on these two studies it is not clear whether
colloids can be considered equal to Xs. Part of the colloidal substrate may be inert, as
was probably the case in the example of Bunch and Griffin (1987), but this was not
considered in these studies.
In addition, parts of the soluble substrate (Sollfrank and Gujer, 1991) and the
settleable matters may belong to the Xs fraction making it rather problematic to
characterise Xs entirely by a physical-chemical method. Finally, if the components Ss,
SI and XI are known and if it is assumed that the biomass concentration is negligible, Xs
can be determined via a simple mass COD balance.
4.1.1.5. Biomass XBH and XBA It is not possible to distinguish biomass concentrations via
a physical-chemical method.
127
4.1.1.6. Nitrogen components SNH, SND, SNO, X ND The nitrogen components can rather
easily be detected by physical-chemical analysis via a combination of standard analyses
of ammonium, nitrite and nitrate and Kjeldahl nitrogen (TKN) on filtered and non-
filtered samples (Henze et al., 1987).
4.1.2. Summary and discussion o/physical-chemical wastewater characterisation

Based on the descriptions and discussions above it can be concluded that a wastewater
characterisation entirely based on physical-chemical characterisation alone will not be
sufficient to obtain an accurate distribution of the organic substrate over the different
ASMI components (Fig. 6A). However, physical-chemical methods alone may be
adequate for the estimation of the nitrogen components (Fig. 6B). In figure 6 the dashed
line indicates the range of uncertainty with respect to the determination of the organic
components.
physical - chemical biological ASM1 ASM1 physical-chemical
total settleable
inert
I total N Particulate TKN
I~
biomass
COD
slowly !X BH
biodegradable SolubieTKN
!~
colloidal
inert
t
readily Ammonium
soluble biodegradable
A
inert S, 8
Nitrite+Nitrate SNO
Fig. 6. Characterisation of ASM1 wastewater components by physical-chemical methods

(A: modified from STOWA, 1995; B: modified from Henze et aI., 1995).
Summarising, the two main problems with respect to determination of the organic
components entirely by physical-chemical means are:
• The reliability of Ss determination based on soluble COD depends very much

on the applied filter size but, even more, on the kind of wastewater under study
since it is possible that part of the particulate substrate is also readily
biodegradable.
• Defining Xs as being the colloids can induce errors because the colloidal
fraction may also contain inert matter. Moreover, parts of the soluble and
settleable fractions may belong to Xs. Thus, it is not possible to separate the
particulate Xs, XI and XBH components adequately.
128
Table 5 summarises the characterisation of wastewater components via physical-

chemical methods, and the assumptions needed, as described in the literature review
above. According to this table it can be seen that with some assumptions and a
combination of a physical-chemical and a biological method for assessment of XI (long-
term BOD test) (Lesouef et al., 1992), it is possible to determine all COD components
(Ss, SI, Xs and Xi). Knowledge of XI allows a determination of Xs via a mass balance of
particulate COD, assuming that XBH is zero. However, it should be kept in mind that the
determination of XI via a long-term BOD test may not be accurate, as discussed above.
Moreover, the assumption that particular COD is not readily biodegradable may be
incorrect.
4.1.3. Biological characterisation

The ASM models are in general biologically defined models. Thus, it is not surprising
that biological wastewater characterisation methods have found wider application and
acceptance than physical-chemical characterisation tests. In the biological methods the
fractionation of organic matter is based on its rate of degradation (Henze, 1992), which
makes the relation to the ASM concepts more direct. It is obvious that mainly the
biodegradable components and the microbial biomass in the wastewater (Ss, X s, SNH,
SND, XND and XBH) can be characterised directly by these methods, whereas the inert
components SI and XI may be determined by a combination of physical-chemical and
biological tests, as already mentioned above (Lesouef et ai., 1992). Typically, a
biological characterisation is based on measurements of the biomass response during
substrate degradation in either a continuous flow-through system or batch type
experiment. This means that the concentration determination of the biodegradable
components is indirect, since the biomass activity has to be interpreted in terms of a
substrate concentration. In principle the consumption of substrate can be measured
directly by measurements of e.g. COD. However, this is typically not very practical due
to problems of sampling and filtration of sludge samples etc. Instead, the biomass
response can be monitored by recording the utilisation of electron acceptors (such as
oxygen or nitrate), or the production of components during substrate degradation (such
as protons, nitrate or carbon dioxide).
A main part of the review on biological characterisation will deal with respirometry.
Respirometry is defined as the measurement and interpretation of the oxygen uptake
rate of activated sludge (Spanjers et al., 1998). In fact the main goal of a WWTP is to
reduce the biochemical oxygen demand of the wastewater, and ASMI was primarily
developed to yield a good description of the sludge production and consumption
patterns of electron acceptors, as described above. Thus, it is not surprisingly that
respirometry has turned into one of the most popular biological characterisation
methods, since the total respiration rate is affected by the concentration of all
aerobically biodegradable components, to which the majority of wastewater
components usually belong. However, nitrate utilisation rates can also be applied for
characterisation of the denitrification potential of a wastewater. Finally, a titrimetric
technique, especially applicable for determination of the ammonium concentration
available for nitrification, will be reviewed.
129
Table 5. Overview of physical-chemical methods for determination of wastewater

components (Fields with grey background indicate that a physical-chemical method is not
applicable)
Component Additional Assumptions

information.
SI
Ss
XI
E86 Ekarna et al., 1986 L92 Lesouef et al., 1992 ST92 Siegrist and Tschui, 1992
G91 Germirli et al., 1991 L85 Levine et ai., 1985 S92 Sollfrank et al., 1992
H87 Henze et ai., 1987 M93 Mamais et ai., 1993 S96 STOWA,1996
T94 Torrijos et ai., 1994
Before the description and discussion on the application of respirometry, nitrate

utilisation rates and titrimetry for wastewater characterisation, the methodology of each
method is described in more detail. Thus, the readers already familiar with these
methodologies can skip these intermediate sections and directly continue reading about
their applications for wastewater characterisation.
4.1.3.1. Respirometry Historically, the determination of the Biochemical Oxygen

Demand (BOD) during an incubation period of 5 to 7 days (BOD 5 or BOD?) has been
widely applied to quantify the effects of pollutants on the oxygen demand of receiving
waters, and was further applied for the characterisation of wastewater. However, due to
the rather arbitrary choice of 5 or 7 days the test result represents a varying part of the
ultimate BOD of different wastewaters, depending on the wastewater composition. For
a more complete analysis of the ultimate oxygen demand of a wastewater the BOD test
130
can be expanded to 20-30 days, typically 28 days. In the BOD tests the oxygen content
is typically only recorded at the start and end of the test without information on the
evolution of the oxygen consumption over time. This means that the test can not give
any information on the different biodegradable fractions.
The test length of 5-7 days or even longer is not very suitable in the frame of
wastewater treatment plant operation. As a consequence the concept of short-term
biochemical oxygen demand (BODsJ was introduced (Vernimmen et ai., 1967). The
concentration of BODs! can be determined via respirometry. As defined above,
respirometry deals with the measurement and interpretation of the oxygen uptake rate,
ro, of activated sludge. In general, the ro may be considered to consist of two
components (Spanjers, 1993): The exogenous oxygen uptake rate (ro,ex), which is the
immediate oxygen uptake needed to degrade a substrate, and the endogenous oxygen
uptake rate (rO,end)' Different definitions of rO,end appear in literature. The definition
applied by Spanjers (1993) is that the rO,end is the oxygen uptake rate in absence of
readily biodegradable substrate, In the context of ASMI it is assumed that rO,end is
associated with the oxidation of readily biodegradable matter produced by (1)
hydrolysis of the slowly biodegradable matter that results from lysis of decayed biomass
and, (2) the use of substrate for maintenance. The integral of the rO,ex profile is a
measure of BODst (Spanjers et ai., 1998).
Contrary to the BODs method, the BODst test is carried out with the same biomass
as in the activated sludge plant under study and may therefore be a more representative
measure of the effect of the wastewater on the particular activated sludge plant under
study. Several attempts have been made to correlate BODs to BODst (Vernimmen et ai.,
1967; Farkas, 1981; Suschka and Ferreira, 1986; Vandebroek, 1986; Ciaccio, 1992;
Vanrolleghem and Spanjers, 1994). However, the success of such a correlation seems to
depend strongly on the type of wastewater, since the wastewater may contain varying
proportions of readily and slowly biodegradable fractions.
Figure 7 illustrates the conceptual idea of respirometry. The degradation of substrate
S] and S2 (Fig. 7A) results in a total exogenous uptake rate rO,ex (Fig. 7B). Figure 7B
illustrates a rather typical respirogram (i.e. a time course of respiration rates) with an
initial peak in rO,ex caused by oxidation of the most readily biodegradable matter, here
Sj, followed by, in this case, one "shoulder" in the rO,ex profile where component S2
continues to be degraded. Thus, in this example the contribution of S] and S2 to the total
rO,ex, and related total BODsl> can easily be distinguished.
However, it will become clear from the "wheel-work" described in table 6
(Vanrolleghem et ai., 1999) that most of the processes in ASMI eventually act on the
oxygen mass balance and may result in more complicated rO,ex profiles.
According to ASMI the total rO,ex of the activated sludge in contact with wastewater is
given in Eq. 6.
r =(I-Y). XBH',umaxH Ss
a,ex H YH Ks +Ss
(6)
+(4.57-YA ). X BA ·,umaxA. SNH
YA KNH +SNH
131
The concentration of Ss and SNH depend on the influent wastewater and also on the rates
at which Xs, SND and XND are degraded. As an example we will follow the arrows from
XBH to So (Table 6): in the mass balance of the heterotrophic biomass XBH (column, c.,
5) the production of XBH by aerobic growth (row, r., 1) is counteracted by the loss of
XBH by heterotrophic decay (r. 4). In this decay process component XBH (c. 5) is
converted to component Xs (c. 4). This production of Xs is counteracted by the loss of
Xs by hydrolysis (r. 7), leading to production of component Ss (c. 2). Ss is subsequently
used for heterotrophic growth (r. 1) where it is converted to component XBH (c. 5) with
concomitant consumption of oxygen So (c. S), i.e. respiration. A similar reasoning can
be made for the processes involving the nitrogen components (SNH, SND and XND) and
autotrophic (nitrifying) organisms (XB,0.
16
52 A
14 ~
,
11: ;
6
10 20 30 40 50
Time (min)
0.45 '._o>~o>"'O>'."' • • • ~""""."'~'~"~~~'~'~"'~'~"'"o> •.••••• ~ ••••••••••0> • • • • • • • ,
0.4
B
0.35 .
~ 0.3
'i3, 0.25'
E
-= 0.2'
~
ci
0.15 .I----~ _____
0.1 .
0.05
10 20 30 40 50
Time (min)
Fig.7. Conceptual respirogram resulting from degradation of substrate S1 and S2
Figure S shows different examples of respirograms collected' in batch experiments

where synthetic substrate (Fig. SA) or different wastewaters (Fig. SB-D) were added to
endogenous sludge. Note that in figure SC-D only the exogenous oxygen consumption
due to substrate oxidation is given, ro,ex> whereas the total ro is given in figure SA-B. It
now becomes clear that the respirograms can differ significantly in shape depending on
the substrate added and may not be as straightforward to interpret as the conceptual
example given in figure 7. Thus, the challenge is to interpret and perhaps divide the
respirogram according to the contribution of rO,ex by different wastewater components.
132
There are two approaches for the determination of model parameters and components:
direct methods focus on specific parameters and components which can directly be
evaluated from the measured respiration rates (Ekama et at., 1986; Spanjers et aI.,
1999), whereas optimisation methods use a (more or less simplified) model that is fitted
to the measured data (Kappeler and Gujer, 1992; Larrea et at., 1992; Wanner et aI.,
1992; Spanjers and Vanrolleghem, 1995; Brouwer et at., 1998; eoen et at., 1998). In
the latter, numerical techniques are used to estimate parameter values that lead to the
smallest deviation between model predicted and measured respiration rates (see figure
4).
Below, examples of respirometric experiments to assess the different wastewater
components will be reviewed and important experimental factors with respect to
wastewater characterisation will be discussed. The overview does not attempt to review
and evaluate different respirometric principles, since a review of these is included in
Spanjers et al. (1998) and Petersen (2000). Different methods may only be included
here to illustrate points that are specifically related to wastewater characterisation.
Readily biodegradable substrate S,

The readily biodegradable substrate is presumably composed of simple and low
molecular soluble compounds, such as volatile fatty acids, alcohols, etc. (Henze, 1992).
The characteristic of these compounds is that they are degraded rapidly and hence result
in a fast respirometric response, e.g. figure 8A.
0.7 40
A
I"'~·""""'\. B
.
0.6
_ 0.5 aD ,
.5 :E
E
~ 0.4
'"
E.'"
20
E. 0.3 ' "
'" 0.2'·
.!?
10 ...
0.1 ~
0
0 2 a 4 5 II
10 15 20 25 30 35
Tim., (h)
Time (min)
Fig.B. A: Typical acetate profile B: Municipal wastewater (after Kappeler and Gujer. 1992)
133
Table 6. Kinetic and stoichiometric relationships for COD removal, nitrification and
denitrification (Vanrolleghem et al., 1999) (cont' on next page)
j Process SI Ss XI Xs XBH XBA Xp

Aerobic hetero-
trophic growth YH
2 Anoxic hetero-
trophic growth YH
3 Aerobic auto-
trophic growth
4 Het. decay fp
-1
5 Aut. Decay I-fp fp

-1
6 Ammonification
7 Hydrolysis
8 Hydrolysis of N
Observed conversion
rates ML·3yI Ii = LIij = LVijPj
Stoichiometric Nomenclature, see text

parameters (see text) All units in ML·3 (COD or N, depending on variable)
134
Table 6. Kinetic and stoichiometric relationships for COD removal, nitrification and
denitrification (Vanrolleghem et aI., 1999) (cont' from previous page)
9 10 11 12 13 Process rate Pi
ML-3yl
SNH XND SALK
_iXB
-iXB
14
l-YH iXB
2.86·14·YH 14
1 . _ _2_ _ iXB
--I
YA 14· YA 14
iXB-fp·i xp
1 ~~-1
14
~I- -1
Nomenclature, see text Kinetic parameters

All units in ML·3 (COD or N, depending on variable) (see text)
135
1.00 , - - - - - - - - - - - - - - , 1.4
c o
o.so 1.2 :~..
c c 1 ~
'"'
.~ 'f!
0.60 ('"-.... =a 0.8
~
li 0.40
~
! "
.E.
j"
06
'?' ' ..
• 0.4
0.20
\ 0.2
0.00 L-_...-L_ _--L.::::::.~_ _--J

o 30 80 90 120 10 20 30 40 50
Time (min)
Tlms(mln)
Fig.B. C: Municipal wastewater (Spanjers and Vanrolleghem, 1995), D: Industrial

wastewater (Coen eta/., 199B)
The most typical batch test for determination of Ss involves the addition of a wastewater
sample to endogenous sludge, and the monitoring of the respiration rate until it returns
back to the endogenous level (Ekama et aI., 1986 among others). The examples shown
in figure 8 are all obtained with such an approach. The respirometric methods may vary
from a very simple lab-scale batch test to more complex methods that may even be
applied on-line. The concentration of readily biodegradable substrate initially present in
the mixture of biomass and wastewater in the experiment is generally calculated
according to Eq. 7.
Ss (0) = _1_. ( fro,ex dt

1-YH
tfin 1 (7)
o
The concentration of Ss in the wastewater is then easily calculated by taking the dilution
into account. The end point tfin of the integration interval is the time instant where Ss is
completely oxidised and where the exogenous respiration rate for Ss becomes zero. The
integral can directly and easily be obtained by determining the area under the rO.ex
profile, e.g. by using a spreadsheet program. An alternative consists of solving the mass
balance equations with a numerical integrator to predict the exogenous respiration rates
for Ss and a given initial value Ss(O). It may be a bit overdone to apply numerical
integration for the profile illustrated in figure 8A, however for more complex profiles
(Fig. 8B-D), the approach may become necessary and more straightforward than direct
calculation, as will be discussed further below.
Notice that knowledge of the heterotrophic yield coefficient YH is needed for the
calculation of Ss from respiration rates (Eq. 7). The yield indicates the COD fraction
that is converted to cell mass. The rest of the COD is used to provide the energy that is
required to drive different synthesis reactions. This energy is made available by
oxidative phosphorylation, which requires a terminal electron acceptor, in this case
136
oxygen. The produced energy is proportional to the mass of electron acceptor utilised,
which in turn is proportional to the COD consumed. As a consequence (1-Y0·COD is
equal to the integral under the rO.ex curve. Evidently, the parameter YH is always
involved when oxygen consumption is converted to substrate equivalents.
The batch test described above is also used to assess other ASMI components and,
likewise, kinetic and stoichiometric parameters. This will be explained further in the
next section on characterisation of sludge kinetics, but this indicates already the
popularity of this test in assessing wastewater components and reaction kinetics.
Apart from the typical batch test as described above, other experimental designs
have also been tried out for the determination of Ss. One example consists of
monitoring the respiration rate of unsettled sewage without inoculum for a relatively
long period, approximately 20 hours (Wentzel et aI., 1995). A respirogram similar to the
one depicted in figure 9 is obtained. The Ss concentration is calculated from the
respiration rates observed between the start of the test up to the time with the
precipitous drop (due to depletion of Ss), with correction for the increasing endogenous
respiration due to the increase of biomass during the test. In addition to YH, knowledge
of the maximum specific growth rate is required, information that can be obtained from
the same test (see Figurel0).
An often-referred continuous flow-through method was developed by Ekama et al.
(1986), see figure 10. This method involves the monitoring of respiration rate in a
completely mixed reactor operated under a daily cyclic square-wave feeding pattern.
The experiment is designed in such a way that the supply of Ss from hydrolysis of Xs
remains constant for a period after the feed is stopped and gives rise to a second ro
plateau. It is hypothesised that the difference in ro plateau values corresponds uniquely
to the Ss that has entered via the influent. Hence, the concentration of readily
biodegradable substrate in the wastewater can be calculated as given in Eq. 8.
(8)
An obvious disadvantage of this method is the length of the experiment (24 h, which is
not including the stabilisation of the continuous reactor used for the test), and the fact
that sufficient Xs is needed in the feed to achieve a constant hydrolysis rate and to
create as such the step change in roo In addition, the method is rather difficult to carry
out in practice (Sollfrank and Gujer, 1991; Wentzel et al., 1995).
A final method for the evaluation of Ss was based on the evolution of the respiration
rates obtained in a continuously fed respirometer during transients between two modes
of operation; a mode of endogenous respiration and wastewater addition respectively
(Spanjers et aI., 1994). In the work of Lukasse et al. (1997) the estimation technique
developed for the determination of Ss in the respirometer of Spanjers et al. (1994) was
further evaluated and improved. In the work of Witteborg et al. (1996) the same
continuously fed respirometer was used but a different estimation of Ss was proposed as
now the measurement of respiration rate was performed under three different
wastewater loading conditions. The wastewater Ss was calculated by numerically
solving a set of mass balances pertaining to different loading conditions of the
respirometer.
137
~~----------------------~ &0
.... .
... .
.. ....
~<lI9"'14O·;a_
..........
lII.A.iIIgo<Og.
80 No
"
...
.'
... ....
... . .. .. ..'
...". ~VlIZI'•
10 F.. Off
Off Fat dOn
o
•
o~~~~~~~~~~~~~
o 284 e o 8 12 18 24
Time (h) Time (h)
Fig.9. Respiration rates measured in Fig.ZO. Respiration rates obtained

a batch experiment for estimation of with the experimental set-up of
/lmaxH and Ks (after Kappeler and Ekama et at. (1986)
Gujer,1992.
Slowly biodegradable substrate Xs

It is assumed that slowly biodegradable substrate Xs is composed of (high-molecular)
compounds ranging from soluble to colloidal and particulate (Henze, 1992). The
common feature of these compounds is that they cannot pass the cell membrane as such,
but have to undergo hydrolysis to low-molecular compounds (Ss), which are
subsequently assimilated and oxidised. The respirometric response on Xs is slower
because the hydrolysis rate is lower than the oxidation rate of Ss.
In a batch test an exponentially decreasing "tail" can frequently be observed in
respirograms (Fig. SB-C). In figure SB, this tailing starts after approximately 0.75
hours. The wastewater concentration of Xs can be assessed in a similar way as above,
Eq. 7 (Sollfrank: and Gujer, 1991; Kappeler and Gujer, 1992). Simultaneously occurring
oxidation processes such as nitrification might interfere and complicate the separation
of the respiration rate due to hydrolysis in the total respiration rate. In that case a
nitrification inhibitor may be used to facilitate the assessment of Xs (Spanjers and
Vanrolleghem, 1995). Alternatively, if the data of such respirometric batch tests are
used in combination with mathematical curve fitting techniques to match the response
of the model to the data, the nitrification part can rather easily be extracted from the
respirogram (Spanjers and Vanrolleghem, 1995).
It has also been proposed to estimate Xs based on a long-term BOD test where Xs is
obtained by subtracting Ss from BOD/(1-YiV (STOWA, 1996). Note that the value of
YH here should be lower than the one applied in Eq. 7, due to internal turnover of
substrate from decayed biomass in long-term tests.
Heterotrophic biomass XBH

In the ASMI report the influent concentration of heterotrophic biomass, XBH, is
assumed to be negligible, as mentioned earlier. However, some wastewaters can contain
a significant concentration of heterotrophic biomass (Henze, 1992), and there may
therefore be a need to quantify this component. A batch test has been proposed where
XBH is assessed from the respirometric response of raw wastewater without inoculum
(Kappeler and Gujer, 1992; Wentzel et at., 1995). The calculation requires knowledge
138
of YH together with two parameters (llmaxH and b0 that can be obtained from the same
data. Respirograms look similar to the one presented in figure 9. The procedure
basically backtracks the amount of heterotrophic biomass originally present in the
wastewater by comparing the original respiration rate with the respiration rate after
significant (hence, well quantifiable) growth ofXBH•
Autotrophic biomass XBA

So far, no procedures were found by which the autotrophic biomass concentration in
wastewater is determined. However, it could be imagined that a similar procedure as the
one developed for XBH is applicable. Thus, by evaluation of the respiration rate for
nitrification, r/i,ex' of the autotrophs present in the wastewater and by comparison to
the respiration rate of a culture with known autotrophic biomass concentration X BA, e.g.
after significant growth, the originally present XBA could be determined.
Ammonium SNH
The concentration of ammonium in wastewater can be determined by using
conventional analytical techniques, as mentioned earlier. However, respirometry also
offers the possibility to deduce SNH from batch measurements in a similar way as Ss and
Xs, provided the test is done with nitrifying activated sludge and the oxygen
consumption for nitrification can be separated from the other oxygen consuming
processes. As follows from table 6, the autotrophic yield coefficient YA is needed to
convert the oxygen consumption for nitrification to a nitrogen concentration by division
by (4.57-Y~, where 4.57 indicates the amount of oxygen needed to oxidise one unit of
ammonium nitrogen. The value of YA is typically 0.24 g COD(biomass)/g N, which
means that the determination of SNH is not very sensitivity to YA since its value is small
compared to 4.57.
Notice that part of the available ammonium may be assimilated into new
heterotrophic biomass, which may be a considerable fraction of the nitrogen in case a
large amount of COD is biodegraded (CODDegraded) simultaneously with the nitrification.
The actual nitrified ammonium nitrogen, denoted NNitr, can be approximated by Eq. 9 in
which iXB is the nitrogen content of newly formed biomass:
NNitr =SNH -i XB .YH ·CODDegraded (9)
From this equation one can easily deduce the original nitrogen concentration when
CODDegraded, and the stoichiometric parameters iXB and YH are given. Note, however that
fitting a model in which carbon and nitrogen oxidation are included to the respirometric
data will automatically take this correction into account (Vanrolleghem and Verstraete,
1993; Spanjers and Vanrolleghem, 1995; Brouwer et al., 1998).
Organic nitrogen SND and slowly biodegradable organic nitrogen X ND

Probably because the ammonification and hydrolysis rates of organic nitrogen
compounds are relatively fast, little attention has been devoted so far to the
establishment of respirometric techniques for SND and X ND quantification. In batch tests
these compounds are typically converted to SNH before the SNH that was originally
139
present in the wastewater is removed by nitrification. Therefore, SND and X ND are not
directly observable in such tests but may be lumped into the fraction of nitrified
ammonium. Still, for some industrial wastewaters the ammonification and hydrolysis
steps may be considerably slower and quantification of these component concentrations
may be required. In such cases, one can imagine a procedure in which the nitrification
respiration rate rC.ex is monitored and interpreted in terms of ammonification and
hydrolysis, similar to the way the respiration resulting from COD degradation is
interpreted in terms of the biodegradation of readily biodegradable substrate and the
hydrolysis process. Subsequently, the amounts of nitrogen containing substrates could
be assessed by taking the integral of rli,ex for the corresponding fractions and dividing
these by (4.57-Y,0. In case simultaneous COD-removal is taking place, correction
should again be made for nitrogen assimilated into new heterotrophic biomass (see
above).
4.1.3.2. Nitrate utilisation rates
Readily or slowly biodegradable substrate SoS. and Xs

The basis for wastewater characterisation via monitoring of nitrate utilisation rates
(rN03) for the determination of the denitrification potential is rather similar to that of
respirometry (Nichols et al., 1985; Ekama et al., 1986; Kristensen et al., 1992; Naidoo
et al., 1998; Sperandio, 1998; Urbain et al., 1998; Kujawa and Klapwijk, 1999). The
application of nitrate utilisation rates for wastewater characterisation within the frame of
ASMI is however not as widespread as respirometry.
The readily biodegradable component Ss (or Xs) is determined by Eq. 10 (similar to
Eq. 7). A typical rN03 profile is given in figure 11, indicating two biodegradable
wastewater fractions.
_ . [tfin!rN03,ex dt
Ss (0) =-2.86
l-YH
1 (10)
o
The factor 2.86 g 02/g N03-N originates from the fact that the theoretical electron
acceptor capacity of nitrate (as N) is 2.86 times that of oxygen (as 0), assuming that
N03-N is converted completely to nitrogen gas N2 (Payne, 1981; van Haandel et al.,
1981). The factor has been verified experimentally by Copp and Dold (1998).
In Eq. 10 it is assumed that the YH of aerobic and anoxic substrate degradation is
equal, as also assumed in ASMI. In a study on a pure denitrifying culture it has
however been reported since long that aerobic yields are larger than anoxic yields
(Koike and Hattori, 1975). It has been theoretically proven, based on the energetics of
the metabolic processes, that anoxic yields indeed are consistently lower than aerobic
ones (Orhon et al., 1996). Indeed similar differences between aerobic and anoxic yield
were obtained experimentally with activated sludge (McClintock et al., 1998; Sperandio
et al., 1999). Thus, to apply nitrate utilisation rates for wastewater characterisation it is
important to correct for this difference in aerobic and anoxic yield since application of
140
aerobic yield values in Eq. 10 will lead to overestimation of the readily biodegradable
wastewater components.
4.1.3.3. Titrimetry The buffer capacity of water samples can be measured accurately by
advanced titration techniques (Van Vooren et al., 1995), and has recently been
successfully applied for the determination of ammonium and phosphorus in low
concentrations (0 - 100 mgll) in effluents, surface waters and manure (Van Vooren,
2000).
Some efforts have been done to characterise VFA concentrations related to
anaerobic processes based on titration procedures and pH measurements (e.g. Munch
and Greenfield, 1998). These techniques may also be applicable for wastewater
characterisation in the frame of ASM2 where one component is defined as the
concentration of fermentation products. This will however not be dealt with any further
in this presentation.
~ ,--------------------------,
2IJ
oL------+--------~...:::,..,,.._._-----'
o 3 6
lime (hours)
Fig.II. Typical profile of rN03 as function of time for determination of Ss and Xs (Urbain et
al., 1998)
Alternative to the classical titration methods (up and down titrations) Ramadori et al.
(1980) proposed to monitor the acid and/or base consumption rate that was needed to
keep the pH constant in an activated sludge sample where pH-affecting biological
reactions occur. This titrimetric method has been successfully applied for the
monitoring of nitrification, which has a clearly defined effect on the pH, and
concentrations of SNH (Massone et aI., 1995; Gernaey et al., 1997). Recently, it has also
been attempted to apply the method for the determination of the total nitrifiable nitrogen
concentration of a wastewater (Yuan et al., 1999).
Ammonium. SNH
A typical cumulative base addition curve and a pH profIle collected during a titration
experiment with nitrifying sludge sampled on-line from a pilot plant are shown in figure
12 (Gernaey et al., 1998). In a first phase, the pH of the sludge sample is increased to
the pH setpoint, and base is added at a maximum rate. This phase took about 2 minutes
for the example of figure 12 For the experiments described here, a pH setpoint ± apH
interval value of 8.2 ± 0.03 was used. Every time the pH of the sludge sample becomes
lower than 8.17 (= pH setpoint minus apH interval), base is added to the sludge. Dosage
141
Petersen B., Gemaey K, Henze M., Vanrolleghem P.A.
of base is repeated until the pH has returned within the pH setpoint ± ilpH interval
range. Here, the nitrification phase is finished after about 25 minutes.
0.6 8.3
0.5
8.1
1
~
0 .4
7.9
"-8 0.3
"...
J:
CI.
Nitrification endpoint
7.7
OJ
~ 0.2
7.5
0.1
0.0 7.3
0 10 20 30 40 50
Time (min)
Fig.l2. Typical cumulative base addition curve (expressed as amount of base dosed per liter
of activated sludge sample) and pH profile obtained during an on-line titration experiment
with a mixed liquor sample. For this example, the nitrification phase is finished after about
25 minutes (Gernaey et aI., 1998).
The analysis of the data can either be via a simple manual interpretation or model-based
(Gernaey et aI., 1998). The simple procedure is based on the detection of the two slopes
(SI and S2) in the cumulative base addition curve, followed by an extrapolation of the
different lines to the Y-axis (Fig. 13). The SNH concentration (mg Nil) and the
nitrification rate rN (mg NIl.min) can be calculated according to Eq. 11 and 12, where
the intercepts Bl and B2 are expressed in meqll units. The factor 0.143 meq/mg N (i.e.,
2 mole W per mole N), is the stoichiometric coefficient relating the amount of acid
(meq) produced per mg of nitrogen nitrified. The slopes S 1 and S2 are expressed in
meqll.min units.
S _ (B2-Bl)
(11)
NH - 0.143
(SI-S2) (12)
r --'---~
N - 0.143
142
0.6
0.5 82
0.4
..
~
....
SO.3
III
0.2
0.1
0.0
0 5 10 15 20 25 30 35 40 45 50
Time (min)
Fig.13. Simple manual interpretation of a typical cumulative base addition curve (Gemaey
et al., 1998).
In the application of Gernaey et al. (1998) the sludge was sampled at the last
compartment of an activated sludge pilot plant thereby reducing the likelihood of
presence of organic substrates. In case ammonification is slower than nitrification it
may be relevant to determine SND, as described above in the section on respirometry.
Thus, the titrimetric method may also be applicable for SND determination. It may be
foreseen, however, that degradation of organic substrates may cause acid or base
consumption effects that may interfere with the determination of SNH according to the
described methodology.
Readily biodegradable substrate Ss.

The titrimetric methodology has also been applied for the determination of readily
biodegradable COD available for denitrification, and within control strategies for
additional carbon dosage (Bogaert et al., 1997). A complicating factor is that depending
on the carbon source denitrification will either produce or consume acid (Bogaert et al.,
1997). Preliminary results (Dhaene, 1996; Rozzi et al., 1997) have indicated that the
method may be used to evaluate Ss in concentrated wastewaters.
4.1.4. Summary and discussion o/biological wastewater characterisation

The capabilities of the different biological methods presented above to directly
determine the ASM1 wastewater components are illustrated in figure 14 (the dashed
lines indicate areas of uncertainties) and summarised in table 7. According to figure 14
it is obvious that the readily biodegradable organic wastewater components, i.e. Ss and
parts of Xs (Fig. 14A), and the nitrogen components SNH and parts of SND and XND (Fig.
14B), can be determined directly via the biological methods. The determination of the
slower biodegradable component Xs can be carried out indirectly via a long-term BOD
test and knowledge of Ss (STOWA, 1996). However, uncertainties may be introduced
143
by long-term BOD tests since significant interference from product formation may
occur during the lengthy test.
Table 7. Overview of biological methods to estimate wastewater component concentrations.
(Fields with grey background indicate that a respirometric method is either not applicable
or not relevant. For an explanation of the references, see table 8)
Component Method Type of Assumptions References

AdditiolUll
experiment
For the determination of SNH it should be remembered that it is in fact the nitrifiable
nitrogen that is determined via the biological methods (as indicated with dashed lines
into the regions of organic nitrogen, since parts of the organic nitrogen may be
hydrolysed making it readily available for nitrification). This is in contrast to the
physical-chemical method where the SNH component is determined via a chemical
analysis of ammonia.
4.1.5. Discussion on physical-chemical vs. biological wastewater characterisation

By definition the total COD in ASMI is sub-divided based on (1) solubility, (2)
biodegradability, (3) biodegradation rate and (4) viability (biomass), as described
earlier. Summarising, the COD components to consider in a wastewater are:
CODtot =SI +Ss +XI +Xs +(X BH ) (13)
In previous sections it has been thoroughly reviewed how to determine these

components by either physical-chemical or biological methods, and different limitations
144
of the methodologies have been underlined and discussed. Furthermore, it is obvious

that the division of the wastewater into model components is to some extent artificial.
For example, a division is made between soluble and readily biodegradable substrate
(Ss) and particulate slowly biodegradable matter (Xs), although it is, for example,
known that some slowly biodegradable substrate may be soluble etc.
Total COD ASM1 Respirometry Nitrate Total N ASM1 Respirometry Titrimetry

Utilisation
Rate
Xs
A
Ss
SI
11 B
Fig. 14. Characterisation of ASM1 wastewater components by different biological methods

(the dashed lines indicate areas of uncertainties). A: COD components; B: Nitrogen
components
It became clear that an application of physical-chemical methods alone is not sufficient

for characterisation of the wastewater into model COD components. These methods
basically only allow to distinguish between soluble and particulate COD and do not
differentiate with respect to biodegradability (non-biodegradable versus biodegradable
matters) and biodegradation rate (readily versus slowly biodegradable substrates).
However, by application of biological characterisation methods it is possible to obtain
knowledge of the biodegradability and biodegradation rate of the wastewater.
Thus, it is obvious that a combination of physical-chemical and biological
characterisation methods is advantageous for the translation of the wastewater
characteristics into the ASM 1 model components. A suggestion for such a combined
approach, based on the literature review above, is presented in figure 15. Here it is
suggested to determine the readily biodegradable substrate (Ss) directly via respiration
tests (respirometry or nitrate utilisation rates). The presence of biomass in the
wastewater may also be determined by respiration tests. The slowly biodegradable
matter (Xs) can be determined via the results of a long-term BOD test. The same kind of
test may provide information on the soluble and particulate inert (SI and XI) matters.
Here, however, the reservation should be repeated that long-term BOD tests may not be
145
very accurate due to possible product formation (Sv and decay, which results in XI.
Therefore, the determination of XI via a long-term BOD test may be questionable.
Indeed, it is proposed by Henze et al. (1987) to determine the influent XI via the
complete model during the calibration of the sludge balance. Subsequently, Xs may be
determined via a COD mass balance as the difference between total COD and the other
components. If it is chosen to determine SI by a long-term BOD test, it may be
advisable to combine it with analyses of the effluent, as proposed in the section about
physical-chemical methods. It is again clear from figure 15 that the borderline
especially between particulate and soluble COD, the differentiation between model
components (Ss and Xs) and the results from short-term respiration and long-term BOD
tests may not be completely consistent.
The nitrogen ASMI components are somewhat easier to determine since they can
basically all be determined via mass balances based on standard chemical analyses of
total nitrogen, Kjeldahl nitrogen, ammonium nitrogen and nitrate nitrogen (see figure
6). It can, however, be advantageous to combine these chemical analyses with
biological methods (respirometry or titrimetry) to obtain the nitrifiable nitrogen as a
measure of SNH (see figure 14) for studies where the focus is specifically on nitrification
capacities.
In a study of STOWA (STOWA, 1996) a similar, but less extensive, study of
physical-chemical versus biological (only respirometric) influent wastewater
characterisation was carried out. In this study guidelines for the COD components were
finally defined based on a more traditional choice of physical-chemical methods
combined with long-term BOD measurements to allow for an easy implementation in
already existing routine analysis programs. It was concluded that respirometry is not yet
at a state where it can easily be applied for routine wastewater characterisation. The
STOWA guidelines for determination of the COD components are summarised in figure
16. Here the concentration of inert soluble matters (Sv is determined as 90% of the
effluent COD for low loaded systems, according to Siegrist and Tschui (1992). For high
loaded systems SI is also determined as 90% of the effluent COD but the effluent BOD
(multiplied by a CODIBOD factor) is subtracted. Ss is determined as the difference
between soluble COD and Sr. Furthermore, the concentration of Xs is based on a long-
term BOD test as the difference between BOD/(I-YIV and Ss, as described above. The
yield coefficient in this long-term test is set to 0.20. Finally, XI is defined as the
difference between particulate COD and the determined Xs. Obviously, in this approach
the division of the wastewater into ASMI components is based on solubility and to
some extent on biodegradability according to physical-chemical methods supplemented
by measurements of the ultimate BOD_ or BODs. The problem with this approach is
that the biodegradation rate of the wastewater is not really considered. This means that
the division of the biodegradable substrate into readily and slowly biodegradable
substrates may not be correct. It should be stressed though that the approach chosen by
STOWA is simple to implement into existing standard measuring routines at full-scale
WWTP's, which is a factor not to be underestimated.
146
Combined
Total COD ASM1 methodology Total COD ASM1 STOWA
XBH Negligible
Respiration tests XBA
Mass balance:
Long-term BOD test? XI particulate COD - Xs
Long-term BOD test
Long-term BOD test

Xs
Xs
Mass balance
soluble COD
Analytically Mass balance:
soluble COD Soluble COD - S,
Respiration tests SS
Ss
Effluent analyses
Effluent analyses 51
SI Long-term BOD test
Fig. IS. Suggested wastewater Fig.I6. The STOWA (1996)

characterisation by combined physical- guidelines for determination of
chemical and biological methods COD components
The STOWA guidelines for nitrogen components are also rather simple and based on
physical-chemical analyses. The SNH component is obtained based on standard analyses
of soluble ammonium nitrogen, and the determination of the organic nitrogen fractions
(SND and X ND) is based on certain fixed fractions of N in organic components. It is
advised that these organic nitrogen fractions are checked regularly based on
measurements of total nitrogen, Kjeldahl nitrogen etc. according to figure 6B.
In this literature review the focus has been on characterisation of the ASM1
wastewater components. However, with the introduction of ASM3 (see table 2), that
also focuses on a description of oxygen consumption, sludge production and N removal,
it is interesting to discuss whether the approaches for wastewater characterisation
applied for ASM1 holds for ASM3 as well.
As described above, there is a shift of emphasis from hydrolysis to storage of
organic matter in ASM3. Furthermore, all Ss is supposed to go through the storage
process (conversion to X STO) before being used for growth. This means a change in how
wastewater characterisation should be viewed, since the separation between Ss and Xs
should now be based on the storage process rather than on the growth process. In ASM3
(Gujer et al., 1999) it is supposed that the soluble (Ss) and particulate (Xs)
biodegradable components can be differentiated with filtration over 0.45 11m membrane
filters, whereas a significant fraction of Xs in ASM1 may be contained in the filtrate of
the influent wastewater. In ASM3 the latter is assumed to be caused by the conversion
of soluble biodegradable COD to storage polymers in the respiration tests. Whether this
may hold in any case seems yet rather unclear. In Gujer et al. (1999) it was recognised
that the model concept of converting all Ss into a storage component is not in
147
accordance with reality. Indeed, it was illustrated by Krishna and van Loosdrecht (1999)
that the difference between feast and famine phases could not be described accurately.
This was caused by the fact that ASM3 does not allow growth on the substrate Ss alone.
Therefore, a new model structure was proposed where growth on external substrate is
allowed in parallel with the storage process. It remains however uncertain how to
differentiate between the amount of Ss that is directed to storage and growth
respectively. Furthermore, the yield coefficient (which is needed to convert
respirometric responses to COD components) in ASM3 is composed of two factors:
YnerYSTO· YH, where YSTO is the storage yield and YH the heterotrophic yield for the
growth process. Also, here it does not seem clear how to differentiate between the two
yields. Basically, concerning the characterisation of COD wastewater components,
more experience will be needed before a wastewater characterisation of the COD
components related to the new storage concept of ASM3 can be proposed.
The characterisation of the nitrogen components in ASM3 is however simplified by
the fact that organic nitrogen components are included in the model as a fraction of the
corresponding COD components. Degradation of the corresponding COD component
results in immediate release of the organic nitrogen as ammonium. The latter was based
on the assumption that the ammonification is fast and the conversion of organic nitrogen
into ammonium therefore hardly affects the model predictions (Gujer et aI., 1999).
Thus, the nitrogen balance includes on the one hand ammonium nitrogen (S~ and
nitrate nitrogen (SNO), which both can be measured easily via standard chemical
analyses, and on the other hand organic nitrogen components. However, typically the
fractions of organic nitrogen in the COD components can be considered to be constant.
4.2. CHARACTERISATION OF SLUDGE COMPOSmON
In this section special attention is only paid to the assessment of the slower varying
sludge characteristics. Knowing the initial value of the concentrations of soluble
components (e.g. ammonia) is not really essential because it has little impact on typical
simulation results with a calibrated model. Hence, the concentrations of the following
particulate, slowly varying components must be assessed: XBH, XBA and XI (+Xp) ,
'assuming that the system is in balance with no accumulation of Xs. Only two
concentrations must be assessed since the sum of the concentrations is equal to the
particulate COD (X) of the sludge that can easily be measured by using traditional COD
analysis (Eq. 14)
X=X 1 +(XP)+XBH +XBA (14)
Below some fast and direct methods for assessing sludge components are summarised.
Notice that the particulate nitrogen components are not considered here as their
concentrations are assumed to be low.
Heterotrophic biomass XBH

One can show that the concentration of heterotrophs in a continuous system in steady
state is equal to:
148
(15)
where ex is the sludge age, eH is the hydraulic retention time, CODDegraded the total
amount of COD removed (taken over a sufficiently long period, e.g. one sludge age), bH
the decay rate coefficient and Y H the yield coefficient. Respirometric methods to
determine the parameters bH and Y H are discussed below, while a respirometric
evaluation of CODDegraded can be performed with the respirometric measurements of
biodegradable COD fractions (Ss, Xs) that was already presented above.
As an alternative, Bjerre et at. (1995) used the method of Kappeler and Gujer (1992)
to determine the concentration of heterotrophs in the mixed liquor. Recently, this
method was thoroughly evaluated by Ubisi et at. (1997).
Autotrophic organisms X BA
In much the same way, the concentration of nitrifying organisms in the activated sludge
can be evaluated by means of a mass balance for the autotrophs (over a sufficiently long
time) (Dupont and Sinkjrer, 1994):
Ox f Aerobic . N Nitr
X BA =YA . _ . _ - - - - (16)
OR l+b A ·OX
where fAerobic is the aerobic fraction of the reactor; NNitr the amount of nitrified nitrogen;
b A the autotrophic decay rate coefficient and Y A the autotrophic yield coefficient. The
methods to determine the parameters b A and Y A are discussed in the next paragraph,
while N nitr can be quantified using the respirometry-based nitrifiable nitrogen evaluation
methods that were given above.
Produced inert suspended organic matter Xp

To determine the produced inert matters, Xp, an evaluation of the mass balance of Xp in
steady state can be made. Assuming that the autotrophic biomass can be neglected, Eq.
17 is obtained:
(17)
The total concentration of inert matters, including the often significant contribution of
suspended inert material from the influent, is given in Eq. 18.
(18)
Respirometry can be involved in calculating this fraction via fp and bH (see below).
149
4.3. CHARACTERISATION OF STOICHIOMETRIC AND KINETIC

PARAMETERS
Similar to the overview of wastewater characterisation the overview on characterisation

of stoichiometric and kinetic parameters will be clarified according to the applied
methodology. The focus will, however, only be on different biological methods since
physical-chemical characterisation is not very relevant when it comes to characterisation
of reactions. As highlighted in the previous section the majority of the processes
involves oxygen consumption, which means that respirometry will again be the
dominating method in the review. However, also other methods such as nitrate
utilisation rates, titrimetry and ammonium uptake rate are powerful to assess some of
the kinetic and stoichiometric parameters.
4.3.1. Respirometry
4.3.1.1. Stoichiometric parameters By definition, determination of stoichiometric

parameters requires the measurement of two factors that are related to the substrate
uptake. One of these factors may be the respiration rate. Theoretically, for ASMI the
following stoichiometric parameters can be evaluated using respirometry: YH, YA, iXB
and fp, though attempts are reported only for the first two.
Heterotrophic yield coefficient YH

This parameter not only influences the estimation of sludge production and oxygen
demand but also has an impact on the value of other parameters whose determination
requires a value for YH (see table 6) An example is the determination of Ss from
respirometric data as described above (Eq. 7). Hence, an accurate value for YH is of
great importance. YH can be determined using respirometry by addition of an amount of
wastewater COD and measurements of the substrate oxidation ro,ex (Sollfrank and
Gujer, 1991; Brands et at., 1994). Eq. 19 is then applied to evaluate YH •
COD degradable - fro,ex (t)dt

Y H = _ _ _ _ _ _0"--_ __ (19)
COD degradable
The amount of degradable COD (CODdegradable) is given by the COD concentration in the
filtered wastewater minus the inert fraction (S,). In the study of Sollfrank and Gujer
(1991) S, was determined as the soluble COD concentration in the effluent.
Brands et at. (1994) and Liebeskind et at. (1996) circumvent the problem of
determining Sj by using a completely biodegradable substrate (acetate) instead of
wastewater. Hence, CODdegradable is known exactly. This approach is, however, doubtful.
First, the choice of acetate is rather arbitrary and there is quite some evidence that the
yield coefficient for acetate differs from the influent wastewaters (Dircks et at., 1999).
Hence, acetate is not really representative for wastewater COD. Moreover, due to the
experimental conditions in the batch reactor, it can be expected that part of the acetate is
150
stored in the cell (Majone et al., 1999). In this case the observed oxygen demand only
represents the needs for transport of the substrate and incorporation in storage material
of the cell, and not for the complete conversion into new biomass. Conclusively, these
procedures for estimation of the heterotrophic yield do not seem without problems.
Autotrophic yield coefficient Y A

A value of 0.24 g biomass COD per g nitrified nitrogen is generally assumed to be a
good theoretical value for Y A. If required it is possible however to determine the actual
Y A from a respirometric batch experiment in which a known pulse of ammonium
(SNH(O» is added to a nitrifying activated sludge sample (Eq. 20).
4.57 ,SNH (0)- Iro.ex (t)dt

YA =_____-'0'--_ __ (20)
SNH(O)
In this approach care has to be taken that no significant net growth of heterotrophs take
place as they would incorporate part of the added ammonium. In the model based data
interpretation applied by Spanjers and Vanrolleghem (1995) correction for
incorporation of SNH into biomass is taken into account directly via the model.
Nitrogen content of the biomass iXB

Obviously, the most likely method for evaluation of iXB would consist of a nitrogen
analysis of the biomass. However, one can imagine (albeit maybe not very realistically)
that nitrogen incorporation into biomass can be assessed using two respirometric
experiments with nitrifying sludge in which different amounts of COD are degraded, the
difference being denoted as ACODDegraded. The reduction in the oxygen consumption for
I
nitrification A rC,ex (t)dt that can be observed for the higher COD loading then allows
a calculation ofiXB (development ofEq. 9).
. H Y ACODDegraded
1XB =---="--- ----:;----- (21)
4.57 - YA A Ir~ex (t)dt
Inert particulate fraction of the biomass fp

Decay of biomass results in a fraction being transformed into inert particulate products.
Typically 20 % of the biomass consists of inert material (Henze et al., 1987). This inert
biological fraction is called f p. The model fp can be calculated starting from the
biological f p with the following implicit equation:
. fp
fp =-1--Y-H--"'-:-(I---fp-:-) (22)
151
If the studied activated sludge has a yield coefficient (estimated for instance by using
respirometry) deviating from the one reported in literature, the fp-value must be adapted
for this. Keesman et al. (1998) theoretically showed that the value of fp can be estimated
directly from a batch test in which only the evolution of the respiration rate and sludge
concentration are monitored over sufficiently long time.
4.3.1.2. Kinetic parameters Basically the kinetic parameters that can be determined via
respirometry are related to aerobic growth, decay and nitrification.
Heterotrophic decay coefficient 1m

The classical respirometric method for determination of b H' described by Henze et al.
(1987) is the protocol proposed by Marais and Ekama (1976) and is the most typical
method applied for the determination of the decay coefficient (e.g. Sollfrank and Gujer,
1991; Kappeler and Gujer, 1992). Sludge is inhibited for nitrification and is aerated in a
non-fed batch reactor. The (endogenous) respiration rate is measured at certain time
instants over a period of several days. Since the endogenous respiration is proportional
with the active biomass concentration, a plot of the logarithm of the endogenous
respiration rate rO,end as function of time describes the exponential biomass decrease as a
straight line with slope b'H.
The death regeneration concept implies that the classical methods for determination
of the decay of biomass based on endogenous decay can not be applied directly. The
parameter based on the endogenous decay concept has to be translated to the death
regeneration concept, similarly to fp (Eq. 22), leading to the ASMI decay coefficient bH
(Eq.23).
(23)
Hence, the stoichiometric parameters YHand fp are necessary for calculation of~.
Vanrolleghem et al. (1992) describe a fast method for estimation of b'H using only
one measurement of the endogenous respiration (in absence of nitrification) in a batch
reactor. By means of Eq. 24 describing endogenous respiration, ~ can be calculated on
condition that fp and XBH are known.
rO,end =(I-f~)·b~ ·X BH (24)
The estimation of b' H can also be based on the fact that the respiration rate for substrate
oxidation is proportional to the heterotrophic biomass concentration (Spanjers and
Vanrolleghem, 1995). If a sufficiently high amount of oxygen So and substrate Ss are
present, rO,ex is not substrate limited and will only be proportional to XBH. Consequently,
the decay of the heterotrophic biomass can be determined by (i) taking a sludge sample
from the aerated and non-fed batch reactor at certain time instants (to, (ii) adding a
sufficient amount of substrate and (iii) measuring the maximum respiration rate.
Assuming that YH and JA.maxH remain constant during incubation, plotting the logarithm
of ro,ex(ttJ as function of time again allows to determine b'H as the slope of the curve
152
obtained via linear regression. In the study of Spanjers and Vanrolleghem (1995) a
model-based interpretation was applied to obtain accurate values of the maximum
respiration rates. However, only two data points were used for the semilog regression,
which does not make the estimated decay coefficients in this study very reliable.
0.70 0.70
0.80 0.110
C" C"
E 0.10 E 0.10
-a, -a,
§.
0.40
0.10
-=
E
0.40
0.80
r
j
I
I
~ 0.20
.
0.20
0.10 0.10
...
~
I
0.00 0.00
0 1IS 10 48 10 75 o 111 80 48 110 75
l1me(mln) l1me(mln)
Fig. 17. Respirograms obtained after injection of a CIN mixture for the simultaneous
determination of bH and bA according to the procedure of Spanjers and Vanrolleghem
(1995). Left: after 1 day incubation, Right: after 7 days
In the study of Avcioglu et al. (1998) a similar procedure was developed, where the
decay rate b'H' was assessed by monitoring the decrease in maximum respiration rate.
Avcioglu et al. (1998) included more data points compared to the study of Spanjers and
Vanrolleghem (1995). It was proposed that this method of determining the decay rate
should be more reliable, since interference of slowly biodegradable substrate, especially
in the initial phase of the traditional test of Marais and Ekama (1976), and inaccuracy of
low endogenous respiration rate measurements were avoided. The latter will, however,
evidently depend on the sensitivity of the applied respirometric method.
Furthermore, in the work of Avcioglu et al., (1998) it was experimentally verified that
the anoxic heterotrophic decay rate was reduced with about 40-50% compared to
aerobic conditions. Other studies confIrm the observation that the heterotrophic decay is
slower under anoxic conditions (McClintock et al., 1988; Siegrist et al., 1999).
Autotrophic decay rate coefficient bl\.

The death regeneration concept is not applied for the autotrophic biomass in ASMI.
However, the approach of monitoring the decrease in rO,end as function of time can not
be applied for the determination of bA since that would require for instance an inhibition
of the heterotrophic biomass. Instead, the method based on the maximum substrate
(here SN0 degradation rate as function of time can be applied similar to the procedure
for the heterotrophic decay coefficient. In fact, in the procedure described by Spanjers
and Vanrolleghem (1995) the heterotrophic and autotrophic decay rate coefficients were
determined simultaneously by addition of a mixture of acetate and ammonium. Figure
17 shows the rO,ex data for the two respirometric tests performed after one and seven
days of sludge incubation, clearly illustrating the decreasing activity.
153
Nowak et al. (1994) pointed to the fact that the release of nitrogen due to decay of
heterotrophic biomass may result in some growth of nitrifying organisms. Hence, an
underestimation of b A would result. To correct this, they proposed the incubation of the
sludge under anoxic conditions to prevent growth of nitrifiers. Daily a sludge sample
was removed from the anoxic reactor and (after aeration) the maximum respiration rate
was determined. It was however observed that the reduction in maximum respiration
rate was significantly smaller (about 50%) under anoxic than aerobic conditions. This
was further confirmed by work on immobilized Nitrosomonas (Leenen et aI., 1997) and
by the findings of Siegrist et al. (1999).
Maximum specific heterotrophic growth rate UmaxH and half-saturation concentration t

&
The maximum heterotrophic growth rate !LmaxH can easily be determined from the
maximum ro,ex (Eq. 25) (Ekama et aI., 1986), assuming that the substrate concentration
is in excess and the yield coefficient and heterotrophic biomass concentration (see
previous section) are known.
(25)
However, the methodology proposed by Ekama et al. (1986) does not provide
information on Ks.
90
80
7G
...0
-•
'"I:;
60
2
...'1(
Fig.IB. A plot of substrate uptake rate versus substrate concentration for estimation of the
parameters for growth, example with vale ric acid (Cech et al., 19B4)
The increase of the substrate uptake rate with increasing Ss concentration is depicted in
figure 18. From such Monod type evolution the maximum specific growth rate !LmaxH
and the half-saturation constant Ks can be determined. In Cech et al. (1984) a
154
respirometric method is described in which a number of measurements are performed,

each of which add one point to figure 18. In this procedure experiments are carried out
with addition of different amounts of wastewater (substrate) to endogenous sludge,
allowing to achieve various substrate uptake rates, i.e. exogenous respiration rates
(ro,eJ, up to a maximum rate.
The parameters J-tmaxH and Ks can, for instance, be found by Lineweaver-Burk
linearization of Eq. 26 that describes the curve in figure 18 (Cech et ai., 1984), although
the statistical quality of this procedure is not optimal (Robinson, 1985).
dS s Ss
dt
= (26)
Ks+SS
The method of Cech et ai. (1984), which was also applied by e.g. Volskay and Grady
(1990), is rather time consuming and the experimental effort is high. As an alternative a
more efficient approach was presented, using a continuously aerated respirometer to
which a single substrate pulse is added (Vanrolleghem et ai., 1990; Kong et ai., 1994).
In this method rO,ex is recorded frequently as the experiment progresses and one
experiment is sufficient for the determination of both J-tmaxH and Ks provided that the
concentration of added substrate is sufficiently high. In this approach a model (Eq. 27 -
28) is fitted to the rO,ex profile for the determination of J-tmaxH and Ks. An example of an
acetate addition is illustrated in figure 19 (obtained from Kong et ai., 1994) where rO,ex
is illustrated together with the corresponding cumulative oxygen consumption and
substrate concentrations as function of time.
dS s IlrnaxH ,X BH Ss
--= (27)
dt YH Ks +Ss
r = (1- Y ). IlrnaxH ,X BH Ss (28)

a,ex H YH Ks +Ss
The heterotrophic kinetic parameters can also be determined based on the cumulative
oxygen uptake profiles rather than oxygen uptake rate data. In the methodology
described by Ellis et al. (1996) and Smets et al. (1996) the kinetics are determined for
specific organic chemicals. However, the procedure is directly applicable for
wastewaters as well.
155
0.80 3.00
C
0.50 2.50 ·e
.....
c =a,
·e 0.40 2.00 E
......
=a, 0.30 rn
1.50 ...:..
E
......
~
0.20 1.00 >:.........
.2
0.10
0.00
0.50
1...
0.00 :;:
.E
-5 0 5 10 15 20
Time (min)
Fig. 19. ro.ex (symbols), cumulative oxygen uptake (increasing line) and substrate
concentration (decreasing line) in batch experiment (Kong et al., 1994).
A batch experiment with high initial substrate (wastewater) to sludge ratio (called the
S(O)/X(O) ratio) was proposed by Kappeler and Gujer (1992). This procedure also
enables estimation of !lmaxH and Ks from a single experiment. An alternative to the
method of Kappeler and Gujer (1992) is to plot the oxygen uptake rate versus the
cumulative oxygen consumption (Smets et al., 1996). Figure 9 shows a respirogram
obtained with such an experiment (Kappeler and Gujer, 1992). Contrary to the
procedures of e.g. Vanrolleghem and Verstraete (1993) biomass growth is significant
and !lmaxH can be assessed directly without knowledge of YH. A plot of the logarithm of
the ro measurements versus time has the slope (!lmaxH - b~. If bH is known, a calculation
of !lmaxH is possible (in the work presented by Kappeler and Gujer (1992), it is assumed
that the decay rate is 5% of the growth rate). Attention has to be paid to the fact that the
high S(O)/X(O) ratio in this experimental set-up (about 4/1) gives rise to significant
growth of the biomass during the experiment. This means that the observed kinetic
characteristics may no longer be representative for the original sludge, due to the risk
that the experimental conditions may have favoured fast growing organisms that
become dominant during the experiment. Novak et al. (1994) gave practical evidence
for this hypothesis by evaluating results from experiments with different S(O)/X(O)
ratios. A 2.5 times higher specific growth rate was obtained at high S(O)/X(O) ratio,
compared to an experiment with a low S(O)/X(O) ratio.
In the work of Grady et al. (1996) the terminology of intrinsic and extant kinetics
was introduced. Intrinsic kinetics refer to the ultimate capacity of the biomass, whereas
extant kinetics refer to the biomass activity prior to the lab-scale experiments, e.g. in the
full-scale plant. This will be discussed further in later sections.
Maximum specific autotrophic growth rate UmaxA and half-saturation concentration KNH
In the studies by Drtil et al. (1993) and Nowak et al. (1994) the above mentioned
methodology of Cech et al. (1984) was applied to evaluate the maximum specific
156
autotrophic growth rate and half-saturation concentration KNH. To assess the ro for
autotrophic activity only, the heterotrophic endogenous respiration was determined by a
separate experiment, where ATU was added, and was subtracted from the total ro
obtained from an ammonium addition. Here too knowledge of Y Aand XBA is needed for
the calculation of ILmaxA. In the work by Nowak et al. (1994) the concentration of XBA
was determined based on full-scale data.
Alternatively, ILmaxA and KNH can be obtained directly from experimental data of a
simple ammonium addition as presented in figure 20. In a study by Spanjers and
Vanrolleghem (1995) a model-based interpretation was applied for the determination of
the nitrification kinetic parameters (Eq. 29), similar to the approach described above for
the kinetic parameters of heterotrophic growth.
(29)
Hydrolysis constants ~~
As far as known the only experimental protocol that enables a determination of both
parameters of the hydrolysis process is the "cyclic square wave feed" experiment
proposed by Ekama et al. (1986). This method has already been described earlier for the
determination of Ss with a typically obtained profile shown in figure 10. To determine
the hydrolysis parameters the data obtained after the drop in respiration rate are
important. If ro remains constant on a plateau value (as is noticed in figure 10 between t
= 12 and t = 15 h), this is related to the hydrolysis that proceeds at maximum rate and
the biomass that is saturated with hydrolysable products (Xs /XBH » Kx). As such,
these data contain the information to assess the value of ~ on condition that the
heterotrophic biomass concentration X BH and the yield coefficient YH are known. With
decreasing Xs also the rate of hydrolysis decreases and the respiration rate is depending
on the value for Kx, allowing its estimation. Estimation of the parameters is best by
means of model optimisation (Henze et al., 1987).
In many cases the dependency of the rate of hydrolysis on the heterotrophic biomass
concentration may be neglected and first order hydrolysis process dynamics are then
obtained (SoUfrank and Gujer, 1991). This assumes that XslXBH« Kx. Sollfrank and
Gujer (1991) proposed a method to determine the first order hydrolysis constant, i.e.
kWKx, using respiration rates measured by dosage of wastewater to a continuous flow
pilot reactor. To simplify the estimation, they suggested to present the respiration rate as
function of the residual amount of substrate. In this plot one is able to isolate a linear
part from which the hydrolysis constant kWKx is deduced (provided YHis known).
For estimation of the frrst order hydrolysis constant ~/Kx Kappeler and Gujer
(1992) performed a batch experiment with an initial COD based S(O)/X(O) biomass ratio
which was 10 times higher than their experiment for determination of the maximum
specific growth rate (S(O)/X(O) = 112). Figure 8B shows the respiration rate data of such
an experiment, from which the slowly biodegradable substrate, Xs, can also be
determined, as described above. Once the readily biodegradable substrate Ss is removed
(in Figure 8B after 0.75 h) the further decrease of the respiration rate is determined by
hydrolysis of Xs. As a consequence, the ro measurements enable to estimate the
hydrolysis rate constant. The authors advise to do this exercise at different biomass
157
concentrations to check for a possible dependency of the hydrolysis rate to the biomass
concentration.
0.10
-"e
c
0.40
=a,
-...
0.30
E
.
~ 0.20
0.10
0.00
0 20 40 80 80
Tlm·emln)
Fig.20. Respirogram obtained after injection of 3.31 mg NH4-N in 1.41 activated sludge
(Spanjers and Vanrolleghem, 1995)
Parameters of "switching functions" KOIu.KoA

Kappeler and Gujer (1992) determined the respiration rate as function of different
oxygen concentrations in the respiration chamber of their respirometer. According to
these authors the concentration of readily biodegradable substrate Ss needs to exceed a
minimal concentration in order to have an accurate determination of KoH. The same
technique can be used for KoA with ammonia as substrate.
Ammonification rate constant k.

So far, no respirometric method has been reported for the determination of the
ammonification rate. However, it is theoretically possible (see table 6) to assess this
parameter from the evolution of the oxygen consumption for nitrification resulting from
ammonified nitrogen, provided ammonification is the rate limiting step.
Simultaneous determination of heterotrophic and autotrophic kinetic parameters.

In the previous sections on determination of heterotrophic and autotrophic growth
kinetics the focus was put on how to determine the kinetic parameters for the
heterotrophic and autotrophic processes separately. However, except for the examples
of Sollfrank and Gujer (1991) and Kappeler and Gujer (1992) the presented examples
mainly dealt with additions of known substrates (acetate as carbon source and
ammonium). The fact is that when dealing with real wastewater and activated sludge
both heterotrophic and autotrophic processes will take place simultaneously, and a
detailed data interpretation of the respirograms and good experimental design will be
needed to "separate" and as such determine the kinetic parameters for the different
processes.
158
Vanrolleghem and Verstraete (1993) proposed an experimental design that enables to

simultaneously measure both heterotrophic and autotrophic maximum respiration rates.
In their approach a mixture of ammonium and acetate was added to endogenous sludge.
The maximum respiration rate for carbon oxidation and nitrification can be derived
from the respirograms on the condition that the two aerobic processes can be clearly
distinguished from each other. The problem with this approach is however that the
kinetic parameters are highly dependent on the nature of the substrate. Thus, the use of a
single compound like acetate to represent a complex substrate like wastewater is
difficult to justify scientifically.
In the study on wastewater by Spanjers and Vanrolleghem (1995) experiments with
municipal wastewater were presented with much lower substrate to biomass ratios
(S(O)/X(O» compared to Kappeler and Gujer (1992). Figure 21 shows a typical
respirogram from an experiment with a S(O)/X(O) of 11200. This respirogram is much
more complicated to interpret than the ones shown so far. First, simultaneous carbon
oxidation and nitrification take place. The only seven minutes lasting initial peak in rO,ex
is assumed to be due to the oxidation of Ss. After some time only nitrification and,
assumingly, oxidation of substrates released by hydrolysis occurs. In this work the
respirograms of the wastewater were interpreted with a more complex ASM1 based
model including degradation of two readily biodegradable substrates SSI and SS2, frrst
order hydrolysis and nitrification. Thus, kinetic parameters for all these processes were
obtained simultaneously. Experiments in the presence of a nitrification inhibitor ATU
were performed to check the contribution of nitrification to the respiration rate. This is
shown in the insert of figure 21, where the rO,ex related to the degradation of Ss and Xs
can be observed.
An approach circumventing ATU addition, suggested by Spanjers and Vanrolleghem
(1995), consisted of the following two-step procedure. First, the nitrification process is
characterised separately via an experiment where only ammonium was added, as
described above and illustrated in figure 20. In a second step, the full model is applied
to fit to the data of figure 21. However, during this step the nitrification parameters are
kept at their values obtained from the separate nitrification experiment, and are thereby
used to "eliminate" the nitrification oxygen consumption in an experiment with addition
of wastewater. The amount of nitrogen in the wastewater sample can be estimated
simultaneously as it determines the length of the nitrification shoulder. Spanjers and
Vanrolleghem (1995) demonstrated that the ATU and model-based elimination of the
nitrification respiration rate lead to similar values for the kinetic parameters and waste
water characteristics.
Another example of a detailed interpretation of a respirometric test with municipal
wastewater addition is given by Brouwer et al. (1998). Here a model including
degradation of two readily biodegradable substrates, hydrolysis and two step
nitrification is applied to interpret wastewater respirograms. The problem encountered
in this study was, however, that not all processes were clearly identifiable from the
respirograms. It was thus suggested that the number of unknown model parameters
should be reduced for this example by including experiments with separate additions of
synthetic substrates, for example ammonium and nitrite. In this way it would be
possible to fix these kinetic parameters in the characterisation of the complex
wastewater, similar to the approach of Spanjers and Vanrolleghem (1995).
159
1.00 .
.
+ATU
-
c
E
0.80
0.80
~
-...
E
0=
0.40
0.20
0.00
0 10 20 30 40 50 80
Tlme(mln)
Fig.21. Respiration rate after injection of 70 ml raw waste water to 1.5 I activated sludge.
Insertion: similar experiment but after addition of ATU (Spanjers and Vanrolleghem, 1995)
Finally, an application with industrial wastewater (no nitrification) was presented by

Coen et al. (1998), where a model-based interpretation approach was applied for the
determination of kinetic parameters and substrate concentrations of simultaneous
degradation of three COD wastewater fractions.
4.3.2. Nitrate utilisation rates

Characterisation of reaction kinetics via analysis of the nitrate utilisation rate is
basically very similar to the methodology based on oxygen respiration rates, and
different studies have dealt with the comparison of ro,ex and rN03,ex (e.g. McClintock et
al., 1988; Kristensen et aI., 1992; Orhon et aI., 1996; Sozen et al., 1998). In ASM1, the
same kinetic expressions are applied for nitrate utilisation processes as for oxygen, with
the only difference that a correction factor 11 is incorporated in the equations for anoxic
processes. This factor allows to describe that only a fraction of the total biomass is
capable of respiring with nitrate and/or that the anoxic rate is lower than the aerobic
one. Typically, one applies the relationship given in Eq. 30 in order to relate ro,ex with
rN03,ex·
1] = 2.86. rN03,ex (30)

rO,ex
160
Correction factors for anoxic growth and hydrolysis n

It has been shown that the value of 11 can vary significantly for different activated
sludge systems. In different studies values have been recorded in the range 0 - 0.95
(Van Haandel et al., 1981; Henze, 1986; Henze et al., 1987; 1995; McClintock et al.,
1988; Kristensen et al., 1992; Sozen et al., 1998; Sperandio et at., 1999). Some theories
were developed based on general mass balances that allowed for an estimation of 11
from wastewater characteristics, treatment plant layout and operation (Henze, 1986). It
was shown that the dominating factor for 11 is the potential inlet fraction of denitrifiers,
which includes the denitrifying fraction of the influent biomass plus the primary
produced anoxic biomass. Based on some practical constraints concerning e.g.
minimum anoxic sludge age and minimum aerobic sludge age to keep both nitrification
and denitrification in the system, it was estimated that in practice 11 might be in the
order of 0.4 - 0.9 (Henze, 1986).
An underlying assumption behind Eq. 30 is that the aerobic and anoxic yields are
equal. As discussed above significant evidence exists that the anoxic yield may be lower
than the aerobic one. In the studies by Orhon et al. (1996) and Sozen et al. (1998) very
high values (> 1) for the conversion factor 11 were related to possible lower anoxic yields
for which correction will be needed. The occurrence of lower anoxic biomass yields
was already discussed in the section about application of nitrate utilisation rates for the
determination of readily or slowly biodegradable substrate Ss and Xs.
4.3.3. Titrimetry
Maximum specific autotrophic growth rate IlmaxA and half-saturation concentration KNH
So far the titrimetric technique, based on pH control and monitoring of the cumulative
amount of base or acid added to keep the pH set-point, proposed by Ramadori et al.
(1980), and introduced in more detail above, has only been applied to the determination
of the nitrification kinetic parameters ~ and KNH • As illustrated in figures 12 and 13
and by Eq. 12, the cumulative amount of base added can be used to calculate the
nitrification rate and thereby provide kinetic information. In the work of Gemaey et al.
(1998) a model-based data interpretation was applied for the estimation of ~ and
KNH • The model is similar to the one applied for the description of respirometric and
nitrate utilisation rate data. The only difference is the stoichiometric coefficient relating
the ammonium degradation to proton production Hp (Eq. 31).
2+ YA ·i XB f.lmaxA ·X BA SNH
(31)
rHp = - 14 . YA KNH +SNH
4.3.4. Summary and discussion of biological characterisation of stoichiometric and

kinetic parameters
The above review on biological characterisation has illustrated that, theoretically, nearly
all parameters can be determined with biological methods. Especially respirometry
stands as a powerful characterisation method but other methods too are useful for the
characterisation of specific processes, e.g. titrimetry for the characterisation of
161
nitrification and application of nitrate utilisation rates for the determination of the
correction factor for denitrification.
One of the challenges in the application of the biological methods is how to interpret
and relate the experimental data to the different processes that may take place
simultaneously. It is obvious that experiments with addition of known and simple
substrate such as ammonium or acetate are easier to interpret in terms of determination
of stoichiometric and kinetic parameters than experiments with real wastewater. For
example, it is difficult to assess the heterotrophic yield YH by experiments with real
wastewater, and in some cases it was therefore suggested to determine it from an
experiment with known substrate in the form of acetate (Brands et al., 1994; Liebeskind
et al., 1996). It has also been suggested to determine the maximum specific growth rate
~axH based on experiments with acetate in respirometric experiments (e.g.
Vanrolleghem and Verstraete, 1993). However, acetate does not represent the actual
wastewater very well. As already stressed above it is generally questionable to use a
single substrate to represent complex wastewaters. Furthermore, it is a known
phenomenon that acetate easily gets directed towards the storage process instead of
directly being consumed for growth (Majone et al., 1999). This means that if such data
are only interpreted in terms of the growth process, the estimated parameters related to
growth will be erroneous. E.g. the stoichiometric growth yield (Y0 will be
overestimated (Dircks et al., 1999). On the other hand, characterisation of the
stoichiometric and kinetic parameters for nitrification can be done by respirometric or
titrimetric experiments with single additions of pure ammonium.
It is of course advantageous if several parameters (kinetic or stoichiometric) and
some wastewater components can be obtained from the same experiment. This was
illustrated in studies with municipal wastewater by e.g. Spanjers and Vanrolleghem
(1995) and Brouwer et al. (1998), and also for an industrial COD removal case (Coen et
al., 1998).
In table 8 (adopted and modified from Vanrolleghem et al., 1999) the experiments
described above for characterisation of stoichiometry and kinetics are concisely
represented. Attention is drawn to
• The method (respirometry, nitrate utilisation rates, titrimetry)

• The type of reactor set-up (continuous or batch experiment) and the additions
performed;
• The requirement for other information collected from other experiments (or
assumed);
• Major assumptions made during the interpretation of the data;
From table 8, it can for example be seen that in the work of Spanjers and Vanrolleghem
(1995) with wastewater (reference SV95 and experiment type "B, WW add.") the
parameters ~axH, Ks, JlmaxA, KNH and ~ and the substrate components Ss, Xs and SNH
could be retrieved from a single experiment.
It will now be attempted to evaluate whether the characterisation approaches of the
kinetic and stoichiometric parameters as reviewed for ASMI can hold for ASM3 too.
As reviewed above, it should be theoretically possible to assess the ammonification
rate from respirometric data, provided that ammonification is the rate limiting step.
162
However, in most applications this is not the case making it difficult to quantify the
kinetics of ammonification. Furthermore, ammonification does not affect the model
predictions significantly, since it is usually a fast process. Thus, with this in mind the
ammonification process was not included in ASM3, thereby also eliminating the need to
determine its kinetic rate.
Another simplification in ASM3 is the way the decay process is described. Instead
of the more complex death regeneration concept it was chosen to describe decay with a
more traditional and simple endogenous decay process. This means that the results from
a simple long-term aeration test (Marais and Ekama, 1976), where the endogenous
respiration rate is monitored over a period of several days, can be applied more directly.
In this way a transformation of the data from the endogenous test to the death
regeneration concept is no longer needed. Furthermore, the exclusion of the death
regeneration concept also resulted in a simplification of the hydrolysis process, since
this process is now only involved in hydrolysis of slowly biodegradable substrate (Xs)
contained in the influent.
However, with the introduction of the storage model concept it becomes difficult to
separate between the kinetics of storage and growth. Already in the discussion of
wastewater characterisation it was pointed out that the yield obtained from a
respirometric test is composed of two factors YnePYSTO'YH' Furthermore, it does not
seem clear how to differentiate between the storage rate and growth rate from e.g. a
respirometric test
4.4. IS CHARACTERISATION VIA LAB-SCALE EXPERIMENTS RELEVANT?
In previous sections the sources of information that can be used for calibration of ASMI
were reviewed and attention was especially focused on how to characterise the different
wastewater components, stoichiometric and kinetic parameters. Different problems
were already highlighted.
The focus is now turned back to calibration of ASMI and the aim of describing a
full-scale WWTP. It should be remembered that the purpose of the model calibration
determines the degree of detail of the information that is needed, e.g. which wastewater
components and parameters need a more accurate determination than others. Even
though it may be possible to characterise some components or parameters, it may not
always be relevant for the actual purpose.
163
Table 8. Overview of biological methods for estimation of ASMl parameters. (Fields with
grey background indicate that respirometry is either not applicable or not relevant)
Method:
R: Respirometry Type of experiment For references see table 8 bis
N: Nitrate respiration test Ac: acetate
A: Ammonia uptake test B: batch reactor
T: Titrimetry StSt: steady state
add.: addition
adds.: additions
C: continuous system
WW: waste water
S: synthetic substrate
164
Table Bbis references cited in Table B
L92 Lesouef et at., 1992

B95: Bjerre et al., 1995
ME76 Marais and Ekama, 1976
B97 Bogaert et at., 1997
M95 Massone et at., 1995
B94 Brands et al., 1994
N98 Naidoo et al., 1998
B98 Brouwer et al., 1998
N94 Nowaketal., 1994
C84 Cech et al., 1984
R97 Rozzi et aI., 1997
D96 Dhaene, 1996
Sm96 Smets et at., 1996
D93 Drtil et al., 1993
S091 Sollfrank and Oujer, 1991
DS94 Dupont and Sinkjaer, 1994
SV95 Spanjers and Vanrolleghem, 1995
E86 Ekamaetal., 1986
S99 Sperandio et aI., 1999
E96 Ellis et al., 1996
S96 STOWA, 1996
097 Oernaey et al., 1997
U97 Ubisi et al., 1997
098 Gernaey et al., 1998
U98 Urbain et al., 1998
H87 Henze et al., 1987
V92 Vanrolleghem et aI., 1992
K092 Kappeler and Oujer, 1992
VV93 Vanrolleghem and Verstraete, 1993
K98 Keesmanetal., 1998
VO Volskay and Grady, 1990
K94 Kong et aI., 1994
We95 Wentzeletal., 1995
K92 Kristensen et at., 1992
Wi96 Witteborg et al., 1996
KK99 Kujawa and Klapwijk 1999
However, the problem is not only whether it is possible to carry out a characterisation
of the different model components or parameters in lab-scale. A probably more relevant
question is whether it is possible to transfer the lab-scale observations to the full-scale
system. Or, to apply the terminology suggested by Gradyetal. (1996), do the lab-scale
experiments provide extant kinetic parameters, i.e. parameters representative for the
biomass prior to the experiments? Furthermore it will be discussed how the relations are
between lab- and full-scale observations, and how the biological processes are presented
in ASMl:
• Transferability between lab-scale and full-scale observations: Are the different

components and parameters that may be determined via lab-scale experiments
representative, i.e. transferable to the full-scale system? That is, do the
experiments provide extant kinetic parameters?
• Transferability between full-scale observations and modelled processes: Are
the full-scale processes described in a biologically realistic way in the model or
are the model processes lumping different biological processes? If so, it may
be impossible to characterise them by any experiment.
• Transferability between model processes and lab-scale observations: Are the
processes defmed in the model reflected by the lab-scale experiments?
165
Petersen B., Gemaey K., Henze M., Vanrolleghern P.A.
Full-scale process
Lab-scale c Model
Fig.22. Schematic representation of discussion on transferability
These conflicts of transferability are illustrated in figure 22, and the discussion is taken
below considering the different wastewater components, kinetic and stoichiometric
parameters. The aim of this discussion is to decide which information source is most
relevant for the different components and parameters. Of course, in principle all
components and parameters can be obtained from the model, e.g. via the default
parameter set or via adjustment of the values during the model calibration exercise.
However, some model processes do not reflect reality completely, although they
enable a mathematical description of the biological observations. The model
components and parameters related to such processes can not be characterised reliably
via either lab-scale or full-scale data and should preferably be tuned during the model
calibration with the full ASMI. Then there are some components and parameters that
readily and reliably can be transferred from a lab-scale experiment. For others the lab-
scale results are difficult to transfer to the model of the full-scale system, and for
instance a mass balance with full-scale data may be more appropriate as information
source. Whether a certain component or parameter should be obtained via lab-scale or
full-scale data or should be tuned directly via the model will depend on what the
component or parameter in question is depending on. In this discussion it is assumed
that the values of the components and parameters can depend on either the actual
biomass in the activated sludge system or the actual WWTP operation. It should be
stressed that only the actual state of the system is considered in this discussion, since
this is what the calibrated model is aimed at describing. Obviously, the biomass
character (e.g. maximum specific growth rate, decay rate etc.) of the WWTP is
determined by both the incoming wastewater and WWTP operation. However, changes
in biomass characteristics caused by changing WWTP operation or wastewater
character are more long-term effects. Description of these effects is not within the scope
of the ASM models. Thus, the actual wastewater considered for the model calibration is
assumed to be representative for the general wastewater composition to which the
biomass has adapted and by which the biomass character is determined.
4.4.1. Kinetic and stoichiometric parameters

Below, the information sources for the most relevant kinetic and stoichiometric
parameters will be discussed in relation to figure 22. Furthermore, the discussion on
whether a parameter is depending on the wastewater, biomass and/or WWTP operation
is summarised in table 9. Finally, the most relevant information source is indicated in
table 9. Brackets in table 9, i.e. (X), indicate that a lab-scale experiment is possible for
166
determination but the transferability of the obtained parameters to the full-scale

situation is uncertain for different reasons, as explained below. Finally, as mentioned
above all parameters can in principle be determined based on the model alone without
additional supporting information.
The maximum specific heterotrophic growth rate, Um.xH
The observed actual specific growth rate in the full-scale system, f..l'maxH, depends on the
sludge age and therefore depends both on the actual wastewater and the WWTP
operation. If the wastewater contains a significant amount of biomass, f..l'maxH will
depend primarily on the wastewater, whereas it will depend on the operation if the
biomass is primarily produced within the plant. On the contrary, the maximum specific
growth rate, f..lmaxH' is the maximum possible specific growth rate of the actual sludge,
and is only influenced by the actual kind of bacteria present. It may be important to
distinguish here between f..lmaxH and the growth rate, f..l, which is influenced by the mixed
liquor substrate concentration. Thus, f..lmaxH is not depending on the wastewater whereas
f..l is.
This means that the problem of transferability between the lab-scale and the full-
scale observations will be insignificant (conflict a in figure 22) if the lab-scale
experiment is carried out under conditions that are comparable to the full-scale system
(e.g. with respect to pH, temperature, ratio between substrate and biomass concentration
etc.). In other words, if the lab-scale experiments are performed in a way that allows
measurement of extant parameter values, then little or no conflict will arise. As
described earlier, the death regeneration concept in the model has the effect that the cell
mass turnover rate increases, resulting in a higher growth rate than if a more traditional
concept of endogenous decay was applied. Thus, this should be taken into account in
the interpretation of lab-scale experiments and in the transferability of results to the full-
scale model (conflict C in figure 22). Similarly the death regeneration model concept
and the way it influences the maximum specific growth rate may not reflect the full-
scale process completely (conflict B in figure 22), but may allow for an adequate
description of observations.
Summarising, the f..lmaxH is one of the most relevant parameters to study in lab-scale
experiments and can be considered to be a biological parameter, which is only
determined by the actual bacteria present (see table 9).
The maximum specific autotrophic growth rate, blmaxA

Although specific bacterial groups undertake nitrification, they adapt to the actual
environment and the bacterial species can therefore vary. Therefore, the discussion on
the maximum specific autotrophic growth rate f..lmaxA is rather similar to the one of f..lmaxH.
Thus, it is possible to determine the value of f..lmaxA from lab-scale experiments, and
transfer the value to the model of the full-scale system.
Half-saturation coefficients: KS~O"KOA and MH

In pure cultures the half-saturation coefficients canbe regarded as pure biological
parameters that give measures of the affinity of the biomass for substrates. However, in
cultures where the bacteria grow in flocks (as in activated sludge), the flock size and
structure playa role in the diffusion of substrate to the cell and thereby on the apparent
value of saturation coefficients. Especially in full-scale systems mixing characteristics
167
will further influence the apparent values. Even in lab-scale tests under simpler mixing
characteristics, mixing may play a role and influence the obtained values of the half-
saturation coefficients. Thus, the different mixing characteristics of the lab-scale and
full-scale system make it difficult to transfer the lab-scale observation to the full-scale
system (conflict A in figure 22). If the flock size decreases due to e.g. a more intensive
mixing in the small batch-scale experiments, the obtained coefficients will be smaller
than required to describe the full-scale behaviour (Henze et ai., 1999). This makes it
difficult to obtain a model relevant value of the half-saturation coefficients from lab-
scale experiments (conflict C in figure 22). The saturation coefficients in ASMI
describing a full-scale situation may therefore be regarded more as model parameters
with the purpose of preventing unrealistically high substrate uptake and growth rates.
Thus, the biological meaning of the model half-saturation coefficients is mixed with the
hydraulics of the system (conflict B in figure 22). Obviously, if a very detailed model is
available to describe the hydraulics of a system accurately, it may be possible to
separate the effects of biomass affinity for a substrate and the hydraulic effects from
mixing. However, usually the hydraulic pattern is approximated by a simple tanks-in-
series model that may be sufficient for a mathematical description but not accurate
enough for a complete elimination of hydraulic effects on the biological parameters.
Table 9. Discussion on relevant information sources for kinetic and stoichiometric
parameters. A bracketed X indicates that a lab-scale experiment is possible for
determination but the transferability of the obtained parameters to the full-scale situation is
uncertain (see text for further explanation)
Dependency Relevant infonnation soun:e

Sludgelbiomass Plant operation Lab-scale Full-scale data Model calibration
experiment Mass balances
I1maxH X X X
/1maxA X X X
Ks,~o X X (X) (X) X
KNH X X (X) (X) X
KoH, KoA X X (X) (X) X
hH,bA X X X
YmaxH X (X) X
YmaxA X (X) X
~ X X
Kx X X
118 X X X X
11h X X
Thus, all half-saturation coefficients of the full-scale system will depend on both the
WWTP operation (mixing) and the actual kind of biomass present. The coefficients can
be determined by lab-scale experiments but the values obtained may not be very
representative. It may therefore be better to estimate these parameters from full-scale
data, via the operational rate of COD removal found by mass balances as function of the
operational range of COD concentrations. The question is of course whether the full-
168
scale data is informative enough for such determinations. Thus, in practice these values
may have to be tuned during the model calibration.
Decay rate of heterotrophs bH and autotrophs b A

The decay rate in a full-scale WWTP is in principle a characteristic of the actual
biomass, and can, similarly to the maximum specific growth, rate be considered as a
biological parameter. However, it may be difficult to obtain a representative value of the
decay rates of a full-scale system from the lab-scale tests presented above (conflict A in
figure 22), since decay and growth due to substrate inflow (and internal production)
take place simultaneously in the full-scale WWTP. On the contrary, decay is typically
investigated under starving conditions (endogenous respiration) in lab-scale
experiments. Furthermore, the decay rate in the full-scale plant is typically influenced
by grazing, i.e. presence of protozoa, which may not be present or may not be able to
survive in the lab-scale experiment.
In ASM1, the death regeneration concept includes both lysis combined with
hydrolysis of released substrate and, subsequently, growth on this substrate. As
discussed earlier, this interaction of different processes makes it difficult to determine
the decay coefficient related to the death regeneration concept (conflict B in figure 22).
However, according to ASMI it is possible to transfer the decay rate obtained from a
lab-scale experiment with decreasing endogenous respiration as function of time for
determination of the endogenous decay rate to the death regeneration model concept
(via Eq. 23, conflict C in figure 22). Obviously, the change in ASM3 to the endogenous
respiration decay concept makes it more straightforward to determine the decay rate of
the model by a lab-experiment.
In conclusion, it is possible to determine the decay rate via lab-scale experiments,
and to convert the obtained value to the death regeneration concept of ASM1, but the
value may to some extent have to be adjusted during the model calibration procedure.
Maximum heterotrophic and autotrophic yield, Y~

The observed yields in a full-scale WWTP, Y'H and Y';:' are depending on the process
operation, i.e. the actual wastewater load and the sludge age. On the other hand, the
actual maximum yields (YHand YA> are depending on the kind of biomass present. For
municipal WWTP's the parameters YH and YA are typically assumed to be rather
constant, indicating that the biomass character is rather similar among different
municipal WWTP's. However, it may still be needed in some cases to determine the
biomass yields. This can be carried out in lab-scale experiments, but there may be some
experimental difficulties, e.g. caused by the possible influence of storage which may be
induced by the conditions in the lab-scale experiment (Majone et al., 1999), as earlier
described (conflict A in figure 22).
In fact the typical maximum heterotrophic yield of 0.67 for municipal wastewater
(Henze et al., 1987) is higher than the yields observed with pure cultures (Heijnen et al.,
1992). The reason for this may be that the model yield covers different processes as
storage, death regeneration etc. and may thereby be considered more as a model yield
(van Loosdrecht and Henze, 1999) (conflict B and C in figure 22). Although the
heterotrophic yield seems influenced by the available electron acceptors (the anoxic
yield is reported to be lower than the aerobic one, Koike and Hattori, 1975; Orhon et al.,
169
1996; McClintock et al., 1998; Sperandio et aI., 1999), the yield may be more
influenced by storage than by the electron acceptor.
Hydrolysis rate kt, and half-saturation coefficient Kx

Although only limited knowledge is available about hydrolysis, the process is needed in
ASMI to describe the degradation of slowly biodegradable organic matter originating
from the influent COD and from internal turnover of substrate in the death regeneration
cycle.
As described above attempts have been made to analyse hydrolysis in lab-scale
experiments. It may be possible to compare the real enzymatic hydrolysis as it takes
place in lab-scale with the full-scale hydrolysis process. However, the real enzymatic
hydrolysis is not the same as the hydrolysis process in the model, as it might also cover
consumption of storage polymers, hydrolysis of decayed biomass (death regeneration),
protozoan activity etc. (conflict B and C in figure 22) (van Loosdrecht and Henze,
1999). Thus, it remains problematic to design an experiment that is representative for
both the model concept and the hydrolysis process as it takes place in full-scale. If this
is compared to the determination of e.g. the maximum specific growth rate, we note that
this parameter also covers many details but still only describes one process, i.e. growth.
In conclusion, the real hydrolysis process is probably determined by the actual
biomass, which produces the enzymes, but for the model calibration of ASMI it does
not seem relevant to attempt to characterise this process via lab-scale tests. Hence, the
hydrolysis as it is described in ASMI should be considered as a model process that has
to be adjusted during the model calibration procedure. It should be remembered that the
definition of hydrolysis has changed in ASM3 and is closer to the real biological
hydrolysis. Thus, a characterisation of the hydrolysis parameters from a lab-scale
experiment will be more relevant for ASM3. The problem remains, however, to design
a good experiment for characterising the real biological hydrolysis.
Correction factors for denitrification Dg and Db

The correction factors for denitrification can be found via a combination of
respirometric and nitrate utilisation rate experiments for the determination of the growth
and hydrolysis process, although some problems may be encountered in the case where
the aerobic and anoxic yields can not be considered equal. It was also referred above
that the correction factors can be determined based on some general mass balances of
the full-scale system (Henze, 1986). Both correction factors will depend on the actual
biomass character. However, no particular conflicts, as indicated in figure 22, are
apparent concerning the correction factor for growth, l1g. Determination of the
correction factor for hydrolysis will suffer from the same problems as indicated above
for the hydrolysis itself, and may therefore also be considered more as a model
parameter.
4.4.2. Relevant kinetic and stoichiometric parameters for lab-scale characterisation

In the discussion on the relevance of characterising the stoichiometric and kinetic
parameters of ASMI via lab-scale experiments, one has to remember that none of the
ASM model processes are pure or microbiologically correct. To some extent they are all
bulk processes. It has clearly been illustrated above that experiments oriented in
170
identifying mechanisms introduced in the model might easily lead to conflict with the
actual model coefficients (van Loosdrecht and Henze, 1999). Thus, although possible, it
may not always be relevant to retrieve the model parameters from lab-scale tests.
Above the discussion was taken on these conflicts between lab- and full-scale
observations and the links to the model processes. Table 9 summarised the dependency
of the parameters on the biomass and WWTP operation, and it was attempted to indicate
the most relevant information source based on these discussions. Notice the difference
to table 8 that listed how the different parameters could be estimated from lab-scale
tests, whereas table 9 indicates whether this is relevant or not, considering that the
parameter should correspond reasonably well both with the full-scale behaviour i.e.
extant parameters are sought, and with the model concepts.
From table 9 it is deduced that it may be relevant to determine the following list of
stoichiometric and kinetic parameters from lab-scale experiments. It is not judged
whether it is always needed to characterise these parameters since that will depend on
the purpose of the model calibration. For the same reason it is not attempted to make an
indicative order of parameter importance.
• J.lmaxH
• J.lmaxA
• llg
• bA
• bH
• (Y~
• (Y~
The yields are included in the list, knowing that they are not easy to determine in lab-
scale tests and that they are usually assumed to be rather constant. However, it should
also be realised that the yield coefficients have an important influence on nearly all the
processes (see table 6), and therefore it would be rather relevant to have a more accurate
determination of these.
The remaining parameters can be determined via either full-scale data or directly via
the model calibration, as indicated in table 9. It is important to notice that the above
parameter list is significantly reduced compared to the list of parameters retrieved from
experiments based on table 8, basically due to the fact that the half-saturation
coefficients and hydrolysis parameters are left out.
4.4.3. Relevant wastewater components for lab-scale characterisation

Only the side of the triangle dealing with the conflict between lab-scale observations
and model concepts (conflict C) outlined in figure 22 is relevant when it comes to
characterisation of wastewater components. Also, the discussion summarised in table 9
is not relevant here, since the wastewater components do not depend on the biomass or
the WWTP operation. Therefore, the discussion on wastewater components is less
extensive here (see also the earlier discussion and summary of wastewater
characterisation methods) and is not divided according to the different components.
As discussed above in the review of wastewater characterisation a conflict may
indeed exist between the need for quantification of some of the ASM1 wastewater
171
components and what is practically obtainable from lab-scale experiments. The origin
of this problem mainly lays in the way the components are defined in ASM 1. The death
regeneration cycle and the hydrolysis processes of ASMI are model processes that are
not directly measurable in lab-scale experiments, as discussed above. Thus, the slowly
biodegradable substrate and inert particulate matter components, Xs and XI respectively,
that are related to these processes, may then also be regarded as model components that
should rather be quantified during the model calibration exercise than through dedicated
experiments. Indeed, it was proposed by Henze et at. (1995) to estimate XI in the
influent via the complete model during the calibration of the sludge balance and,
subsequently, estimate Xs from the difference between total COD and the other COD
components, as discussed earlier. A determination of the heterotrophic biomass (XB~ in
the wastewater is possible via lab-scale experiments, as described above. However, in
most cases the X BH present in wastewater is not of great importance, since the growth
rates are so high that wash-out of X BH never occurs in practice. Thus, an inclusion of
XBH in the Xs component does not affect the modelling significantly, although it will
affect the value of the heterotrophic yield coefficient (a slightly smaller yield may need
to be chosen) (Henze et at., 1999). On the contrary, the presence of autotrophic biomass
(XB,0 in the wastewater may be of importance to prevent wash out of the nitrifiers. The
concentration of XBA can in principle be detennined via lab-scale experiments, but in
practice the procedure may not be straightforward and X BA may rather be adjusted
during the model calibration.
In general there is no need for a detailed characterisation of the nitrogen components
since the main part of nitrogen in wastewater is ammonium without any coupling to the
organic matter (Henze et at., 1999). An exception to this may, however, exist for some
industrial wastewaters. Thus, the wastewater components relevant to be characterised
separately via lab-scale experiments are listed below. Again, an indicative order of
importance is not aimed for, since this will depend on the actual case.
• SNH
• Ss
• SI
• (SND, X ND)
The relevance of determining the inert soluble matter (SI) is linked to the determination
of the soluble readily biodegradable substrate (Ss) since SI may be needed for the mass
balance of soluble COD.
5. Biological experimental constraints
In the previous section the wastewater components and the stoichiometric and kinetic
parameters that are considered most relevant to be determined in lab-scale experiments
were listed. This list was compiled on the basis of considerations that the component or
parameter resulting from the lab-scale experiment should be relevant to full-scale
behaviour and fit within the model concepts.
172
In this last section, we will further zoom in on the problem of transferability between
lab-scale results and full-scale behaviour, i.e. the problem of obtaining extant kinetic
parameters. As discussed above care should be taken in the transfer of results derived
from lab-scale experiments to a model of the full-scale system. Summarising, the reason
for problems with transferability are on the one hand differences in biological
experimental conditions between lab-scale and full-scale experiments (conflict A in
figure 22) and, on the other hand, differences in the models used (conflict C in figure
22).
At the experimental level the lab-scale behaviour may not equal the full-scale
behaviour due to, for instance, differences in feeding pattern resulting in other
concentration profiles, differences in environmental conditions such as pH, temperature
or mixing behaviour, or differences in sludge history. One of the most discussed
biological experimental factors is the ratio between initial substrate concentration (So)
and initial biomass concentration (Xo). This S(O)/X(O) ratio is considered to be one of
the important factors determining (1) the response of the sludge with a certain
wastewater or substrate and (2) whether the experimental response is sufficiently
informative for adequate interpretation. The first point is of a more basic nature since it
has been observed that the S(O)/X(O) ratio directly influences the behaviour of the
sludge, leading to different characteristics (Chudoba et ai., 1992; Grady et ai., 1996,
Pollard et ai., 1998). The second point is more related to the practical identifiability of
model parameters, i.e. it affects the quality of the experimental data (Spanjers and
Vanrolleghem, 1995; Sperandio and Paul, 2000). For instance, if S(O)/X(O) is very high
the measured response, e.g. respiration rate, may be too small and the experiment may
take too long. On the other hand, if S(O)/X(O) is very low the respirometric response
may be too short for a reliable measurement, or it may be swamped into the endogenous
respiration rate. Below, special attention is paid to the frrst point, where the S(O)/X(O)
problem will be discussed in more detail.
At the modelling level the results from lab-scale experiments may be described with
a model different from the model used to describe the full-scale behaviour. Although
not obvious at first sight, the use of a simple model for interpretation of the lab-scale
data increases calculation speeds significantly, resulting in, for instance, a faster and
more straightforward parameter estimation. Problems arise when the model uses
different concepts that may not allow to transfer the estimated parameters from one
model to the other, e.g. the death regeneration versus endogenous respiration concept
(Yuan and Stenstrom, 1996).
5.1. TRANSFERABILITY BETWEEN MODEL CONCEPTS: AN EXAMPLE
In ASMI the death regeneration concept is applied, whereas the model to describe the
lab-scale results may only include the degradation of substrates, i.e. decay and death
regeneration are omitted, because they are considered insignificant in relation to the
time scale used in the experiment (Spanjers and Vanrolleghem, 1995). In ASMI oxygen
is consumed for growth on incoming substrate plus growth on substrate produced due to
death regeneration, whereas in a lab-scale model one may only consider that oxygen is
consumed for growth on incoming substrate. This is illustrated in figure 23.
In this figure the line illustrates substrate uptake rate, rs, as function of time and the
values at the left hand side of the y-axes indicate the corresponding substrate
173
concentrations Ss. The left figure illustrates how the substrate uptake rate is interpreted
in the lab-scale (batch) experiments whereas the right figure gives the ASMI
interpretation. In both cases the total oxygen consumption rate is the same, but it is
interpreted differently in the lab and full-scale model. In the lab-scale model oxygen is
consumed to degrade the incoming substrate and the substrate concentration will
eventually go to zero. Apart from oxygen for substrate degradation (ro,e,.), oxygen is
also used for endogenous respiration (rO,end)' In ASMI substrate will also be degraded.
However the concentration will not reach zero since there will be some production of
substrate from the death regeneration process. Thus, according to the ASMI model
concept oxygen will be consumed for degradation of both the incoming substrate and
the produced substrate. In figure 23 it is assumed for clarity that the concentration of the
produced substrate is 10 mg CODn. This slightly higher substrate availability in ASMI
means that the contribution of the observed total ro to degradation of incoming substrate
is lower in ASMI than in the lab-scale model. As a consequence the estimated
maximum growth rate, which is proportional to the maximum ro, will be lower in the
batch system. This is illustrated with the size of the double arrow at the right hand side
of both graphs in figure 23. Also, the value of the half-saturation coefficient Ks will be
underestimated in the batch model compared to ASMI. In the batch model this is
illustrated by a Ks value of 50 whereas it may be 55 in ASMI.
~ ~ ~ ~
Max. Substrate Uptake Rate 110
Max. Substrate Uptake Rate
(~mruJ
(~m"')
50 55 ~--.-~
Time Time
Fig. 23. Illustration of difference in interpretation of substrate uptake rate in lab-scale

(endogenous respiration, left) model versus ASMl (death regeneration, right).
As discussed earlier, it is possible to derive analytical transformations between both

model concepts for the decay and growth rates, the yield and the fraction of inert
material produced (Henze et at., 1987). However, a transformation for Ks is more
complicated.
5.2. REVIEW AND DISCUSSION OF S(O)/X(O) RATIO
Depending on the experimental conditions the organic substrate (COD) uptake rate in
both lab- and full-scale may consist of different responses. This is illustrated in figure
24. In this concept COD is produced from decay (flow 1). Maintenance (flow 2) is
defined as the external substrate requirements to maintain the organisms in their current
174
state. Note the difference here to endogenous respiration, which can be defined as the
respiration in absence of external substrate (for a detailed review see van Loosdrecht
and Henze, 1999). However, here it is assumed that external substrate is present.
Growth (flow 3) is divided in two; (i) increase in biomass due to production of cell
constituents (e.g. proteins etc.) but without cell multiplication, (ii) increase in biomass
caused by cell multiplication. Storage (flow 4) is defined as the accumulation of
polymers, e.g. poly-hydroxy-alkanoates and glycogen. Energy spilling (flow 5) (Zeng et
al., 1995) is defined as substrate waste that may take place when the organisms are
exposed to very high substrate concentrations. In such cases the organisms may not be
able to regulate the catabolism rate to the needs for anabolism, resulting in inefficient
use of substrate and possible excretion of metabolites. Figure 24 illustrates the possible
COD flows in the single organisms. Depending on the experimental conditions one of
the flows may dominate in a single organism (Fig. 24). The same experimental
conditions also provoke a particular distribution of COD over the different organisms.
Competition may eventually lead to a shift in the population (Novak et al., 1994).
Organism 4
Organism 3
Organism 2
Organism!
Decay Maintenance Growth Storage Spilling

Lysis
~
increase of cell size Cell division
Fig.24. Different flows of external COD in the organisms
As mentioned above the S(O)/X(O) ratio is considered to be one of the determining

factors for the way the organisms respond in a system. However, even though the
importance of this ratio has been recognised, only few references that deal with the
subject in more detail can be found (Chudoba et al., 1992; Novak et al. 1994; Zeng et
al., 1995; Grady et al., 1996; Liu, 1996). They all deal with the subject from a more
theoretical point of view without much experimental support, and there is still a lack of
both qualitative and especially quantitative explanation of the exact role of the
S(O)/X(O) ratio. The discussion on the effect of S(O)/X(O) can be considered from a
reaction stoichiometry or reaction kinetics point of view.
5.2.1. Effect of S(O)/X(O) on stoichiometry

Both Chudoba et al. (1992) and Liu (1996) explain the importance of the S(O)/X(O) ratio
from a thermodynamic point of view based on the observations that the observed yield
(Y' FV decreased with increasing S(O)/X(O) ratio (Fig. 25).
In the work of Chudoba et al. (1992) substrate (COD) profiles versus time were
measured. Here it was assumed that autocatalytic growth would cause substrate uptake
175
at an increasing rate whereas substrate uptake at a constant rate was assumed to be an

indirect evidence of storage. It was hypothesised that at low S(O)/X(O) ratio the main
response is storage (flow 4 in figure 24) since the energy level in the cell will be too low
to trigger cell multiplication, resulting in less substrate being oxidised (Daigger and
Grady, 1982) and thereby a higher Y'H.
At high S(O)/X(O) ratios on the contrary, the growth response where cell multiplication
(flow 3 in figure 24) is dominating results in lower observed yields (Chudoba et at.,
1992). However, the lower Y'H at higher S(O)/X(O) ratios may as well be explained by
less energy being required for growth without associated cell mUltiplication (flow 3)
and without the involvement of storage (flow 4). A second possible explanation of the
data of Chudoba et al. (1992) is that the contribution of endogenous respiration to the
total amount of oxygen consumed is higher at high S(O)/X(O) ratio. This could also
result in lower Y' H, since the experiments at high S(O)/X(O) ratios take longer time and
therefore the amount of decayed biomass (flow 1) is higher.
..
0.8 r····················································· .............................................................,
0.7 A
0.8 ••
8
~ 0."
!3 0.4
~
S 0.3
Ii
> 0.2
0.1
.SofXo :(mgCODlmgMlSS)
Fig.25A. literature data of Y aba as function of S(O )!X(O) ratio Rao and Gaudy, 1966; Data
digitised from Liu (1996).
0.45
0.4· B
0.35
"~ 0.3
£ 0.25
..l!'
~
0.2
l o..is
>
M
0.05
~--T--~-~-~
10 20 30 40 50 60
SolXo (mgCOD/mgMLSS,
Fig.25B. literature data of Yaba as function of S(O)!X(O) ratio.Chudoba et al.• 1969; Data
digitisedjrom Liu (1996).
176
A still different explanation of the decreasing observed yield with increasing S(O)IX(O)
is found in the work ofLiu (1996), who presented an attempt to quantify the importance
of S(O)IX(O). Here the decrease in Y' H is explained by an increase in energy spilling
(flow 5) with increasing S(O)IX(O) (Liu, 1996). However, the problem in verifying this
approach is to define at which S(O)IX(O) energy spilling will start to take place. In the
study of Liu (1996) the ratio is assumed to be 1. The proposed model was tested on
literature data, but the S(O)IX(O) ratios of all the literature data used in the study were
higher than 1, making the evidence for the model incomplete. It should be noted that
none of these studies attempted to explain the observed behaviour with a more complex
model, such as ASM1.
0.9 c
0.8
(""
~ D.6
. • •
; 0.5
gOA
J 0.3
D.2
0.1
O~----r-----~--~----~----~
o 10 15 20 25
SdXo (mgCODfmg"LSS,
Fig.25C. Literature data of Yob. as function of S(O)IX(O) ratio. Chang et al., 1993; Data
0.6·
0.5
D
c
~ 0 .•
i 0.3
IJ 0.2
0.1
O~ __~~--~--~--~--~__~~
o •
SofXc (ingCODfmgMLSS)
Fig.25D. Literature data of YObs as function of S(O)IX(O) ratio. Chudoba et al., 1991. Data
177
5.2.2. Effect ojS(O)IX(O) on kinetics

Another way of looking at the influence of S(O)JX(O) is from a kinetic point of view
focusing on the physiological, i.e. enzymatic, state and adaptation. In order to describe
these phenomena, the concept of the machinery necessary for protein synthesis (PSS)
has been introduced (Grady et al., 1996). This should basically be understood as
follows. If the organisms are adapted to grow under substrate limited conditions, its PSS
will not be sufficient to quickly increase the growth rate if the substrate limitation is
removed. Thus, the PSS and eventually the specific growth rate will gradually increase
during time, until the maximum possible value according to the new conditions, i.e.
physical adaptation has taken place. It has been stated that the synthesis of storage
polymers requires less physiological adaptation than the growth response (Daigger and
Grady, 1982). Thus, this would mean that if a substrate limitation is removed, as
described above, a storage response may be triggered as a fast response and as an
alternative mechanism when the growth response is too slow.
A simple example of physiological adaptation is illustrated in figure 26 where three
pulses of acetate were added consecutively to a sludge sample (Vanrolleghem et al.,
1998). Each of the three responses is characterised by a fast start-up of about two
minutes. These two minutes are assumed to be the time needed by a cell to take up fresh
substrate and oxidise it (Vanrolleghem et al., 1998). In the first two responses a more
gradual increase of ro is observed for about 10 minutes, presumably due to an increased
conversion capacity (e.g. enzyme activation or synthesis). In the third response (after 40
minutes) this capacity has become constitutive. Starvation of the culture for one night
turned the capacity down (the organisms ''forgot'') and a similar behaviour could be
observed when acetate was added again (results not shown).
1 Acetate Acetate
0.8
'2
'E
0:::: 0.6
C)
E
--; 0.4
,§
0.2
o 10 20 30 40 50 60
Time (min)
Fig.26. ro, .. profiles obtained by 3 additions of acetate to an activated sludge sample

(Vanrolleghem et aI., 1998).
In both Chudoba et al. (1992) and Liu (1996) the applied S(O)JX(O) ratios are very high
(above 1), whereas in the example of figure 26 the S(O)JX(O) ratio was very low (below
1120). It is commonly assumed that it is necessary to work under low S(O)JX(O) ratios
178
(Chudoba et al., 1992; Novak et al., 1994; Spanjers and Vanrolleghem, 1995; Gradyet
al., 1996). Indeed, if the S(O)/X(O) ratio is high this may result in a change of maximum
specific growth and substrate removal rate due to physiological adaptation, which
eventually may result in changes of the proportions among slow-growers and fast-
growers leading to population shifts (Novak et al., 1994). The kinetics measured under
such conditions will more represent the ultimate capabilities of the organisms (intrinsic
kinetics), whereas kinetics measured in experiments performed under low S(O)/X(O)
ratio may be more representative of the physiological state of the cells prior to the
experiments (extant kinetics) (Grady et al., 1996). In the example of Kappeler and
Gujer (1992) a very high S(O)/X(O) ratio was applied resulting in overestimation of the
growth rates due to shift in biomass composition towards fast-growers. In addition,
population shifts will also take place if the substrate source is changed.
5.2.3. Discussion on S(O)/X(O) ratio

As illustrated above the discussion on the effect of the S(O)/X(O) ratio is looked at from
many angles and it seems difficult to draw a coherent picture. However, instead of
focusing on a threshold value for the S(O)/X(O) ratio it may be more relevant to consider
the following factors in the discussion of what kind of response can be expected in a
lab-scale experiment:
• AS: how big is the change of substrate concentration in the lab-scale system
compared to the full-scale system, i.e. to what extent are organisms subjected
to a drastic change in their environmental conditions.
• Time: for how long is AS maintained, i.e. what is the time frame of the
experiments.
• History: how strong is the history of the sludge, e.g. starvation periods prior to
the experiment
These three factors should be understood as follows. If AS is low and the experiment is
performed over short-term, the risk for changing the response of the sludge compared to
the full-scale system is probably low and extant parameters can be obtained. If AS is
high and the time is short the risk for excess substrate uptake not resulting in immediate
growth increases (maybe induction of storage or spilling). Finally, if AS is high and the
experiment is performed over long-term the risk for physiological adaptation due to
enzymatic changes is increasing, eventually leading to a population shift. The specific
growth rate may increase during the experiment resulting in an increase in growth
response and a decrease in excess substrate uptake response, i.e. the initial stress
reaction such as storage or energy spilling will decrease as the organisms get adapted to
the new environment. Thus, somehow a compromise between AS and time is needed.
Furthermore, the history of the sludge will play a role in the experimental designs,
since for example starvation periods prior to the experiment will result in an initial
slower response of the sludge. It is, however, not really clear if for example starvation
periods can lead to an initial different response.
The above discussion on S(O)/X(O) focused on heterotrophic organisms and their
response to a carbon substrate. However, the discussion can easily be extended to
autotrophic organisms where the substrate is ammonium. In this case a too high AS may
179
result in inhibition of the nitrification process. However, the risk for a population shift
may be lower since the nitrifying group of organisms is supposed to be rather uniform
in character. Still, adaptations to new environments will take place and the bacterial
species can vary.
6. Summary
In this extensive review numerous aspects of activated sludge model calibration have
been touched upon. As an introduction the industry-standard Activated Sludge Model
No. 1 was introduced to set the scene and it was compared to the more recent update
ASM3. The wastewater and sludge fractions considered in these models were described
and the processes taking place among them were given. All these items are focused
upon when calibrating such model.
In a next section an overview was given on the descriptions of calibration
procedures that were found in literature. Surprisingly, it is not possible to find a single
paper where a comprehensive overview is given. The information is only available as
''bits and pieces" and is scattered in a vast amount of literature. The information sets
that are typically required were presented and a 10-step calibration procedure was
proposed.
The multitude of methods for model calibration was structured along three lines: (1)
wastewater characterisation, (2) sludge composition analysis and (3) stoichiometric and
kinetic parameters.
The wastewater characterisation is typically done either by physical-chemical or
biological characterisation methods. Whereas the former appear the easiest to apply,
even in routine lab analysis, their results are not directly related to the model concepts
and, moreover, the results need to be augmented with specific characteristics obtained
from biological characterisation methods. Among these biological methods attention
was particularly given to the respirometric tests as they form the core technique, but
nitrate utilisation tests and the upcoming titrimetric tests were presented as well. For the
extraction of the model-related information, either direct or model-based analysis is
needed. Whereas the former is really simple, the latter allows extracting multiple
characteristics from a single experiment.
For the sludge composition analysis, mainly in-out mass balancing methods are
being used. The estimation of stoichiometric and kinetic parameters is typically based
on dedicated batch experiments using respirometers. Special attention was drawn to the
simultaneous estimation of parameters from well-designed single experiments.
Especially for this, model-based analysis is required. It is also noteworthy that these
more complex approaches not only lead to stoichiometric and kinetic parameter
estimates, but typically also lead to estimates on wastewater composition.
In the last section of this review attention was focused upon the problem of
transferring the results of the specific tests to a model apt to describe the full-scale
behaviour. It was indeed argued that quite some estimation results give a near-perfect
description of what happened in the batch test. However this result could not be applied
in the practical situation because, for instance, the insufficiently modelled mixing
characteristics have to be lumped into the biological parameters of the full-scale model.
180
Still, it was attempted to point towards the parameters whose values can most likely be
assessed realistically from lab-scale tests and transferred to the full-scale model.
All in all, this review has led to the belief that a considerable potential exists for
efficient characterisation of Activated Sludge Models, provided that precautions are
taken with respect to constraining the experimental conditions. The PhD thesis of
Petersen (2000) was entirely devoted to this question. The thesis focused on the design
of optimal experiments that not only lead to high-information content data sets with
good identifiability properties, but that also take into account the biological constraints
to guarantee transferability of calibration results to the full-scale model.
Acknowledgement
The work reported in this paper was supported (in part) by FWD-project G.0286.96 of
the Fund for Scientific Research (Belgium), and by the Flemish Institute for the
Promotion of Scientific-Technological Research in Industry (IWT, Brussels, Belgium).
References
Avcioglu E., Orhon D. and Sozen S. (1998) A new method for the assessment of heterotrophic endogenous
respiration rate under aerobic and anoxic conditions. Water Sci. Technol., 38(8-9), 95 - 103.
Bjerre H.L., Hvitved-Iacobsen T., Teichgraber B. and te Heesen D. (1995) Experimental procedures
characterizing transformations of wastewater organic matter in the Emscher river, Germany. Water Sci.
Technol., 31(7), 201- 212.
Boero V.I., Eckenfelder W.W. Ir. and Bowers AR. (1991) Soluble microbial product formation in biological
systems. Water Sci. Technol., 23, 1067 - 1076.
Bogaert H., Vanderhasselt A, Gemaey K., Yuan Z., Thoeye C. and Verstraete W. (1997) New sensor based
on pH effects of denitrification process. J. Environ. Engineering., 123, 884 - 891.
Bortone G., Cech I.S., Germirli F., Bianchi R, and TiIche A (1994) Experimental approaches for the
characterisation of a nitrification/denitrification process on industrial wastewater. Water. Sci. Technol.,
29(7), 129 - 136.
Brands E., Liebeskind M. and Dohmann M. (1994) Parameters for dynamic simulation of wastewater
treatment plants with high-rate and low-rate activated sludge tanks. Water Sci. Technol., 30(4), 211 -
214.
Brouwer H., Klapwijk A. and Keesman J. (1998) Identification of activated sludge and wastewater
characteristics using respirometric batch-experiments. Water Res., 32, 1240 - 1254.
Bunch B. and Griffin D.M. Jr. (1987) Rapid removal of colloidal substrate from domestic wastewater. 1.
Water Pollut. Control Fed., 59, 957 - 963.
Cech 1.S., Chudoba 1. and Grau P. (1984) Determination of kinetic constants of activated sludge
microorganisms. Water Sci. Technol., 17, 259 - 272.
Chang J., Chudoba P. and Capdeville B. (1993) Determination of the maintenance requirements of activated
sludge. Water Sci. Technol., 28,139 -142.
Chudoba 1. (1969) Residual organic matter in activated sludge processes effluents V-effluent of the initial
food-to-microorganisms ratio. Sci. Paper, Inst. Chern. Technol. Prague FI-F5, 23 - 34.
Chudoba J. (1985) Quantitative estimation in COD units of refractory organic compounds produced by
activated sludge microorganisms. Water Res., 19,37 - 43.
Chudoba P., Chevalier J.J., Chang J. and Capdeville B. (1991) Effect of anaerobic stabilisation of activated
sludge on its production under batch conditions at various SdXo. Water Sci. Technol., 23, 917 - 926.
Chudoba P., Capdeville B. and Chudoba 1. (1992) Explanation of biological meaning of the SolXo ratio in
lab-scale cultivation. Water Sci. Technol., 26(3-4),743 - 751.
Ciaccio L.L. (1992) Instrumental determination of energy oxygen and BODs. Water Sci. Technol., 26(5-6),
1345-1353.
181
Coen F., Petersen B., Vanrolleghem P.A., Vanderhaegen B. and Henze M. (1998) Model-based
characterisation of hydraulic, kinetic and influent properties of an industrial WWTP. Water Sci. Technol.,
39(1), 195 - 214.
Coen F., Vanderhaegen B., Boonen I., Vanrolleghem P.A. and Van Meeuen P. (1997) Improved design and
control of industrial and municipal nutrient removal plants using dynamic models. Water Sci. Technol.,
35(10), 53 - 61.
Copp I.B. and Dold P.L. (1998) Confirming the nitrate-to-oxygen conversion factor for denitrification. Water
Res., 32, 1296 - 1304.
Daigger G.T. and Grady C.P.L. (1982) The dynamics of microbial growth on soluble substrates. A unifying
theory. Water Res., 16, 365 - 382.
De Clercq B., Coen F., Vanderhaegen B. and Vanrolleghem P.A. (1999) Calibrating simple models for mixing
and flow propagation in waste water treatment plants. Water Sci. Technol., 39(4), 61 - 69.
de la Sota A., Larrea L., Novak L., Grau P. and Henze M. (1994) Performance and model calibration of R-D-
N processes in pilot plant. Water Sci. Technol., 30(6), 355 - 364.
Dhaene R. (1996) Ben nieuwe biosensor voor het denitrificatieproces gebaseerd op een pH-regelaar. M.Sc.
thesis at Hogeschool West-Vlaanderen, Kortrijk, Belgium. (in Dutch)
Dircks K., Pind P.F., Mosbrek H. and Henze M. (1999) Yield determination by respirometry - The possible
influence of storage under aerobic conditions in activated sludge. Water SA, 25, 69 - 74.
Dold P. (1980) A general model for the activated sludge process. Prog. Wat. Tech., 12(6),47 -77.
Dupont R. and Sinkjrer O. (1994) Optimisation of wastewater treatment plants by means of computer models.
Water Sci. Technol., 30(4),181-190.
Drtil M., Nemeth P. and Bodik I. (1993) Kinetic constants of nitrification. Water Res., 27, 35 - 39.
Dupont R. and Sinlgrer O. (1994) Optimisation of wastewater treatment plants by means of computer models.
Water Sci. Technol., 30(4), 181 - 190.
Ekama G.A., Dold P.L. and Marais G.v.R. (1986) Procedures for determining influent COD fractions and the
maximum specific growth rate of heterotrophs in activated sludge systems. Water Sci. Technol., 18(6),91
-114.
Ellis T.G., Barbeau D.S., Smets B.F. and Grady C.P.L. Ir. (1996) Respirometric techniques for determination
of extant kinetic parameters describing biodegradation. Water Environ. Res., 38, 917 - 926.
Farkas P. (1981) The use of respirography in biological treatment plant control. Water Sci. Technol., 13, 125
-131.
Funamizu N. and Takakuwa T. (1994) Simulation of the operating conditions of the municipal wastewater
treatment plant at low temperatures using a model that includes the lAWPRC activated sludge model.
Water Sci. Technol., 30(4), 150 - 113.
Germirli F., Orhon D. and Artan N. (1991) Assessment of the initial inert soluble COD in industrial
wastewaters. Water Sci. Technol., 23(4-6), 1077 - 1086.
Gemaey K., Bogaert H., Massone A., Vanrolleghem P. and Verstraete W. (1997) On-line nitrification
monitoring in activated sludge with a titrimetric sensor. Environ. Sci. Technol., 31,2350 - 2355.
Gemaey K., Vanrolleghem P.A. and Verstraete W. (1998) On-line estimation of Nitrosomonas kinetic
parameters in activated sludge samples using titration in-sensor-experiments. Water Res., 32, 71 - 80.
Grady C.P.L., Smets B.F and Barbeau D.S. (1996) Variability in kinetic parameter estimates: a review of
possible causes and a proposed terminology. Water Res., 30,742 -748.
Gujer W., Henze M., Mino T. and van Loosdrecht M.C.M. (1999) Activated sludge model No.3. Water Sci.
Technol., 39(1),183 -193.
Haider S. (2000) CSB-Elimination in A-stufen und ihre Auswirkung auf die Stickstoffelimination von AB-
Anlagen unter dem Gesichtspunkt der mathematischen Modellierong. PhD. Thesis. TU Wien, Austria. (in
preparation)
Heijnen I.I., van Loosdrecht M.C.M. and Tijhnis L. (1992) A black box mathematical model to calculate
auto- and heterotrophic biomass yields based on Gibbs energy dissipation. Biotechnol. Bioeng., 40, 1139
-1154.
Henze M. (1986) Nitrate versus oxygen utilisation rates in wastewater and activated sludge systems. Water
Sci. Technol., 18(6), 115 -122.
Henze M. (1992) Characterization of wastewater for modelling of activated sludge processes. Water Sci.
Technol. 25(6), 1 - 15.
Henze M., Grady C.P.L. Ir., Gujer W., Marais G.v.R. and Matsuo T. (1987) Activated Sludge Model No.1.
lAWQ Scientific and Technical Report No. I, London, UK.
Henze M., Gujer W., Mino T., Matsuo T., Wentzel M.C.M. and Marais G.v.R. (1995) Activated Sludge
Model No.2. lAWQ Scientific and Technical Report No.3, London, UK.
182
Henze M., Gujer W., Mino T., Matsuo T., Wentzel M.C., Marais G.v.R. and van Loosdrecht M.C.M. (1999)
Activated sludge model No. 2D, ASM2D. Water Sci. Technol., 39(1), 165 - 182.
Henze M., Harremoes P., 1a Cour Janssen J. and Arvin E. (1997) Biological and chemical wastewater
treatment, 2.edition, Springer, Berlin.
Holmberg A (1982) On the practical identifiability of microbial growth models incorporating Michaelis-
Menten type nonlinearities. Mathematical Biosciences, 62, 23 - 43.
Jeppsson U. (1996). Modelling aspects of wastewater treatment processes. Ph.D. thesis: Department of
Industrial Electrical Engineering and Automation, Lund Institute of Technology, Sweden. pp. 428.
Kappeler J. and Gujer W. (1992) Estimation of kinetic parameters of heterotrophic biomass under aerobic
conditions and characterization of wastewater for activated sludge modelling. Water Sci. Technol., 25(6),
125 -139.
Keesman K.J., Spanjers H. and van Straten G. (1998) Analysis of endogenous process behaviour in activated
sludge. Biotechnol. Bioeng., 57,155 -163.
Koike 1. and Hattori A (1975) Growth yield of a denitrifying bacterium, Pseudomonas denitrificans, under
aerobic and denitrifying conditions. J. Gen. Microbiol., 88, 1 - 10.
Kong Z., Vanrolleghem P.A and Verstraete W. (1994) Automated respiration inhibition kinetics analysis
(ARlKA) with a respirographic biosensor. Water Sci. Technol., 30(4), 275 - 284.
Krishna C. and van Loosdrecht M.C.M. (1999) Substrate flux into storage and growth in relation to activated
sludge modelling. Water Res., 33, 3149 - 3161.
Kristensen H.G., Elberg Jl'lrgensen P. and Henze M. (1992) Characterisation of functional micro-organism
groups and substrate in activated sludge and wastewater by AUR, NUR and OUR. Water Sci. Technol.,
25(6),43 - 57.
Kristensen H.G., 1a Cour Janssen J. and Elberg Jl'lrgensen (1998) Batch test procedures as tools for calibration
of the activated sludge model- A pilot scale demonstration. Water Sci. Technol., 37(4-5), 235 - 242.
Kujawa K. and Klapwijk B. (1999) A method to estimate denitrification potential for predenitrificaiton
systems using NUR batch test. Water Res., 33(10), 2291 - 2300.
Larrea L., Garcia-Heras J.L., Ayesa E. and Florez J. (1992) Designing experiments to determine the
coefficients of activated sludge models by identification algorithms. Water. Sci. Technol., 25(6), 149 -
165.
Leenen E.J., Boogert AA van Lammeren AA, Tramper J. and Wijffels R.H. (1997) Dynamics of artificially
immobilized Nitrosomonas europaea: Effect of biomass decay. Biotechnol. Bioeng., 55, 630 - 641.
Lesouef A, Payraudeau M., Rogalla F. and Kleiber B. (1992) Optimising nitrogen removal reactor
confignrations by on-site calibration of the lAWPRC activated sludge model. Water Sci. Technol., 25(6),
105 -123.
Levine AD., Tchobanoglous G. and Asano T. (1985) Characterisation of the size distribution of contaminants
in wastewater: treatment and reuse implications. J. Water Pollut. Control Fed., 57(7), 805 - 816.
Liebeskind M., Schiipers D., Bornemann C., Brands E., Freund M. and Rolfs T. (1996) Parameter
determination and model fitting - two approaches for modelling processes in wastewater treatment plants.
Water Sci. TechnoL, 34(5-6), 27 - 33.
Liu Y. (1996) Bioenergetic interpretation on the SoIXo ratio in substrate-sufficient lab-scale culture. Water
Res., 30, 2766 - 2770.
Ljung L. (1987) System Identification - Theory for the User. Prentice-Hall, Englewood Cliffs, New Jersey.
Lukasse LJ.S., Keesman K.J. and van Straten G. (1997) Estimation of BODst, respiration rate and kinetics of
activated sludge. Water Res., 31, 2278 - 2286.
Majone M., Dircks K. and Beun J.J. (1999) Aerobic storage under dynamic conditions in activated sludge
processes. The state of the art. Water Sci. Technol., 39(1), 61 - 73.
Mamais D., Jenkins D. and Pitt P. (1993) A rapid physical-chemical method for the determination of readily
biodegradable soluble COD in municipal wastewater. Water Res., 27,195 -197.
Marais G.v.R. and Ekama G.A. (1976) The activated sludge process. Part 1 - Steady state behaviour. Water
SA, 2, 163 - 199.
Massone A, Gernaey K., Rozzi A., Willems P. and Verstraete W. (1995) Ammonium concentration
measurements using a titrimetric biosensor. Med. Fac. Landbouww. Univ. Gent, 60, 2361 - 2368.
McClintock S.A., Sherrard J.H., Novak J.T. and Randall C.W. (1988) Nitrate versus oxygen respiration in the
activated sludge process. J. Water Pollut. Control Fed., 60, 342 - 350.
Melcer H. (1999) Full scale experience with biological process models - calibration issues. Water Sci.
Technol., 39(1), 245 - 252.
Mino T., San Pedro D.C., Yamamoto S. and Matsuo T. (1997) Application of the lAWQ activated sludge
model to nutrient removal process. Water Sci. Techno!., 35(8), 111 - 118.
183
Miinch E.v. and Greenfield P.F. (1998) Estimating VFA concentrations in prefennenters by measuring pH.
Water Res., 32,2431 - 2441.
Naidoo V., Urbain V. and Buckley C.A. (1998) Characterisation of wastewater and activated sludge from
European municipal wastewater treatment plants using the NUR test. Water Sci. TeclmoL, 38(1), 303 -
310.
Nichols H.A., Pitman A.R. and Osborn D.W. (1985) The readily biodegradable fraction of sewage: its
influence on phosphorus removal and measurement. Water Sci. TeclmoL, 17, 73 - 87.
Novak: L., Larrea L. and Wanner J. (1994) Estimation of maximum specific growth rate of heterotrophic and
autotrophic biomass: A combined technique of mathematical modelling and batch cultivations. Water Sci.
Teclmol., 30(11),171-180.
Nowak 0., Franz A., Svardal K., Muller V. and Kulm. (1999) Parameter estimation for activated sludge
models with help of mass balances. Water Sci. Teclmol., 39(4),113 -120.
Nowak 0., Schweighofer P. and Svardal K. (1994) Nitrification inhibition - A method for the estimation of
actual maximum growth rates in activated sludge systems. Water Sci. Teclmol., 30(6), 9 - 19.
Orhon D., Artan N. and Cimsit Y. (1989) The concept of soluble residual product fonnation in the modelling
of activated sludge. Water Sci. Teclmol., 21(4-5), 339 - 350.
Orhon D., Sozen S. and Artan N. (1996) The effect of heterotrophic yield on assessment of the correction
factor for the anoxic growth. Water Sci. Teclmol., 34 (5-6), 67 - 74.
Payne W.J. (1991) Denitrification. Wiley-Interscience, New York.
Pedersen J. and Sinkjrer O. (1992) Test of the activated sludge models capabilities as a prognostic tool on a
pilot scale wastewater treatment plant. Water Sci. Teclmol., 25(6), 185 - 194.
Petersen B. (2000) Calibration, identifiability and optimal experimental design of activated sludge models.
PhD Thesis. BIOMATH Department, Ghent University, Belgium.
Pollard P.C., Steffens M.A., Biggs C.A. and Lant P.A. (1998) Bacterial growth dynamics in activated sludge
batch assays. Water Res., 32, 587 - 596.
Ramadori R., Rozzi A. and Tandoi V. (1980) An automated system for monitoring the kinetics of biological
oxidation ofanunonia Water Res., 14, 1555 -1557.
Rao B.S. and Gaudy A.F.Jr (1966) Effect of sludge concentration on various aspects of biological activity in
activated sludge. J. Water. Pollut. Control. Fed., 38, 794 - 812.
Rickert D.A. and Hunter J.V. (1971) General nature of soluble and particulate organics in sewage and
secondary effluent. Water Res., 5,421 - 436.
Robinson J.A. (1985) Determining microbial parameters using nonlinear regression analysis: Advantages and
limitations in microbial ecology. Adv. Microb. Ecol., 8, 61 - 114.
Rozzi A., Massone A. and Alessandrini A. (1997) Measurement of rbCOD as biological nitrate demand using
a biosensor: Preliminary results. In: Proceedings EEROIEFB International Symposium Environmental
Bioteclmology ISEB3. April 21-24, Ostend, Belgium.
Siegrist H. and Tschui M. (1992) Interpretation of experimental data with regard to the activated sludge model
no. 1 and calibration of the model for municipal wastewater treatment plants. Water Sci. Teclmol., 25(6),
167 -183.
Siegrist H., Brunner I., Koch G., Linh Con Phan and Van Chieu Le (1999) Reduction of biomass decay rate
under anoxic and anaerobic conditions. Water Sci. Teclmol., 39(1), 129 -137.
Smets B.F., Jobbagy A., Cowan R.M. and Grady C.P.L. Jr. (1996) Evaluation of respirometric data:
Identification of features that preclude data fitting with existing kinetic expressions. Ecotoxicology and
Environmental Safety, 33, 88 - 99.
Sollfrank U. and Gujer W. (1991) Characterisation of domestic wastewater for mathematical modelling of the
activated sludge process. Water Sci. Teclmol., 23, 1057 - 1066.
Sollfrank U., Kappeler J. and Gujer W. (1992) Temperature effects on wastewater characterization and the
release of soluble inert organic material. Water Sci. Teclmol., 25(6), 33 - 41.
Sozen A., Ubay Cokgor E., Orhon D. and Henze M. (1998) Respirometric analysis of activated sludge
behaviour -II. Heterotrophic growth under aerobic and anoxic conditions. Water Sci. Teclmol, 32(2), 476
-488.
Spanjers H. (1993) Respirometry in activated sludge. Ph.D. thesis, Landbouwuniversiteit Wageningen, the
Netherlands, 199 p.
Spanjers H. and Vanrolleghem (1995) Respirometry as a tool for rapid characterisation of wastewater and
activated sludge. Water Sci. Teclmol., 31(2), 105 - 114.
Spanjers H., Olsson G. and Klapwijk A. (1994) Determining influent short-term biochemical oxygen demand
and respiration rate in an aeration tank by using respirometry and estimation. Water Res., 28, 1571 -
1583.
184
Spanjers H., Takacs I. and Brouwer H. (1999) Direct parameter extraction from respirograms for wastewater
and biomass characterisation. Water Sci.Techno!., 39(4), 137 -145.
Spanjers H., Vanrolleghem P.A, Olsson G. and Dold P. (1998) Respirometry in control of the activated
sludge process. International Association on Water Quality, London, UK.
Sperandio M. (1998) Developpment d'une procedure de compartimentation d'une eau residuaire urbaine et
application a la mod61isation dynamique de procedes boues activ6es. PhD thesis, Institut National des
Sciences Appliqu6es de Toulouse, France, 221 p.
Sperandio M., Urbain V., Audic J.M. and Paul E. (1999) Use of carbon dioxide evolution rate for determining
heterotrophic yield and characterising denitrifying biomass. Water Sci. Techno!., 39(1), 139 - 146.
Sperandio M. and Paul E. (2000) Estimation of wastewater biodegradable COD fractions by combining
respirometric experiments in various SdXo ratios. Water Res., 34, 1233 - 1246.
Stokes L., Takacs I., Watson B. and Watts J.B. (1993) Dynamic modelling of an A.S.P. sewage works - A
case study. Water Sci. Techno!., 28(11-12),151- 161.
STOWA (1996) Methoden voor influentkarakterisering - Inventarisatie en richtlijnen. STOWA Report 80-96.
STOWA, Utrecht, The Netherlands.
Suschka J. and Ferreira E. (1986) Activated sludge respirometric measurements. Water Res., 20, 137 - 144.
Torrijos M., Cerro R.M., Capdeville B., Zeghal S., Payraudeau M. and Lesouef A (1994) Sequencing batch
reactor: A tool for wastewater characterisation for the IAWPRC mode!. Water Sci. Techno!., 29(7), 81 -
90.
Ubisi M.F., Jood T.W., Wentzel M.e. and Ekama G.A (1997) Activated sludge mixed liquor heterotrophic
active biomass. Water SA, 23, 239 - 248.
Urbain V., Naidoo V., Ginestet P. and Buckley C.A (1998) Characterisation of wastewater biodegradable
organic fraction: accuracy of the nitrate utilisation rate test. In Proceedings of the Water Environmental
Federation 71 st Annual Conference and Exposition, October 3 - 7, Orlando, Florida (USA) , 247 - 255.
Vandebroek R. (1986) Study and development of a microcomputer controlled sensor for the determination of
the biodegradability and the toxicity of wastewaters: The RODTOX. PhD. Thesis. Faculty of Agricultural
Sciences. University of Gent, Belgium. pp. 171.
Vanderhasselt A, Aspegren H., Vanrolleghem P.A and Verstraete W. (1999) Settling characterisation using
on-line sensors at a full-scale waste water treatment plant. Water SA, 25, 453 - 458.
van Haandel AC., Ekama GA, Marais G.v.R. (1981) The activated sludge process - part 3. Single sludge
denitrification. Water Res., 15, 1135 - 1152.
van Loosdrecht M.C.M. and Henze M. (1999) Maintenance, endogenous respiration, lysis, decay and
starvation. Water Sci. Techno!., 39(1), 107 - 117.
Vanrolleghem P.A and Coen F. (1995) Optimal design of in-sensor-experiments for on-line modelling of
nitrogen removal processes. Water. Sci. Techno!., 31(2),149-160.
Vanrolleghem P.A and Dochain D. (1998) Model Identification. In: Advanced Instrumentation, Data
Interpretation and Control of Biotechnological Processes, Kluwer Academic Publishers, Dordrecht, The
Netherlands, 251 - 318.
Vanrolleghem P.A and Spanjers H. (1994) Comparison of two respirometric principles for the detennination
of short-term biochemical oxygen demand. In: Proceedings 49th Purdue Industrial Waste Conference.
Lewis Pub!., Chelsea, Michigan, 177-188.
Vanrolleghem P.A. and Verstraete W. (1993) Simultaneous biokinetic characterization of heterotrophic and
nitrifying populations of activated sludge with an on-line respirographic biosensor. Water Sci. Techno!.,
28(11-12),377 - 387.
Vanrolleghem P.A, Dries D. and Verstraete W. (1990) RODTOX: biosensor for rapid determination of the
biochemical oxygen demand and the on-line monitoring of the toxicity of wastewaters. In: Proceedings
Sib European Congress on Biotechnology. Copenhagen, Denmark, July 8 - 13 1990. Vo!. 1, 161-164.
Vanrolleghem P.A, Gernaey K., Coen F., Petersen B., De Clercq B. & Ottoy J.-P. (1998) limitations of
short-term experiments designed for identification of activated sludge biodegradation models by fast
dynamic phenomena. In: Proceedings 7th IFAC Conference on Computer Applications in Biotechnology
CAB7. Osaka, Japan, May 31 - June 4 1998.
Vanrolleghem P.A., Spanjers H., Petersen B., Ginestet P. and Takacs I. (1999) Estimating (combinations of)
Activated Sludge Model No.1 parameters and components by respirometry. Water Sci. Technol, 39(1),
195 - 215.
Vanrolleghem P.A., Van hnpe J.F., Vandewalle J. and Verstraete W. (1992) Advanced monitoring and control
of the activated sludge process: On-line estimation of crucial biological variables in a structured model
with the RODTOX biosensor. In: Modeling and Control of Biotechnical Processes. Eds. Karim M.N. and
Stephanopoulos G., Pergamon Press, Oxford. 355-358.
185
Petersen B., Gemaey K., Henze M., Vanrolleghem P.A
Van Vooren L. (2000), Buffer capacity based multipurpose hard- and software sensor for environmental
applications. PhD thesis, BIOMATH department, Ghent University, Belgium.
Van Vooren L., Willems P., Ottoy J.P., Vansteenkiste G.C. and Verstraete W. (1995) Automatic buffer
capacity based sensor for effluent quality monitoring. In: Proceedings IAWQ Conference on Sensors in
Wastewater Technology, October 25 - 27 , Copenhagen, Denmark.
Vernimmen A.P., Henken E.R. and Lamb J.C. (1967) A short-term biochemical oxygen demand test. 1. Water
Pollut. Control Fed., 39, 1006 - 1020.
Volskay V.T.Jr., Grady C.P.L.Jr. and Tabak H.H. (1990) Effect of selected RCRA compounds on activated
sludge activity. Res. J. Water Pollut. Control Fed., 62, 654 - 664.
Wanner 0., Kappeler J. and Gujer W. (1992) Calibration of an activated sludge model based on human
expertise and on a mathematical optimisation technique - A comparison. Water Sci. Technol, 25(6), 141 -
148.
Weijers S.R., Kok J.J., Preisig H.A., Buunen A and Wouda T.W.M. (1996) Parameter identifiability in the
IAWQ model no. 1 for modelling activated sludge plants for enhanced nitrogen removal. In: Proceedings
6th European Symposium on Computer Aided Process Engineering ESCAPE-6, Rhodos, May 1996. pp.
6.
Wentzel M.C., Mbewe A and Ekama G.A (1995) Batch test for measurement of readily biodegradable COD
and active organism concentrations in municipal waste waters. Water SA, 21, 117 - 124.
Witteborg A, van der Last A, Hamming R. and Hemmers I. (1996) Respirometry for determination of the
influent Ss-concentration. Water Sci. Technol., 33(1), 311 - 323.
Xu S. and Hultman B. (1996) Experiences in wastewater characterisation and model calibration for the
activated sludge process. Water Sci. Technol. 33(12), 89 - 98.
Yuan W. and Stenstrom M.K. (1996) The modelling of biomass decay in aerobic activated sludge systems:
death-regeneration versus endogenous respiration. In: Proceedings 69th Annual WEF Conference and
Exposition. 73 - 82.
Yuan Z., Bogaert H. and Verstraete W. (1999) A titrimetric respirometer measuring the total nitrifiable
nitrogen. In: Proceedings 13th Forum on Applied Biotechnology, September 23 - 24, Gent, Belgium.
Med. Fac. Landbouww. Univ. Gent, 64(5a), 73 - 80.
Zeng AP. and Deckwer W.D. (1995) A kinetic model for substrate and energy consumption of microbial
growth under substrate-sufficient conditions. Biotechnol. Prog., 11,71-79.
186
OPTIMIZATION AND CONTROL OF NITROGEN REMOVAL ACTIVATED
SLUDGE PROCESSES: A REVIEW OF RECENT DEVELOPMENTS
ZHIGUO YUAN, .rURG KELLER AND PAUL LANT

The Advanced Wastewater Management Centre, The University of
Queensland, St Lucia, QW 4072, Australia, Fax: +61 733654726;
email: zhiguo@cheque.uq.edu.au
Abstract
The optimisation of biological nitrogen removal processes has attracted a lot of research
in the past few years. Considerable achievements in not only optimised process
operation and control but also improved process designs have resulted. In this paper, we
review these new developments in light of the progress they represent towards the
solution of the fundamental problems with biological nitrogen removal. It is emphasized
that, while being able to fmd the optimal or sub-optimal tradeoffs between different
objectives, on-line process control optimises a process within the constraints imposed
by the process design. The integration of innovative process design and optimised
process control represents the solution to the fundamental problems with biological
nitrogen removal.
1. Introduction
We are witnessing an enormous growth in biological nitrogen removal from

wastewater. Nitrogen removal presents specific challenges beyond traditional COD
(carbon) removal. The optimisation of biological nitrogen removal processes has
attracted a lot of research in the past few years. Considerable achievements in not only
optimised process operation and control but also improved process designs have
resulted. This paper aims at reviewing these new developments in light of the progress
they represent towards the solution of the fundamental problems with biological
nitrogen removal, and, on the basis of this, identifying new directions for the future
research.
Biological nitrogen removal takes place via the following two steps:
Biological nitrification, by which the ammonium nitrogen, either directly

contained in the wastewater or ammonified from the incoming organic
187
Zhiguo Yuan, Jiirg Keller and Paul Lant
nitrogen by heterotrophs is oxidized to nitrate nitrogen by autotrophs

(nitrifiers) under aerobic conditions; and
• Biological denitrification, by which the nitrate nitrogen is reduced to molecular
nitrogen by heterotrophs using COD as the electron donor, in the absence of
dissolved oxygen.
• Compared to the traditional COD removal activated sludge process, biological
nitrogen removal activated sludge systems have the following fundamental
differences:
• Unlike a COD removal process, which requires only the heterotrophic bacteria
functioning as long as sufficient oxygen is supplied, the nitrogen removal
process requires two types of bacteria: autotrophs and heterotrophs, which
function under conflicting conditions. Autotrophs function aerobically,
denitrifiers, the portion of the heterotrophs that denitrify, function anoxically
(dissolved oxygen is absent but nitrate or nitrite is present). A biological
nitrogen removal plant has to be operated so that both aerobic and anoxic
conditions are present in the plant and under proper control.
• Autotrophs grow slowly and therefore require a long sludge retention time.
This causes over-growth of heterotrophs and over-accumulation of inert solids,
incurring large capital cost.
• In addition to the disturbance that an ordinary COD removal plant receives, for
instance the fluctuation of influent flow rate and substrate concentrations, the
wastewater composition imposes a severe disturbance to the operation and
control of nitrogen removal plants. The removal of nitrogen depends on the
availability of COD. Unfavourable COD to nitrogen ratio in the influent often
limits the nitrogen removal efficiency of a biological nitrogen removal plant.
• An inherent problem with biological nitrogen removal is that denitrification
should naturally be preceded by nitrification, while the latter is always
accompanied by aerobic COD oxidation. A large fraction of the influent COD,
which is fed or carried over to the aerobic zone, is oxidized aerobically and is
therefore not available for denitrification. The situation becomes even worse
with simultaneous biological phosphorus removal, due to the added
competition for COD by phosphorus accumulating organisms.
The above features make the operation of a biological nitrogen removal plant difficult.
However, they have also offered greater possibilities for performance improvement by
means of optimised process design and on-line process monitoring and control, as has
been witnessed by the achievements made in the past few years.
The organization of the paper is shown in figure 1. In section 2, the single-sludge
systems, which have been overwhelmingly used for biological nitrogen removal, are
analysed, to identify their shortcomings and the opportunities for performance
improvement using on-line process control and improved process designs. The three
most important control problems identified, namely aeration control, external carbon
dosage control arid SRT control, are then addressed in sections 3 to 5, respectively.
Performance improvement by means of improVed process designs is discussed in
sections 6 and 7. Some novel process designs are presented and analysed. Conclusions
are given in section 8.
188
Optimization and control of nitrogen removal activated sludge processes: a review of recent developments
Section 2:
Elementary analysis of biological nitrogen
removal systems
Section 3: Section 4: Section 5:

Aeration control External carbon SRTcontrol
dosage control
Fig.l. Structure of the paper
2. Elementary analysis of biological nitrogen removal systems
Single-sludge systems, which have been predominantly used for biological nitrogen
removal, are analysed in this section. Their volume requirement, treatment capacity and
influent COD utilization efficiency for nitrate reduction are discussed. The analysis
aims at identifying opportunities for performance improvement of biological nitrogen
removal systems by means of on-line process control and improved process designs.
2.1. SYSTEM ANALYSIS
2.1.1. SRT and volume requirement

A characteristic of a single sludge system is that all solids, including all types of
bacteria as well as inert solids, are mixed together. Therefore they all have the same
retention time (RT) in the system. This identical time is termed as the sludge retention
time (SRT) or sludge age.
Nitrifiers grow slowly, and additionally, are sensitive to environmental changes (pH,
temperature, toxic and inhibitory compounds, etc.). Therefore, the SRT of a biological
nitrogen removal plant should be designed sufficiently long in order to secure the
nitrification process. This results in equally long retention times of heterotrophs and
inert solids, causing a large increase in the amount of MLSS (mixed liquor suspended
solids) and hence ofthe size ofthe plant. The dependency ofthe MLSS concentration in
a biological nitrogen removal plant on SRT is shown in figure 2, which is obtained
using the stoichiometric and kinetic parameters and the typical domestic wastewater
composition given in Henze et al. (1987). As shown, the amount of MLSS increases
almost linearly with SRT. The declining fraction of active biomass in MLSS indicates
that the increase of MLSS is mainly caused by inert solids.
189
-
8.0 0.5
...... 0.4 0
1/1
..J 6.0 U U 1/1
25 C III III
0
0 0.3 8 1t
.. all .2
E
~ 4.0 GI ...... .CI
m
..J nitrifier concentration
0.2 ~~~
- 0'-
:s 2.0 zoti
0.1 ~III
0.0 0.0
5 10 15 20
SAT (days)
Fig.2. Autotrophic biomass and MLSS concentrations and the fraction of active biomass in
MLSS, as afunction of SRT of a single-sludge system
2.1.2. Anoxic fraction and volume requirement

In a single-sludge system, each type of solids goes through all of the existing conditions
in the plant (aerobic, anoxic and anaerobic in case of simultaneous phosphorus
removal). At any moment, only a fraction of nitrifiers and denitrifiers are functional. In
terms of SRT design, this implies an even longer SRT than required by a fully aerobic
nitrification plant. Denoting the latter as Ox,N, the SRT of a nitrification-denitrification
single-sludge system (abbreviated as an N and DN plant below), denoted as Ox,NDN here,
is generally designed as (Henze et al., 1995),
()
() X,NDN =~
I-a
(1)
where a is the fraction of the anoxic and anaerobic volume. Therefore, the SRT and
hence the size ofthe plant is further increased by a factor of lI(l-a) due to the existence
of anoxic/anaerobic volume. O;oN = Ox,NDN (l-a) is often termed as the aerobic sludge
age of an Nand DN plant.
2.1.3. Anoxic fraction and nitrification capacity

Using equation (1) to design the SRT of an Nand DN plant, it is expected that the N
and DN plant has the same nitrification capacity as an aerobic nitrification plant with an
SRT of Ox,N' However, this is not true. Based on the fact that the two plants have the
same nitrification capacity only when the amount of nitrifiers present in the aerobic
volume of the Nand DN plant equals the total amount of nitrifiers in the aerobic
nitrification plant, the following relationship is obtained (by making mass balances for
nitrifiers in the two plants),
190
(l-a)OXNDN (l-a)OX,N
OX,N,eq ' =
= I + abA0 X,NDN (2)
where bA is the autotrophic decay rate; Ox,N,eq is the SRT of an aerobic nitrification plant
that has the equivalent nitrification capacity to an N and DN plant with an SRT of Ox,NDN
and an aerobic SRT of Ox,N. For simplicity of derivation and expression, it was assumed
that nitrifiers decay at the same rate under aerobic and anoxic conditions. Also assumed
was that the amount of assimilated nitrogen was independent of sludge age. A few
examples of the relationships among Ox,NDN, Ox,N and Ox,N,eq, calculated from equation
(2), are shown in table I (bA =0.ld- 1 was used in the calculation).
Table 1. Reduction of the nitrification capacity of a single-sludge biological nitrogen

removal plant as a function of the anoxic fraction a. Bx.NDN is the sludge age, 8x,N is the
aerobic sludge age and Bx.N,eq is the sludge age of an aerobic nitrification plant with an
equivalent nitrification capacity.
a 0.1 0.3 0.5

8x,NDN (days) 10 20 30 10 20 30 10 20 30
8x,N (days) 9 18 27 7 14 21 5 10 15
8x,N,eq (days) 8.2 15 20.8 5.4 8.8 11 3.3 5 6
Obviously, the nitrification capacity of a single-sludge nitrification-denitrification plant

is dramatically reduced by the presence of the anoxic conditions in the system. The fact
that the equivalent sludge age is significantly smaller than the aerobic sludge age
implies that the real nitrification capacity of the plant is significantly smaller than what
it is designed for. Furthermore, for any given a, there exists a maximally achievable
Ox,N,eq, which can be obtained with equation (2) with Ox,NDN = 00,
I-a
oX,N,eq,max =ab- (3)
A
For example, an N and DN plant with a =0.5 can never achieve the same nitrification
capacity of a fully aerobic plant with an SRT>lO days (assuming bA = 0.ld- 1). This
seriously limits the applicability of single-sludge systems at low temperatures.
2.1.4. Anoxic fraction and influent COD utilization efficiency for nitrate reduction
A single-sludge biological nitrogen removal plant is able to use the influent COD for
denitrification. Its influent COD utilization efficiency for nitrate reduction is analysed
below using a pre-denitrification system as an example.
Influent biodegradable COD (bCOD) consists of two parts: soluble bCOD and
particulate bCOD with fractions of p and 1-P, respectively. When contacting the sludge,
the latter is normally entrapped on sludge flocks. As the particulate bCOD has to be
hydrolysed before being degraded, the degradation of this part of bCOD proceeds
slowly. It is reasonable to assume that the particulate bCOD is equally available for both
anoxic and aerobic reactors. The fractions that are taken anoxically or aerobically
191
depend on the volume fractions, provided that the electron acceptors are readily
available in the respective reactors. In contrast, soluble bCOD is usually more available
for the anoxic reactor than for the aerobic reactor in a pre-denitrification system.
Nevertheless, a significant part of the soluble bCOD is still washed to the aerobic
reactor due to its relatively large affinity constant (half-saturation coefficient). It is
assumed here that a 1-17 fraction of the soluble influent COD is leaked to the aerobic
reactor.
Therefore, of the incoming bCOD, a fraction of 17f3+f»..1-fJ) is initially 'removed'
with nitrate as the electron acceptor and the rest (a fraction of (l-17)f3+(l-a)(l-fJ») is
initially 'removed' with oxygen as the electron acceptor.
Part (with a fraction of 1-YH ) of the initially 'removed' bCOD is oxidized to carbon
dioxide. The rest (fraction YH ) is assimilated into biomass cells or built as cell storage
products (see e.g. Majone et al., 1998), part of which is oxidized later via endogenous
respiration. YH is the short-term yield factor, which can be rather high. In IAWQ ASMI
(Henze et al., 1987), where cell assimilation is assumed, a value of 0.67 is
recommended. In IAWQ ASM3 (Gujer et al., 1998), where COD storage is used as the
mechanism for instant COD removal, a value of 0.8 is recommended. With the same
reasoning as done for particulate COD, it can be assumed that the cell COD is equally
available for aerobic and anoxic oxidation. The fraction of influent bCOD that is
oxidized via endogenous respiration is Y H - YH,ob., where YH,obs is the observed yield
factor of the plant, which is,
(4)
where bH , Bx and jp are, respectively, the heterotrophic biomass decay rate, SRT in the
plant and the fraction of inert COD contained in biomass cells (Henze et al., 1987).
Based on the above analysis, the fraction of influent bCOD used for nitrate
reduction (called the utilization efficiency) in a pre-denitrification system is obtained as,
(5)
A graphical representation of equation (5) is shown in figure 3. In the calculation,

Bx=15 days, YH = 0.75, bH =0.2,jp=0.2 and fi=0.3 were used.
The following observations are made:
• The influent COD utilization efficiency for nitrate reduction is generally low.
For example, when the anoxic fraction is 0.3, the efficiency is about 20%.
• The efficiency increases linearly with the anoxic fraction a.
• Factor 11 does not have much influence on the COD utilization efficiency.
192
~ 0.5
-
c -+-1')=0.4
.!! 0.4
oS:! - l I I - 1')=0.6
\ I) -*-1')=0.8
c 0.3
0
i 0.2
I:;:;
:::s 0.1 ~
Q
0 0.0
0
0.1 0.2 0.3 0.4 0.5
Anoxic fraction
Fig.3. Influent bCOD utilization efficiency for nitrate reduction as a function of anoxic
fraction (a.) and the fraction of the soluble influent bCOD consumed in the anoxic zone (1'))
2.1.5. Alkalinity and pH

Nitrification and denitrification have an opposite pH effect. The pH in a single-sludge
biological nitrogen removal system is thus self-regulating. The alkalinity produced by
denitrification also partially compensates its consumption by nitrification. This is
considered a nice feature of a single-sludge biological nitrogen removal system.
2.2. OPTIMIZATION OPPORTUNITIES
2.2.1. Optimisation by on-line process control

The anoxic fraction a is obviously an important operating parameter for a biological
nitrogen removal plant. A larger a increases the availability of COD for nitrate
reduction (Fig. 3) and hence improves nitrate removal, while a smaller value increases
the nitrification capacity (Table 1), improving ammonia removal. Therefore,
manipulating parameter a on-line is important for achieving the highest degree of total
nitrogen removal.
SRT is another important operating parameter of a biological nitrogen removal
plant. Minimizing SRT on-line results in a significant reduction of the amount of
suspended solids in the system, allowing the plant to take higher loads. The SRT control
system, when integrated into the plant design, allows a less conservative design of the
plant and therefore reduces the capital cost of a biological nitrogen removal plant.
Nitrate recirculation flow in a pre-denitrification biological nitrogen removal plant is
designed to recirculate nitrate from the aerobic zone to the anoxic zone. However, this
parameter on its own is not an effective on-line control handle for nitrate removal as it
influences the 'COD utilization for nitrate reduction' via its limited impact on
parameter 17, which has been shown above to be not very influential to the COD
utilization efficiency. As an alternative, external carbon addition has been developed for
193
a number of years as an effective control handle for the denitrification process in a

biological nitrogen removal system. The availability of external carbon sources has also
improved the control authority of nitrate recirculation because it guarantees that all the
recirculated nitrate is removed in the anoxic zone, unlike in the case of using influent
COD as the sole carbon source for denitrification.
As will be discussed in sections 3 to 5, a lot of research has been devoted in the past
few years to the above control problems. Considerable achievement has been obtained.
2.2.2. Optimisation by improved process design

While on-line process control is able to find the optimal or sub-optimal tradeoffs among
different objectives, it plays a limited role in solving the fundamental problems
associated with a single-sludge biological nitrogen removal system, namely its large
volume demand, low nitrification/denitrification capacity and low utilization efficiency
of influent COD for nitrate reduction.
The analysis given in section 2.1 indicates that a multi-sludge system may be
superior to a single-sludge one, as the separation of nitrifiers and denitrifiers allows
different conditions being provided to each type of bacteria.
Indeed, some multi-sludge systems have already been in use for nitrogen removal
for quite a long time (Henze et at., 1995). The most popular configuration is to have a
separate denitrification reactor treating the effluent from the aerobic nitrifying plant.
While solving the problems caused by the co-existence of anoxic and aerobic conditions
in a single reactor, the plant does not use influent COD for denitrification. The addition
of external carbon to the denitrification reactor increases the operational cost
dramatically. In addition, just like an ordinary single-sludge biological nitrogen removal
plant, autotrophs, heterotrophs and inert solids are mixed together in the nitrifying plant.
A large volume is thus still demanded. Low pH and alkalinity in the nitrifying plant
may represent another problem of the design.
There also exist plants where COD removal and nitrification are separated, for
instance an activated sludge system with an SRT that does not allow nitrification to take
place, followed by a nitrification reactor (usually a biofilter). While this type of design
greatly reduces the size of the plant and hence the capital cost, nitrate removal has again
to be done with external carbon source (Marsman et at., 1997). Although the nitrate-rich
effluent from the nitrification reactor can theoretically be recycled to the COD removal
reactor for denitrification (Balmer et at., 1998), the settlers, which are in the
recirculation loop, have to be expanded in order to accommodate the largely increased
hydraulic load, offsetting the savings gained by using smaller reactors.
Some novel multi-sludge systems have been developed in recent years. In these
systems, autotrophs and heterotrophs grow at different locations, making it possible to
provide each type of bacteria the required working conditions. Superior to the
conventional multi-sludge systems as described above, they are able to make efficient
use of influent COD for denitrification. In addition, the pH and alkalinity in these
systems are self-regulating just like in a single-sludge system.
Another novel process design consists of altering the retention times of different
particulate components in sludge, by means of providing nitrifiers from a side-stream
unit, such that nitrifiers have a significantly longer retention time than inert solids. The
194
system may therefore contain the same amount of nitrifiers as a conventional system,
but significantly less inert solids. The volume requirement is thus significantly reduced.
As will be analysed in sections 6 and 7, these systems have great potential to solve
some of the fundamental problems with single-sludge systems.
2.3. CONCLUSIONS
Mass balance analysis has revealed some inherent constraints of single-sludge

biological nitrogen removal systems. These include its large volume demand and hence
large capital cost, limited influent COD utilization efficiency for nitrate removal and
low nitrification capacity. Aeration, SRT and external carbon source addition have been
identified as effective control handles for the on-line optimisation of these systems. It is
pointed out that on-line process control optimises the processes within the constraints
imposed by process designs. The elimination of these constraints by applying novel
process designs allows further optimisation of biological nitrogen removal processes.
3. Aeration Control
Optimising the anoxic fraction a on-line by means of aeration control has been one of
the major research areas. Different from the conventional work on aeration control,
which was mainly focussed on designing appropriate control loops to control the DO at
the pre-selected set-points (see e.g. Olsson, 1976; Ko et at., 1982; Olsson et at., 1985;
Holmberg et at., 1989; Marsili-Libelli, 1989), the recent work has been devoted to the
on-line determination of the optimal aeration phase length/aerobic volume and the
optimal DO set-points for the local DO control loops (see Olsson and Newell, 1999).
A wide variety of on-line measured signals have been used in designing the control
systems. Substantially different control strategies have thus resulted. A detailed
discussion of these strategies is given in this section.
3.1. AERATION PHASE LENGTH CONTROL BASED ON ORP AND PH

MEASUREMENT
Aeration control in Sequencing Batch Reactors (SBR) and in intermittently aerated

continuous systems using Oxidation Reduction Potential (ORP) and/or pH as the
measured signals has been studied by many researchers in the past decade (see e.g.
Charpentier et at., 1987,1989; Wareham et at., 1993, 1994; Lo et at., 1994; Demuynck
et at., 1994; Sasaki et at., 1996; Bertanza, 1997; Plisson-Saune et at., 1996; Wouters-
Wasiak et at., 1994; Zipper et at., 1998; Al-Ghusain et at., 1994; 1995; Hao and Huang,
1996; Hamamoto et at., 1997; Yu et at., 1997, 1998). The control systems designed are
inferential ones, due to the fact that ORP and pH are indirect measures of the
nitrification and denitrification processes.
Figure 4 depicts typical ORP and pH profiles in an alternating aerobic-anoxic nitrogen
removal bioreactor with excess aerobic and anoxic periods. The ammonia nitrogen,
nitrate nitrogen and DO profiles are also shown in the figure.
195
Zbiguo Yuan, Jiirg Keller and Paul Lant
12 -.---------------------T 8.0
aeration off
arrrronia valley .- nitrate apex
9 7.5
aeI ......
I
r_ --Y---
I. _-----\
\ ~
L
I-
f-_
c~ 6
E
CD __
1:1)
itf \ _-~-£"-A - ~,tI
- -A 7.0
~ C A
:t::: 0 3 4. ..
.
co:
- til iI 6.5 ::J:
S.= D-
s:! c 0
> 8
Ec 6.0
--8 -3
~8
-6 ---ORP II1II N03-N 5.5
ton
aeration •• - •••• DO A NH4-N
----pH
-9 5.0
time (hour)
FigA. ORP, NHrN. NOrN, DO and pH profiles in a nitrogen removal bioreactor

(illustrative, adaptedfrom Wareham et al., 1994; Hao and Huang, 1996)
As illustrated, the ORP in the reactor rises when aeration is switched on and drops when
it is switched off. Two bending points may occur on the ORP curve: the 'ammonia break:
point', caused by a sharp DO rise due to the depletion of ammonia nitrogen in the mixed
liquor; and the 'nitrate break: point' or the 'nitrate knee', caused by the depletion of
nitrate in the reactor. Therefore, the two bending points correspond to the ends of
nitrification and denitrification, respectively.
The pH in the bioreactor also varies periodically. Unlike ORP, whose variation is
caused by the presence/absence of oxygen and nitrate, the variation of pH is caused by
the biochemical reactions of nitrification and denitrification. When aeration is switched
on, nitrification takes place, resulting in a decrease in the mixed liquor pH until the end
of nitrification. Aeration then brings pH up to a higher value, resulting in a bending
point, called the 'ammonia valley' (see figure 4) on the pH curve at the end of
nitrification. When aeration is switched off, denitrification takes place, resulting in an
increase in pH until the end of denitrification. This is then followed by a decrease of pH
caused by the anaerobic process. Another bending point, called the 'nitrate apex' is thus
formed on the pH curve at the end point of denitrification. A more detailed analysis of
the pH curve can be found in Hao and Huang (1996).
For both ORP- and pH-based aeration control systems, two types of control strategies
have been studied: namely the absolute value based control strategy and the bending
point based control strategy.
196
3.1.1. Aeration control based on absolute values ojORP and pH

The absolute value based control strategy was initiated by the fact that ORP and pH
vary in a certain range during an aerobic-anoxic cycle. The basic idea is to pre-select
two limits and to switch on/off the aeration when the limits are exceeded.
For the ORP-based controller, aeration is switched off when the measured ORP
exceeds the upper limit, and switched on when the lower limit is exceeded (see e.g.
Charpentier et al., 1987, 1989; Wouters-Wasiak et al., 1994; Zipper et al., 1998).
Ideally, the two limits should be chosen as the two ORP values of the two bending
points, which enables complete nitrification and denitrification with minimum cycle
time. A too low/high upperllower limit results in incomplete nitrification/denitrification,
while a too high/low upperllower limit results in unnecessarily long phase lengths,
reducing the treatment capacity of the plant. Charpentier et al. (1987, 1989) reported
full-scale applications of this strategy. Satisfactory nitrogen removal was achieved.
By merging the two limits into one, Lo et al. (1994) and Bertanza (1997) studied an
ORP set-point control strategy. Aeration is controlled such that the ORP is maintained
at the set-point. Unlike the two limits strategy, which results in aerobic-anoxic cycle
with typically a period of a few hours, the set-point strategy results in simultaneous
nitrification and denitrification (Bertanza, 1997). They reported a nitrogen removal rate
of 81-89% on a full-scale plant.
pH-based controllers work in a similar way. Al-Ghusain et al. (1994, 1995) applied
such a controller to an aerobic-anoxic sludge digestion process, which achieved a
nitrogen removal rate of 50%.
A major problem with the absolute value (either ORP or pH) based control strategy
is the determination of the limits or set-point. It is commonly known that ORP as
measured is not a true thermodynamic parameter. It is merely an indication of the
overall oxidative-reductive state of the system, the absolute ORP value per se does not
impart any process significance (AI-Ghusain et al., 1994). In addition, the measured
ORP values depend on the initial treatment of platinum probes, and thus depend on the
surface characteristics of the metal. Therefore, the limits and set-point required by the
controller are site and probe specific and must be determined individually. Another
problem with ORP measurement is the drifting of the signal. Significant drift of the
signal may occur within a relatively short period. Hao and Huang (1996) reported a drift
in the ORP value from about 120mV to -50mv in 40 hours. This makes it necessary to
calibrate the limits on-line. However, no work on this aspect has been reported. An idea
may be to include calibration cycles into the operation, during which excess aerobic and
anoxic periods are applied. The 'ammonia break point' and the 'nitrate knee' detected
suggest the new limits.
The pH value does not suffer the same problem. A pH value has the same meaning
regardless the system measured or the probe used. However, the choice of appropriate
limits is not at all straightforward, given the complicated variation of pH during an
aerobic-anoxic cycle (Fig. 4). pH values of 6 and 8 were chosen in Al-Ghusain et al.
(1994, 1995) as the limits for the aerobic-anoxic sludge digestion processes studied
therein. Such a large range does not seem to be applicable to an intermittently aerated
wastewater treatment process. The pH in such a process likely varies in a much
narrower range during one operation cycle, which takes typically a few hours (see Fig.
4). Indeed, guidelines for the determination of the two limits are still to be established.
197
3.1.2. Aeration control based on bending points of ORP and pH

The bending point based control strategies are designed based on the detection of the
bending points on the ORP (the 'ammonia break point' and the 'nitrate knee') or the pH
(the 'ammonia valley' and the 'nitrate apex') curves. Aeration is switched off when the
'ammonia break point' or the 'ammonia valley' is detected, and is switched on when the
'nitrate knee' or the 'nitrate apex' is detected. In this way, the lengths of the aerobic and
anoxic phases are controlled to be just sufficient for complete nitrification and
denitrification, respectively. Plisson-Saune et al. (1996) used the two ORP bending
points to control a lab-scale plant treating domestic wastewater. 89% of the influent
nitrogen was removed. Al-Ghusain et al. (1994; 1995) used the two pH bending points
to control a lab-scale sludge digestion process. Nearly complete nitrogen removal was
achieved.
The 'ammonia break point' on the ORP curve, which appears only when the DO is
subject to a sharp rise from a low level to a significantly higher one at the end of
nitrification (Wouters-Wasiak et al., 1994), is usually difficult to identify. Wareham et
al. (1993, 1994) used an aeration strategy based on the 'nitrate knee' alone: aeration is
switched on when the 'nitrate knee' is detected and the aerobic phase is set to be equal to
the previous anoxic phase. For an alternating aerobic-anoxic-anaerobic biological
nitrogen and phosphorus reactor, Sasaki et al. (1996) proposed to control the aeration
phase such that the 'nitrate knee' appears at the specified time. This was achieved by
using a feedback controller: the present aerobic period = the previous aerobic period +
K*(seCknee_time - actuaCknee_time), where K, a positive constant, was the feedback
gain. A two-month pilot plant study achieved 93% of total N removal and 90% of total
Premoval.
The concerns about the reliability of the bending-point detection have initiated the
idea of combining ORP and pH signals for the aeration control (Yu et al., 1997, 1998;
Hamamoto et al., 1997). In Yu et al. (1997, 1998), the derivatives of ORP and pH with
respect to time were calculated simultaneously. The combination of the two derivative
signals, together with an ANN (artificial neural network) predictor, which predicts the
time and magnitude of the bending points, significantly improved the reliability of the
bending point detection and hence the performance of the controller. Both the treatment
capacity and the nitrogen removal rate of the controlled plant were significantly
improved compared to a control plant with fixed phase lengths. Hamamoto et al. (1997)
adopted a different approach. The on-line measured ORP and pH signals, together with
DO and water level data, were fed to a fuzzy controller, which inferred the aeration
lengths based on linguistic rules.
With a bending point based control strategy, nitrification and denitrification come to
their ends in the aerobic and anoxic phase, respectively. This is not necessarily an
optimal strategy. For an intermittently aerated continuous system, high effluent
ammonia and nitrate peaks may appear alternatively, resulting in high effluent nitrogen
concentration, when the plant is over-loaded with nitrogen. It would be more desirable
to have the phases switched before ammonia/nitrate reaches a too low concentration that
limits the nitrification/denitrification rate. For an SBR, the cycle time is often not at the
disposal of the controller, as the plant has to take what it receives. Given the cycle time,
a complete nitrification phase may lead to a too short denitrification phase and vice
versa, resulting in effluent nitrate or ammonia peaks. Again, it is more desirable to have
198
the phases switched before ammonia/nitrate reaches a too low concentration so that a
better compromise between nitrification and denitrification is found.
3.2. AERATION CONTROL BASED ON RESPIROMETRY
Notwithstanding that respirometers and respirometry-based control of activated sludge

have attracted a large amount of research in the past few decades with considerable
progress (Spanjers et al., 1998; Vanrolleghem et al., 1998), few respirometry-based
aeration control strategies have been developed in recent years for nitrogen removal
processes.
An OUR bending point based aeration control strategy has been suggested by
several researchers (Olsson and Andrews, 1978; Demuynck et ai., 1994; Johansen et al.,
1997; Klapwijk et al., 1998). The idea is to switch off the aeration when the bending
point corresponding to the end of nitrification is detected on the OUR profile.
An obvious problem of this strategy is the reliability of the bending point detection. The
nitrification bending point may not be readily detectable on the OUR profile. Another
problem is that the OUR profile can hardly tell when to terminate the anoxic phase. For
an intermittently aerated continuous plant that receives influent only during anoxic
phase, Klapwijk et al. (1998) suggested that the anoxic phase be terminated when a
sharp rise of OUR is measured using an on-line continuous-flow respirometer. The
underlying assumption is that the consumption rate of the readily biodegradable COD
by denitrification is higher than its feeding rate as long as nitrate is present. In this case,
a sharp rise of OUR during the anoxic phase indicates an accumulation of readily
biodegradable COD caused by the cease of denitrification due to depletion of nitrate.
The applicability of the strategy is restricted by the feeding regime required.
Brouwer et al. (1998a) proposed a simple but more promising feedforward control
strategy to control the aerobic volume of a continuous activated sludge plant. A batch
experiment based respirometer was used to characterize the wastewater composition
and the sludge kinetics. The respirometer allowed the estimation of, among other
variables and parameters, the concentration of the influent nitrogen that was to be
nitrified in the treatment plant, and the maximum ammonia oxidation rate of the sludge.
The aerobic volume in the plant required to completely nitrify the incoming nitrogen
was then calculated as: Aerobic volume = influent flow rate* concentration of the
influent nitrogen to be nitrified/maximum ammonia oxidation rate.
Compared to nutrient sensors (which will be discussed below), respirometers, when
used to measure ammonia concentration, have the disadvantage of providing discrete
and considerably delayed signals. The measurement delay imposed is proportional to
the concentration measured, and typically ranges from one to a few hours. In addition, a
respirometer is not able to measure nitrate concentration. These probably explain the
limited achievement made so far using respirometer in controlling the aeration to a
nitrogen removal plant. However, respirometers have the advantage of being able to
providing sludge kinetic parameters (see e.g. Vanrolleghem et al. 1995; Brouwer et al.,
1998b), which is obviously valuable for the control system design. More research in this
direction is demanded.
199
3.3. AERATION CONTROL BASED ON AMMONIA AND NITRATE

MEASUREMENT
With the continuous improvement of reliability, accuracy and ease of maintenance of

ammonia and nitrate sensors (Thomsen and Kenneth, 1996; Londong and Wachtl,
1996), aeration control using nutrient sensors has been studied by many researchers and
full scale applications have been reported.
-----------
R2
-----------
NH.... ---,------
Fig.5. Flow path, aeration status and typical variations of ammonia and nitrate in a six-
phose BioDenipho® process cycle (after Bundgaard et al., 1989). N: nitrification; DN:
denitrification; solids curve: ammonia nitrogen concentration; dashed curve: nitrate
nitrogen concentration; circled points: switching points
3.3.1. Objective functions

Two types of objective functions have been used in the control system design. One
(called Type I Objective in the sequel) is to control the measured ammonia nitrogen
concentration (sensor normally located at the outlet of the bioreactor) at a pre-selected
set-point or within two predefmed boundaries (Balslev et al., 1996; Hoen et al., 1996;
Husmann et al., 1998). The underlying idea is to control the effluent ammonia at a
satisfactory level and at the same time to minimize aeration thus reducing the effluent
nitrate concentration by increased denitrification on the one hand, and reducing the
aeration cost on the other hand.
Another (called Type I1 Objective in the sequel) is to control the aeration such that
the effluent total nitrogen is minimized (Thornberg et al., 1993; Sorensen et al., 1994;
Nielsen and Onnerth, 1995; Potter et al., 1996; Sorensen, 1996; Onnerth et al., 1996;
200
Leeuw and van 't Oever, 1996; Lukasse et al., 1998; Isaacs and Thornberg, 1998;
Steffens and Lant, 1999). With such an objective, the effluent ammonia and nitrate
concentrations are comprised. Either nitrification or denitrification can be favoured in
defining the objective function. However, nitrification should usually be favoured,
given the fact that the amount of nitrifiers present in the system is determined by the
amount of ammonia oxidized, while the amount of heterotrophs is independent of the
amount of nitrate removed. An elevated effluent ammonia content reduces the amount
of nitrifiers, and hence the nitrification capacity of the plant. Compromising ammonia
removal for nitrate removal may eventually be detrimental to nitrate removal, as the
system may require a smaller anoxic fraction. Another reason for favouring ammonia
removal is that ammonia is toxic to water lives.
3.3.2. Aeration control of alternating systems

Aeration control to the alternating BioDenipho® (or BioDenitro™) plants has been well
studied using conventional feedback control systems based on ammonia and nitrate
measurements. The BioDenipho® is a rather complicated activated sludge nitrogen and
phosphorus removal process, with nitrification and denitrification accomplished in a
semi-batch manner by periodically changing the path of flow through two parallel
aeration tanks that are periodically aerated (Bundgaard et al., 1989). The aeration
control of this system is addressed in some details here because of the pioneering role it
played in the application of nutrient sensors to the control of wastewater treatment
systems, and its general applicability to other types of intermittently aerated systems. In
addition, this section is also intended to clarify the relationships among the various
aeration control systems reported in literature for this type of systems.
A typical BioDenipho® cycle comprises six phases, which are shown in figure 5
(Bundgaard et al., 1989). The variation of nitrate and ammonia nitrogen concentrations
in the two aeration tanks is also indicated in the figure. The aeration control laws
typically consist of the following 'switch-point' rules (potter et al., 1996):
• Rule 1: Transition from Phase A to Phase B takes place when the ammonia
concentration in reactor Rl reaches NHmax.
• Rule 2: Transition from Phase B to Phase C takes place when the nitrate
concentration in reactor Rl reaches NOmin.
• Rule 3: Transition from Phase C to Phase D takes place when the ammonia
concentration in reactor R2 reaches NHmin.
The transitions from D to E, E to F and F to A mirror the above rules. Obviously, Type
II Objective as discussed in the previous section has been adopted in the design.
To improve the robustness to external disturbance (transient loading, temperature and
pH variations, etc.), the concept of 'criteria function' has been proposed (Thornberg et
al., 1993). The idea is to determine the switching points NHmi" and NOmin on-line, so
that they are adapted according to the reactor status. An example of the criteria
functions is shown in equation (6) (potter et al., 1996),
201
Zhiguo Yuan, JUrg Keller and Paul Lant
N0min = aNH4 - N + p
(6)
NHmin = '}N03 - N +0
where ~ p, rand oare predefmed parameters, and NH4-N and NOrN are the measured
ammonia and nitrate nitrogen concentrations, respectively. The importance of the
criteria function can be illustrated with the following example. When reactor Rl
receives a high nitrogen load in Phase B, a high NOmin results so that Phase B is
terminated earlier, leaving more time for Rl to nitrify the accumulated ammonia.
Some phases in a BioDenipho® may be dropped out, resulting in slightly different
phase length control problems. Isaacs and Thornberg (1998) studied the phase length
control of a four-phase BioDenipho® process, with Phase C and Phase F (see figure 5)
left out. Due to the absence of Phase C and Phase F, Rule 2 and Rule 3, as discussed
earlier, apply simultaneously, resulting in a conflicting situation. This was resolved by
merging the two rules: transition from Phase B to Phase D takes place when the
conditions in both Rule 2 and Rule 3 are satisfied. This implies that the reactor, which
fIrst completes its task, is made to wait for the other reactor before the roles of the two
reactors are switched. Thornberg et al. (1993) studied another type of four-phase
process, where Phase A and Phase D were left out. In this process, Rule 1 was no longer
applicable. The phase lengths were controlled by Rule 2 and Rule 3. Several full-scale
applications of the control system were reported in Thornberg et al. (1993). The total
effluent nitrogen concentrations were significantly reduced, accompanied by
considerable savings of energy consumption.
In parallel to controlling the aerobic-anoxic phase lengths, the DO set-points during
the aerobic phases were also studied. A typical control law is to control DO at an on-
line determined set-point, with lower and upper boundaries (Thornberg et al., 1993;
Sorensen et al., 1994; Nielsen and Onnerth, 1995; Sorensen, 1996). The DO set-point is
calculated proportionally to the measured ammonia concentration.
3.3.3. Aeration control ofpre-denitrification systems

The above control systems have also been successfully applied to pre-denitrification
systems. By applying intermittent aeration to the aerobic zone and controlling the
aerobic and anoxic phase lengths and the DO set-points using the control system
discussed above, 30% energy consumption and 100% external carbon source were
saved on a full-scale plant with 17,000PE, accompanied by slightly reduced effluent
total nitrogen content (Nielsen and Onnerth, 1995; Onnerth et al., 1996). Similar
aeration control strategies were also studied by Balslev et al. (1996) on a pilot-scale pre-
denitrification plant.
Husmann et al. (1998) studied the aeration control of a step-feed full-scale biological
nitrogen removal plant (60,000 PE) using a simple feedback control system. The plant
consisted of two anoxic and aerobic zones, with a configuration of anoxic-aerobic-
anoxic-aerobic. The influent wastewater was fed to the two anoxic zones with a ratio
that was determined on-line. Type I Objective (see the previous section) was employed
in the control system design. A very simple control law was used to control the aeration
to the aerobic zones:
202
• the DO set-points in the aerobic zones are reduced in steps by 0.5 mgIL
starting from the fIrst aerobic zone until a concentration of 0.5 mgIL is reached
in both aerobic reactors, when the measured ammonia concentration in the
second aerobic zone is lower than the lower limit of the targeted ammonia
range (0.8 mg NIL in the reported practice).
• the DO set-points in the aerobic zones are increased in steps by 0.5 mgIL
starting from the second aerobic zone until a concentration of 2.0 mgIL is
reached in both aerobic reactors, when the measured ammonia concentration in
the second aerobic zone is higher than the upper limit of the targeted ammonia
range (1.3 mg NIL in the reported practice).
• the second anoxic zone is aerated and all wastewater is directed to the fIrst
anoxic zone when the measured ammonia concentration exceeds an extreme
value (2.0 mg NIL in the reported practice).
The control action output intervals are obviously important to the stability of the control
system. SuffIcient time must be given before a new control action is taken.
Compared to the reference lane, where DO in the two aerobic zones were constantly
controlled at 2.5 mgIL, and two-third of the wastewater were fed to the fIrst anoxic
zone, the effluent total nitrogen was reduced by 50% in summer and 33% in winter. The
aeration cost was also saved by 16%.
3.3.4. Model-based control

Some researchers have also studied model-based aeration control in the past few years
(see e.g. Hoen et al., 1996; Lukasse et al., 1998; Steffens and Lant, 1999). The designs
are typically based on a simplifIed model consisting of ammonia and nitrate dynamics.
Model predictive control was employed by Hoen et al. (1996) to control the aerobic
volume of a single-sludge post-denitrifIcation plant. The aerobic volume was controlled
such that the effluent ammonia concentration was maintained within the target range
(Type I Objective). Instead of directly using the measured ammonia concentration in the
control action calculation, as used by Husmann et al. (1998) (see the previous section),
the predicted effluent ammonia concentration was used. This approach obviously had
the advantage of more prompt control, provided that the predictions were suffIciently
accurate. The model used in the prediction was a semi-mechanistic, non-linear one
consisting of ammonia and nitrate dynamics, which was obtained by simplifying a
mechanistic model for nitrifIcation and denitrifIcation. The model involved three time-
varying rate coeffIcients that were identifIed on-line with an observer.
The approach was only illustrated with two simulation case studies. Its applicability to
real plants remains unclear. The critical part of the approach is the estimation of the
three rate coeffIcients, which are lumped factors and may therefore vary signifIcantly.
The estimation results in the simulation studies were not reported.
Lukasse et at. (1998) applied receding horizon optimal control design to the aeration
control of a completely mixed system. By choosing the intermittent aeration regime, the
control problem was simplified as to minimize, within the predicting horizon, the
deviations of the predicted effluent ammonia and nitrate nitrogen concentrations from
their targeted values (Type II Objective).
203
The model used in the prediction was a linear, semi-mechanistic one consisting of
ammonia and nitrate dynamics, obtained by neglecting other biological processes than
nitrification and denitrification, and assuming zero order nitrification and denitrification
rates.
The control system was demonstrated on a pilot plant study with the predicting
horizon being one measurement interval (20 minutes) (Lukasse et al.,1998). In a related
study, the control system was compared to a few other controllers using simulation
(Lukasse, 1999). This controller failed to outperform one with a conventional feedback
control based on ammonia measurement alone.
Steffens and Lant (1999) evaluated, by means of a simulation study, a few model-
based control designs by applying them to the control of the DO set-points in the two
aerobic reactors of a pre-denitrification biological nitrogen removal plant. The designs
evaluated included linear quadratic control (LQC), dynamic matrix control (DMC) and
non-linear optimal control (NOC). They used three criteria to compare the controllers:
operating costs, discharge costs and a process performance indicator. The last indicated
the extent of capacity creep that the process could withstand, which directly relates to
savings in deferred capital expenditure. They concluded that for all disturbance
scenarios examined, the model-based controllers outperformed the base case and PI
controllers. The major factor being that all the investigated model based controllers
provided scope for increased throughput, whereas the base case controller failed to meet
the license specs and the PI controller was operating at the constraint.
3.4. CONCLUSIONS
Aeration control has been proven to be an effective means for optimising the nitrogen
removal efficiency in a biological nitrogen removal system.
Compared to other types of sensors, nutrient sensors support the direct control of the
ammonia and nitrate nitrogen concentrations in the system. The control systems
designed based on these sensors therefore exhibit more flexibility in making comprise
between nitrification and denitrification. With the continued improvement of the
reliability and ease of maintenance of nutrient sensors, it can be expected that the
nutrient sensor based aeration control systems will be more widely used, especially in
large biological nitrogen removal plants.
While several researchers have studied model-based aeration control systems,
limited achievement has been made. No applications have been reported so far. The
bottleneck is to obtain simple process models that are identifiable, applicable to control
system design and yet characterizing the processes reasonably well.
Aeration control does not change the fact that only a fraction of nitrifiers and
denitrifiers are functional at any given moment in a single-sludge biological nitrogen
removal plant.
4. External COD dosage optimisation and control
To reject the disturbance of low influent COD to N ratio, external COD addition has
been developed as an effective means of controlling denitrification processes (see e.g.
Isaacs et al. 1994). Addition of external COD to the anoxic zone/phase significantly
204
increases the denitrification rate, and therefore enhances nitrate removal. This section
aims to review the research on external carbon sources and the control of their addition
to a biological nitrogen removal plant.
4.1. EXTERNAL CARBON SOURCES
Methanol, ethanol, and hydrolysate from fermentation of primary sludge have been the
main external carbon sources used for denitrification. They are either added to the
anoxic zone of a single-sludge system (pre- or post-denitrification) as supplement to the
influent COD, or to the denitrification tank of a two-sludge post-denitrification system.
Investigating the effectiveness of different carbon sources is obviously important for the
choice of the most appropriate carbon source.
Several researchers have made comparative studies on the effects of methanol and
ethanol as external carbon sources for denitrification (Christensson et al., 1994; Hallin
et al., 1996; Nyberg et al., 1996; Hallin and Pell, 1998). The properties compared
included the specific denitrification rates they support, the time the sludge needs to
adapt to the carbon source, the response time of the effluent nitrate to the addition of the
carbon source, and the ability of the adapted sludge to denitrify with other types of
carbon sources. The results are summarized in table 2.
The comparison shown in table 2 suggests that ethanol should be a better external
carbon source than methanol, especially when added to the denitrification zone of a
single-sludge biological nitrogen removal system. In the latter case, the influent and
external carbon is used concomitantly for denitrification and the external one is
provided only when the influent COD to N ratio is low.
Hydrolysate from fermentation of primary organic solids has also been used as
external carbon source for denitrification. Compared to using methanol and ethanol,
where 'clean' carbon is used to remove waste (nitrate) and, as a side effect, to generate
new waste (sludge production), using hydrolysate is more environmentally friendly. It
may also be more cost-effective if the operational cost for the fermentation can be
maintained low.
The hydrolysate is most often generated in a separate fermentor where desirable
conditions are provided (e.g. Aesoy and Odegaard, 1994; Brinch et al., 1994; Charlton,
1994). A yield factor (unit mass soluble organics generated per unit mass volatile
suspended solids added) of 0.06-0.25 has been reported in literature (Aesoy and
Odegaard, 1994; Brinch et al., 1994; Skalsky and Daigger, 1995; Rabinowitz and
Barnard, 1997). A large fraction of the soluble COD generated is volatile fatty acid, of
which acetic acid forms a big part, making the hydrolysate a desirable carbon source for
denitrification or for phosphorus removal. The denitrification rate using hydrolysate has
been found to be similar to that of using acetate (Kristensen and Jorgensen, 1990; Isaacs
and Henze, 1995) or ethanol (Aesoy et al., 1998). Full-scale applications of using
hydrolysate for improved denitrification have been reported (see e.g. Brinch et al.,
1994; Chalton, 1994; Rabinowitz and Barnard, 1997).
Hydrolysate may also be generated in the primary clarifier with fermentation under
the sludge blanket that is obtained by extending the sludge retention time in the clarifier
(Barnard, 1984). Christens son et al. (1998) reported an increase of 10 mgIL of readily
biodegradable COD in the primary effluent, however at the price of a significantly
increased suspended solids concentration (40%) in the primary effluent. The latter was
205
Zbiguo Yuan, Iiirg Keller and Paul Lant
caused by the high sludge blanket maintained in the clarifier (estimated to be half the
height of the clarifier). Obviously, this approach is useful only when a small amount of
extra readily biodegradable COD is needed.
Table 2: Comparison of methanol and ethanol as carbon sources for denitrification
Specific Adaptation time Response time Denitrification capacity

denitrification rate with other carbon
Methanol relatively low one sludge age days lower capacity with
carbon other than
ethanol'
Ethanol relatively high one sludge age hours higher capacity with a
(2-3 times higher large vapety of carbon
than methanol) sources
Comments denitrification adaptation time is methanol is not methanol adapted sludge
with ethanol caused by change desirable for has a lower capacity of
requires a much of microbiology in intermittent using influent COD for
smaller volume sludge dosage denitrification
References Christensson et al. Hallin et al. Christensson et al. Hallin and Pell (1998)
(1994); Nyberg et (1996); Nyberg et (1994); Nyberg et
al. (1996) al. (1996) al. (1996); Hallin
and Pell (1998)
•compared with the reference sludge to which no external carbon was added
Primary sludge may also be used as external carbon source (see e.g. Kurata et at.,
1996). Due to its high inert solid content, adding primary sludge inevitably results in a
significant increase of MLSS in the system, which may not be allowed in all cases.
4.2.VTILIZATION EFFICIENCY OF EXTERNAL COD
The supplementary external carbon may be added directly to the anoxic zone of a
single-sludge biological nitrogen removal system (with either pre- or post-
denitrification configuration), or to a separate post-denitrification reactor designed to
further denitrify the effluent from the mainstream biological nitrogen removal system.
In the latter case, a two-sludge system results. As will be analysed below, the utilization
efficiency of the dosed COD for nitrate reduction is significantly different in the two
cases.
Due to the small affinity constant (half-saturation coefficient) of the external carbon
sources, it is reasonable to assume that the carbon is 'removed' instantly after the
addition. This is especially true when the dosage rate is properly controlled (see the next
section). For a pre-denitrification system, the instant removal implies negligible leakage
of external carbon in soluble form.
In analogy to the reasoning made in section 2, the fractions of the totally added
external COD that is used for nitrate reduction in the single-sludge and two-sludge
systems, denoted as Y01Ie and Ytwo> respectively, are calculated as follows,
206
yOM = 1-(I-a)YH -aYH,obS

(7)
y two = 1- YH •obS
where ais the anoxic fraction, YH is the short-term yield and YH,obsis the observed yield
as defined in equation (4). Equation (7) indicates that the two-sludge post-denitrification
system has a significantly higher COD utilization efficiency. With 8x = 15 days,
YH = 0.75, bH =0.2 d- 1 andjp:=0.2, Yone (for a=0.1, 0.3 and 0.5) and Ytwo are calculated
and shown in table 3. Yone toytwo ratios are also shown in the table.
Table 3: COD utilization efficiencies for nitrate reduction in a two-sludge post-
denitrification system, and a single-sludge pre-denitrification system with different anoxic
fractions (0.1,0.3 and 0.5)
a 0.1 I 0.3 I 0.5

%ne 0.25 I 0.39 I 0,48
x.., 0.7
'Yone/x.., 0.36 I 0.55 I 0.68
The above analysis indicates that significantly more external carbon is needed for the
removal of the same amount of nitrate in the single-sludge system than in the two-
sludge system. Note that the extra amount of carbon demanded by a single-sludge
system is oxidized aerobically, incurring added aeration cost.
Adding the supplementary COD to the mainstream system has the advantage of not
requiring a separate reactor. However, when the supplementary COD is required in a
large amount, the great savings of carbon source and aeration cost may justify the
construction of a separate denitrification tank. Denitrification using external carbon in a
separate tank also avoids influencing the microbiology in the mainstream system. The
latter may reduce the capability of the heterotrophs to denitrify using influent COD (see
the previous section).
When the external carbon is added to the main-stream reactor, an option is to add the
carbon into a second anoxic zone that is near the end of the reactor (the volume of the
frrst one can be reduced as less nitrate is to be removed there) (Nyberg et at., 1996).
This option allows the reduction of the nitrate recirculation flow, and therefore the
reduction of oxygen that is transferred into the anoxic zone. This improves the carbon
utilization efficiency. Adding a second anoxic zone allows reducing the effluent nitrate
concentration to levels that is difficult to achieve with pre-denitrification only.
4.3. EXTERNAL CARBON DOSAGE CONTROL SYSTEMS
The dosing rate of external carbon to a biological nitrogen removal plant is important.
The dosage should guarantee a satisfactory nitrate removal, and in the mean time,
should be at a minimum level (except when excess hydrolysate is available). Dosing too
much increases the operational cost considerably due to higher carbon source
consumption, higher sludge production and increased oxygen demand. Several control
strategies have been proposed. Control systems have been designed for both single-
sludge pre-denitrification systems (Londong, 1992; Isaacs et at., 1995; Hoen et at.,
207
1996; Yuan et al., 1996, 1997; Lindberg and Carlsson, 1996; Zeghal et al., 1997;
Steffens and Lant, 1999) and two-sludge post-denitrification systems (Puznava et al.,
1998).
4.3.1. Control of external carbon dosage to recirculating biological nitrogen removal

systems
Depending on the controlled variable chosen in the design, two strategies have been
proposed for the control of external carbon dosage to a pre-denitrification system. One
is to control the effluent nitrate nitrogen concentration below a pre-specified limit
(Londong, 1992; Hoen et al., 1996) or at a set-point (e.g. Steffens and Lant, 1999)
(Strategy 1). The other is to control the nitrate nitrogen concentration in the
denitrification zone at a low set-point (Yuan et al., 1996, 1997; Lindberg and Carlsson,
1996; Zeghal et al., 1997) (Strategy /1).
Strategy /
Londong (1992) and Hoen et at. (1996) proposed to add external carbon to the anoxic
zone when the measured/predicted effluent nitrate nitrogen concentration exceeded its
limit. The dosage rate in these periods was controlled such that the COD to N ratio to
the anoxic zone was at a pre-selected value. The controller is obviously a feedforward
one. In addition to the measurement of effluent nitrate concentration, the controller also
requires measuring the influent readily biodegradable COD concentration. This strategy
has also been studied and evaluated using model-based and PID feedback controllers
(Lindberg 1998; Steffens and Lant,1999), with some simulation results reported.
While some improvement to the nitrate removal can generally be expected with this
strategy, it does not guarantee a minimum dosage of the external carbon source. The
dosage rate is determined on the basis of effluent nitrate concentration, regardless
whether or not the addition improves the nitrate removal. Addition of external carbon
obviously does not increase the denitrification rate when the nitrate concentration in the
anoxic zone is zero. The full-scale experiment reported in Regan et al. (1998) showed
that the effluent nitrate nitrogen was reduced by only 0.05-0.01 mg per mg methanol
dosed due to a too low nitrate nitrogen concentration in the denitrification zone.
Strategy II
The strategy of controlling the nitrate nitrogen concentration (SNO,AN) at a low set-point
by manipulating the external carbon dosage rate represents a solution to the above
problem (Yuan et al., 1996,1997; Lindberg and Carlsson, 1996; Zeghal et al., 1997,
1998). Controlling SNO,AN at a low but non-zero set-point guarantees the effectiveness of
external carbon on the one hand, and an (almost) complete removal of the recirculated
nitrate on the other hand, thus preventing insufficient denitrification in the anoxic zone.
Obviously, the strategy is not able to control the effluent nitrate concentration at a set-
point. This concentration varies with the influent nitrogen loading. However, as will be
shown below, the average effluent nitrate nitrogen concentration can be controlled at a
given level using this control strategy and by applying an appropriately determined
constant nitrate recirculation flow rate.
208
The choice of the set-point for SNO,AN, denoted as SNO,AN,sp, is obviously important for the
minimization of carbon addition. The denitrification rate depends on both COD and
nitrate concentrations. A lower SNO,AN,sp requires a higher COD concentration in the
anoxic zone in order to maintain a sufficient denitrification rate for the removal of the
recirculated nitrate, resulting in a larger leakage of COD (both influent and external) to
the aerobic zone. A higher SNO,AN,sp results in a lower removal rate of the recirculated
nitrate. A higher nitrate recirculation flow has to be used in order to keep the average
effluent nitrate at the required level, causing more COD leakage to the aerobic zone.
The arguments above suggest that the amount of carbon that leaks to the aerobic zone,
and hence is no longer available for denitrification, is minimal with some intermediate
value of SNO,AN. Yuan et al. (1997) investigated the determination of the optimal SNO,AN,sp
using the IAWQ ASMI (Henze et at., 1987). By investigating the dependency of the
required carbon dosage rate on SNO,AN,sp, it was suggested that 1 mg NIL be chosen as
the set-point. It was also shown that the optimal set-point is rather insensitive to model
parameters and loading conditions.
The response of SNO,AN, to the carbon dosage rate was also analysed in Yuan et at.
(1997) by means oflinear zing ASMI (Henze et at., 1987). It was found that the control
channel is approximately a first order system, which indicates that a proportional
feedback controller with a high gain could be used to control the carbon dosage without
causing oscillations in the controlled variables or loosing stability of the closed-loop
system, provided that perfect measurement of SNO,AN is available. The conclusion was
validated by simulation studies using ASMI (Henze et at., 1987). SNO,AN was tightly
controlled at its set-point.
Taking into consideration the noise and delay that are normally associated with the
nitrate measurement, another two controllers were designed in Yuan et at. (1997). Both
employed a feedforward component to release the feedback gain. One required the
measurement of the nitrate concentration in the aerobic zone to provide the feedforward
information, the other used a constant feedforward based on the average COD and
nitrogen loading to the plant. In the latter case, a non-linear feedback gain was designed
to improve the control accuracy. Both controllers were validated by simulation studies.
The latter was further evaluated by a full-scale experiment with satisfactory results.
Zeghal et at. (1997) developed a similar controller for a pre-denitrification Biostyr®
up-flow floating bioftlter system. Analysis showed that the response of SNO,AN (at the
end of the anoxic zone) to the carbon dosage can be characterized by a first order
transfer function plus a delay. A PI controller was thus designed to manipulate the
carbon dosage rate so that SNO,AN was controlled at Img NIL.
A model-based carbon dosage control system to control SNO,AN at a set-point was studied
by Lindberg and CaIsson (1996). The controller was designed based on a linear model
using the generalized minimum-variance approach (Clarke and Gawthrop, 1975). The
parameters involved in the controller were estimated on-line using a recursive least
square algorithm. In addition to SNO,AN" the controller also required the measurement of
the influent COD and the nitrate concentration in the nitrate recirculation flow. The
latter two provided information not only to the feedforward component of the controller,
but also to the parameter estimation algorithm. The controller was validated by both
simulation and pilot plant studies.
209
As mentioned, Strategy IT does not directly control the effluent nitrate nitrogen
concentration. Although it is able to keep average effluent nitrate nitrogen at the
required level, it by no means guarantees that the instant effluent nitrate will be lower
than its limit. The problem can be solved by adding a nitrate recirculation control loop,
which increases recirculation flow when the effluent nitrate limit is exceeded (Londong,
1992). The multiple-loop controller guarantees that the right amount of nitrate is
recirculated and that all the recirculated nitrate is removed.
4.3.2. Control of external carbon dosage to alternating biodenipho® systems

Isaacs et al. (1995) studied the control of external carbon dosage to an alternating
BioDenipho® system. A typical cycle of the BioDenipho® process has been shown in
Fig. 5. The system studied by Isaacs et al. (1995) consisted of only four phases (phase C
and Phase F in figure 5 were not present). External carbon was dosed during the anoxic
phase of a reactor.
The objective was to remove all nitrate initially present in a reactor when transiting
from an aerobic phase to an anoxic phase with minimum carbon dosage. The latter was
equivalent to maximize the usage of influent COD for denitrification. Therefore the
control strategy used was to control the dosage rate such that the denitrification was just
completed at the end of the anoxic phase (Isaacs et al., 1995). A feedforward controller
was designed, and demonstrated by simulation and pilot plant studies.
4.3.3. Control of external carbon to two-sludge post-denitrification systems

The control of carbon dosage to post-denitrification system presents a simpler problem,
compared to the case of single-sludge pre-denitrification systems, due to the absence of
the disturbance from influent COD. A simple solution is to use a feedforward controller,
which keeps the COD to nitrate ratio to the reactor at an appropriately chosen constant
value. A feedback component may also be included to adjust the dosage rate based on
the nitrate concentration in the effluent of the denitrification reactor.
Puznava et al. (1998) evaluated three controllers on an upflow floating biofilter
denitrification process. A feedback controller was designed to control the effluent
nitrate nitrogen concentration at a set-point (2 mg NIL). A feedforward controller was
designed to control the inlet COD to nitrate ratio. The combination of the two resulted
in the third controller. The latter two performed equally well, while the performance of
the feedback controller was less satisfactory owing to a long process delay.
4.4. CONCLUSIONS
External carbon addition has been proven to be an effective means for improving nitrate
removal in a biological nitrogen removal system. The type of the carbon source, the
location where it is added and the addition rate are important for the efficient use of the
carbon sources:
• Ethanol has been shown to be a better (but more expensive) external carbon
source than methanol, especially when added to the anoxic zone of a single-
sludge biological nitrogen removal system.
210
• When a large amount of external carbon is needed, it may be necessary to

build a separate denitrification reactor, where the external carbon is supplied to
further reduce the nitrate content in the effluent from the mainstream system.
Significant savings of the carbon source can be expected, compared to the case
where the carbon is directly added to the mainstream system.
• The carbon addition rate needs to be controlled on-line. A good strategy is to
control the outlet nitrate concentration from the denitrification zone/reactor at a
low set-point.
However, the control system does not increase the utilization efficiency of the influent
COD for nitrate removal.
5. SRT optimisation and control
Notwithstanding the importance of SRT to a nitrogen removal process, the on-line

control of this parameter has not attracted much research. In this section, the impact of
SRT on a biological nitrogen removal plant will be analysed. This will be followed by a
brief summary of an on-line SRT controller, which was designed to minimize the SRT
without risking the nitrification process (Aquafin and Severn Trent Water, 1998).
5.1. IMPACT OF SRT ON A BIOLOGICAL NITROGEN REMOVAL PLANT
The sludge retention time (SRT) is an important design and operating parameter for a
biological nitrogen removal plant. As mentioned in a previous section, the SRT of a
biological nitrogen removal plant should be designed and operated sufficiently long in
order to secure the nitrification process. Applying long SRT also has the advantage of
reducing the sludge production. On the other hand, a short SRT offers several
advantages.
A short SRT reduces the total amount of MLSS in a treatment plant. As has been
shown in figure 2, the amount of MLSS, and hence the size of the plant, increase almost
linearly with SRT. Obviously, significant capital cost can be saved if a shorter SRT can
be used in the design stage of a biological nitrogen removal plant. At the level of
operation, a smaller MLSS concentration implies a smaller loading to the secondary
settler, which is beneficial to the sludge and water separation in the settler. This further
implies that the plant may be able to take higher hydraulic loading, reducing the number
of bypassing the wastewater directly to receiving waters during wet weather periods. A
smaller MLSS concentration also reduces aeration cost due to decreased endogenous
respiration.
Moreover, a short SRT is beneficial for phosphorus removal, when it is
accomplished simultaneously with nitrogen removal (van Loosdrecht et ai., 1998). The
optimal SRT for phosphorus removal reported in literature falls in the range of 5 to 12
days (see e.g. Choi et ai., 1996; Chuang et ai., 1997; Nolasco et ai., 1998).
In addition, a shorter SRT results in 'younger' sludge. As shown in figure 2, the
active biomass to MLSS ratio increases significantly with the decrease of SRT. There
has also been evidence that a smaller SRT may result in the washout of nitrite oxidizers
so that nitrification ends up with nitrite (see e.g. Hellinga et ai., 1998; van Loosdrecht
211
and Jetten, 1998). Denitrification from nitrite requires less COD and less oxygen to
oxidize ammonia.
However, a shorter SRT leads to a higher sludge production rate.
5.2. SRT MINIMIZATION VIA SURPLUS SLUDGE WASTE FLOW CONTROL
A control system to optimise SRT on-line by means of manipulating the surplus sludge
waste flow has been reported (Aquafin and Severn Trent Water, 1998). The strategy
employed was to minimize the SRT without risking the nitrification process.
/lA,max,esti
/lA,max estimation
lorO C
a ON/OFF
CN
SNO
Qw
PLANT SNO,AE
*
product 24 hour moving ~H

average
lorO b
ON/OFF
S~H d long term average

with forgetting
Fig.6. Structure of a patent pending SRT control system (after Aquafin and Severn Trent
Water, 1998). SNO,/IN, SNO,AE are the nitrate nitrogen concentrations at the ends of anoxic and
aerobic reactors, respectively; SNH and SNB,sp are the effluent ammonia nitrogen
concentration and its set-point, respectively; CN is the nitrification capacity of the sludge; K
is a feedback gain.
The following four loops were used to optimise SRT (see figure 6):
• The feedforward loop a calculates the nominal surplus sludge waste flow based
on the maximum specific growth rate of the nitrifiers (JlA,max) , such that a
certain amount of spare nitrification capacity is provided to the plant. The
amount of the spare capacity is determined according to the variation of the
influent nitrogen load. In general, this loop generates a waste flow rate that is
significantly higher than what would be used without the control system.
JlA,max is estimated on-line using the on-line measured nitrate data (Nowak et
al., 1994; Yuan et al., 1999).
212
• The ON/OFF feedback loop b switches off the surplus sludge waste flow when
the effluent ammonia concentration (flow proportional daily average) goes
exceeds a certain level. This loop protects against estimation errors made in the
feedforward loop on the one hand, and responds to an abnormally high
nitrogen load to the plant on the other hand.
• The ON/OFF feedback loop c switches off the surplus sludge waste flow when
the nitrification capacity of the plant decreases below a certain percentage of
the moving average of this variable. This loop protects the system against a
sharp drop of the nitrification capacity due to toxicity incidents or abnormally
low nitrogen load.
• Finally, the outer proportional feedback loop d corrects control errors of the
inner loops. It adjusts the waste flow rate generated by the feedforward loop
using a proportional feedback loop with a low gain.
The control system has been evaluated using simulation and pilot plant studies and
implemented into a Nitrogen Removal Control Kit (NRC-Kit) (Aquafin and Severn
Trent Water, 1998).
5.3. CONCLUSIONS
The sludge retention time of a biological nitrogen removal plant can be minimized on-
line by means of monitoring the nitrification process. The minimization of SRT results
in less MLSS in the system, allowing the plant to take a load that is higher than
designed.
However, on-line SRT control does not change the fact that all types of solids in the
biological nitrogen removal plant have the same retention time. The resulting SRT,
though minimized, still leads to a large accumulation of inert solids in the plant.
6. Side-stream nitrifier supplies
The following two sections review the optimisation of nitrogen removal processes by
means of innovative process designs. Different from the on-line process control
technology, which optimises a process within the constraints imposed by the process
design, this approach favours changing, or eliminating the constraints of traditional
designs, and thus presents more fundamental solutions to the problems.
A large fraction of the capital cost of a biological nitrogen removal plant is caused by
the fact that all particulate components in activated sludge (autotrophs, heterotrophs and
inert solids) have the same retention time. Providing nitrifiers the required retention
time (RT) results in equally long RTs of heterotrophs and inert solids. The accumulation
of the inert solids in the system is in fact responsible for the large volume demand.
Schemes have been developed to alter the retention times of different components in
the system such that the RTs of active biomass are extended to the desired level without
raising the RTs of inert solids to the same level (RT decoupling). The volume
requirement is thus significantly reduced.
213
6.1. SHORTENING THE RTS OF INERT SOLIDS VIA SLUDGE STORAGE
Yuan et al. (1998, 2000) investigated the properties of the plant as shown in figure 7.
Different from an ordinary biological nitrogen removal plant, the plant contains a
surplus sludge storage tank. The idea is to design the main stream plant with an SRT
that allows the plant to treat the ordinary loads (including diurnal variations), while
keeping the sludge that is needed for treating shock nitrogen load and/or inhibitory/toxic
influent in the storage tank. During ordinary load periods, the surplus sludge is wasted
to the surplus sludge storage tank, which is properly aerated. The 'overflow' of that tank
then goes to the sludge treatment. During periods of nitrogen shock loads and/or
inhibitory/toxic influent, the sludge is pumped back into the main stream to temporarily
enhance the nitrification capacity.
surplus sludge return
influent
to sludge
treatment
aeration tank
surplus
sludge
sludge recycle storage
Fig.7. An activated sludge wastewater treatment plant with a surplus sludge storage tank
(after Yuan etal., 1998)
The surplus sludge storage tank is designed as follows. In order to give the plant shown
in figure 7 (called the new plant), with a main stream SRT = Ox,maim the same capability
to treat nitrogen shocks as an ordinary plant with a SRT = Ox,trad>Ox,maim the SRT in the
storage tank Ox,st, which is defined as Ox,st = Vs/Qw where Vst is the volume of the
storage tank and Qw is the waste flow rate, should be designed as (Yuan et al., 1998),
(J -(J
(J - X ,trod X ,main
(8)
x,st- b(J .+1
A X,mam
where bA is the decay coefficient of the autotrophic biomass. For instance, in order to
provide a plant with Ox,main = 10 days the same capability to counter nitrogen shocks as a
traditional plant with Ox,trati = 15 days, Ox,st is required to be 2.5 days (bA = 0.1 d- 1
assumed).
Yuan et al. (1998) mathematically proved that the overall RTs of the four particulate
components, namely autotrophs, heterotrophs, inert solids from influent and inert solids
214
produced by biomass decay, in the whole plant, denoted as ~,overall' (}A,overalh ~,overall and
(}p,overall, respectively, satisfy,
(} A,overall (} X,trati
(} H ,overall > (} X,trati

(9)
(} I ,overall = + (} X
(} X ,main ,Sf < (} X,trati
(} P ,overall < (} X ,main + (} X ,Sf < (} X,trati
Equation (9) clearly indicates that the active biomass has longer RTs than the inert
solids. For example, when Ox.main = 10 days and Ox,Sf = 2.5 days, (}A,overall, ~,overall' ~,overall
and (}P,overall are 15, 17.5, 12.5 and 11 days, respectively. This implies that, with such a
design, one may extend the retention time of the active biomass to the desired level
without rising the retention times of the inert solids to the same level. It was estimated
that the savings on the reactor volume are typically around 20% (Yuan et at., 1998).
The concept has been fully verified on a pilot plant by Yuan et at. (2000). Furthermore,
it was observed that the decay rate of the nitrifiers in the storage tank could be
maintained at an extremely low level by controlling the DO at a low level.
6.2. SIDE STREAM NITRIFICATION OF REJECT WATER
The reject water produced by sludge thickening and dewatering contains high ammonia
content. When recycled to the secondary treatment, the reject water typically represents
10-30% of the total nitrogen load to the plant (Hellinga et at., 1998; Jeavons et at.,
1998; Rosen et at., 1998). Recycling the reject water in an unbalanced manner may
cause significant fluctuations in effluent ammonia concentration (Jeavons et at., 1998).
The added oxygen demand may also cause an oxygen limitation situation in the aeration
tank (Hellinga et at., 1998), which demands expansions of the tank. To solve the
problems, processes with side-stream nitrification of the reject water have been
developed and applied to full-scale biological nitrogen removal plants (Hellinga et at.,
1998; Jeavons et at., 1998; Mossakowska et at., 1997; Rosen et at., 1998; Wett et at.,
1998). The surplus sludge waste from the side stream nitrification reactor, which is
nitrifier rich, is often recycled to the mainstream reactor (Hellinga et at., 1998; Jeavons
et at., 1998).
Kos (1998) studied the process using dynamic simulation. He concluded that, due to the
supply of nitrifiers from the side-stream reactor, the mainstream system demands a
significantly smaller SRT than what would otherwise be needed to achieve the same
degree of nitrification. A more theoretical analysis of the process is developed below.
Assuming that the mainstream plant has an SRT of Ox,maim while that of the side
stream nitrification reactor is Ox,ssn' Further assuming that:
• the daily oxidized nitrogen in the mainstream system is RN (mass Niday).

Note that RN is smaller than the influent nitrogen loading rate because part of
the loaded nitrogen is assimilated into biomass cells, and that,
• the daily oxidized nitrogen in the side stream reactor is RN,rej = o*RN (mass
Niday).
215
Mass balance analysis shows that, by wasting the surplus sludge of the side stream
system to the main stream one, the amount of nitrifiers contained in the latter IS
equivalent to that of a plant receiving the same influent but operated with an SRT of,
(10)
18
....... -+-&=0.1
(II
main stream SAT=10 days
~ -%-&=0.2
'C 16
....... ............... &=0.3
..
Ii:In
c
GI
14
~
~
12
M
10
0 5 10 15 20 25 30
Side stream SRT (days)
Fig.B. The equivalent retention time of nitrifiers in a plant with main stream SRT Ox,main =10
days and with a side stream nitrification reactor treating reject water, as a function of the
side stream SRT (Ox,sm) and the nitrogen strength of the reject water (OJ
A graphical representation of equation (10) is shown in figure 8. Obviously, the side-

stream process is more beneficial when 8 is higher and when the side stream SRT
(Ox,ssn) is lower. While parameter 8is generally not at the disposal of the designer, Ox,ssn
should be designed as small as possible. Fortunately, the reject water usually has high
temperature, making short SRT in the side stream possible. The SHARON (Single
reactor High Activity Ammonia Removal Over Nitrite) developed in Hellinga et at.
(1998) to treat reject water has an SRT of 1.5 days. The aerobic sludge age is only 1
day.
As the wasted sludge from the side stream reactor contains negligible inert solids,
feeding the sludge does not significantly alter the retention time of inert solids in the
mainstream reactor. Therefore, a regime of different retention times for different
components is formed in the mainstream system.
6.3. CONCLUSIONS
Supplying nitrifiers from a side-stream system results in shorter retention times of inert
solids than nitrifiers in a biological nitrogen removal system, allowing a significant
reduction of the volume of a biological nitrogen removal plant, and hence the capital
cost. The techniques provide a low-cost option for upgrading a COD removal plant to
nitrogen removal.
216
Control systems are required for the operation of these plants. For example, the SRT
control system presented in section 5 can be used to minimize the SRT of the side-
stream nitrification reactor treating reject water.
7. Novel multi-sludge systems
7.1. ATTACHED GROWTH PROCESSES
Providing a surface area inside reactors for the biomass to grow on has been employed
in many different ways and for several decades (e.g. trickling filters, fixed bed filters
etc). However, new processes have been emerging in recent years that try to overcome
some of the disadvantages of the existing systems, such as blocking or channelling.
These processes use biomass-growth-supporting media, either fixed or as suspended
carriers, in the reactors. The systems have been developed as high-rate COD removal
processes or as an economic means for upgrading COD removal plants to nutrient
removal. The latter is of particular interest since they provide an alternative to the
single-sludge activated sludge systems whereby a two-sludge process can be established
without additional reactors and clarifiers. As such, they offer a possible solution of the
fundamental difficulties associated with single-sludge BNR systems. The attached
growth processes (AGP) are analysed here in terms of their volume requirement,
treatment capacity, and influent COD utilization efficiency for nitrate reduction as well
as other properties.
An AGP typically has a pre-denitrification configuration, with the media added to
the aerobic reactor (Emori et al., 1994; Rusten et al., 1994; Sen et aI., 1994; Deguchi
and Kashiwaya, 1994; Morper, 1994; Takizawa et aI., 1996; Mishima et al., 1996;
Randall and Sen, 1996; Chuang et al., 1997; Matsumura et al., 1997; Aravinthan et al.,
1998; van Benthum et al., 1998b), or to both the aerobic and anoxic reactors (Deguchi
and Kashiwaya, 1994; Rusten et al., 1995a, 1995b; Takizawa et al., 1996; Su and
Ouyang, 1996; Kim et aI., 1997; Welander et al., 1997; Welander et al., 1998;
Aravinthan et al., 1998; Zhang et al., 1998). This results in two different types of
attached growth processes, one with autotrophs growing on the media but heterotrophs
growing in suspension (sometimes also called hybrid systems), and the other with both
autotrophs and heterotrophs growing on the media.
To simplify the discussion, the latter is used as an example for the analysis, a basic
structure of which is shown in figure 9. Growth media is used in all three zones. In the
anoxic (AN) zone, heterotrophs oxidize influent COD using nitrate as electron acceptor.
In the aerobic/anoxic (AA) zone, heterotrophs oxidize COD using either nitrate or
oxygen as electron acceptor. Oxygen is supplied when nitrate is not present. In the
aerobic (AE) zone, autotrophs oxidize ammonia nitrogen.
An important feature of the system is that autotrophs and heterotrophs are physically
separated. Autotrophs grow neither in the AN zone nor in the AA zone due to either the
absence of oxygen or failure to compete with heterotrophs for oxygen (Hem et al.,
1994; Boller et al., 1994; van Benthum et aI., 1998a). Similarly, heterotrophs could
hardly grow in the AE zone because of the lack of COD. Zhang et al. (1998) reported
that the number of nitrifiers in the AE zone is about 102 to 103 times higher than that of
217
Zhiguo Yuan, liirg Keller and Paul Lant
heterotrophs. Bacteria normally grow in the suspended phase as well. However, the
amount of bacteria in suspension is small compared with that on biofilm. Many systems
use growth support media in all tanks also in order not to have any sludge recycle, and
the clarifier is only used to separate the waste sludge from the effluent.
The system shown in figure 9 is analysed below in terms of its volume requirement,
treatment capacity and influent COD utilization for nitrate reduction.
influent nitrate recirculation

effluent
~
clarifier
Anoxic Aerobic/ Aerobic

(AN) Anoxic(AA) (AE)
sludge recycle (optional) sludge waste
Fig. 9. An attached growth process with polypropylene pellets used in both aerobic and
anoxic zones as bacteria-growth-supporting media
7.1.1. SRT and volume requirement

Growing on the supporting media, bacteria typically have a longer retention time than in
a conventional activated sludge system. Therefore more bacteria are maintained in the
system. However, they are accommodated in a more compact space due to the higher
solids density in a biofilm system compared to the suspended flock processes. This has
in fact been the most appealing feature of AGP's. The system also requires a much
smaller settler, particularly if there is no sludge recycle.
7.1.2. Treatment capacity

The treatment capacity of an AGP was found to be significantly higher than that of a
conventional activated sludge system. The separation of heterotrophs and autotrophs is
likely responsible for the increased treatment capacity. Both nitrifiers and denitrifiers
are only present in the zones of the reactor where they are required. This leads to a
situation where the full nitrification and denitrification capacity of the plant can be used
simultaneously, e.g. all nitrifiers in the system are functional at all times. Comparison
has shown that the maximum nitrification rate in the AE zone of an AGP can be a few
times higher than a conventional activated sludge system receiving the same nitrogen
load (see e.g. Deguchi and Kashiwaya, 1994; Randall and Sen, 1996). A similar
increase of the denitrification rate was also observed by Deguchi and Kashiwaya
(1994).
7.1.3. Influent COD utilization efficiency for nitrate reduction
218
As analysed in section 2, COD is leaked to the aerobic zones of a conventional activated

sludge system in three ways, particulate COD which is adhered to sludge flocks, cell
COD and soluble COD. Since most of the particulate and cell COD is kept in the
biofilm in an AGP, the COD leakage in the first two ways is dramatically reduced. The
leakage of soluble COD is also reduced because of the increased denitrification rate in
the AN and AA zones. Kim et al. (1997) reported that 95% of the influent suspended
solids and 80% of influent BOD are removed in an anoxic biofilter. Liu et al. (1998)
reported a similar BOD removal rate (85%).
The better availability of influent COD for denitrification implies an AGP system
requires a lower influent COD to nitrogen ratio than does a conventional activated
sludge system.
7.104. Comparison with other multi-sludge systems

Attached growth systems are often preferred to many other multi-sludge systems
because of the significantly lower complexity, particularly compared to systems with
multiple clarifiers. Furthermore, the capability of using influent COD for denitrification
as outlined above and the self-regulation of pH and alkalinity by the aerobic-anoxic
recycle are additional advantages over systems where nitrification and denitrification
are performed separately.
7.1.5. Aeration in an attached growth system

The main drawback of AGP systems has been identified as its high aeration cost (see
e.g. Welander et al., 1998). The reason for this is that a high bulk DO concentration is
needed to drive the diffusion of oxygen into the biofilm. It has been reported that bulk
DO concentrations below 3-4mgIL start limiting the nitrification rate (Takizawa et al.,
1996; Aravinthan et al., 1998; Welander et al., 1998).
However, the added aeration cost due to the increased bulk DO concentration is
partially compensated by a lower oxygen uptake rate (OUR). The OUR in the AE zone
of the system is significantly smaller than that in a conventional activated sludge
reactor, due to the lack of or reduced heterotrophic activities (COD oxidation and
endogenous respiration). On-line aeration control may further reduce the aeration cost.
The bulk DO concentration should be controlled at the minimum level achieving
complete nitrification under the specific loading. The fact that the nitrification rate is
linearly dependent on the bulk DO concentration in a rather large range (Rusten et al.,
1994; Mishima et al., 1996; Aravinthan et al., 1998; Welander et al., 1998) makes the
DO set-point an effective control handle for the improvement of nitrogen removal.
7.2. COD PRESERVATION FOR DENITRIFICATION
Similar to the attached growth systems discussed above, the DEPHANOX and similar
processes (Wanner et al., 1992; Kuba et al., 1993, 1996; Bortone et al., 1994, 1996;
Sorm et al., 1996, 1997; Jun et al., 1997) are other types of multi-sludge systems
developed in recent years. Furthermore, these systems allow for simultaneous
phosphorus removal using the same COD as for nitrate removal.
The basic structure of a DEPHANOX system is shown in figure 10 (Wanner et al.,
1992; Bortone et al., 1994, 1996; Sorm et al., 1996, 1997). In the anaerobic reactor (1),
219
the influent particulate COD is entrapped on the sludge flocks. A large fraction of the
soluble COD is also 'taken' into the sludge via different mechanisms: adsorption,
absorption or anaerobic storage. In the presence of phosphorus accumulating organisms
(PAO), the short chain fatty acids are taken up by PAO and stored as intracellular
products (PHA), accomplished by the release of phosphate from poly-phosphate. Jun et
at. (1997) reported that, after 30 minutes anaerobic contact, more than 90% of the total
COD and soluble COD were separated from the liquid phase. The intermediate settler
(2) following the anaerobic reactor separates the organic substrate-rich activated sludge
from the ammonia-rich supernatant. The supernatant then goes to an aerobic bioftlm
reactor (3) where nitrification takes place, while the settled sludge bypasses the
nitrification phase, entering the anoxic reactor (4) together with the effluent from the
bioftlm reactor. In the anoxic reactor (4), denitrification takes place. The organic
substrate contained in the sludge is oxidized by heterotrophs, including a large portion
of PAO, using nitrate as the electron acceptor. Denitrification by PAO is accompanied
by the simultaneous phosphorus uptake (see e.g. Mino et at., 1998; Meinhold et ai.,
1998). Note that, in this case, PHA stored by PAO is used for both P-uptake and nitrate
reduction. The aerobic reactor (5) allows nitrogen gas stripping from the sludge before
the latter enters the final settler (6). It also further improves P uptake and removes any
residual COD. Kuba et at. (1993, 1996) and Jun et at. (1997) implemented a similar
process using SBRs.
Similar to the attached growth systems, the DEPHANOX process is also a two-
sludge system and in fact uses attached growth media for the nitrification reactor. With
nitrification accomplished in a separate reactor, the process has similar features in terms
of volume requirement and treatment capacity. Through stimulating PAO to denitrify,
the DEPHANOX process further improves the influent COD utilization efficiency.
A reported problem of the DEPHANOX process is the high effluent ammonia
concentration (see e.g. Kuba et at., 1996). When bypassing the settled sludge from
settler (2) to reactor (4), considerable nitrogen in both soluble and particulate form is
bypassing the nitrification phase. Although part of the nitrogen is assimilated into
biomass cells during heterotrophic growth, a large fraction is directly discharged to the
effluent, resulting in high nitrogen content in the effluent.
Influent effluent
sludge bypass
sludge
sludge recycle waste
Fig.10: The DEPHANOX process: 1. Anaerobic reactor; 2. Intermediate settler; 3. Fixed-

film nitrification reactor; 4. Anoxic reactor; 5. Re-aeration reactor; 6. Final settler (after
Wanner et al., 1992)
220
Kuba et al. (1996) suggested solving the problem by reducing the ratio between the
bypassing flow and the supernatant flow (Settler 2). However, the approach has limited
effect. Reducing the ratio has no impact on the amount of the particulate nitrogen that is
fed to reactor (4). A better approach may be to add attached-growth media into reactor
(5), which will support the growth of nitrifiers, despite the small aerobic sludge age of
the suspended sludge.
7.3. CONCLUSIONS
The novel multi-sludge systems presented in this section have provided promising
solutions to the fundamental problems of single-sludge BNR systems. More research is
required to further verify these concepts and to refine the designs. On-line control of
these systems should also be studied as it offers likely significant advantages In
managing the highly variable influent loads of typical domestic treatment plants.
8. Conclusions
Biological nitrogen removal is a complex process because it involves two conflicting

processes, each process being performed by different bacteria. In most cases, these
bacteria co-exist in 'one sludge'. Elementary mass balance analysis has revealed the
following fundamental problems of single-sludge biological nitrogen removal systems:
• The co-existence of aerobic and anoxic conditions in the system reduces the
treatment capacity of the system, as only a fraction of nitrifiers and denitrifiers
are functional at any given moment. This is particularly a problem for
nitrification as the number of functioning nitrifiers is usually the limiting
factor.
• The co-existence of aerobic and anoxic conditions also results in low
utilization efficiency of influent COD for nitrate reduction, as a large fraction
of influent COD is oxidized aerobically, making it difficult to have a high
degree of nitrogen removal from wastewater with low COD to nitrogen ratio.
• The long SRT required by nitrifiers results in over-growth of heterotrophs and
over-accumulation of inert solids. Large reactor and settler volumes are thus
required, with a large increase in capital cost.
In this paper, we have shown that there are two major schools of thought for addressing
the problems, which adopt either incremental or revolutionary solutions. The
'incremental' approach is to use process control technology to attempt to get the process
to perform to its capability, within the constraints imposed by the process designs. The
revolutionary approach favours changing, or removing, the problem by eliminating the
design constraints. This is being achieved by innovative process designs.
On-line process control has been extensively studied. Put simply, the control
problem is how to determine the optimum between nitrification and denitrification on-
line, given continuous variations in loading. We have investigated control in terms of
the available and effective 'control handles', namely aeration, COD dosage and SRT.
The control strategies have been classified in terms of the measured variables used.
221
These variables may be considered as either inferential variables, such as ORP, pH and
respirometry, which are indirect measurements used to infer the key variables, or direct
measurements such as ammonia, nitrite and nitrate nitrogen concentrations.
Advances over the last ftfteen years in the control of biological nitrogen removal
have been shaped by two major advances: analyser technology and models. The
availability of robust on-line nutrient analyser technology has influenced the control
work signiftcantly, with more and more workers focussing on using these direct
measurements for control. The second major influence has been the development and
uptake of models of nitrogen removal. This has resulted in a large growth in the
application of models to the control system design. The models have been extensively
used in the analysis of the processes, which provides valuable information for the
control system designs. Simulation using the models has provided an economic means
to the preliminary veriftcation of the designs. However, model-based control of
biological nitrogen removal is still in its infancy. The difficulty has been and still is to
obtain simple yet accurate models applicable to the design of control systems.
Optimisation of the design of biological nitrogen removal systems has been shown a
critical issue for the performance improvement of biological nitrogen removal systems.
We have shown that there is a growing amount of contemporary work looking at
innovative process designs such as attached growth, multiple sludge and side-stream
nitrifter supplying systems. The common link with these concepts is the desire to
exploit the biomass behaviour, rather than treat it as a constraint as is the case with the
control work discussed above. We believe that this area is where the next major
advances in biological nitrogen removal operation will occur.
The integration of the innovative process designs with on-line process control will
result in much more efficient biological nitrogen removal systems in the future.
Acknowledgment
The authors would like to thank Prof. Peter Vanrolleghem from the BIOMATH
Department, University of Gent, Belgium, Tekn. Lic. Christian Rosen from the lEA,
Lund University, Sweden and Mr. James Lennox from the AWMC, the University of
Queensland for the fruitful discussions. They also would like to thank ir. Herwig
Bogaert from Aquaftn N.V., Belgium for his permission of using unpublished materials
in this paper. The frrst two authors thank CRC for Waste Management and Pollution
Control Ltd., Australia, for the ftnancial support provided.
References
Aesoy, A., Odegaard, H., Bach, K., Pujol, R. and Hamon, M. (1998). Denitrification in a Packed Bed Biofilm
Reactor (Biofor)- Experiments with Different Carbon Sources. Wat. Res 32(5): 1463-1470.
Al-Ghusain, I., Huang, J., Hao, O. and Lim, B. (1994). Using pH as Real-Time Control Parameter for
Wastewater Treatment and Sludge Digestion Processes. Wat. Sci. Tech. 30(4): 159-168.
Al-Ghusain, I., and Hao, J. (1995). Use of pH as Control Parameter for Aerobic/Anoxic Sludge Digestion. J.
Envir. Engrg. 121(3): 225-235.
Aquafin, N. V. and Severn Trent Water (1998). An Automatic Controller for the Surplus Sludge Waste Flow
in Nitrifying Activated Sludge Wastewater Treatment Plants. Patent Proposal BE0980377.
222
Aravinthan, V., Takizawa, S., Fujita, K. and Komatsu, K. (1998). Factors Affecting Nitrogen Removal from
Domestic Wastewater Using Immobilized Bacteria. Wat. Sci. Tech. 38(1): 193-202.
Balmer, P., Ekfjorden, L., Lumley, D. and Mattsson, A. (1998). Upgrading for Nitrogen Removal Under
Severe Site Restrictions. Wat. Sci. Tech. 37(9): 185-192.
Balslev, P., Lynggaard-Jensen, A. and Nickelsen, C. (1996). Nutrient Sensor Based Real-Time On-Line
Process Control of a Wastewater Treatment Plant Using Recirculation. Wat. Sci. Tech. 33(1): 183-192.
Barnard, 1. (1984). Activated Primary Tanks for Phosphate Removal. Water S. A. 10(3): 121-126.
Bertanza, G. (1997). Simultaneous Nitrification-Denitrification Process in Extended Aeration Plants: Pilot and
Real Scale Experiences. Wat. Sci. Tech. 35(6): 53-61.
Boller, M., Gujer, W. and Tschui, M. (1994). Parameters Affecting Nitrifying Biofllm Reactors. Wat. Sci.
Tech. 29(10-11): 1-11.
Bortone, G., Malaspina, F., Stante, L. and Tilche, A. (1994). Biological Nitrogen and Phosphorus Removal in
an AnaerobidAnoxic Sequencing Batch Reactor with Separated Biofllrn Nitrification. Wat. Sci. Tech.
30(6): 303-313.
Bortone, G., Saltarelli, R., Alonso, V., Sorm, R., Wanner, J. and Tilche, A. (1996). Biological Anoxic
Phosphorus Removal- the DEPHANOX Process. Wat. Sci. Tech. 34(1-2): 119-128.
Brinch, P., Rindel, K. and Kalb, K. (1994). Upgrading to Nutrient Removal by Means of Interual Carbon from
Sludge Hydrolysis. Wat. Sci. Tech. 29(12): 31-40.
Brouwer, H., Bloemen, M., Klapwijk, B. and Spanjers, H. (1998a). Feedforward Control of Nitrification by
Manipulating the Aerobic Volume in Activated Sludge Plants. Wat. Sci. Tech. 38(3): 245-254.
Brouwer, H., Klapwijk, A. and Keesman, K. (1998b). Identification of Activated Sludge and Wastewater
Characteristics Using Respirometric Batch-Experiments. Wat. Res. 32(4): 1240-1254.
Bundgaard, E., Andersen, K. and Petersen, G. (1989). Bio-Denitro and Bio-Denipho Systems-Experiences and
Advanced Model Development: The Danish Systems for Biological N and P Removal. Wat. Sci. Tech.
21: 1727-1730.
Charlton, J. (1994). Biological Nutrient Removal Applied to Weak Sewage. Wat. Sci. Tech. 29(12): 41-48.
Charpentier, J., Florenz, M. and David, G. (1987). Oxidation-Reduction Potential (ORP) Regulation: A Way
to Optimise Pollution Removal and Energy Savings in Low Load Activated Sludge Process. Wat. Sci.
Tech. 19: 645-655.
Charpentier, J., Godart, H., Martin, G. and Mogno, Y. (1989). Oxidation-Reduction Potential (ORP)
Regulation as a Way to Optimise Aeration and C, N and P Removal: Experimental Basis and Various
Full-scale Examples. Wat. Sci. Tech. 21(10): 1209-1223.
Choi, Y., Shin, E. and Lee, Y. (1996). Biological Phosphorus Removal from Wastewater in a Single Reactor
Combining Anaerobic and Aerobic Conditions. Wat. Sci. Tech. 34(1-2): 179-186.
Christensson, M., Lie, E. and Welander, T. (1994). A Comparison Between Ethanol and Methanol as Carbon
Sources for Denitrification. Wat. Sci. Tech. 30(6): 83-90.
Christensson, M., Lie, E., Johansson, P. and Welander, T. (1998). Increasing Substrate for Polyphosphate-
Accumulating Bacteria in Municipal Wastewater Through Hydrolysis and Fermentation of Sludge in
Primary Clarifiers. Water Environ. Res. 70(2): 138-145.
Chuang, S., Ouyang, C., Yuang, H. and You, S. (1997). Effects of SRT and DO on Nutrient Removal in a
Combined AS-Biofllm Process. Wat. Sci. Tech. 36(12): 19-27.
Clarke, D., and Gawthrop, P. (1975). Self-tuning Controller. Proc. lEE 122: 929-934.
Deguchi, H., and Kashiwaya, M. (1994). Study on Nitrified Liquor Recycling Process Operations Using
Polyurethane Foam Sponge Cubes as a Biomass Support Medium. Wat. Sci. Tech. 30(6): 143-149.
Demuynck, C., Vanrolleghem, P., Mingneau, C., Liessens, J. and Verstraete, W. (1994). NDBEPR Process
Optimisation in SBRs: Reduction of Extemal Carbon-Source and Oxygen Supply. Wat. Sci. Tech. 30(4):
169-179.
Ekama, G., and Marais, G.v.R. (1984). Biological Nitrogen Removal. Pretoria, Water Research Commission.
Emori, H., Nakamura, H., Sumino, T., Takeshima, T., Motegi, K. and Tanaka, K. (1994). High Rate and
Compact Single Sludge Pre-Denitrification Process for Retrofit. Wat. Sci. Tech. 30(6): 31-40.
Gujer, W., Henze, H., Mino., T. and van Loosdrecht, M. (1998). Activated Sludge Model No.3. Wat. Sci.
Tech. 39(1): 183-193.
Hallin, S., Lindberg, C.-F., Pell, M., Plaza, E. and Carlsson, B. (1996). Microbial Adaptation, Process
Performance and a suggested Control Strategy in a Pre-denitrification System with Ethanol Dosage. Wat.
Sci. Tech. 34(1-2): 91-99.
Hallin, S., and Pell, M. (1998). Metabolic Properties of Denitrifying Bacteria Adapting to Methanol and
Ethnol in Activated Sludge. Wat. Res. 32(1): 13-18.
223
Hamamoto, Y., Tabata, S. and Okubo, Y. (1997). Development of the Intermittent Cyclic Process for
Simultaneous Nitrogen and Phosphorus Removal. Wat. Sci. Tech. 35(1): 145-152.
Hao, 0., and Huang, J. (1996). Alternating Aerobic-Anoxic Process for Nitrogen Removal: Process
Evaluation. Water Environ. Res. 68(1): 83-93.
Hellinga, c., Schellen, A., Mulder, J., van Loosdrecht, M. and Heijnen J. (1998). The SHARON Process: an
Innovative Method for Nitrogen Removal from Ammonium-rich Waste Water. Wat. Sci. Tech. 37(9):
135-142.
Hem, L., Rusten, B. and Odegaard, H. (1994). Nitrification in a Moving Bed Biofilm Reactor. Wat. Res.
28(6): 1425-1433.
Henze, M., Grady Jr, c., Gujer, W., Marais, G., Matsuo, T. (1987). Activated Sludge Model No. I. London,
lAWQ.
Henze, M., Harremoes, P., la Cour Jansen, 1. and Arvin, E. (1995). Wastewater Treatment. London, Springer.
Hoen, K., Schuhen, M. and Kohne, M. (1996). Control of Nitrogen Removal in Wastewater Treatment Plants
with Predenitrification, Depending on the Actual Purification Capacity. Wat. Sci. Tech. 33(1): 223-236.
Holmberg, U., Olsson, G. and Andersson, B. (1989). Simultaneous DO control and Respiration Estimation.
Wat. Sci. Tech. 21: 1185-1195.
Husmann, M., Ortb, H., Schlegel, S. and Teichgraber, B. (1998). Application of Process Control for Improved
Nitrogen RemovaL Wat. Sci. Tech. 38(3): 263-269.
Isaacs, S., Henze, M., Soeberg, H. and Kummel, M. (1994). External Carbon Source Addition as a Means to
Control an Activated Sludge Nutrient Removal Process. Wat. Res. 28(3): 511-520.
Isaacs, S., and Henze, M. (1995). Controlled Carbon Source Addition to an Alternating Nitrification-
Denitrification Wastewater Treatment Process Including Biological P Removal. Wat. Res. 29(1): 77-89.
Isaacs, S., Henze, M. and Kummel, M. (1995). An Adaptive Algorithm for External Carbon Addition to an
Alternating Activated Sludge Process for Nutrient Removal from Wastewater. Chern. Engng. Science
50(4): 617-629.
Isaacs, S., and Thornberg, D. (1998). Rule Based Control of a Periodic Activated Sludge Process. Wat. Sci.
Tech. 38(3): 281-289.
Jeavons, J., Stokes, J., Upton, 1. and Bingley, M. (1998). Successful Sidestream Nitrification of Digested
Sludge Liquors. Wat. Sci. Tech. 38(3): 111-118.
Johansen, N., Andersen, J. and la Cour Jansen, J. (1997). Optimum Operation of a Small Sequencing Batch
Reactor for BOD and Nitrogen Removal Based on On-line OUR-Calculation. Wat. Sci. Tech. 35(6): 29-
36.
Jun, H. B., Lee, S. H., Seo, I. S. and Moon, M. J. (1997). Denitrification by the Preserved Influent COD in a
Separated Sequencing Batch Reactors System. Biological Nutrient Removal 3, Brisbane, Australia.
Kim, Y., Mikawa, T., Tanaka, K. and Emori, H. (1997). Development of Novel Anaerobic/Aerobic Filter
Process for Nitrogen Removal Using Immobilized Nitrifier Pellets. Wat. Sci. Tech. 36(12): 151-158.
Klapwijk, A., Brouwer, H., Vrolijk, E. and Kujawa, K. (1998). Control of Intermittently Aerated Nitrogen
Removal Plants by Detection Endpoints of Nitrification and Denitrification Using Respirometer Only.
Wat. Res. 32(5): 1700-1703.
Ko, K., Mcinnis, B. and Goodwin, G. (1982). Adaptive Control and Identification of the Dissolved Oxygen
Process. Automatica 18(6): 727-730.
Kos, P. (1998). Short SRT (Solids Retention Time) Nitrification ProcesslFlowsheet. Wat. Sci. Tech. 38(1):
23-29.
Kuba, T., Smolders, G., van Loosdrecht, M. and Heijnen, J. (1993). Biological Phosphorus Removal from
Wastewater by Anaerobic-Anoxic Sequencing Batch Reactor. Wat. Sci. Tech. 27(5-6): 241-152.
Kuba, T., van Loosdrecht, M. and Heijnen, J. (1996). Phosphorus and Nitrogen Removal with Minimal COD
Requirement by Integration of Denitrifying Dephosphatation and Nitrification in a Two-Sludge System.
Wat. Res. 30(7): 1702-1710.
Kurata, G., Tsumura, K., Nakamura, S., Kuwahara, W., Sato, A. and Kanaya, T. (1996). Retrofit of Biological
Nutrient Removal Process Assisted by Numerical Simulation with Activated Sludge Model No.2. Wat.
Sci. Tech. 34(1-2): 221-228.
Larose, A., Perrier, M. and Comeau, Y. (1997). Respirometric Control of the Anaerobic Duration of an SBR
Bio-P Process. Wat. Sci. Tech. 36(5): 293-300.
Leeuw, E., and van't Oever (1996). Process Selection, Design and Operation of the EDE WWTP. Wat. Sci.
Tech. 33(12): 57-63.
Lindberg, C.-F, and Carlsson, B. (1996). Adaptive Control of External Carbon Flow Rate in an Activated
Sludge Process. Wat. Sci. Tech. 34(3-4): 173-180.
224
Lindberg, C.-F. (1998). Multivariable Modelling and Control of an Activated Sludge Process. Wat. Sci. Tech.
37: 149-156.
Liu, J., Li, W., Wang, X., Liu, H. and Wang, B. (1998). Removal of Nitrogen from Coal Gasification by
Nitrosofication and Denitrosofication. Wat. Sci. Tech. 38(1): 39-46.
Lo, c., Yu, C., Tam, N. and Traynor, S. (1994). Enhanced Nutrient Removal by Oxidation-Reduction
Potential (ORP) Controlled Aeration in a Laboratory Scale Extended Aeration Treatment System. Wat.
Res. 28(10): 2087-2094.
Londong, J. (1992). Strategies for Optimised Nitrate Reduction with Primary Denitrification. Wat. Sci. Tech.
26(5-6): 1087-1096.
Londong, J., and Wachtl, P. (1996). Six Years of Practical Experience with the Operation of On-Line
Analysers. Wat. Sci. Tech. 33(1): 159-164.
Lukasse, L., Keesman, K., Klapwijk, A. and van Straten G. (1998). Optimal Control ofN-Removal in ASPs.
Wat. Sci. Tech. 38(3): 255-262.
Lukasse, L. (1999). Control and Identification in Activated Sludge Processes. Wageningen, Ph.D. Thesis,
Wageningen Agricultural University: 155p.
Majone, M., Dircks, K. and Beun, J. J. (1998). Aerobic Storage Under Dynamic Conditions in Activated
Sludge Process. The State of the Art. Wat. Sci. Tech. 39(1): 61-73.
Marsili-Libelli, S. (1989). Modelling, Identification and Control of the Activated Sludge Process. Advances in
Biochemical EngineeringlBiotechnology. A. Fiechter (Ed.). Berlin Heidelberg, Springer-Verlag. 38: 89-
148.
Marsman, E., Roeleveld, P. and Rensink, J. (1997). High Nutrient Removal in the Three-Sludge Sewage
Treatment System: Results and Economic Evaluation. Wat. Sci. Tech. 35(10): 129-136.
Matsumura, M., Yamamoto, T., Wang, P., Shinabe, K. and Yasuda, K. (1997). Rapid Nitrification with
Immobilized Cell Using Macro-Porous Cellulose Carrier. Wat. Res. 31(5): 1027-1034.
Meinhold, 1., Filipe, C. D. M., Daigger, G. T. and Isaacs, S. (1998). Characterisation of the Denitrifying
Fraction of Phosphate Accumulating Organisms in Biological Phosphate Removal. Wat. Sci. Tech. 39(1):
31-42.
Mino, T., van Loosdrecht, M. and Heijnen, J. (1998). Microbiology and Biochemistry of the Enhanced
Biological Phosphate Removal Process. Wat. Res. 32(11): 3193-3207.
Mishima, K., Nishimura, T., Goi, M. and Katsukura, N. (1996). Characteristics of Nitrification and
Denitrification of the Media-Anaerobic-Anoxic-Oxic Process. Wat. Sci. Tech. 34(1-2): 137-143.
Morper, M. (1994). Upgrading of Activated Sludge Systems for Nitrogen Removal by Application of the
LINPOR-CN Process. Wat. Sci. Tech. 29(12): 167-176.
Mossakowska, A., Reinius, L. and Hultman, B. (1997). Nitrification Reactions in Treatment of Supernatant
from Dewatering of Digested Sludge. Wat. Environ. Res. 69(6): 1128-1133.
Nielsen, M., and T. Onnerth (1995). Improvement of a Recirculating Plant by Introducing STAR Control.
Wat. Sci. Tech. 31(2): 171-180.
Nolasco, D., Daigger, G., Stafford, D., Kaupp, D. and Stephenson, 1. (1998). The Use of Mathematical
Modelling and Pilot Plant Testing to Develop a New Biological Phosphorus and Nitrogen Removal
Process. Water Environ. Res. 70(6): 1205-1215.
Nowak, 0., Schweighofer, S. and Svardal, K. (1994). Nitrification Inhibition-A Method for the Estimation of
Actual Maximum Autotrophic Growth Rates in Activated Sludge Systems. Wat. Sci. Tech. 30(6): 9-19.
Nyberg, U., Andersson, B. and Aspegren H. (1996). Long-term Experiences with External Carbon Sources for
Nitrogen Removal. Wat. Sci. Tech. 33(12): 109-116.
Olsson, G. (1976). State of the Art in Sewage Treatment Plant Control. AIChE Symposium Series 72(159):
52-76.
Olsson, G. and Andrews, J. F. (1978) The Dissolved Oxygen Profile - A Valuable Tool for Control of the
Activated Sludge Process. Wat. Res., 12,985-1004.
Olsson, G. and Newell, B. (1999) Wastewater Treatment Systems: Modelling, Diagnosis and Control. IWA
Publishing, London.
Olsson, G., Rundqwist, L., Eriksson, L. and Hall, L. (1985) Self-tuning Control of the Dissolved Oxygen
Concentration in Activated Sludge Systems. in Advances in Water Pollution Control (Drake, R. A. R.
ed.), IAWPRC, Houston, Texas, USA, Pergamon Press, 473-480.
Onnerth, T., Nielsen, M. and Stamer, C. (1996). Advanced Computer Control Based on Real and Software
Sensors. Wat. Sci. Tech. 33(1): 237-245.
Plisson-Saune, S., Capdevil1e, B., Mauret, M., Deguin, A. and Baptiste, P. (1996). Real-Time Control of
Nitrogen Removal Using Three ORP Bending-Points: Signification, Control Strategy and Results. Wat.
Sci. Tech. 33(1): 275-280.
225
Potter, T., Koopman, B. and Svoronos, A. (1996). Optimisation of a Periodic Biological Process for Nitrogen
Removal from Wastewater. Wat. Res. 30(1): 142-152.
Puznava, N., Zeghal, S. and Reddet, E. (1998). Simple Control Strategies of Methanol Dosing for Post-
Denitrification. Wat. Sci. Tech. 38(3): 291-297.
Rabinowitz, B. a. B., 1. L. (1997). The Use of Primary Sludge Fermentation in Biological Nutrient Removal
Processes. Biological Nutrient Removal 3, Brisbane, Australia.
Randall, C., Barnard, J. and Stensel, H. (1992). Desigu and Retrofit of Wastewater Treatment Plants for
Biological Nutrient Removal. Western Hemisphere, Technomic Publishing Company.
Randall, C., and Sen, D. (1996). Full-Scale Evaluation of an Integrated Fixed-Film Activated Sludge (lFAS)
Process for Enhanced Nitrogen Removal. Wat. Sci. Tech. 33(12): 155-162.
Regan, J., Koopman, B., Svoronos, S. and Lee, B. (1998). Full-Scale Test of Methanol Addition for Enhanced
Nitrogen Removal in a Ludzack-Ettinger Process. Water Environ. Res. 70(3): 376-381.
Rosen, B., Ullman, A. and Raguarsson, N. (1998). Upgrading for Nitrogen Removal, Using a Combination of
SBR (Sequencing Batch Reactor) Technique and Unloading of Existing Biological Stage. Wat. Sci. Tech.
37(9): 17-24.
Rusten, B., Siljudalen, J. and Nordeidet, B. (1994). Upgrading to Nitrogen Removal with the KMT Moving
Bed Biofilm Process. Wat. Sci. Tech. 29(12): 185-195.
Rusten, B., Hem, L. and Odegaard, H. (1995a). Nitrification of municipal wastewater in moving-bed biofilm
reactors. Water Environ. Res. 67(1): 75-86.
Rusten, B., Hem, L. and Odegaard, H. (1995b). Nitrogen removal from dilute wastewater in cold climate
using moving-bed biofilm reactors. Water Environ. Res. 67(1): 65-74.
Sasaki, K., Yamamoto, Y., Tsumura, K., Ouchi, S. and Mori, Y. (1996). Development of 2-Reacor
Intermittent-Aeration Activated Sludge Process for Simultaneous Removal of Nitrogen and Phosphorus.
Wat. Sci. Tech. 36(1-2): 111-118.
Sen, D., Mitta, P. and Randall, C. (1994). Performance of Fixed Film Media Integrated in Activated Sludge
Reactors to Enhance Nitrogen Removal. Wat. Sci. Tech. 30(11): 13-24.
Skalsky, D., and Daigger, G. (1995). Wastewater Solids Fermentation for Volatile Acid Production and
Enhanced Biological Phosphorus Removal. Water Environ. Res. 67: 230.
Sorensen, J., Thornberg, D. and Nielsen, M. (1994). Optimisation of a Nitrogen-Removing Biological
Wastewater Treatment Plant Using On-Line Measurements. Water Environ. Res. 66(3): 236-242.
Sorensen, J. (1996). Optimization of a Nutrient-Removing Wastewater Treatment Plant Using On-Line
Monitors. Wat. Sci. Tech. 33(1): 265-273.
Sorm, R., Bortone, G., Saltarelli, R., Jenicek, P., Wanner, 1. and Tilche, A. (1996). Phosphate Uptake under
Anoxic Conditions and Fixed Film Nitrification in Nutrient Removal Activated Sludge System. Wat. Res.
30(7): 1573-1584.
Sorm, R., Wanner, J., Saltarelli, R., Bortone, G. and Tilche, A. (1997). Verification of Anoxic Phosphate
Uptake as the Main Biochemical Mechanism ofthe "DEPHANOX" Process. Wat. Sci. Tech. 35(10): 87-
94.
Spanjers, H., Vaurolleghem, P., Olsson, G. and Dold, P. (1998). Respirometry in Control of the Activated
Sludge Process: Principles. London, 1AWQ.
Steffens, M., and Lant, P. (1999). Multivariable Control of Nutrient-Removing Activated Sludge Systems.
Wat. Res. (in press).
Su, J., and Ouyang C. (1996). Nutrient Removal Using a Combined Process with Activated Sludge and Fixed
Biofilm. Wat. Sci. Tech. 34(1-2): 477-486.
Takizawa, S., Aravinthan, V. and Fujita, K. (1996). Nitrogen Removal from Domestic Wastewater Using
Immobilized Bacteria. Wat. Sci. Tech. 24(1-2): 431-440.
Thomsen, H. a. K., K. (1996). Nand P On-Line Meters: Requirement, Maintenance and Stability. Wat. Sci.
Tech. 33(1): 147-157.
Thornberg, D., Nielsen, M. and Andersen, K. (1993). Nutrient Removal: On-Line Measurements and Control
Strategies. Wat. Sci. Tech. 28(11-12): 549-560.
van Benthum, W., Garrido, J., Mathijssen, J.,Sunde, J., van Looscrecht, M. and Heijnen, J. (1998a). Nitrogen
Removal in Intermittently Aerated Biofilm Airlift Reactors. J. Envir. Engrg., ASCE 124(3): 239-248.
van Benthum, W., Derissen, B., van Loosdrecht, M. and Heijnen, 1. (1998b). Nitrogen Removal Using
Nitrifying Biofilm Growth and Denitrifying Suspended Growth in a Biofilm Airlift Suspension Reactor
Coupled with a Chemostat. Wat. Res. 32(7): 2009-2018.
van Loosdrecht, M., and Jetten, M. (1998). Microbiological Conversion in Nitrogen Removal. Wat. Sci. Tech.
38(1): 1-7.
226
van Loosdrecht, M., Brandse, F. and Vries, A. (1998). Upgrading of Waste Water Treatment Processes for
Integrated Nutrient Removal-the BCFS Process. Wat. Sci. Tech. 37(9): 209-217.
Vanrolleghem, P., Van Daele, M. and Dochain, D. (1995). Practical Identifiability of a Biokinetic Model of
Activated Sludge Respiration. Wat. Res. 29(11): 2561-2970.
Vanrolleghem, P., Spanjers, H., Petersen, B., Ginestet, P. and Takacs, I. (1999). Estimating (Combination of)
Activated Sludge Model No.1 Parameters and Components by Respirometry. Wat. Sci. Tech., 39(1), 195-
214.
Wanner, J., Cech, J. and Kos, M. (1992). New Process Design for Biological Nutrient Removal. Wat. Sci.
Tech. 25(4-5): 445-448.
Wareham, D., Hall, K. and Mavinic, D. (1993). Real-Time Control of Wastewater Treatment Systems Using
ORP. Wat. Sci. Tech. 28(11-12): 273-282.
Wareham, D., Mavinic, D. and Hall, K. (1994). Sludge Digestion Using ORP-Regnlated Aerobic-Anoxic
Cycles. Wat. Res. 28(2): 373-384.
Welander, U., Henrysson, T. and Welander, T. (1997). Nitrification of Landfill Leachate Using Suspended-
Carrier Biofilm Technology. Wat. Res. 31(9): 2351-2355.
Welander, U., Henrysson, T. and Welander, T. (1998). Biological Nitrogen Removal from Municipal Landfill
Leachate in a Pilot Scale Suspended Carrier Biofilm Process. Wat. Res. 32(5): 1564-1570.
Wett, B., Rostek, R., Rauch, W. and Ingerie, K. (1998). pH-Controlled Reject Water Treatment. Wat. Sci.
Tech. 37(12): 165-172.
Wouters-Wasiak, K., Hednit, A., Audic, J. and Lefevre, F. (1994). Real-Time Control of Nitrogen Removal at
Full-Scale Using Oxidation Reduction Potential. Wat. Sci. Tech. 30(4): 207-210.
Yu, R., Liaw, S., Chang, C., Lu, H. and Cheng, W. (1997). Monitoring and Control Using On-line ORP on the
Continuous-Flow Activated Sludge Batch Reactor System. Wat. Sci. Tech. 35(1): 57-66.
Yu, R., Liaw, S., Chang, C. and Cheng, W. (1998). Applying Real-Time Control to Enhance the Performance
of Nitrogen Removal in the Continuous-Flow SBR System. Wat. Sci. Tech. 38(3): 271-280.
Yuan, Z., Bogaert, H., Vanrolleghem, P., Thoeye, C., Vansteenkiste, G. and Verstraete, W. (1996). Carbon
Dosage Control for Predenitrification Processes. Proceedings Workshop Modelling, Monitoring and
Control of Wastewater Treatment Plants. Med. Fac. Landbouww. Univ. Gent, 6114a, 1733-1743.
Yuan, Z., Bogaert, H., Vanrolleghem, P., Thoeye, C., Vansteenkiste, G. and Verstraete, W. (1997). Control of
External Carbon Addition to Predenitrifying Systems. J. Envir. Engrg. 123(11): 1080-1086.
Yuan, Z., Bogaert, H., Vansteenkiste, G. and Verstraete W. (1998). Sludge Storage for Countering Ammonia
Shock Loads and Toxicity Incidents. Wat. Sci. Tech. 37(12): 173-180.
Yuan, Z., Bogaert, H., Devisscher, M., Vanrolleghem, P. and Verstraete, W. (1999). On-line Estimation of the
Maximum Specific Growth Rate of Nitrifiers in an Activated Sludge System. Biotechnology and
Bioengineering 65: 265-273.
Yuan, Z., Bogaert, H., Leten, J. and Verstraete, W. (2000). Reducing the Size of a Nitrogen Removal
Activated Sludge Plant by Shortening the Retention Time of Inert Solids via Sludge Storage. Wat. Res.
34(2): 539-549.
Zeghal, S., Puznava, N. Subra, J. P. Sauvegrain, P. and Vignoles, C. (1997). Methanol Dosing Feedback
Control for Denitrification. Biological Nutrient Removal 3, Brisbane, Austalia.
Zeghal, S., and Puznava, N., Subra, J. and Sauvegrain, P. (1998). Process Control for Nutrients Removal
Using Lamella Sedimentation and Floating Media Filtration. Wat, Sci. Tech. 38(3): 227-235.
Zhang, M., Tay, J., Qian, Y. and Gu, X. (1998). Coke Plank Wastewater Treatment by Fixed Biofilm System
for COD and NH3-N Removal. Wat. Res. 32(2): 519-527.
Zhao, H., Isaacs, S., Soeberg, H. and Kummel, M. (1994). A Novel Control Strategy for Improved Nitrogen
Removal in an Alternating Activated Sludge Process - PART I. Process Analysis. Wat. Res. 28(3): 521-
534.
Zhao, H., Isaacs, S., Soeberg, H. and Kummel, M. (1994). A Novel Control Strategy for Improved Nitrogen
Removal in an Alternating Activated Sludge Process - PART II. Control Development. Wat. Res. 28(3):
535-542.
Zipper, T., Fleischmann, N. and Haberl, R. (1998). Development of a New System for Control and
Optimisation of Small Wastewater Plants Using Oxidation-Reduction-Potential. Wat. Sci. Tech. 38(3):
307-314.
227
PART 2
WASTE GAS BIOFILTRATION
PERFORMANCE AND CHARACTERISATION OF A MEMBRANE
BIOLOGICAL AIR FILTER FOR SPACE APPLICATIONS
JAAP VAN DER WAARDE1, ARJAN VAN DER WERF1,

MAURICE HENSSEN1, BERT GEURKINK1, KLAAS VAN DER
MARELi, PIET PAUL2 AND MARC GENr
1Bioclear Environmental Biotechnology, Groningen, The Netherlands,
Tel: +31 505718455, FAX: +31505717920, email: waarde@bioclear.nl
2 STORK Engineers and Contractors, Amsterdam, The Netherlands
Summary
A membrane Biological Air Filter (BAF) is designed for the degradation of low
concentrations of various organic contaminants in indoor air. The BAF showed stable
performance during a 15 month test in which a near complete removal of most organic
volatile contaminants was observed. Even at extremely low concentrations (unto a few
JLglm3) good biodegradation efficiencies are obtained. Molecular and physiological
methods to detect and identify bacteria showed that the biodegradation process is
concentrated in the biofilm and that a mixed bacterial population is present growing on
all added organic components.
1. Introduction
Environmental quality assurance and recycling of raw materials are essential elements
in manned spacecraft missions. Indoor air quality plays an important role, since the
limited amount of air in the spacecraft is continuously recycled. Both astronaut
activities and materials onboard the spacecraft are sources of air contamination. A wide
range of volatile organic contaminants has been observed in these closed environments,
including aliphatic hydrocarbons, alcohols, aldehydes, aromatic hydrocarbons and
chlorinated aliphatic components. Most of these components are biodegradable but the
concentrations are usually low (mglm3 ). A biological system to remove these
components needs to meet several criteria: active at low concentrations; combined
removal of a wide range of organic contaminants; removal to levels below the space
maximum allowable concentration (SMAC), usually below 1 mglm3 ; stable activity
over long periods (months); small volume and low weight; no contact between bacteria
and astronauts. A biological air fllter (BAF) (Fig. 1) has been designed to purify air in
manned spacecraft and meets these criteria (Binot and Paul, 1989).
231
Jaap VanDer Waarde et aI
filtermodute
membrane fibres
{b'iO'fiim~~wter sulfat$}
Gasflaw Git$fIOw
inlet outlet
detail
membrane
Fig. 1: Principle oftke Biological Air Filter (BAF)
This system has shown efficient removal of aromatic components like toluene, xylenes
and chlorobenzene. Improved biodegradation of chlorobenzene was observed under
mixed substrate conditions (Keuning et ai, 1991). Air from a model spacecraft training
unit during a training session was efficiently purified using the BAF. Removal of
toluene, isopropanol and acetone to levels well below the SMAC values was
demonstrated (Binot et ai, 1994).
Finally, a model BAF system has been designed to test the effects of space
conditions (Il-gravity) on biodegradation. Efficient removal of 1,2-dichloroethane has
been demonstrated under space conditions with this system (van der Waarde et ai,
1997). In this report the results are described from long term biodegradation studies in
the BAF with low concentrations of mixtures of contaminants.
2. Materials and methods
The following bacterial strains were used (growth substrate in brackets): Pseudomonas
marginalis GJ8 (p-xylene), Xanthobacter autotrophicus GJ10 (1,2-dichloroethane),
Pseudomonas fluorescens GJ31 (toluene), Pseudomonas putida GJ40 (chlorobenzene),
Ancylobacter aquaticus AD20 (l,2-dichloroethane), Pseudomonas fluorescens BC20
(hexane) and Pseudomonas putida BCG2 (p-xylene). All strains are environmental
isolates. Growth was maintained on mineral medium MMY supplemented with 5 mM
substrate (Oldenhuis et ai, 1991). Maximum growth rates (Ilmax) of the selected strains
232
Perfonnance and characterisation of a membrane biological air filter for space applications
are determined for various substrates by adding 4 mM substrate to a 1% inoculate and

following the growth by measuring the optical density (at 450 om). The BAF system
consists of a multilayer membrane cassette (12 Accurel membranes), with a volume of
50 ml (25 ml gas phase and 25 mlliquid phase) and a membrane area of 0.086 m2• The
BAF was inoculated with a mixture of the selected bacterial strains and operated during
400 days. The BAF system was fed with a gas flow of 65 mlImin (HRT of 23 sec.). The
liquid loop (total volume 0.5 1) was recycled with a flow of 400 ml/h. Biofilter removal
efficiency was monitored by GC analysis of influent and effluent gas flow once to twice
per week. Low concentrations contaminants « 2 mglm3) were determined by leading
the gas flow through activated carbon, followed by desorption with CS 2 and GC
analysis. Water samples for FISH (Fluorescent In Situ Hybridisation) analysis were fixed
in 36% formaldehyde (final concentration 5%), diluted to 50% ethanol and stored at -20°C
until analysis. Sample (20 ~) was air dried on Vectabond coated microscope slides and
washed in 50, 80 and 96% ethanol for 3 min each. FISH analysis was performed according
to a protocol adapted from Stahl et al. (1991). The probes EUB338 (a general probe
directed against all Eubacteria) (Amann et al. 1990b), ALF (directed against the ex-subclass
of bacteria) (Manz et al.I992), BET (directed against the l3-subclass of bacteria) (Manz et
al.1992), GAM (directed against the ,,(-subclass of bacteria) (Manz et a1.1992) and DEL
(directed against the &.subclass of bacteria) (Amann et al. 1990a) were hybridised to
separate samples in a final concentration of 8 ngl~ each with formamide at 45°C
overnight. Slides were viewed with an Olympus fluorescence microscope with Fluorplan
100* objective and Olympus U-MNB filter set.
3. Results and discussion
The BAF research was focused on biodegradation of a mixture of volatile organic

components that were model components for contaminated air in closed manned
spacecraft. The following components were selected: hexane (poorly water-soluble
aliphatic), benzene, toluene, ethylbenzene and xylenes (common, non-chlorinated
aromatic compound), chlorobenzene (chlorinated aromatic compound) and 1,2-
dichloroethane (1,2-DCE, chlorinated aliphatic compound).
These components were selected since they can form problems in air biofiltration
and chlorinated components may have an effect on pH levels in the system.
3.1. PHYSIOLOGICAL CHARACTERISATION
Several bacterial strains that had been isolated from environmental sources were tested
for their capacity to biodegrade these components and their maximum growth rates
were determined (Table 1). The specific maximum growth rates of the pure cultures on
the tested substrates ranged from 0.05 h-I (BC20 with hexane) to 0.56 h-I (GJ31 with
toluene). It is clear from these data that BTEX biodegradation can be performed by a
range of bacterial strains, but biodegradation of hexane or chlorobenzene is dependent
on the presence of a single strain, strain BC20 or GJ31 respectively.
233
Jaap VanDer Waarde et al
Table 1 Physiological characterisation and maximum growth rates (hOI) of bacterial strains
used
Benzene Toluene Ethyl- p-Xylene Chloro- 1,2-DCE n-Hexane

benzene benzene
GJ8 + 0.2661
GJ10 + + 0.087 1
GJ31 + 0.555 + 0.2241
GJ40 0.308 0.440 1 0.365

AD20 0.077 1
BC20 0.0501
BCG2 + 0.085 1
+: growth detected, no growth rate determined

- : no growth detected
1: originally isolated on this substrate.
Table 2 Biofilter removal efficiency and elimination capacity (EC)
Component Ingoing BAF perfonnance BAF perfonnance BAF perfonnance

Concen- (gas flow = (gas flow = after 6 weeks
tration 65 mlIrnin) 195 mlIrnin) starvation
(gas flow =
65 mlImin)
[mglm3] Eff. EC Eff. EC Eff. EC
[%] [glm3.h] [%] [glm3.h] [%] [glm3.h]
1,2-DCE 70-200 30-40 2-5 50-60 22-26 100 4-7

Benzene 30-80 75-100 2-6 60-80 22-27 100 2-4
Toluene 100-190 90-100 6-11 70-100 19-22 100 2-14
Ethylbenzene 30-120 90-100 3-9 80-100 19-23 100 1-5
p-Xylene 40-80 90-100 2-7 90-100 13-15 100 1-5
Chlorobenzene 90-130 90-100 4-10 90-100 24-26 100 6-10
Hexane 40-200 0-10 0.5-3 4-10 0.5-1.5 10-20 0.5-1.5
Eff. : Removal efficiency = (1-(effluent concentration I influent concentration»*100%
EC : Elimination capacity in g componentl(m3 reactor*hour)
234
3.2. PERFORMANCE OF THE BIOLOGICAL AIR FILTER (BAF)
After inoculation of the BAF at bacterial densities of 104/ml for each bacterial strain, an
adaptation period was observed in which the elimination capacity improved from 50%
to 90-100% for most components. Biofilter performance was stable after 130 days of
operation, hexane was poorly removed by the biofilter (Table 2). In this period the
ingoing concentrations fluctuated between 20 and 200 mg/m3 • Bioillter capacities are
relatively low due to the low influent concentrations and low gas flow.
Raising the load of the BAF for two weeks by a threefold increase of the gas flow
results in an increase in the elimination capacity. The effect of starvation on BAF
performance was determined by switching the gas flow off for 6 weeks. After feeding
the system again the removal efficiency for most components is near 100%, hexane is
poorly removed (Table 2). This indicates that the bioillter retains its activity and
biodegradation capacity throughout the 6 weeks starvation period and completely
removes most of the organic contaminants upon re-establishing the feed to the system.
3.3. MOLECULAR CHARACTERISATION
A substrate depletion test was performed with a sample from the liquid loop of the
BAF, using chlorobenzene and benzene as substrate. It was found that biodegradation
occurs after a lag phase of 8 hours for chlorobenzene and 5 hours for benzene (data not
shown), clearly indicating that the bacterial cells in the liquid phase are not adapted to
active biodegradation. Samples from both the substrate depletion test and the liquid loop
of the BAF were analysed for the presence of bacteria using FISH analyses. Group
specific probes were used to discriminate between the used bacterial strains. Cell
activity was based on the brightness of the FISH signal. The sample from the liquid
loop does contain bacterial cells but most cells are inactive as evidenced by a low EUB
signal (Table 3).
Table 3. FISH analyses of bacteria in samples from the liquid loop and a substrate
depletion test
EUB338 ALF BET GAM DEL

Liquid loop +(20%) ± (10%) +(5%)
Substrate depletion test ++ (70%) + (20%) ++ (50%)

no signal;
± poor signal
+ clear signal
++ strong signal
between the brackets percentage of the cells that responds to FISH analysis
Bacteria from the AD20 type are not present or active, since no ALF positive cells were
detected. Strain Gn 0 is probably present, since the probe BET shows a positive
response. A sample from the substrate depletion test shows much higher numbers of
cells and more cells are active than a sample from the liquid loop of the BAF. This
indicates that biodegradation in the BAF module is caused by the bacteria in the bioillm
235
Jaap Van Dec Waarde et al
on the membrane, and that bacteria in the liquid phase are poorly active and do not
significantly contribute to the biodegradation process.
Six of the seven inoculated strains hybridise to the EUB probe, only strain GJ8
could not be detected. ALF hybridises with strain AD20, BET with strain GJlO and
GAM with strains GJ31, GJ40, BC20 and BCG2. The DEL probe hybridises with none
of the inoculated strains (data not shown).
3.4. PERFORMANCE OF THE BAF AT EXTREMELY LOW CONCENTRATIONS

OF VOLATILE CONTAMINANTS
Extremely low concentrations of volatile contaminants were dosed to the system to

determine the air purification efficiency at concentrations that are representative for
indoor air in manned spacecraft. The removal of these components was determined once
a week during 16 weeks.
Table 4. Biofilter performance at extremely low concentrations (range in 16 measurements)
Component Influent concentration cange [llgfm3] Removal efficiency [%]
Toluene 2-46 40-100

Xylenes 5-57 16-100
Chlorobenzene 5-90 90-100
I-Octene <100 14-100
n-Buty1acetate <70 96-100
Limonene <114 80-100
From these experiments it is clear that extremely low concentrations (J.tglm3) of volatile
organic components can efficiently be removed by the BAF. Good removal was found
during prolonged periods for limonene, toluene, l-octene, butylacetate, chlorobenzene
and xylenes. The large variation in removal efficiencies is caused by inaccuracies in the
dosing system and in the analysing method at these low concentrations. Limonene,
l-octene and butylacetate were dosed to the system without addition of new bacterial
strains. This implies that the biofilm in the BAF is able to quickly adapt to new
contaminants in the air.
4. Conclusion
The biological air filter (BAF) has shown to be effective in removing low
concentrations of volatile contaminants from air. Removal capacities range between 2
and 26 glm3.h at ingoing concentrations of 20 to 200 mglm3 and at retention times of 8-
23 seconds. Even at extremely low concentrations (up to concentrations of a few J.tglm3)
good biodegradation efficiencies are obtained.
236
Molecular and physiological methods to detect and identify bacteria show that the
biodegradation process is concentrated in the biofilm and that a mixed bacterial
population is present growing on all added organic components.
References
Amann, RI., B.I. Binder, RI. Olson, S.W. Chrisholm, R Devereux and D.A. Stahl. (1990a) Combination of
16S rRNA-targeted oligonucleotide probes with flow cytometry for analysing mixed microbial
populations. Appl. Env. Microbiol. 56,1919-1925.
Amann, RI., L. Krumholz and D.A. Stahl. (1990b) Fluorescent-oligonucleotide probing of whole cells for
detenninative, phylogenetic, and environmental studies in microbiology. J. Bacteriol. 172, 762-770.
Binot, RA. and P.G. Paul. (1989) BAF - an advanced ecological concept for air quality control. SAE
technical paper series no. 891535.
Binot, RA., R.J. Breukers, P.G. Paul and D. Jager. (1994) BAF-EXEMSf92: Testing of the biological air
filter for air quality control during a manned space mission simulation. SAE technical paper series no.
941343.
Keuning, S., D. Jager, P.G. Paul, and R.A. Binet (1991) Biodegradation study with space cabin contaminants to
determine the feasibility of Biological Air Filtration (BAP) in space cabins. ESA SP-324.
Manz, W., R Amann, W. Ludwig, M. Wagner and K.-H. Schleifer. (1992) Phylogenetic oligodeoxynucleotide
probes for the major subclasses of proteobacteria: problems and solutions. Syst. Appl. Microbiol. 15, 593-600.
Oldenhuis, R, J. Y. Oedzes, J.J. van der Waarde and D.B. Janssen. (1991) Kinetics of chlorinated hydrocarbon
degradation by Methylosinus trichosporium OB3b and toxicity of trichloroethylene. Appl. Environ.
Microbiol. 57, 7-14.
Stahl, D.A. and R Amann. (1991) Development and application of nucleic acid probes. Nucleic acid
techniques in bacterial systematics. Eds. E. Stackebrandt and M. Goodfellow. Wiley 11<1, New York, 205-
248.
Waarde, J.J. van der, H. Dorenbos, E. Dijkhuis, S. Keuning, P.G. Paul, C. Klabbers and R.A. Binot. (1997)
Determination of the space influence on the kinetics for biodegradation of organic volatile contaminants. ESA
SP-400.
Wagner, M., R Amann, H. Lemmer, and K.H. Schleifer. (1993) Probing activated sludge with
oligonucleotides specific for proteobacteria: inadequacy of culture-dependent methods for describing
microbial commuuity structure. Appl. Environ. Microbiol. 59, 1520-1525.
237
BIOFILTRATION FOR WASTE GAS HANDLING
FREDERIC THALASSOI, MARiA C. VEIGAz AND

CHRISTIAN KENNESz
lDepartment of Biotechnology, CINVESTAV, Av. IPN 2508,
CP 07000 Mexico City, Mexico 2University of La Coruna, Chemical
Engineering Laboratory, Campus da Zapateira, E-J5071-La Coruna,
Spain. Fax: 34981167065. e-mail: Kennes@udc.es
1. Introduction
Facing the increasing concern about the environment, biological gas cleaning has been
developed since the 1920's [1, cited by 2]. Nevertheless, it is only since a few decades
ago that biological gas cleaning is being accepted as a competitive alternative to the
more conventional physico-chemical treatment technologies. Nowadays, it is commonly
used for the cleaning of a wide variety of gaseous pollutants. Since the beginning, three
main groups of biological gas cleaning technologies were considered: biofiltration,
bioscrubbing and trickling filtration. These technologies differ by the presence or
absence of a carrier material and of a mobile liquid phase (Table 1). Trickling filters and
bioscrubbers are quite similar concerning the presence of a mobile liquid phase serving
as nutrient source for the microorganisms. Conversely, biofiltration is characterized by
the use of an organic carrier ensuring the nutrient supply and by the absence of a mobile
liquid phase or by the use of an inert carrier but with intermittent supply of nutrients.
This chapter presents the main characteristics of these three technologies although it
mainly focuses on biofiltration with organic carriers, which is the most largely used
process [3].
Table 1: Characteristics of the three main biological gas-cleaning technologies.
Reactor Mobile phases Carrier Active biomass

Bioscrubbers Liquid and gas Dispersed
Trickling filters Liquid and gas Inert Fixed
Biofilters Gas OrganicJInert Fixed
2. Bioscrubbing
In bioscrubbing the pollutant is fIrst absorbed in a liquid phase, which is then treated in
a second stage in a bioreactor (Fig. 1). The main advantages of this technology are: (i)
239
Frederic Thalasso, Maria C. Veiga and Christian Kennes
the wash-out of the reaction products, avoiding their potential inhibitory effects, (ii) an
easy control of the biological process thanks to the control of the composition of the
liquid medium, and (iii) a good adaptation capacity of the microbial biomass to the
composition of the gas to be cleaned.
Treated air
Fig. 1: Bioscrubber.
The major drawback of this technology is the requirement to dissolve the gaseous
pollutants in an aqueous phase, which generates problems of gas transfer. Bioscrubbing
is therefore of interest for gaseous pollutants having a low air/water partition
coefficient. This is of major importance since most of the target pollutants are volatile
and poorly water-soluble. According to Kok as well as van Groenestijn and Hesselink
[4, 5], bioscrubbing is of interest for the treatment of pollutants having an air/water
partition coefficient lower than 0.01. This is probably why bioscrubbing is less in focus
than biofiltration, although several examples of successful applications have been
reported [6, 7] and the interest of bioscrubbing has been pointed out for the treatment of
several Volatile Organic Compounds (VOC) [8]. Furthermore, recent developments [9,
10] foresee a new interest in this technology, among others because the utilisation of
bioscrubbing for biological desulphurisation of very large gas flow rates (up to 2 106 m3
per hour) seems to be feasible.
3. Trickling f'dters
The trickling filtration technology consists in the use of an inert carrier on which a
biofilm grows. The polluted gas passes through the carrier material, co- or counter-
currently to the mobile liquid phase, which ensures the nutrient supply to the
microorganisms (Fig. 2). In some cases, the liquid supply is limited and the reactor
design and operation are similar to organic-biofilters. For that reason, the border
between the conventional biofilter and the trickling filter is relatively narrow. Several
authors have used different names for sometimes very similar systems: biofilter [11,
12], trickling biofilter [5], biotrickling filter [13, 14] or trickle bed air biofilter (TBAB)
[15, 16].
240
Biofiltration for waste gas handling
Treated air
1:i~_ _I'"""""""u-~utrient solution
Inorganic carrier
Fig. 2: Tricklingfilter
The main carriers used are plastic or ceramic structured packing [17, 18, 19, 20, 21],
Celite [22, 23], perlite [24], vermiculite [25] activated carbon [26, 27, 28], polyurethane
foam [29, 30], lava stone [31, 32] or mixtures of them [33].
Both co- and counter-current operation have been used. The counter-current
operation has the advantage of a more uniform activity and biomass growth but suffers
from higher pressure drops [34].
This technology presents similar advantages as bioscrubbing: (i) wash-out of the
reaction products, (ii) easy control of the biological process and (iii) good adaptation
capacity of the active biomass. Like bioscrubbing, the major drawback of this
technology is the problem of gas transfer resulting from the need to dissolve the gaseous
pollutants in an aqueous phase. According to Kok and van Groenestijn and Hesselink
[4,5], trickling filters are of interest for the treatment of compounds having an air/water
partition coefficient lower than 0.1. Concerning that aspect, different publications [20,
26, 35, 36] have pointed out the interest to restrain the liquid supply to the minimal
microbial needs obtaining better gas treatment efficiencies with a limited liquid supply.
On the other hand, this limitation of liquid supply can lead to a decrease of the wetted
area of the filter carrier, which can be approximated to the active area [18]. This stresses
on the importance of a careful carrier design and of the homogeneity of the liquid
supply.
Another drawback, specific to trickling filters, is the potential excessive biofilm
growth on the carrier material. This microbial growth progressively reduces the empty
volume of the carrier and results in an increase of the pressure drops. This well
described phenomenon [2, 37, 38], can lead to the complete clogging of the filter-bed
[39]. Although not systematically observed [17], it reinforces the importance of careful
carrier design. Methods have been developed to restrain clogging problems, either by
limiting microbial growth [21, 22], by regular filter-bed washing [22, 40, 41, 42, 36] or
by reducing the liquid supply [35]. An extensive review of the characteristics and
efficiency of all such methods has been published very recently [43].
241
Frederic Thalasso, Marfa C. Veiga and Christian Kennes
Until the late nineties, trickling filters were indubitably less in focus than biofilters.
According to a literature overview, about 70% of the papers, presented results obtained
with filter beds composed mainly of organic material. After 1998, a clear change was
observed and nowadays, inorganic carriers are used in about half of the lab-scale
biofilters.
4. Biotiltration with organic packing materials
In conventional biofilters, the polluted gas stream flows through a filter-bed generally
composed of organic matter (peat, compost, sawdust, etc.) and serving both as a support
for the active biomass and as a reservoir of nutrients for microorganisms (Fig. 3). A
very important characteristic of this process is the absence of a mobile liquid phase.
Thanks to this characteristic, biofilters are very efficient for the treatment of poorly
water-soluble pollutants. According to Kok [4] and van Groenestijn and Hesselink [5],
biofiltration is of interest for the treatment of pollutants having an air/water partition
coefficient lower than 1. Several examples of successful applications have appeared in
the literature [44, 45, 46, 47] and nowadays some industrial plants can treat up to
200,000 m3 per hour.
The nature of the filter-beds is quite variable. According to Clark and Wnorowski
[48], almost any organic material presenting a "satisfactory structure and composition"
could be used. Thirteen important characteristics of good biofilter media have been
listed by Bohn [49], including physical, chemical and biological parameters. The most
important physical characteristics are: (i) a high surface area, for an optimum microbial
development, (ii) a low bulk density for an easy and cheap carrier operation and (iii) a
high void fraction to limit pressure drop and clogging problems. In addition to these
physical characteristics, the natural presence of a large number of different bacteria and
fungi in the carrier as well as a balanced chemical composition are of major importance
in order to favour microbial adaptation and activity inside the biofilter-bed.
Water lNutrien~
+- Water
Gas humidifier
! .
Treated arr
Fig. 3: Biofilter, design and control.
Peat, soil and compost are the most commonly used filter-bed materials. Although peat
and soil are often well characterized, they are sometimes relatively complex mixtures.
242
The word "compost" derives from the Latin "compositum", meaning mixture and refers
to decaying organic matter. Many raw materials may be used as compost: yard trash
[50], digested sewage sludge [50, 51], forest sub-products [51, 52], wastewater
"biosolids" [52], manure [53], or the organic fraction of municipal waste [54]. Many
other carrier materials are mentioned in the literature, among others, bark, saw dust or
dried wastewater sludge as shown in table 2. A new tendency consists in the use of
specially designed nutrient beads for the immobilization of the active biomass [55].
The carrier materials are ever more frequently prepared by mixing organic and
inorganic matter. Approximately 85 % of the papers published in the early eighties
concerned pure organic matter while over the last five years this percentage declined to
less than fifty percent The main objective of incorporating an inorganic material, such
as perlite [33, 52, 56], glass beads [57], polyurethane foam [33] or polystyrene [58, 59],
into the filter-bed is to reduce the pressure drop and to limit channelling problems.
Sometimes, activated carbon and buffer compounds are also incorporated to minimize
the fluctuation ofthe pollutant concentration [60, 61] or to stabilize the pH of the filter-
bed [52, 56, 62, 63].
The filter-beds are simple or multi-stage and may reach up to four stages [33,50,51,
52]. According to Ottengraf et al. [64], "the treatment of a waste gas containing
different biodegradable and xenobiotics compounds may advantageously take place in a
multi-staged filter-bed". This allows the presence of different environmental conditions
in a same biofilter and allows to optimise the degradation of each of the compounds to
be removed. Usually, the height of filter beds is between 0.5 and 1 meter but they can
reach up to 2 or 3 meters high in some cases [7, 65, 66].
Since organic carriers provide nutrients necessary to the microorganisms, there is no
need for any external nutrient supply initially. Bohn [67] underlines that the nutrient
supplying power of the filter media are usually high enough for the pollutant loads
generally applied. This is confirmed, for instance by Cardenas-Gonzalez et al. [68] who
found no significant nitrogen and total carbon decrease in a compost biofilter after 5
years operation. Nevertheless, Dalouche et al. [70] underline the lack of phosphorus in
peat and Morgenroth et at. [51] reported a problem of nutrient limitation during hexane
degradation in a compost biofilter. To limit such a risk, some authors adopted the
alternative to discontinuously spray a nutrient solution on top of the filter-bed [12, 57].
Even temporarily using tap water seems to be suitable to ensure high biodegradation
efficiencies [71]. Although nutrients are available in organic filter beds, the additional
intermittent supply of nutrients to the reactor allows increasing biofilter performance
and long-term stability [72].
243
Table 2: Main materials used as biofiltration carriers (updatedfrom [124]).
Carrier Volume ratio Reference
Compost 1 11,12,73,74,75,76
Compost - yard trash 1/3, - 61,77
Compost - Wood chips -,317 78, 79
Compost - perlite 111 52,51,80,81
Compost - diatomaceous earth 82
Compost - polystyrene III 58,83
Compost - chaff 111 50
Compost - manure 69
Pellets: compost + Inorganic 2/1 84
polymer
Peat 1 85,86,87,88,89
Peat balls 1 90
Peat - polyuretbane foam 7/3 33
Peat - polyurethane foam - 2/2/1 33
vermiculite
Peat - polyuretbane foam - perlite 2/3/5 33
Peat - perlite 2/3 33,56,91
Peat - pine bark 90
Peat - compost 111 92
Peat - glass beads 211,4/1 57,93
Wood bark 1 73,94
Wood waste - perlite 111 95
Soil 1 7,44,47,82,96
Soil - sand - peat - compost 2012/3/3 97
Seaweeds 1 70
Sawdust - manure 53
Dry wastewater sludge 98
One of the most significant parameters of the filter-bed is its moisture content, which is
of major importance for the microbial activity. Ensuring a high enough moisture level is
important, because of the high sensitivity of the microorganisms to the water activity
[99, 100]. However, too high a moisture content may generate serious problems such as
the formation of stagnant zones with diffusion limitation and possible anaerobic
conditions [75] or increased pressure drops [94, 101]. As clearly presented by Wang et
al. [12] the maximum degradation capacity of the biofilter is obtained at a moisture
content of about 35% when compost is used as filter-bed and with a moisture content of
about 40% when peat is used. Wang et at. [12] also underline that a drying period could
provoke irreversible effects by modifying the structure of the filter-bed, which was also
confirmed by Bohn [67]. An adequate average value seems to be in the range of 30 to
80% for peat, compost and wood sub-products [7, 47]. For soil-beds, the optimal
moisture content seems to be in a 10-20 % range [47, 65, 96, 102]. Bohn and Bohn
recently published an extensive review on moisture in biofilters [103].
The control of the filter-bed moisture content is currently done either by means of a
spray system directly spraying water on the filter-bed, and/or by indirectly controlling
its moisture content by means of the humidity of the polluted gas fed to the biofilter. In
the case of direct spraying, Van Lith et al. [101] underlined the importance of a small
droplet size in order to reach a near homogenous liquid distribution and to limit the
mechanical impact of the droplets on the carrier material. Sprayers commercially
244
available can produce droplets of less than 40 10-6 m diameter [20, 35]. Suchlike small
droplets present also the advantage of an extremely high gaslliquid transfer area (above
150 m2 per liter of liquid injected) and therefore of a quick liquid saturation. It is
important to underline that some authors do not specify if the humidity of their fIlter-
bed is given on a dry [12, 54] or on a wet [50, 51, 52, 57, 58, 104] basis. Although both
are used in the literature, it is likely that the majority of the results are given on a wet
basis.
Obviously, temperature is also an important parameter. Although microbial activity is
possible in a 0 - 95°C range, the optimal temperature for the filter-beds corresponds to
the mesophilic temperature, i.e. about 30°C [12, 65, 95, 105]. Some examples indicate
that the temperature may be lower without any significant microbial deactivation in the
range of 10 to 20°C [85, 106], or even at 2°C as presented by Ebinger et al. [102] for
propane treatment in a soil-bed system, or as observed by Elsgaard [107] in a biofilter
treating ethylene. In this context, it is important to stress that a temporary biological
deactivation of the filter bed does not necessarily result in the complete loss of the
treatment capacity, because of the existence of adsorption phenomena in some fIlter-
beds. Consequently, biofllters can cope with temporary and relatively significant
temperature decreases, even to temperatures below O°C [108]. Additionally, some
examples of bioflltration at thermophilic temperatures have recently been presented
[109, 110]. Finally, it should be recalled that biodegradation processes are exothermic
and may lead to significant increases of the filter-bed temperature [111, cited by 74].
Obviously, such temperature increases result in a decrease of the relative humidity of
the gas phase, which can provoke a drying effect of the filter-bed [112] especially if
channels are present.
The pH of the filter-bed is also an important parameter. A good carrier material
should preferably have a near neutral pH [2, 113] and must present a good buffering
capacity [49]. This is of major importance since numerous processes generate acidic,
basic or toxic products such as hydrochloric acid during the biodegradation of
chlorinated compounds [44], sulphuric acid during biodegradation of hydrogen sulphide
[7, 85, 78] or methyl sulphide [54] and basic compounds during ammonia elimination
[70]. Devinny and Hodge [27] present results of acetaldehyde and acetic acid generation
during ethanol degradation when applying overloads. Under non-regulated pH
conditions, the elimination capacity usually drops concomitantly with pH decrease, as
observed by Bronnenmeier et al [114] and by Hartmans and Tramper [115].
According to Bohn [49], soil presents a higher buffering capacity than compost
which, according to Smet et al. [54] is five times more buffered than wood bark. On the
other hand peat is characterized by its naturally low pH, around 3 to 4 [70, 87].
Therefore, a pre-treatment of the filter-bed is sometimes a prerequisite in order to
increase its pH value or to improve its buffering capacity. Several compounds are used
to control the pH of the filter-bed: lime [75, 105, 113] and powdered oyster shell [51,
52, 80, 81] mixed with the filter-bed during the pre-treatment or Ca(OHh [88, 116]
NaHP04, NaHC0 3 [50] and NaOH [57] either mixed with the filter-bed or sprayed on
top of it during the process.
Peculiar cases of pH resistance have been reported by Windsperger et al. and by
Yang and Allen [77, 117] with an optimum H2S degradation rate obtained at pH 3.2,
and by Shinabe et al. [19] observing no inhibition of H2S degradation at pH 1 in a
245
Frederic Thalasso. Maria C. Veiga and Christian Kennes
trickling filter. During batch studies of H2S removal by Thiobacillus thiooxidans

Shinabe et al. [19] found an optimal pH of 2.5. It is important to stress that during
hydrogen sulphide degradation, inorganic sulphate compounds do accumulate in the
filter-bed and can reach up to 10% by weight [112]. Similarly, the formation of up to
100 mole nitrites and nitrates per m3 carrier during ammonia degradation has been
reported [118, 119].
The pressure drop created by the gas phase passing through the filter-bed can represent
a non-negligible part of the treatment costs. The pressure drop depends on numerous
factors, mainly the nature of the filter-bed and its moisture content. Soil induces the
highest pressure drop, followed by compost, peat and finally by wood bark [2, 75, 94,
120, 121]. Obviously, the size of the particles forming the filter-bed is of major
importance since for a same filter-bed material, small particles can provoke a pressure
drop more than 10 times higher than observed with large particles [112]. The effect of
the moisture content in organic-biofilters has been shown, among others, by Shoda [7]
and by Van Langenhove et al. [94] and can result in a pressure drop increase of up to
100 %. This once again indicates the importance of a good moisture control.
5. Applications
5.1. GENERAL ASPECTS
Biofiltration of waste gases was originally applied to a rather limited range of sources
and compounds, mainly odours usually quantified by means of Odour Units (OU) rather
than through the identification of specific compounds.
Over the past two decades, the technology has been extended to a much wider
variety of sources and pollutants. Nowadays, more than thousands gas-phase bioreactors
are being operated in Europe. In Germany and neighbouring countries, odours emitted
from wastewater treatment plants are solved with biofilters in almost two third of the
cases [3]. A non-exhaustive list of pollutants biodegradable in conventional and
trickling biofilters is given in table 3. Theoretically, any pollutant biodegradable under
aerobic conditions could be degraded in gas-phase biofilters. Even compounds requiring
anaerobic environments for their biological removal have been degraded in gas-phase
lab-scale biofilters. Perchloroethylene (PCE), which needs strict anaerobic conditions
for biodegradation [122] appeared to be biodegradable in a biofilter fed with PCE-
polluted air [123]. The removal of hardly biodegradable contaminants in biofilters may
require the inoculation of selected specialized microorganisms [124]. For the treatment
of waste gases polluted with non-recalcitrant compounds in biofilters packed with
natural organic carriers, inoculation is often not necessary since indigenous
microorganisms will rapidly grow and adapt to the contaminants. However, seeding the
reactor usually allows to shorten the start-up phase [125], although inoculated strains
may finally be overgrown by indigenous strains [124].
246
· Biofiltration for waste gas handling
Table 3. List of some pollutants for which biodegradation in gas-phase biofilters has been
shown.
Acetaldehyde Dioxane Methylmethacrylate

Acetone Ethane Methyl-Tertiary-Butyl-Ether
(MTBE)
Acetonitrile Ethanol Nitric oxide
Acetylene Ethene Nitrobenzene
Acrylonitrile Ethylacetate Phenol
Ammonia Ethylaldehyde a-pinene
Benzene Ethylbenzene Propane
1,3-Butadiene Ethylene Propanol
Butanal Fonnaldehyde Propionaldehyde
Butanol Heptane Styrene
Butyl acetate Hexane Terpenes
Butyric acid Hydrogen sulphide Tetrachloroethene
Carbon monoxide Isobutylacetate Thiocyanates
Carbon disulphide Isopentane Toluene
Chlorobenzenes Isopropanol Trichloroethane
Cresol Methane Trichloroethene
Cyclohexane Methanethlol Trichloromethane
Dichloroethane Methanol Triethylamine
Dichloromethane Methoxypropylacetate Trimethylamine
Diethylether Methyl-Ethyl-Ketone (MEK) Vinyl acetate
DimethyIsulphide Methyl-Iso-Butyl-Ketone (MmK) Vinyl chloride
Dimethyldisulphide MethyImercaptan Xylenes
In the first studies undertaken and published on biofiltration, bioreactor performances

were relatively poor and rather low elimination capacities of only a few grams per cubic
meter per hour were usually reported which was, however, acceptable for full-scale
applications at wastewater treatment plants or composting facilities. The improvement
of bioreactor design and a better understanding of the phenomena affecting their
performance have allowed reaching much higher elimination capacities together with
high removal efficiencies (Table 4). Thus, nowadays biofiltration may be used in a
much wider range of sectors and applications. Although it has recently been shown to
be suitable for the treatment of high gas flow rates and high loads, the treatment of low
pollutant concentrations released at low flow rates, i.e. low loads, still remains a major
application field of biofitration. High efficiencies have often been obtained in lab-scale
experiments (Table 4). Nevertheless, it is important to point out that in some of the
reported lab-scale studies high elimination capacities were maintained only for short
periods of time, while full-scale systems must operate under stable conditions for long
periods of at least several years. It should also be mentioned that in lab-scale
experiments, a few authors report high maximal elimination capacities corresponding,
however, to low removal efficiencies. At industrial scale, such situation is generally not
viable, and whenever high loads are to be treated, relatively high removal efficiencies
are required in order to avoid or minimize air pollution.
247
Table 4: Some typical recent examples of relatively high elimination capacities (EC) and
removal efficiencies (RE) reached in biofilters and biotrickling filters for typical Volatile
Inorganic and Organic Compounds (VICs and VOCs) (more examples are available in
[124]).
Packing Compound Max. EC (glm3.h) / Ref.

Corresponding RE (%)
BIOFILTERS:
Fuyolite Ammonia 22.8(') / > 99 [126]
Compost/dolomite Diruethylsulphide 70/85 [127]
Perlite Ethylbenzene 85/99 [24]
CompostIHog fuellPerlite Hydrogen sulphide 120/30 [128]
Compost/Glass beads Phenol 700/25 [57]
CompostIWood chips u -Pinene + Methanol 40 + 250/ > 90 (each) [129]
Perlite Styrene 80/99 [130]
Perlite Toluene 73/99 [24]
Perlite Toluene + Ethylbenzene + 120 (total) / > 99 [125]
o-Xylene 0:1:1)
Perlite o-Xylene 64/99 [24]
BIOTRICKLING FILTERS:
Coal particles Acrylonitrile 465 / 95 [131]
Activated carbon Hydrogen sulphide 142/90 [132]
Coal particles Pentane + styrene 20 + 55 / 80 [133]
Sintered glass Propionaldehyde 500/ > 80 [134]
Pall Rings Toluene 83 / 35 [135]
*gNikg dry packing. d (addition of nutrients 4xlhJday)
5.2. INDUSTRIAL SCALE APPLICATIONS
Although the use of conventional biofilters with organic filter-beds remains the most
classical and most widely accepted alternative, a few pilot-scale and industrial-scale
bioreactors packed with inert carriers have recently been installed as well. Inert carrier
materials are in most cases packed in biotrickling filters. However, recent studies have
proven that the intermittent (weekly or even monthly) addition of a nutritive aqueous
phase to conventional biofilters packed with inert carriers may result in a very high
reactor performance [136]. Full-scale systems are already being operated under such
conditions.
Table 5 shows a list of compositions of some gaseous effluents successfully treated
in pilot- or full-scale biofilters. Over the past decade the number of applications
increased fast and it would be difficult to cite them all. Most examples published in the
literature and some non-published full-scale applications indicate that bioreactor
technologies are suitable above all when dealing with waste gases containing either
single compounds or simple mixtures of pollutants. However, good results have
sometimes also been obtained in the case of more complex mixtures as, for example, in
the biofiltration of petroleum hydrocarbons released to the atmosphere from soil vapour
extraction processes. The treatment of mixtures of pollutants usually provokes a
decrease of the biodegradation rates of each compound of the mixture in comparison
with the biodegradation rates of individual compounds [124]. However, sometimes very
248
similar elimination capacities may be reached for pollutants fed individually to a

biofilter or in mixture. This was the case for a waste gas containing toluene,
ethylbenzene and o-xylene [24]. Still in other situations, the presence of a given
pollutant may stimulate the removal of another one in case of co-metabolic processes.
The most recalcitrant pollutant will limit the overall removal rate, meaning that the
biofilter will be oversized for the most easily biodegradable pollutants.
Regarding the application range of full-scale biofilters, several parameters must be
considered. The flow rates may range from relatively low values of a few hundreds
cubic meters per hour [137] up to reported values of more than 200,000 m3/h [138].
Some projects on the treatment of waste gas flow rates greater than 1,000,000 m3/h have
been studied [124]. Pilot-scale and industrial-scale biofilters treating flow rates below
hundred cubic meters per hour have been described [139, 140]. However, this is less
usual.
Volumetric biofilter loading rates are between approximately 5 and 600 m3/m3 .h
[124]. Higher values are possible and have been published for bioscrubbers; reaching
1,428 m3/m3 .h in a bioscrubber treating a methanol polluted waste gas [5].
In most applications, total pollutant concentrations in the feed are of a few hundreds
mg per cubic meter but full-scale biofilters have also already been used to treat lower
concentrations, below 10 mglm3 , or higher concentrations of a few fams per cubic
meter. For the treatment of waste gases containing more than 3-5 glm , biofiltration is
normally not suitable [124]. Wright et al. [141] describe the treatment of air
contaminated with up to 2.7 glm3 of petroleum hydrocarbons in a pilot-scale compost
biofilter and with relatively high removal efficiencies. Several authors have shown that
when feeding high concentrations of pollutants, the removal efficiency will drastically
decrease. Jorio et al. [142] observed that feeding 6.2 glm3 toluene or 8.2 glm3 xylenes
resulted in, respectively, approximately 60% and 23% removal efficiencies at empty
bed residence times (EBRT) of 102 s for toluene and 78 s for the xylene isomers. The
best results are reached when combining low pollutant concentrations and relatively
high gas flow rates. Biodegradation of very low pollutant concentrations, of only a few
ppbv, is sometimes also difficult and may result in low removal efficiencies as well
[81].
Although optimal values for the empty bed residence time (EBRT) are most often
between 20 s and one minute, above 90% removal efficiency has been reported in full-
scale applications at EBRT below 20 s and even below 10 s in some cases [2]. Lower
loads, i.e. lower pollutant concentrations, allow using lower EBRT. At lab-scale, EBRT
of several minutes have sometimes been used [143], but at industrial scale such high
values are not recommended. Nevertheless, EBRT between 6 and 22 minutes have been
reported in pilot- and full- scale biofilters at sites where polluted air from soil vapour
extraction was treated [139]. Highly recalcitrant VOCs require longer EBRT than easily
biodegradable compounds. The optimal EBRT will also depend on the selected filter-
bed.
The largest above-ground biofilters built for the treatment of waste gases are of the
order of 3000-4000 m3 [138], while the area necessary for typical open bed soil
biofilters may reach 3000 m2 • If necessary, full-scale biofilters may be operated in series
or in parallel.
249
Frederic Thalasso, Marla C. Veiga and Christian Kennes
It has already been mentioned previously in this chapter that pressure drop is an
important parameter since it affects biofilter performance. At industrial scale, it is also
very important because pressure drop and costs will increase simultaneously. High
pressure drops may be reached after several years of operation mainly as a result of
compaction when using organic filter-beds or because of excess biomass accumulation
on inert packing materials. The presence of a liquid phase in trickling biofilters will also
lead to higher head losses and the need of more power of fans or blowers used for
feeding the waste gas to the reactor. Filter-bed replacement is generally required after 4-
5 year operation for organic carriers. Inert carriers may be used for longer periods of
time depending on the operating conditions.
Table 5. Examples of mixtures ofpollutants biodegradable in biofilters
Source of contamination Pollutants
Coating industry Toluene, VOCs

Commercial bakery Ethanol, methane, ethylacetate, aliphatics, VOCs
Composting Odours, ammonia, hydrocarbons
Fibreglass industry Styrene, acetone
Flavour industry Odours, flavours
Flexograpbic printing Alcohols, acetone
Food industries Odours
Foundry Ethanol, VOCs
Foundry Phenol, ammonia
Lacquering industry Toluene, ethylbenzene, xylenes, butyl acetate
Latex production Styrene, butadiene
Metal foundry Benzene
Oil production Odours
Pharmaceutical industry Alcohols, acetone, dichloroethane
Plastic dashboard manufacturing Styrene, butylacetate
Plastic resins Styrene
Polluted soil Gasoline
Pulp and paper industry Mixture of sulphur compounds
Slaughterhouses Odours
Rayon manufacturing H2S, CS2
Tobacco industry NH3, nicotine, odours
Wastewater treatment Odours
Wood industries Formaldehyde,phenol,methanol
6. Other biological gas treatment technologies
To be complete, other technologies described in the literature should be mentioned. The

use of membranes are for instance presented by Fischer, Hartmans et al., Reij et al. and
more recently by Parvatiyar et al. and Ergas et al. [144, 145, 146, 147, 148], an
interesting external loop airlift bioreactor was described by Ritchie and Hill [149], a
new technology using a double liquid phase was proposed by Cesario et al. [150, 151,
152], a biorotor reactor was used by Buisman et al. [153] for the removal of hydrogen
sulphide and an horizontal flow biofilter has been proposed by Lee et al. [154]. A
combination of photochemical oxidation and biodegradation seems a promising
250
alternative for some specific applications [155]. Nevertheless, these technologies have
only been tested on a lab-scale or pilot-scale.
7. Conclusions
Biological gas cleaning is, in some aspects, very different from the main biological
cleaning technologies currently used for water and/or soil treatment. Although the
biological mechanisms involved are quite similar, the gaseous state of the pollutant to
be treated involves new engineering concepts and has led to a new application field of
environmental biotechnology.
The new engineering concepts induced by the development of biological gas treatment
processes comprises mainly (i) the possibility of handling quite high flow rates, with (ii)
very short contact times, in (iii) technically simple reactors that (iv) ensures a synergy
between absorption and degradation processes. Nowadays, it is possible to treat up to
200,000 m3 waste gas per hour with a contact time of about 30 seconds in reactors
formed only of an organic filter-bed and a spraying system as major features.
The new application field opened by biological gas cleaning is characterized by the
possibility to treat volatile and/or non-water soluble pollutants, even at relatively low
concentrations. Biological gas cleaning is therefore a technology that complements the
more traditional treatment processes. It can be used with pollutants directly emitted to
the atmosphere from the production source, and with pollutants, which have a
significant environmental impact even at low concentration.
As a first relatively recent and obvious application field of biological gas cleaning
one should mention the treatment of industrial off-gases. Biofilters or other bioreactors
are therefore currently used in, chemical, food and beverages or mechanical industries
among others (Table 5). Secondly, biological gas treatment may be used as a polishing
step in combination with other treatment technologies. For instance, it is employed in
wastewater treatment plants to avoid the emission of volatile pollutants transferred from
the liquid to the gas phase. Thirdly, biological gas treatment can be used for site
remediation. Indeed, combined to soil vapour extraction or stripping, biological gas
treatment allows the elimination of the volatile fraction of soil pollutants. So, the
problems of handling large amounts of pollutants in on-site soil remediation
technologies or the problem of the low microbial kinetics in in-situ soil remediation can
partly be overcome. Finally, thanks to the recent developments in the removal of
metallic compounds and radionuclides, this gas treatment technology can be applied in
the remediation of sites polluted by non-organic airborne compounds.
In conclusion, after a few decades of development, biological gas treatment is now
seen as an increasingly valuable technology, which can play an important role in air
pollution control. It is a promising and significant tool, which assists mankind in
cleaning up our environment.
Acknowledgements
Part of this chapter has been prepared thanks to funds from project PPQ2001-0557 to
CK.
251
References
[1] Bach, H. (1924). Gesundheits-Ingenieur, 46,370-376.
[2] Ottengraf, S.P.P. and Diks, R (1992). Review paper: process technology of biotechniques. In
Biotechniques for air pollution abatement and odour control policies, A J. Dragt and 1. van Ham (Eds.),
Elsevier, Maastricht, The Netherlands, 17-32.
[3] Frechen, F.-B. (1994). Odour emissions of wastewater treatment plants-recent German experiences. Wat.
Sci. Techno!., 30, 35-46.
[4] Kok, H., (1992). Bioscrubbing of air contaminated with high concentration of hydrocarbons. In
Biotechniques for air pollution abatement and odour control policies, A. J. Dragt and J. van Ham (Eds),
Elsevier, Maastricht" 77-82.
[5] van Groenestijn, J.W. and HesseJink, P.G.M. (1993). Biotechniques for air pollution contro!.
Biodegradation, 4, 283-30l.
[6] Paul, E. (1987). Experiences with large-scale biological scrubber. In Biological treatment of industrial
waste gases, state of the art and comparison with physico-chemical processes, Heidelberg, Germany" oral
14.
[7] Shoda, M. (1991). Methods for the biological treatment of exhaust gases. In Biological degradation of
wastes, A. M. Martin (Ed.), Elsevier, London, 31-46.
[8] Neal A.B. and Loehr RC. (2000). Use of biofilters and suspended-growth reactors to treat VOCs. Waste
Manage., 20, 59-68.
[9] Anonymous (1995). The effective and reliable removal of both sulphur compounds and heavy metals from
water and (flue) gas. Paques b.v., Postbus 52, 8560 AB Balk, NL.
[10] Dijkman, H. (1995). Biological gas desulphurisation, Med. Fac. Landbouww. Univ. Gent, 60/4b, 2677-
2684.
[11] Ottengraf, S.P.P. (1987). Biological systems for waste gas elimination, Trends Biotechno!., 5, 132-136.
[12] Wang, Z., Govind, R and Bishop, D.F. (1996). Review of biofilttation - Effect of support media on
Biofilter performance. In 89th annual meeting & exhibition Air & Waste Management Assoc., Nashville,
Tennessee, 96-wp87 A05.
[13] Seignez C; Atti A; Adler N and Peringer P (2002). Effect of biotrick1ing filter operating parameters on
chlorobenzenes degradation. J. Environ. Eng., 128,360-366.
[14] Le Cloirec, P.; Humeau, P. and Ramirez-Lopez, E.M. (2001). Biotreatment of odours: control and
performances of a biofilter and a bioscrubber. Water Sci. Techno!., 44, 219-226.
[15] Smith, F.L.; Sorial, G.A.; Suidan, M.T.; Biswas, P. and Brenner, R.C. (2002). Development and
demonstration of an explicit lumped-parameter biofilter model and design equation incorporating Monod
kinetics. J. Air Waste Manage. Assoc., 52, 208-219
[16] Chang, K.S.; Lu, C.Y. and Lin, M.R. (2001). Treatment of volatile organic compounds from
polyurethane and epoxy manufacture by a trickle-bed air biofilter. J. Biosci. Bioeng., 92,126-130.
[17] Diks, RM.M., Ottengraf, S.P.P. and Vrijland, S. (1994). The existence of a biological equilibrium in a
trickling filter for waste gas purification, Biotechno!' Bioeng., 44,1279-1287.
[18] Pedersen, AR. and Arvin, E. (1995). Removal of toluene in waste gases using a biological trickling filter.
Biodegradation, 6, 109-118.
[19] Shinabe, K., Oketani, S., Ochi, T. and Matsumura, M. (1995). Characteristics of hydrogen sulphide
removal by Thiobacillus thiooxidans KSI isolated from a carrier-packed biological deodorization system.
J. Ferment. Bioeng., 80, 592-598.
[20] Thalasso, F., Ancia, R, Willocx, B., L'Hermite, Ph., Naveau, H. and Nyns, E.-J. (1993). The "Mist-
Foam" concept: A concept for biological treatment of gaseous organic compounds. In Characterisation
and control of odours and VOC in the process industries, Vigneron, S., Hermia, 1. and Chaouki, J. (Eds.),
Elsevier, Amsterdam, The Netherlands, 419-429.
[21] Weber, F.J. and Hartmans, S. (1996). Prevention of clogging in a biological trickle-bed reactor removing
toluene from contaminated air. Biotechno!' Bioeng., 50, 91-97.
[22] Smith, F.L., Sorial, G.A., Suidan, M.T., Breen, AW., Biswas, P. and Brenner, R.C. (1996). Development
of two biomass control strategy for extended, stable operation of highly efficient biofilters with high
toluene loadings. Environ. Sci. Techno!., 30, 1744-175l.
[23] Speitel, G.E. and McLay, D.S. (1993). Biofilm reactors for treatment of gas stream containing
chlorinated solvents. J. Env. Eng., 119,658-678.
[24] Kennes, C., Cox, H.H.J., Doddema, H.J. and Harder, W. (1996). Design and performance ofbiofilters for
the removal of alkylbenzene vapours. J. Chern. Techno!. Biotechno!., 66, 300-304.
252
[25] Garcia-Pena, E.I.; Hernandez, S.; Favela-Torres, E.; Auria, R. and Revah, S. (2001). Toluene biofiltration
by the fungus Scedosporium apiospermum TBl. Biotechnol. Bioeng., 76, 61-69.
[26] De Heyder, B., Overmeire, A., Van Langenhove, H. and Verstraete, W. (1994). Ethene removal from a
synthetic waste gas using a dry biobed, Biotechnol. Bioeng., 44, 642-648.
[27] Devinny, 1.S. and Hodge, D.S. (1995). Formation of acidic and toxic intermediates in overloaded ethanol
biofilters. 1. Air Waste Manage. Assoc., 45,125-131
[28] Kirchner, K., Hauk, G. and Rehm, H.l. (1987). Exhaust gas purification using immobilised monocultures
(biocatalyst), Appl. Microbiol. Biotechnol., 26, 579-587.
[29] Moe, W.M. and Irvine, R.L. (2001). Effect of nitrogen limitation on performance of toluene degrading
biofilters. Wat. Res., 35,1407-1414
[30] Moe WM and Irvine RL (2001). Polyurethane foam based biofilter media for toluene removal. Water Sci.
Technol., 43, 35-42.
[31] Pol A.; van Haren F.J.l.; den Camp 1.H.M.O. and van der Drift, C. (1998). Styrene removal from waste
gas with a bacterial biotricking filter. Biotechnol. Lett., 20,407-410.
[32] Chitwood, D.E. and Devinny, 1.S. (2001). Treatment of mixed hydrogen sulphide and organic vapours in
a rock medium biofilter. Water Environ. Res., 73, 426-435.
[33] Shareefdeen, Z., Baltzis, B.C., Oh, Y.-S. and Bartha, R. (1993). Biofiltration of methanol vapours.
Biotechnol. Bioeng., 41, 512-524.
[34] Lu, C.S.; Lin, M.R. and Chu, C.H. (2002). Effects of pH, moisture, and flow pattern on trickle-bed air
biofilter performance for BTEX removal. Adv. Environ. Res., 6, 99-106.
[35] Thalasso, F., Naveau, H. and Nyns, E.-I. (1996). Effect of dry periods in a "Mist-Foam" bioreactor
designed for gaseous substrate. Environ. Technol., 17,909-913.
[36] Zhu, X.Q.; Alonso, C.; Suidan, M.T.; Cao, H.w.; Kim, B.J. and Kim, B.R. (1998). The effect of liquid
phase on VOC removal in trickle-bed biofilters. Water Sci. Technol., 38, 315-322.
[37] Le Cloirec, P., Fanlo, l-L. and Degorce-Dumas, 1. (1991). Traitement des odeurs et desodorisation
industrielle. Innovation 128, Paris, 1991.
[38] Iorio H; Bibeau L . and Heitz M (2000). Biofiltration of air contaminated by styrene: Effect of nitrogen
supply, gas flow rate, and inlet concentration. Environ. Sci. Technol., 34, 1764-177l.
[39] Okkerse, W.J.H.; Ottengraf, S.P.P.; Osinga-Kuipers, B. and Okkerse, M. (1999). Biomass accumulation
and clogging in biotrickling filters for waste gas treatment. Evaluation of a dynamic model using
dichloromethane as a model pollutant. Biotechnol. Bioeng., 63,418-430.
[40] Alonso, C.; Zhu, X.Q.; Suidan, M.T.; Kim, B.R. and Kim, B.J. (2001). Mathematical model of
biofiltration ofVOCs: Effect of nitrate concentration and backwashing. 1. Environ. Eng., 127,655-664.
[41] Sorial, G.A.; Smith, F.L.; Suidan, M.T. and Brenner, R.C. (2001). Removal of ammonia from
contaminated air by trickle bed air biofilters. 1. Air Waste Manage. Assoc., 51, 756-765.
[42] Lim, 1.S.; Park, S.l.; Koo, 1.K. and Parle, H (2001). Evaluation of porous ceramic as microbial carrier of
biofilter to remove toluene vapours. Environ. Technol., 22, 47-56.
[43] Kennes, C. and Veiga, M.C. (2002). Inert filter media for the biofiltration of waste gases-Characteristics
and biomass control. ReNiews. Environ. Sci. Bioffechnol, 1,201-214.
[44] Bentz, R. (1987). Biological waste treatment: experience of a chemical company in Switzerland. In
Biological treatment of industrial waste gases. State of the art and comparison with physico-chemical
processes, Dechema, Heidelberg, oral 13.
[45] Hereth, H. (1987). Biological waste air treatment in a gelatine plant In Biological treatment of industrial
waste gases. State of the art and comparison with physico-chemical processes, Dechema, Heidelberg, oral
1l.
[46] Lebeault, 1.-M. (1993). Le traitement des effluents gazeux. Biofutur, September, 29-31.
[47] Prokop, W. and Bohn, H. (1985). Soil bed system for control of rendering plant odours. In 78th annual
meeting of the Air Pollution Control Association, Detroit, Michigan, 6, 85-79.
[48] Clarle, R. and Wnorowski, A. (1992). Biofilters for sewer pump station vents: influence of matrix
formulations on the capacity and efficiency of odorant removal by an experimental biofilter. In
Biotechniques for air pollution abatement and odour control policies, A. 1. Dragt and 1. van Ham (Eds.),
Elsevier, Maastricht, The Netherlands, 183-186.
[49] Bohn, H. L. (1996). Biofilter media. In 89th annual meeting & exhibition Air & Waste Management
Assoc., Nashville, Tennessee, 96-WP87A.01.
[50] Tang, H.-M., Hwang, S.-l. and Hwang, S.-C. (1996). Waste gas treatment in biofilters. 1. Air Waste
Manage. Assoc., 46, 349-354.
[51] Morgeuroth, E., Schroeder, E.D., Chang, D.P.Y. and Scow, K.M. (1996). Nutrient limitation in a
compost biofilter degrading hexane. 1. Air Waste Manage. Assoc., 46, 300-308.
253
[52] Veir, 1. K, Schroeder, E.D., Chang, D.P.y' and Scow, KM. (1996). Interaction between Toluene and
Dichloromethane degrading populations in a compost biofilter. In 89th annual meeting & exhibition Air
& Waste Manage. Assoc., Nashville, Tennessee, 96-WP87AJJ7.
[53] Elias A; Barona A; Arreguy A; Rios J; Aranguiz I; Penas J (2002). Evaluation of a packing material for
the biodegradation of H 2S and product analysis. Process Biochem., 37 (8), 813-820.
[54] Smet, E., Chasaya, G., Van Langenhove, H. and Verstraete, W. (1996). The effect of inoculation and the
type of carrier material used on the biofiltration of methyl sulphides. Appl. Microbiol. Biotechnol., 45,
293-298.
[55] Ibrahim, M.A.; Mizuno, H.; Yasuda, Y.; Fukunaga, K and Nakao K (2001). Removal of mixtures of
acetaldehyde and propionaldehyde from waste gas in packed colunm with immobilized activated sludge
gel beads. Biochem. Eng. J., 8,9-18.
[56] Baltzis, B.C. and Shareefdeen, Z. (1994). Biofiltration of VOC mixtures: Modeling and pilot Scale
experimental verification. In 87th annual meeting & exhibition Air & Waste Management Assoc.,
Cincinnati, Ohio, A907.
[57] Zilli, M., Fabiano, B., Ferraiolo, A and Converti, A (1996). Macro-kinetic investigation on phenol
uptake from air by biofiltration: Influence of superficial gas flow rate and inlet pollutant concentration.
Biotechnol. Bioeng., 49, 391-398.
[58] Deshusses, M.A, Hamer, G. and Dunn, I.J. (1995). Behaviour of biofilters for waste air biotreatment. 2.
Experimental evaluation of a dynamic model. Environ. Sci. Technol., 29,1059-1068.
[59] Arulneyam, D. and Swarninathan, T. (2000). Biodegradation of ethanol vapour in a biofilter. Bioprocess
Eng., 22, 63-67
60] Weber, F. and Hartmans, S. (1995). Use of activated carbon as a buffer in biofiltration of waste gases with
fluctuating concentration of toluene. Appl. Microb. Biotechnol., 43, 365-369.
[61] Abumaizar, R.J.; Kocher, W. and Smith, E.H. (1998). Biofiltration of BTEX contaminated air streams
using compost-activated carbon filter media. J. Hazard Mat., 60, 111-126.
[62] Tholander, P. (1987). Biological processes for odour abatement in industrial processes. In Biological
treatment of industrial waste gases. State of the art and comparison with physico-chemical processes,
Dechema, Heidelberg, poster 10.
[63] Dragt, A and Ottengraf, S.P.P. (1987). Process engineering aspects and new developments in biological
air pollution control technology.. In Biological treatment of industrial waste gases, state of the art and
comparison with physico-chemical processes, Heidelberg, Germany, oral 21.
[64] Ottengraf, S.P.P., Meesters, J.J.P., van den Oever, AH.C. and Rozema, H.R. (1986). Biological
elimination of volatile xenobiotic compounds in biofilters. Bioprocess Eng., 1, 61-69.
[65] Bohn, H. (1975). Soil and compost filters for malodorous gases. J. Air Pol. Control. Assoc., 25, 953-955.
[66] Pomeroy, R.D. (1982). Biological treatment of odorous air, J. WPCF, 54,1541-1545.
[67] Bohn, H.L. (1993). Biofiltration: Design principles and pitfalls. In 86th annual meeting & exhibition Air
& Waste Manageme. Assoc., Denver, Colorado, 93-TP-52AOl.
[68] Cardenas-Gonzalez, B.; Ergas, SJ.; Switzenbaum, M.S. and Phillibert, N. (1999). Evaluation of full-scale
biofilter media performance. Environ. Prog., 18, 205-211.
[69] Cardenas-Gonzalez, B.; Ergas, S.J. and Switzenbaum, M.S. (1999). Characterization of compost
biofiltration media. J. Air Waste Mauage. Assoc., 49, 784-793
[70] Dalouche, A, Lemasie, M., Le Cloirec, P., Martin, G. and Besson, G. (1989), Utilisation de biofiltres
pour l'epuration de gaz charges en composes azotes et soums. In Man and his ecosystem. 8th world
clean air congress, eds. Brasser L.J. and W.C. Mulders, The Hague, The Netherlands, 1989,379-384.
[71] Kennes, C., Cox, H.HJ., Veiga, M.C. and Doddema, HJ. (1995). Continuous removal of benzene related
compounds from waste gases. Med. Fac. Landbouww. Univ. Gent., 60, 279-284.
[72] Delhomenie, M.C., Bibeau, L., Roy, S., Brzezinski, R. and Heitz, M. (2001). Influence of nitrogen on the
degradation of toluene in a compost -based biofilter. J. Chern. Technol. Biotechnol. 76, 997-1006
73] Van Langenhove, H. and Schamp, N. (1987). Inhibitory effect of sulphur dioxide on biofiltration of
aldehydes. In Biological treatment of industrial waste gases. State of the art and comparison with
physico-chemical processes, Dechema, Heidelberg, poster 8.
[74] Holubar, P. and Braun, R. (1995). Biofiltration - bottlenecks in biological air purification and possible
future solutions, Med. Fac. Landbouww. Univ. Gent, 60/4b, 2303-2312.
[75] Ottengraf, S.P.P. and Van den Oever, AH.C. (1983). Kinetics of organic compound removal from waste
gases with a biological filter, Biotechnol. Bioeng., 25, 3089-3102.
[76] Rands, M., Cooper, D., Woo, C.-P., Fletcher, G. and Rolfe, K (1981). Compost filters for H2S removal
from anaerobic digestion and rendering exhausts, 1. WPCF, 53, 185-189.
254
[77] Yang, Y., and Allen, E.R. (1994). Biofiltration control of hydrogen sulfide. 1. Design and operational
parameters. J. Air Waste manage. Assoc., 44,863-868.
[78] Allen, E.R. and Phatak, S. (1993). Control of organo-sulphur compound emissions using biofiltration -
Methyl mercaptan. In 86th annual meeting & exhibition Air & Waste Manage. Assoc., Denver, Colorado,
93-WA-52B.03.
[79] Nicolai, R.E. and Janni, KA. (2001). Biofilter media mixture ratio of wood chips and compost treating
swine odours. Water Sci. Technol., 44,261-267.
[80] Ergas, S.1., Kinney, K, Fuller, M.E. and Scow, KM. (1994) Characterization of a compost biofiltration
system degrading dichloromethane. Biotechnol. Bioeng., 44, 1048-1054.
[81] Ergas, S.1., Schroeder, E.D., Chang, D.P.Y and Morton, R.L. (1995). Control of volatile organic
compound emissions using a compost biofilter. Water Environ. Res., 67, 816-821.
[82] Hodge, D. S., Medina, V.F., Islander, R.L. and Devinny, J.S. (1991). Treatment of Hydrocarbon fuel
vapours in biofilters. Environ. Techno!., 12, 656-662.
[83] Deshusses, M.A. and Hamer, G. (1993). The removal of volatile ketone mixtures from air in biofilters.
Bioprocess. Eng., 9,141-146.
[84] Solla Martin, C. (1995). Evaluaci6n de nuevos materiales de relleno para la depuraci6n de aire en
biofiltros. B.S. thesis, University of Santiago de Compostela, Spain.
[85] Cho, K, Hirai, M. and Shoda, M. (1992). Enhanced removal efficiency of malodorous gases in a pilot-
scale peat biofilter inoculated with Thiobacillus thioparus DW44. J. Ferm. Bioeng., 73, 46-50.
[86] Dalouche, A., Gillet, M., Lemasle, M., Martin, G. and Orain, L. (1981). Biodesodorisation des effluents
gazeux. Pol. Atm., 23, 317-322.
[87] Hirai, M., Ohtake, M. and Shoda, M. (1990). Removal kinetics of hydrogen sulfide, methanethiol and
dimethyl sulfide by peat biofilters, J. Ferm. Bioeng., 70, 334-339.
[88] Morales, M., Perez, F., Auria, R. and Revah, S. (1994). Toluene removal from air stream by biofiltration,
Adv. Bioproc. Eng., 405-411.
[89] Yoon, I.K and Park, C.H. (2002). Effects of gas flow rate, inlet concentration and temperature on
biofiltration of volatile organic compounds in a peat-packed biofilter. J. Biosci. Bioeng., 93,165-169.
[90] Rothenbiihler, M., Heitz, M., Beerli, M. and Marcos, B. (1995). Biofiltration of volatile organic
emissions in reference to flexographic printing processes. Water Air Soil Poll., 83, 37-50.
[91] Baltzis, B.C. and Androutsopoulou, H. (1994). A study on the response of biofilters to shock-loading. In
87th annual meeting & exhibition Air & Waste Management Assoc., Cincinnati, Ohio, 1994, A905.
[92] Don, J. and Feenstra, L. (1984). Odour abatement through biofiltration. In Characterisation and control of
odoriferous, SBF (Ed.), SBF, Louvain-Ia-Neuve, Belgium, 337-349.
[93] ZiIli, M.; Palazzi, E.; Sene, L.; Converti, A. and Del Borghi, M. (2001). Toluene and styrene removal
from air in biofilters. Proc. Biochem., 37,423-429.
[94] Van Langenhove, H., Wuyts, E. and Schamp, N. (1986). Elimination of hydrogen sulphide from odorous
air by a wood barkbiofilter, Wat. Res., 20,1471-1476.
[95] Lee, B.D., Apel, W.A., Walton, M.R. and Cook, L.L. (1996). Treatment of methanol contaminated air
streams using biofiltration. In 89th annual meeting & exhibition Air & Waste Manage. Assoc., Nashville,
Tennessee, 96-RP87C.03.
[96] Bohn, H. and Bohn, R. (1988). Soil beds weed out pollutants. Chern. Eng., Apri125, 73-76.
[97] Frye, R.J., Welsh, D., Berry, T.M., Stevenson, B.A. and McCallum, T. (1992). Removal of contaminant
organic gases from air in closed systems by soil. Soil BioI. Biochem., 24,607-612.
[98] Faulo, J.-L., Degorce-Dumas, J. R., Kowal, S. and Le Cloirec, P. (1991). Procede biologique de
traitement de gaz, biofiltres et application 11 la desodorisation de gaz, Patent 91.053346.
[99] Atlas, R. (1989). Microbiology, Fundamentals and Applications. Macmillan, New York.
[100] VanDemark, P. and Batzing, B. (1987). The microbes: An introduction to their nature and importance.
Benjamin/Cummings, Menlo Park, California.
[101] Van Lith, c., David, S.L. and Marsh, R. (1990). Design criteria for biofilters. Trans IChemE, 68, 127-
132.
[102] Ebinger, M.H., Bohn, H.L. and Puis, R.W. (1987). Propane removal from propane-air mixtures by soil
beds. J. Air Poll. Control. Assoc., 37, 1486-1489.
[103] Bohn, H.L. and Bohn, KH. (1999). Moisture in biofilters. Environ. Prog., 18, 156-161
[104] Deshusses, M.A., Hamer. G. and Dunn, U. (1996). Transient-state behaviour of a biofilter removing
mixtures of vapours ofMEK and MffiK from air. Biotechnol. Bioeng., 49, 587-598.
[105] Jol, A. and Dragt, A. (1988). Biotechnological elimination of volatile organic compounds in waste
gases, J. Dechema Biotech. Conf., 2, 373-389.
255
[106] Kleis, G., Schelchshom, l and Vinke, A. (1987). Desulphurisation of H2S containing gases by
biological oxidation. In Biological treatment of industrial waste gases. State of the art and comparison
with physico-chemical processes, Dechema, Heidelberg, poster 3.
[107] Elsgaard, L. (2000). Ethylene removal at low temperatures under biofilter and batch conditions. Appl.
Environ. Microbiol., 66, 3878-3882.
[108] Lehtomiiki, J., Torronen, M. and Laukkarinen, A. (1992). A feasibility study of biological waste-air
purification in a cold climate. In Biotechniques for air pollution abatement and odour control policies, A.
J. Dragt & J. van Ham (Eds), Elsevier, Maastricht, The Netherlands, 131-134.
[109] Lee, B.D.; Apel, W.A. and Smith, W.A. (2001). Oxygen effects on thermophilic microbial populations
in biofilters treating nitric oxide containing off-gas streams. Environ. Prog., 20, 157-166
[110] Matteau, Y. and Ramsay, B. (1999). Thermophilic toluene biofiltration. J. Air Waste Manage. Assoc.,
49, 350-354.
[111] Plas, C., Holubar P., Moser, K., Ploder W. and Braun, R. (1994). Die Bilanzierung von wasser und
kohlenstoffbei der biofiltration, VOI-Berichte, 1104,273-278.
[112] Yang, Y. and Allen, E.R. (1994). Biofiltration control of hydrogen sulfide. 2. Kinetics, biofilter
performance and maintenance. J. Air Waste Manage. Assoc., 44,1315-1321.
[113] Leson, G. and Winer, A.M. (1991). Biofiltration: An innovative air pollution control technology for
VOC emissions. J. Air Waste Manage. Assoc., 41,1045-1054.
[114] Bronnenmeier, R., Fitz, P. and Tautz, H. (1994). Reinigung von Lackiererei-Abluft mit einem
Gittertrager-Biofilter. VOl Berichte, 1104, 203-215.
[115] Hartmans, S. and Tramper, J. (1991). Dichloromethane removal from waste gases with a trickle-bed
bioreactor. Bioproc. Eng., 6, 83-92.
[116] Acuna, E., Auria, R., Pineda, J., Perez, F., Morales, M. and Revab, S. (1996). Studies on the
microbiology and kinetics of a biofilter used to control toluene emissions. In 89th annual meeting &
exhibition Air & Waste Management Assoc., Nashville, Tennessee, 96-WP87 A.03.
[117] Windsperger, A., Buchner, R. and Stefan, K. (1990). Rienigung lOsungmittelhaitiger abluft mit
biofiltem. Staub-Reihaltung der Luft, 50, 465-470.
[118] Hunik, J.H., Meijer, H.lC. and Tramper, J. (1992). Kinetics of Nitrosomonas europea at extreme
substrate, product and salt concentration. Appl. Microbiol. Biotechnol., 37, 802-807.
[119] Smits, M.C.J., Hoek, A.P., Osinga, B. Ottengraf, S.P.P and Wijngaard, M.H. (1995). Removal of
anunonia and odour from composting facility waste gas with a bio-trickling filter. Med. Fac.
Landbouww. Univ. Gent., 60, 2313-2320.
[120] Martin, G., Gaid, K., Lemasle, M. and Nogrix, P. (1979). Etude d'un procede biologique de
desodorisation, Pol. Atm., 21, 31-34.
[121] Smit, F. and Derber, H. (1987). Biofiltration - an economical and efficient waste gas treatment method.
In Biological treatment of industrial waste gases. State of the art and comparison with physico-chemical
processes, Dechema, Heidelberg, Germany, poster 15.
[122] Kennes, C., Veiga, M.C. and Bhatnagar, L. (1998). Methanogenic and perchloroethylene-dechlorinating
activity of anaerobic granular sludge. Appl. Microbiol. Biotechnol. 50,484-488.
[123] Kim, J.O. (1997). Gaseous TCE and PCE removal by an activated carbon biofilter. Bioproc. Eng. 16,
331-337.
[124] Kennes, C. and Thalasso, F. (1998). Waste gas biotreatment technology. J. Chern. Technol. Biotechnol.
72,303-319.
[125] Veiga, M.C. and Kennes, C. (2001). Parameters affecting performance and modeling of biofilters
treating alkylbenzene-polluted air. Appl. Microbiol. Biotechnol. 55, 254-258.
[126] Kim, N.J., Sugano, Y., Hirai, M. and Shoda, M. (2000). Removal characteristics of high load anunonia
gas by a biofilter seeded with a marine bacterium, Vibrio alginolyticus. Biotechnol. Lett. 22, 1295-1299.
[127] Smet, E., Van Langenhove, H. and Philips, G. (1999). Dolomite limits acidification of a biofilter
degrading dimethyl sulphide. Biodegradation. 10, 399-404.
[128] Wani, A.H., Lau, A.K. and Branion, R.M.R. (1999). Biofiltration control of pulping odors - hydrogen
sulfide: performance, macrokinetics and co-existence of effects of organo-sulphur species. J. Chern.
Technol. Biotechnol. 74, 9-16.
129 Mohseni, M. and Allen, D.G. (2000). Biofiltration of mixtures of hydrophilic and hydrophobic volatile
organic compounds. Chern. Engin. Sci. 55,1545-1558.
[130] Cox, H.H.J., Moerman R.E., van Baalen S., van Heiningen W.N.M., Doddema H.J. and Harder W.
(1997). Performance of a styrene-degrading biofilter containing the yeast Exophiala jeanselmei.
Biotechnol. Bioeng. 53, 259-266.
256
[131] Lu, C., Lin, M.-R. and Lin, J. (2000). Removal of acrylonitrile vapours from waste gases by a trickle-
bed air biofilter. Biores. Technol. 75, 35-41.
[132] Guey, C., Degorge-Dumas, J.R. and Le C1oirec, P. (1995). Hydrogen sulfide removal on biological
activated carbon. J. Odours VOC's. 1, 144-145.
[133] Lu, C., Lin M.-R. and Wey I. (2001). Removal of pentane and styrene mixtures from waste gases by a
trickle-bed air biofilter. J. Chern. Technol. Biotechnol. 76, 820-826.
[134] Kirchner, K., Wagner, S. and Rehm, H.-J. (1992). Exhaust gas purification using biocatalysts (fixed
bacteria monocultures) - the influence of biofim diffusion rate (02) on the overall reaction rate. Appl.
Microbiol. Biotechnol. 37,277-279.
[135] Cox, H.H.J. and Deshusses, M.A. (1999). Biomass control in waste air biotrickling filters by protozoan
predation. Biotechnol. Bioeng. 62: 216-224.
[136] Prado, O.J., Mendoza J.A., Veiga M.C. and Kennes C. (2002). Optimization of nutrient supply in a
downflow gas-phase biofilter packed with an inert carrier material. Appl. Microbiol. Biotechnol. (in
press).
[137] Windsperger, A. (1991). Use of biofilters for the purification of gases containing solvents. Radex
Rundschau. 3-4, 457-464.
[138] Huber, 1. (1992). Planung, Durchfiihrung und erste Erfahrungen zum Biofilter
Tierkorperbeseitigungsanlage Plattling. In Dragt and van Ham (eds.), Biotechniques for air pollution
abatement and odour control policies. Elsevier, Amsterdam, 161-165.
[139] Leson, G. and Smith, B.J. (1997). Petroleum environmental research forum field study on biofilters for
control of volatile hydrocarbons. J. Environ. Eng. 123: 556-562.
[140] Swanson, W.J. and Loehr R.C. (1997). Biofiltration: Fundamentals, design and operation principles and
applications of biological APC technology. J. Environ. Eng. 123,538-546.
[141] Wright, W.F., Schroeder, E.D., Chang, D.P.Y., and Romstad, K. (1997). Performance of a pilot-scale
compost biofilter treating gasoline vapours. J. Environ. Eng. 123,547-555.
[142] Jorio, H., Kiared, K., Brzezinski, R., Leroux, A., Viel, G. and Heitz, M. (1998). Treatment of air
polluted with high concentrations of toluene and xylene in a pilot-scale biofilter. J. Chern. Technol.
Biotechnol. 73,183-196.
[143] Kraislas, S., Than Pham, Q., Amal, R., Jiang, J.K. and Heitz, M. (2ooo). Effect of inlet mass loading,
water and total bacteria count on methanol elimination using upward flow and downward flow biofilters.
J. Chern. Technol. Biotechnol. 75, 299-305
[144] Fischer, K. (1992). Vergleichende Untersuchungen: Biofilter- oder Biomembranvefahren zur reinigung
losemittelhaltiger industrieabluft. In Biotechniques for air pollution abatement and odour control policies,
A. J. Dragt andJ. van Ham (Eds.), Elsevier, Maastricht, The Netherlands, 97-102.
[145] Hartmans, S., Leenen, E.1.T.M and Voskuilen, G.T.H. (1992). Membrane bioreactor with porous
hydrophobic membranes for waste-gas treatment In Biotechniques for air pollution abatement and odour
control policies, A. J. Dragt and J. van Ham (Eds.), Elsevier, Maastricht, The Netherlands, 103-106.
[146] Reij, M.W., de Gooijer, K.D., de Bont, J.A.M. and Hartmans, S. (1995). Membrane bioreactor with a
porous hydrophobic membrane as a gas-liquid contactor for waste gas treatment. Biotechnol. Bioeng., 45,
107-115.
[147] Parvatiyar, M.G., Govind, R. and Bishop, D. F. (1996). Treatment of trichloroethylene (TCE) in a
membrane biofilter. Biotechnol. Bioeng., 50, 57-64.
[148] Ergas, S.1., Shumway, L., Fitch, M.W. and Neeman, J.J. (1999). Membrane processes for biological
treatment of contaminated airstreams. Biotechnol. Bioeng. 63, 431-44.
[149] Ritchie, B.J. and Hill, G.A. (1995). Biodegradation of phenol-polluted air using and extemalloop airlift
bioreactor. J. Chern. Technol. Biotechnol., 62, 339-344.
[150] Cesario, M.T., Beeftink, H.H. and Tramper, J. (1992). Biological treatment of waste gases containing
poorly soluble pollutants. In Biotechniques for air pollution abatement and odour control policies, A.J.
Dragt and J. van Ham (Eds.), Elsevier, Maastricht, The Netherlands, 135-140.
[151] Cesario, M.T., Beeftink H.H. and Tramper, J. (1994). Removal of poorly water-soluble compounds
from waste gases using water-immiscible organic solvents. In 6th. Eur. Congr. Biotechnol., L.
Alberghina, L. Frontali, and P. Sensi (Eds.), Elsevier, Amsterdam, 1207-1210.
[152] Cesario, M.T., Beeftink, H.H. and Tramper, J. (1995). Feasibility of using organic water-immiscible
solvents in biological waste-gas treatment. Bioproc.Eng. 12, 55-63.
[153] Buisman, C.J., Wit, B. and Lettinga, G. (1990). Biotechnological sulphide removal in three
polyurethane carrier reactors: Stirred reactor, biorotor reactor and upflow reactor. Wat. Res., 24, 245-251.
[154] Lee, D.H.; Lau, A.K. and Pinder, K.L. (2001). Development and performance of an alternative biofilter
system. J. Air Waste Manage. Assoc., 51, 78-85.
257
[155] Van Groenestijn, J.W., Doddema, R., Kok, R. and Koster, T.P.M. (1994). Combined photochemical and
biological treatment of off-gases. VDI Berichte 1104, 313-324.
258
BIOREACTORS FOR THE TREATMENT OF INDUSTRIAL WASTE GASES
CONTAINING FORMALDEHYDE AND OTHER ALIPHATIC COMPOUNDS
OSCAR J. PRADO, MARTA EIROA, MARiA C. VEIGA AND

CHRISTIAN KENNES
University of La Corufia, Chemical Engineering Laboratory, Campus da
Zapateira, E -15071 - La Corufia, Spain. Fax: 34 981167065. e-mail:
kennes@udc.es
1. Introduction
1.1. GAS PHASE BIOREACTORS
Over the past decades, bioreactors have proven to be efficient and cheap systems for the
abatement of a variety of common air pollutants. Among their main advantages, one
should mention their high efficiency, minimal side-effects on health and on the
environment and their relatively low cost. Three basic types of bioreactors can be
distinguished [1]:
• Bioscrubber (Fig. Ia): composed of an absorption column, where the pollutants

are absorbed in a liquid phase, and a stirred tank bioreactor, in which
biodegradation takes place.
• Trickling bioftlter (Fig. Ib): consists of a fixed ftlm bioreactor, which is
continuously fed a liquid medium.
• Biofilter (Fig. Ic): similar to the previous one, but no continuous supply of
liquid medium is used (liquid supply can be periodical or simply non-existent).
Conventional biofilters based on organic or natural ftlter beds have been used for
several decades for the treatment of polluted air streams mainly at wastewater treatment
plants and composting facilities. Much more recently, the application ofbioftltration has
been extended to new sources, among others industrial waste gases. The design of
biofilters has been improved, new ftlter beds have been tested and new types of
bioreactors have been developed. The present chapter describes the use of biofilters for
the treatment of industrial waste gases containing a mixture of formaldehyde and other
aliphatic compounds as methanol. Very little has been published on this topic and the
first papers dealing with such industrial waste gases were published only a few years
ago, in the nineties.
259
S.N Agathos and W Reineke (eds.), Biotechnologyfbr the Environment:
6scar J. Prado, Marta Eiroa, Marfa C. Veiga and Christian Kennes
.J!t:
Clean air Aqueous
Clean air Clean
phase
air1't (optional)
I
Polluted
~. . .
}}}:Y:
W""Md
nutrients
air ;:;:;:;:;:;:;:;:;:
--7 (b) (e)
Polluted L..----ti-----I
air
Bioreactor
t Water
t1' Polluted
Absorption Sludge
column drain drain air
Fig. 1: Bioreactor schemes: (a) Bioscrubber; (b) Trickling biofilter; (c) Biofilter
In this chapter, after reviewing the literature available on formaldehyde biodegradation,

case studies on biofiltration of formaldehyde-containing waste gases will be described,
before presenting our own preliminary data on the treatment of formaldehyde
containing waste gases in biofilters and biotrickling filters packed with different inert
carriers. The possibility of treating waste gases containing both formaldehyde and
methanol is described.
1.2. FORMALDEHYDE AS INDUSTRIAL AIR POLLUTANT
Formaldehyde (HCHO) is a common compound in the chemical industry used in a wide

variety of processes and frequently found in wastes, causing environmental pollution. It
is a colourless gas at normal temperature and pressure. Formaldehyde presents a
characteristic pungent odour and it is irritating to the mucous membranes at
concentrations above 20 mg/L [2]. It is readily soluble in water, alcohols, and other
polar solvents, but has a low degree of solubility in non-polar fluids. Some physical and
chemical properties of formaldehyde are shown in table 1.
Table 1: Physical and chemical properties offormaldehyde (modified from [2 J)
Relative molecular mass 30.03

Relative gas density (air = I) 1.075
Boiling point ("C) -19
Melting point ("C) -118
Inflammation temperature (0C) 430
Henry's constant (Pa·m3/mol) 0.02
Formaldehyde is present in the environment originating from natural processes and

man-made sources. It is formed in large quantities in the troposphere by the oxidation of
hydrocarbons. Minor natural sources include the decomposition of plant residues.
Formaldehyde is produced industrially in large quantities and used in many
applications, as in glue production, wood products, preservatives, permanent press
fabrics, paper product coatings and certain insulation materials. Building products made
260
Bioreactors for the treatment of industrial waste gases
with formaldehyde resins can emit formaldehyde gas. These products include
particleboard used as sub-flooring or shelving, fibreboard in cabinets and furniture,
plywood wall panels and foamed insulation. Incomplete combustion, cigarette smoking
and burning wood, kerosene and natural gas also release formaldehyde. In our
laboratory, formaldehyde biodegradation in waste gases is studied because of its
presence in gaseous effluents of industrial resin/glue producing industries. Such
resins/glues are mainly used in wood industries. Formaldehyde and urea are major raw
materials in their production.
Formaldehyde can show adverse effects on humans exposed to high concentrations
of the pollutant. Symptoms of formaldehyde exposure include nausea, vomiting,
abdominal pain or diarrhoea. Formaldehyde is considered to present a carcinogenic risk;
it has caused cancer in laboratory animals. It can react with microbial DNA and RNA
molecules, as well as proteins, resulting in cell damage.
2. Biodegradation of formaldehyde
In spite of its inhibitory effect to microorganisms, formaldehyde is known to be

biodegradable under aerobic conditions. Background information on formaldehyde
biodegradation is given below. Most studies deal with batch experiments in aqueous
media, although some authors have also used continuous liquid phase reactors.
Bonastre et al. [3] studied the biological degradation of formaldehyde using the
activated sludge process in order to carry out kinetic studies. In their work, the
experimental system consisted of reaction vessels containing a medium stirred
magnetically and aerated with compressed air. The tests were performed in a
thermostatic bath at three different temperatures of 15, 25, and 35°C. The initial
formaldehyde concentrations ranged from 100 to 2300 mgIL.
An analysis of the experimental data was performed, observing a satisfactory fit
between the data and the Vanillin's kinetic model, corresponding to Equation 1.
(1)
In Equation 1, Il is the specific rate of substrate consumption, Ilmax is the maximal

specific rate of substrate consumption, S is the substrate concentration, So is the initial
substrate concentration, X is the biomass concentration and nand Ks are kinetic
constants.
The data were minimised using equation 1 and the following maximal specific rates
of substrate consumption were obtained «mgIL)sI(mgIL)x-h): between 0.4 and 0.8 at 15
DC, between 1.0 and 2.2 at 25°C and between 4.0 and 6.0 at 35 dc. The relation between
the maximal specific rate of substrate consumption and temperature followed an
Arrhenius correlation (Equation 2).
1
In,llmax = -9413.05-+32.13 (2)
T
261
6scar J. Prado. Marta Eiroa, Maria C. Veiga and Christian Kennes
where ~ is expressed in h- 1 and T in °C.

Different authors have shown the biodegradability of formaldehyde by pure microbial
cultures. Adroer et al. [4] reported formaldehyde biodegradation by a strain of
Pseudomonas putida. Formaldehyde-using bacteria were isolated from sludge of an
industrial wastewater treatment plant by successive enrichments in a salt medium
containing formaldehyde. The isolated microorganisms were used in a fluidised bed
bioreactor treating an industrial wastewater containing formaldehyde. From this
bioreactor the microorganism used in this work was selected as a formaldehyde-using
bacterium. The isolated strain was classified as belonging to the species Pseudomonas
putida. Batch assays were carried out in a shaking bath at 30°C, containing a cell
suspension (30 - 180 mgIL) and formaldehyde (250 - 500 mg/L).
Pseudomonas putida was grown batch wise on formaldehyde in order to study the
degradation of formaldehyde by growing cells. The results indicated that the
biodegradation of formaldehyde led to the simultaneous appearance of formic acid and
methanol. The biodegradation of these metabolites started after exhaustion of
formaldehyde in the medium. In order to study the nature of formaldehyde dismutase
formation, experiments were performed using resting cells obtained from cultures
previously grown in different media. The enzyme activity was measured and the results
are shown in table 2. Dismutase specific activities were different, being higher for the
cells grown in a medium containing formaldehyde as sole carbon source. The results
indicate that the presence of formaldehyde stimulates the production of the enzyme.
Table 2: Formaldehyde dismutase specific activities in resting cells o/Pseudomonas putida
A2 previously grown in different media [4J
Medium Specific activity (J.lmollmg protein·min)

Glucose 3.5
Glucose and fonnaldehyde 5.2
Fonnaldehyde 6.0
Tryptone and yeast extract 1.7
Azachi et al. [5] isolated a bacterium from soil collected at a storage site for
formaldehyde near a chemical plant that uses formaldehyde in the production of glue.
The strain, called MA-C, was identified as Halomonas sp. and was found to be a highly
formaldehyde-resistant halotolerant bacterium. Soil samples were used to inoculate 250
mL flasks with 50 mL medium containing 10 % NaCl, other salts, 5.0 gIL sodium
succinate, 0.5 gIL yeast extract and formaldehyde.
Halomonas sp. MA-C was able to grow in the presence of formaldehyde
concentrations of up to 75 - 100 mgIL in a salt medium. At formaldehyde
concentrations of 125 and 150 mgIL, growth was significantly inhibited. During growth,
formaldehyde disappeared from the medium. At a formaldehyde concentration of 150
mgIL, growth was limited but formaldehyde was still transformed to a significant
extent, probably owing to a high formaldehyde dehydrogenase activity of the cells. In
order to follow the fate of formaldehyde metabolised by strain MA-C, 14C-Iabelled
formaldehyde was added to cultures to test whether formaldehyde was incorporated into
the cells or oxidised to carbon dioxide or to non-volatile dissolved products. The
amount of labelled formaldehyde incorporated into the cells and that remaining in the
262
Bioreactors for the treatment of indnstrial waste gases
culture supernatant were determined. No significant incorporation of radioactivity into

the cells was observed, and the radioactivity of the culture supernatant slowly declined.
The main product of formaldehyde transformation by Halomonas sp. MA-C is probably
carbon dioxide.
The presence of formaldehyde dehydrogenase in cell extracts of strain MA-C was
tested. Activities measured were higher in cells grown for 24 hours in the presence of
20 mgIL formaldehyde than in cells that had not been exposed to formaldehyde during
at least 10 transfers. Activities obtained were between 650 and 850 nmol reduced
NAD/mg protein·min at room temperature in extracts of cells grown in the presence of
20 mg/L formaldehyde. In cell extracts prepared from cells grown in similar media, but
in the absence of formaldehyde, the activities obtained were between 280 and 480 nmol
reduced NAD/mg protein' min.
Kaszycki and Koloczek [6] investigated formaldehyde and methanol biodegradation
by the methylotrophic yeast Hansenula polymorpha. Hansenula polymorpha is a
methylotrophic yeast that can use both formaldehyde and methanol as single carbon
source. The formaldehyde and methanol metabolism found in methylotrophic yeasts
consists of a complex enzymatic pathway comprising both energy-yielding dissimilation
reactions as well as the assimilation of carbon into cell structural components.
The Hansenula polymorpha cells were grown, at 37°C and pH 5.0, on methanol in
order to induce the enzymes of the methylotrophic pathway. In experiments of
formaldehyde biodegradation, the cell culture was additionally treated with 300 mgIL
formaldehyde about 8 hours before the start of each experiment. This step was found to
be necessary to preadapt the cells to the toxic environment and to induce the enzymes
directly involved in formaldehyde utilisation. The experiments were performed in a
simplified medium, which resembled the mineral content of many wastewaters
originated from chemical industries, and was close to the salt solution serving as a
medium for methylotrophic yeasts. The medium contained 200 mgIL (NlLthS04, 300
mg/L KCl, 30 mglL H3P04 , and 25 mgIL yeast extract.
In a simplified environment of a model wastewater solution, Hansenula polymorpha
cells were able to grow and metabolise formaldehyde at concentrations typical for
wastewaters of chemical industries. The yeast culture inoculated at a low cell density of
3.0x105 cells/mL was able to grow and enter the exponential phase on initial
formaldehyde levels of about 400 mgIL. Above this concentration the toxic effect of
formaldehyde prevented the cells from proliferation and led to the final loss of survival.
The capability of methylotrophic yeasts to assimilate exogenously supplied
formaldehyde into cell components is in contrast to formaldehyde-resistant bacteria,
where formaldehyde is oxidised directly into carbon dioxide.
In other experiments a culture grown to the late logarithmic phase was used at a high
cell density of 1.1x107 cells/ml to study the maximum biodegradation potential of
formaldehyde. The yeast cells were fully viable and able to degrade formaldehyde
present at initial concentrations of up to 700 mgIL. At initial concentrations over 700
mg/L, formaldehyde or some of its by-products became toxic for the Hansenula
polymorpha culture. Above this concentration toxicity prevented the cells from
proliferation and led to the final loss of survival, leading to the complete loss of
biodegradation capabilities. The highest rate of formaldehyde biodegradation was
approximately 400 mgIL h.
263
6scar 1. Prado, Marta Eiroa, Maria C. Veiga and Christian Kennes
Yamazaki et al. [7] studied the biodegradation of formaldehyde by a formaldehyde-

resistant bacterium isolated from coastal seawater. The bacterium, designated as a DM-
2 strain, was cultured aerobically at 28°C and pH 6.8. The medium contained 30 giL
NaCl, 5 gIL tryptone, 5 gIL yeast extract, 6 gIL Na2HP04, 3 gIL KH2P04, 1 gIL ~CI,
1.2 gIL MgS04, and 0.1 gIL CaCho Cells of the DM-2 strain were precultured until the
late logarithmic growth phase in either the absence or the presence of 200 or 400 mgIL
formaldehyde and were used for biodegradation experiments. In these experiments cells
were added to medium containing either 200 or 400 mglL formaldehyde as the sole
carbon source.
Cells of DM-2 precultured in the presence of 200 or 400 mglL formaldehyde were
able to completely degrade 200 mgIL formaldehyde within 15 hours. DM-2 cells
precultured in the absence of formaldehyde were also able to degrade formaldehyde;
however, more than 70 % of the substrate remained even after 20 hours. The
degradation of 400 mglL formaldehyde was also investigated and the ability of
formaldehyde degradation was dependent on the preculture conditions. Production of
formaldehyde dehydrogenase may be induced by formaldehyde during the growth
phase; consequently, cells of DM-2 precultured in the presence of formaldehyde
showed high formaldehyde degradation activity.
The effect of the cell concentration on the formaldehyde degradation rate was also
studied. Higher cell concentrations resulted in increased formaldehyde degradation
rates. The highest formaldehyde degradation rate reached in this study was 45 mgIL h at
a concentration of the DM-2 strain corresponding to an optical density of 1.2 at 660 nm.
Hidalgo et at. [8] reported formaldehyde removal in a synthetic medium and in
industrial wastewater by Rhodococcus erythropolis UPV-1. Rhodococcus erythropolis
strain UPV-1 was isolated from a phenol-polluted site in an estuary. Batch experiments
were performed at 30°C in 250 mL flasks containing 100 mL medium. Continuous
formaldehyde feed without variation of the culture volume was achieved.
The disappearance of formaldehyde was studied for different initial concentrations.
The specific rate of formaldehyde disappearance remained constant over the selected
concentration range. For different initial biomass concentrations these specific rates
remained almost constant, decaying slightly at the highest biomass concentration tested.
Bacterial growth was negligible in all cases.
When a repeated and discontinuous feed approach was used, Rhodococcus
erythropotis UPV-1 was able to remove several consecutive doses of 20 mg
formaldehydelL from the culture medium. The last pulses of formaldehyde took longer
to be completely removed than the previous ones, showing some kind of cumulative
effect.
A continuous formaldehyde delivery system, which allowed formaldehyde to diffuse
into the medium at a constant rate of 0.41 mglL h, was used. At this rate, the
formaldehyde was completely removed over long periods of time.
3. Case studies on biological abatement of formaldehyde in industrial waste gases
Over the past decades, the increasing concern about environmental contamination has
led to a fast development of technologies suitable for treating recalcitrant compounds
both in gas and in aqueous phases. As described above, in the case of formaldehyde, a
264
Bioreac,tors for the treatment of industrial waste gases
number of studies concerning its (bio) elimination have been published recently, mainly
for aqueous systems. Very little has been published on the treatment of gaseous
effluents containing formaldehyde, but results reported on aqueous phase reactors
suggest that its biodegradation should be possible in biofilters even in presence of other
compounds as methanol as found in waste gases from synthetic resin producing
industries. In the next pages we present background information and a general
compendium of experiences obtained with industrial gas-phase bioreactors used for
formaldehyde biodegradation in different applications, before presenting our own
results and experience.
3.1. BACKGROUND
Ferranti [9] has reported data on the performance of formaldehyde-degrading biofilters,

at pilot- and industrial-scale. In his studies he used an inert hydrophilic support called
BioKeyTM (Corain Impianti Engineering & Contracting S.r.l, Rome, Italy), consisting of
spherical pellets with a nominal size of 2 cm. Before starting the experiments, this filter
bed was fed appropriate nutrients in aqueous phase in order to allow colonisation of the
packing material by the mixed microbial population inoculated. Start-up of a pilot-scale
reactor, with a total filter-bed volume of 1.33 m3 and 75 cm high, was performed with a
continuous air flow rate of 100 Nm3/h, and with a formaldehyde concentration around
30 mglNm3• The empty bed residence time was 48 s. The biofilter was equipped with a
prescrubber, in which the gas was humidified up to saturation and cooled at the wet
bulb temperature. Moreover, the bed was periodically sprinkled. All of this was done in
order to keep the adequate temperature and moisture content in the filter bed. The
relative humidity of the bed remained between 45 % and 50 % during all the
experiment, while the pH of the water sprinkled was between 7.0 and 8.0. Under these
conditions, the biofilter almost immediately reached a high removal efficiency of
around 60 %, which gradually improved on the subsequent days, reaching, on day 14,
the minimum desired removal efficiency of 95 %. This pilot-scale biofilter was operated
for 20 weeks, with an inlet formaldehyde concentration of 45 ± 13 mglNm3• The total
gas flow rate was increased from 100 to 300 Nm3/h immediately after start-up, and then
stepwise increased to a value of 550 Nm3/h on day 35 of operation. The gas temperature
had a constant value of 35°C, and its relative humidity was approximately 100 %. After
a short start-up phase, this reactor worked continuously with a removal efficiency above
97 %, with a peak of 99.98 % on day 49, when the inlet concentration was 40.33
mglNm3 • On day 59 the formaldehyde inlet concentration increased up to 317.5
mglNm3 • Such an important modification of that parameter did not lead to a significant
decrease of the reactor's efficiency, which dropped only to 88.6 %. This proves the
adaptability of the biofilter to shock loads. After a transition period of two weeks the
system recovered its usual efficiency. With this exception, the reactor worked during all
the experiment with a removal efficiency always above 95 %. The total pressure drop
remained below 50 mm H 20.
The results obtained with this pilot-scale biofilter were used to build an industrial plant
located in Viadana (Italy) [9]. The plant was initially designed for the treatment of
80000 Nm3/h of total exhaust air with a formaldehyde concentration between 30 and 50
mglNm3 , but after 9 months operation the flow rate was increased to 100000 Nm3/h.
The system consisted basically of a prescrubber and four bioreactors connected in
265
parallel, with two beds of filter material each. Each one of the eight filtering beds,
which had an area of 33 m2 and a height of 80 cm, was fed individually. The specific
velocity of the exhaust gas was 306 Nm3/m2h before the flow rate was increased, with
an empty bed residence time of 7.5 s. After that, the specific velocity was increased to
383 Nm3/m2h, with a residence time of 6 s. A removal efficiency over 90 % was
expected. In this case, the start-up of the plant was performed without inoculation of
microorganisms in the filter bed, resulting in the need of an additional week to reach a
nearly-constant removal efficiency of around 91 %, with peaks of 98 %. Hence,
approximately three weeks were needed for the start-up of this reactor. Formaldehyde
concentrations at the inlet (Cin) and at the outlet (Cout) of the bioreactor were measured
during 878 days (Fig. 2).
80
.,.-.
E 60
iE
..... 40
"5
0
0
C 20 ++--~~~--~~~------------~----------+25000
(3
0 ~~~~~~~~~~~~~~--+O
0 200 400 600 800 1000
t (days)
-+-Cin _Gout ~Row rate
Fig. 2: Performance of the Viadana bioreactor (rrwdifiedfrom [9])
As can be seen in figure 2, both the air flow rate and the inlet formaldehyde
concentration were relatively variable. However, the removal efficiency of the system
was high during all the experiment, with a mean value of 90.65 % and even reaching
98.8 % during certain periods, corresponding to mean elimination capacities around 17
glm3h. These results prove that the conditions employed allowed maintaining an
efficient formaldehyde removal.
Another pilot plant designed for formaldehyde elimination was started-up by the same
group in Pomponesco (Italy) [9]. In this case, the pollutant was not fed continuously to
the reactor, because the production was stopped during the weekends. Hence, the air
sup~ly to the biofilter was also interrupted. The inlet gas flow rate was initially 290
Nm Ib, and was increased to 500 Nm31b after a few days of operation. The gas residence
time during these periods was, respectively, 16.6 and 9.5 s. The gas temperature was
around 80 DC, and its absolute humidity was about 30 g of water per Kg dry air.
Formaldehyde concentration at the inlet was lower than in the previous case, between
7.8 and 15.6 mgINm3• Under these conditions, the efficiency of the biofilter was very
high during all the study, which lasted more than two months. Values were constantly
266
over 98 %, reaching 99.7 % after six and a half weeks operation. Interruptions on
weekends did not negatively affect its performance. Data obtained from this pilot plant
were, once again, used for the development of an industrial-scale bioreactor located in
Pomponesco (Italy). This reactor was designed to treat a total flow rate of 120000
Nm3th of exhaust air polluted with formaldehyde [9]. The system employed presented
four lines of toxic supply, with very variable flow rates. The mean formaldehyde
concentration was about 20 mglNm3 , and, as shown in figure 3, the efficiency remained
high during the 15 months operation, varying between 82 and 98 %.
40
--
..":ea
., 100000 i5'
E 30
~ 75000
-
.§. 20 III
::::J 50000 Z
0 3
0
C
10 25000 ~
(3
0 0
0 100 200 300 400 500
t(days)
-+-Qn _Caut ---.- Flow rate
Fig. 3: Performance of the Pomponesco bioreactor (modified from [9])
Other studies have been published on the removal of formaldehyde but in presence of
other pollutants. Doronina et al. [10] developed a laboratory-scale biofilter for the
treatment of formaldehyde, methanol and methylamine. The pollutants were fed in
aqueous solution, and immediately mixed with a large air flow to produce contaminated
air. The reactor was filled with a polyacrylamide fibrous carrier and a washed and
boiled porous ceramic carrier (Ceramsite). The biocatalyst, consisting of an
Ethylobacterium extorquens VKM V-1837D culture, was grown in a nutritive medium
containing methanol as carbon source, before its inoculation into the biofilter. The
inoculation was carried out by means of paper strips in which the bacteria were retained.
These paper strips were transferred from flasks to an agar medium or to a liquid
medium for incubation under optimal conditions, and then inoculated directly into the 2
L biofilter. The selected filter bed allowed the immobilisation of as much as 100 mg dry
biomass per g polyacrylamide fibrous carrier and up to 10 mg dry biomass per g
Ceramsite. A fibrous polyacrylamide layer located on top of the biofilter allowed a
uniform distribution of air and contaminants. As mentioned above, the latter were
supplied in the form of a liquid medium, which contained 420 - 520 mg/L formaldehyde
and 10 mg/L methanol and which was initially fed at a flow rate of 80 mLth. During
this first phase, which lasted three days, the removal efficiency of formaldehyde
exceeded 99 %, while methanol was completely eliminated. The flow rate was
267
6scar 1. Prado, Marta Eiroa, Maria C. Veiga and Christian Kennes
gradually increased during the next days, resulting in a slight decrease in the
formaldehyde removal rate (Fig. 4), although the methanol removal efficiency remained
constant and high at a value of 100 %, proving that the method chosen for the start-up
phase was adequate, although it may not exactly represent the real industrial-scale
situations since the feed was prepared by mixing an aqueous and a gas phase.
100
.......
:::e
~
>- 80
CJ
C
CII
'0 60
:=CII
40
'ii
>
0
E 20
CII
II:
0
80 120 200 400
Row rate (m L./h)
Fig. 4: Formaldehyde removal efficiency as a function of liquid flow rate [101
After the start-up phase, the nutritive medium containing 950 mg/L formaldehyde was
supplied to the biofilter at a constant flow rate of 20 mLlh, obtaining an elimination
efficiency of 99 %. In another series of experiments, the formaldehyde concentration
was increased stepwise to 2000 mg/L. Results showed that, during the first two days
operation with a liquid medium flow rate of 150 mL/h, the removal efficiency was high
(about 95 %), but it tended to decrease, reaching 60 - 70 % on the fourth day. Data of
formaldehyde utilisation as a function of its concentration are represented in figure 5.
Three weeks after that experiment, the composition of the polluted air was switched
from formaldehyde to a mixture of methanol and methylamine, at a constant liquid flow
rate of 150 mL/h and a gas flow rate of 240 Llh. Inlet methanol concentrations ranged
from 400 to 1890 mg/L, while methylamine concentrations ranged from 400 to 1300
mglL. In the case of methanol, removal efficiencies higher than 97 % were found even
at the highest inlet methanol concentration. This means that microorganisms rapidly
adapt from formaldehyde to methanol utilisation, which is not very surprising since
methanol is an intermediate metabolite formed during the biodegradation of
formaldehyde by some microorganisms, as already mentioned above. However,
adaptation to methylamine consumption was slower. At an initial methylamine
concentration of 400 mg/L, only 47.5 % was degraded on the first day operation. After
3 days, 76.3 % was degraded, and after 5 days the removal efficiency had reached 99.3
%. A subsequent increase in the pollutant concentration up to 700 mg/L resulted in 99.6
268
% degradation of methylamine. Another increase in the methylamine supply up to 1300

mg/L performed two days later did not affect the removal efficiency. These results
prove that certain microorganisms (e.g., Methylobacterium extorquens) grown in a
medium containing methanol may be suitable for the elimination of formaldehyde,
methanol and methylamine, although in the case of methylamine an adaptation period
may be necessary for a correct performance .
-
.-..
t. 100
90
• •• • •
80
70
60
o 500 1000 1500 2000
Formaldehyde concentration (mglL)
Fig. 5: Formaldehyde removal efficiency as afunction offormaldehyde concentration in the

liquid phase [10]
Huckschlag [11] reported the case of a series of biological systems designed for the
treatment of a mixture of formaldehyde and phenol in different industrial reactors. One
of those reactors, used for the treatment of waste gases from a fibre glass-producing
industry, was fed an exhaust gas flow rate between 330 and 650 m31h. Phenol inlet
concentrations usually varied between 10 and 15 mg/m3, while concentrations around
0.3 mg/m3 were found in the outlet. This means that the removal efficiency for phenol
was between 97 and 98 %. In the case of formaldehyde, its concentration in the exhaust
gas was approximately 10 mg/m3, while in the outlet gas it was around 2 mg/m3, which
corresponds to a removal efficiency of 80 %. These results clearly show that a mixture
of low concentrations of phenol and formaldehyde may be efficiently treated by means
of a biological system.
The same study also included data obtained from an industrial plant where a highly
fluctuating amount of phenol, formaldehyde and ammonia was produced and treated
biologically. Table 3 shows the concentration range of these substances both in the inlet
and in the outlet gas stream.
269
Table 3: Phenol, formaldehyde and ammonia concentration in the inlet and in the outlet gas
[111
Compound Inlet gas concentration (mglm3) Outlet gas concentration (mglm3)

Phenol 33 -160 1-2
Formaldehyde 6-40 0.5-7
Ammonia 4-42 2-4
Tautz and Rutenfranz [12] used a bioscrubber for formaldehyde, methanol and
ethyleneglycol abatement at pilot-scale, comparing the degree of reduction of each
substance between the suspension phase (in the absorption column) and the fixed phase
(in the regeneration unit). The results show that, in the case of formaldehyde, between 9
and 13 % removal took place in the aqueous phase, being the rest degraded in the
bioreactor. The other chemicals were degraded only in the bioreactor. The bioscrubber
was operated for more than 70 days with an elimination efficiency always above 89 %,
reaching 100 % in some cases.
Mackowiak [13] also reported interesting results regarding formaldehyde abatement
in a chipboard production industry. A pilot-scale biofilter packed with compost and
wood chaff was used for the treatment of variable flow rates of polluted air, between
400 and 1450 m3/h, with an unspecified toxic load. Some experiments were made in
order to determine the effect of different classical organic packing materials as bark,
peat and compost, on the pressure drop registered in the biofilter. Their results showed
that, at gas loads below 350 m 3/m2 h, bark led to the lowest pressure drop, while at gas
loads above that value, peat gave the lowest pressure drop. Regarding formaldehyde
biodegradation, removal efficiencies between 53 and 93 % were found, with an average
value around 80 %.
All these experiments prove that a formaldehyde-degrading biofilter may be started-
up in a short period of time, needing only a basic medium and little work if some
considerations are taking into account. First, an optimum environment is required to
favour the development of the microbial population. This means that all physical and
chemical parameters affecting bioreactor performance (temperature, moisture, pH,
nutrient concentrations, etc.) must be within a range that allows the optimum activity of
the biocatalyst. Moreover, the gradual increase of the pollutant concentration, when
possible, will avoid poisoning of the microorganisms in the first stages of operation.
The presence of other volatile organic compounds in the waste gas may, sometimes,
significantly reduce the overall removal efficiency.
3.2. OPTIMIZATION OF THE REMOVAL OF FORMALDEHYDE IN BIOFILTERS

AND BIOTRICKLING FILTERS
In our own laboratory, the treatment of waste gases typical from synthetic resin
producing industries is being studied. The actual air stream contains at least four
different compounds, among which formaldehyde has been identified as the dominant
one, followed by methanol. In a first study, three biofilters and one trickling biofilter
were used for the treatment of formaldehyde in mixture with methanol. As described
above (see section 2), methanol is a potential intermediate metabolite formed during the
aerobic biodegradation of formaldehyde by Pseudomonas sp. To start-up the reactors,
270
each one of them was inoculated with an aerobic sludge obtained from the wastewater
treatment plant of a resin-producing industry. No adaptation step was necessary since
the sludge was obtained from a wastewater already containing formaldehyde and
nitrogen compounds as major pollutants, allowing induction of the enzymes involved in
the biodegradation of the organic compound (see also section 2). By the way, studies
undertaken in batch and continuous liquid phase reactors have proven the ability of that
sludge to degrade mixtures of formaldehyde and methanol [14, 15]. Some of the most
relevant characteristics of the sludge used for the inoculation are shown in table 4.
Table 4: Physical and chemical parameters of the sludge used for the inoculation
Parameter Mean value (± Standard Deviation)

Density (gIL) 1039.4
Optical Density (diluted 50 x) 0.371 (± 0.003)
P-P04•3 concentration (mgIL) 153.2 (± 3.5)
N-NH/ concentration (mgIL) 171.5 (± 4.9)
N-N02- concentration (mgIL) 2.99 (± 0.04)
N-N03- concentration (mgIL) 0.00 (± 0.00)
Total Suspended Solids (gIL) 9.95 (± 0.21)
Volatile Suspended Solids (gIL) 8.75 (± 0.00)
Chemical Oxygen Demand (mgIL) 233.5 (± 13.7)
pH 7.58 (± 0.01)
All three biofilters were inoculated the same way: 2.3 L sludge was added to the
bioreactors and, after three hours, the liquid was drained off through the bottom of the
reactor, leaving time enough for the biomass to attach to the filter bed. It is important to
mention that each of these biofilters was packed with a different packing material:
volcanic earth, perlite and activated carbon. In the case of the trickling biofilter, only
volcanic earth was used. For its inoculation 750 mL sludge was continuously
recirculated in a trickling mode through the top of the reactor at a constant flow rate of
3.0 Llh. The medium was recirculated without any treatment. After the inoculation, the
gas flow rate was set at 0.15 m31h. Contrary to previously reported studies (see section
3.1.), nutrients were not added to the reactors, although a very fast start-up was
observed. Results showed that all four reactors presented a high removal efficiency of
formaldehyde and methanol already immediately after inoculation. The initial loads of
methanol and formaldehyde were relatively low. Removal efficiencies of 100 % were
found for methanol in all reactors already on the first day of operation, while the
formaldehyde removal efficiency varied from 49.0 % in the perlite-biofilter to 68.6 % in
the trickling biofilter. During the first eight days operation, the formaldehyde removal
efficiency remained high in all bioreactors, being the trickling biofilter the system that
showed the highest elimination rate (Table 5). After that, the removal efficiency of
formaldehyde decreased in all reactors. 1 L of the nutrient solution described by Kennes
et al. [16] was added to each reactor on day 22 of operation, but it had only little effect
on formaldehyde removal.
271
6scar I. Prado, Marta Eiroa, Maria C. Veiga and Christian Kennes
Table 5: Formaldehyde removal in allfour reactors during the first week operation
Volcanic earth-biofilter Perlite-biofilter Activated carbon-biofilter Biotrickling filter

Load (gIm3h) 1.2 ± 0.3 1.3 ± 0.6 1.2 ± 0.5 2.1 ± 0.3
Efficiency (%) 63.2± 10.9 55.8 ± 15.3 57.0 ± 14.0 68.1 ± 2.3
Since methanol was better degraded than formaldehyde, its concentration was increased
frrst while maintaining the formaldehyde load relatively low. The methanol inlet
concentration was daily modified from the frrst day of operation, in order to simulate a
real industrial situation. Table 6 shows methanol loads and elimination efficiencies of
all biofilters during the frrst month operation.
Table 6: Methanol removal in all four reactors during the first month operation
Volcanic earth-biofilter Perlite-biofilter Activated carbon-biofilter Biotricklin~ filter

Max. Methanol
Load 1050.8 610.6 944.2 1634.5
(gIm3h)
Removal
Efficiency at 96.2 98.4 89.5 99.2
Max. Methanol
Load (%)
Mean
Methanol Load 397.8 ± 428.5 170.4 ± 218.2 300.3 ± 384.9 433.9 ± 635.3
(gIm3h)
Mean Removal
Efficiency 94.5 ±7.3 95.2±7.2 92.9± 14.5 89.3 ± 19.0
(%)
As shown in table 6, all biofilters were able to remove high methanol loads, above 1
Kglm3h, during the first month operation, even though the toxic load changed
significantly from one day to another. Results suggest that the presence of methanol in
the mixture of pollutants could have a significant influence on formaldehyde removal.
Methanol seems to be a more accessible carbon source for the microorganisms, which
may explain the relatively low formaldehyde removal. The formaldehyde concentration
is presently being increased to study the effect of higher loads of both formaldehyde and
methanol.
Acknowledgements
The research described in this chapter is being funded by projects PR404E 2000/6-0 and
PPQ 2001-0557. The doctoral research of 6JP and ME is financed by Ph.D. fellowships
of the Xunta de Galicia and the Spanish Ministry of Education and Culture,
respectively.
272
References
[1] Kennes, C. and Thalasso, F. (1998). Waste gas biotreatment technology. J. Chern. Techno!. Biotechno!'
72:303-319.
[2] Walker, J. F. (1964). Formaldehyde. American Chemical Society Monograph Series. Reinhold Publishing
Corporation. New York, Amsterdam, London.
[3] Bonastre, N.; de Mas, C. and Sola, C. (1986). Vavilin equation in kinetic modeling of formaldehyde
biodegradation. Biotechno!. Bioeng. 28:616-619.
[4] Adroer, N.; Casas, C.; de Mas, C. and Sola, C. (1990). Mechanism of formaldehyde biodegradation by
Pseudomonas putida. App!. Microbio!. Biotechno!. 33:217-220.
[5] Azachi, M.; Henis, Y.; Oren, A.; Gurevich, P. and Sarig, S. (1995). Transformation of formaldehyde by a
Halomonas sp. Can. J. Microbio!. 41:548-553.
[6] Kaszycki, P. and Koloczek, H. (2000). Formaldehyde and methanol biodegradation with the
methylotrophic yeast Hansenula polymorpha in a model wastewater system. Microbio!. Res. 154:289-
296.
[7] Yamazaki, T.; Tsugawa, W. and Sode, K. (2001). Biodegradation of formaldehyde by a formaldehyde-
resistant bacterium isolated from seawater. App!. Biochem. Biotechno!. 91-93:213-217.
[8] Hidalgo, A; Lopategi, A; Prieto, M.; Serra, J. L. and Llama, M. J. (2002). Formaldehyde removal in
synthetic and industrial wastewater by Rhodococcus erythropolis UPV-1. App!. Microbio!' Biotechno!.
58:260-263.
[9] Ferranti, M. M. (2001). Formaldehyde biological removal from exhaust air in the composite panel board
industry from pilot tests to industrial plant. 35 th International Particleboard Composite Materials
Symposium. Pullmann Washington State, U.S.A. April 2-5.
[10] Doronina, N. V.; Ezhov, V. A and Trotsenko Y. A (1996). Aerobic biodegradation of formaldehyde,
methanol and methylamine by immobilized Methylobacterium extorquens cells. App!. Biochem.
Microbio!. 33:138-141.
[11] Huckschlag, W. (1992). Biotechnologische Behandlung Phenol und Formaldehydhaltiger Abluft. In
Dragt, A J. and van Ham, J. (Eds) Biotechniques for Air Pollution Abatement and Odour Control
Policies. Elsevier Science Publishers BV, Amsterdam, The Netherlands, pp. 279-286.
[12] Tautz, H. and Rutenfranz, C. (1992). Biologischer Abbau toxischer Substanzen - Verfahrensauswahl und
Betriebserfahrungen mit einer Biowiischer-pilotanlage. Chern. Ing. Tech. 64:192-194.
[13] Mackowiak, J. (1992). Abscheidung von Formaldehyd aus der Abluft im Biofilter. In Dragt, A. J. and
van Ham, J. (Eds) Biotechniques for Air Pollution Abatement and Odour Control Policies. Elsevier
Science Publishers BV, Amsterdam, The Netherlands, pp. 273-278.
[14] Cant6, M.; G6mez, J.; Kennes, C. and Veiga, M. C. (1998). Integrated anoxic-aerobic treatment of
wastewaters from a synthetic resin producing factory. In Proceedings of the European Conference on
New Advances in Biological Nitrogen and Phosphorus Removal for Municipal or Industrial Wastewaters.
Narbonne, France, October 12-14.
[15] Eiroa, M.; Kennes, C. and Veiga, M. C. (submitted). Simultaneous nitrification and formaldehyde
biodegradation in aerobic assays.
[16] Kennes, C.; Cox, H. H. J.; Doddema, H. J and Harder, W. (1996). Design and performance ofbiofilters
for the treatment of alkylbenzene vapours. J. Chern. Techno!. Biotechno!. 66:300-304.
273
INDEX
activated sludge .11, 13, 14, 17, 19,20,23,24,25,26,36,57,58,60,61,65,66,69,70,
71,72,73,74,84,85,87,101,102,108,109,110,115,118,119,125, 126,127,129,
131,139,140,141,142,143,149,151,152,158,160,161, 166, 167, 178, 180, 181,
182,183,184,185,186,187,188, 194, 199,201,213,214,217,218,219,220,237,
254,261
aeration ...... 20, 25, 74, 77, 78, 79,80,117,130,154,163,184,188,195,196,197,198,
199,200,201,202,203,204,207,211,215,219,220,221
ammonification ....................................................... 20, 106, 110, 139, 143, 148, 158, 162
artificial neural network ................................................................. 15, 17,87,88,93,198
ASM1 .... 11, 12, 13, 14, 17, 18, 19,20,21,23,24,25,26,36,37,58,59,60,62,81,83,
101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 114, 115, 118, 121, 122, 123,
124, 128, 129, 131, 137, 138, 140, 143, 144, 145, 146, 147, 150, 152, 153, 159, 160,
162,163,164,165,166,168,169,170,171,173,174,177,192,209
ASM3 .. 102, 103, 106, 107, 109, 110, 111, 112, 113, 114, 115, 147, 148, 162, 163, 169,
170,180,192
attached growth systems ...................................................................................... 219,220
ATU ............................................................................................................. 157,159,160
autotrophic biomass .... 19, 20, 63,103,104,107,108,121,139,149,153,172,184,214
batch experiment... ........................................ 132, 138, 151, 156, 157, 162, 180, 199,261
batch test... .................................................... 110, 130, 136, 137, 138, 139, 152, 180, 183
benchmark .................................................... 73, 74, 75, 77, 78, 79,80,81,82,83,84,85
biodegradation17, 102, 103, 111, 140, 144, 145, 146, 182, 185,231,232,233,235,236,
237,243,245,246,247,248,250,254,259,260,261,262,263,264,265,268,270,
272,273
biofiltration .. 229,233,239,240,242,244,245,247,248,249,252,253, 254, 255, 256,
259,260
biological characterisation ........................................................... 119,129,145, 161, 180
biological nitrogen removal. 101, 187, 188, 189, 191, 193, 194, 195,202,204,205,206,
207,208,210,211,213,214,215,216,221,222
bioprocess ....................................................... 1,11,13,14,15,26,28,31,37,57,60,61
bioremediation ................................................................................................................. 1
bioscrubbers ......................................................................................................... 239, 249
calibration ...... 65, 67, 69, 72, 91,101,115,116,117,118,119,120,121,122,123,124,
146,163,166,168, 169, 170, 171, 172, 180, 181, 182, 183, 184, 186, 197
control.... 11, 12,13,16,17, 18, 19,20,21,23,24,25,26,28,36,37,58,59,60,61,65,
73,74,75,77, 78, 79,80,81,82,83,84,85,87,88,90,91,92,94,95,96,98,99,
100,143,161,182,185,187,188,189,193,194,195,196, 197, 198, 199,200,201,
202,203,204,205,207,208,209,210,211,212,213,217,219,221,222,224,237,
240,241,242,244,245,246,251,252,253,254,255,256,257
control strategies .. 17, 60, 61, 73, 74, 75, 78, 79, 81, 82, 84, 85,143,195,196,198,199,
202,207,221
correction factors ................................................................................................. 109, 170
275
decay .. 22, 23, 25, 58, 62, 63, 67,102,103,106,108,109,110,111,114,115,117,125,
127,132,134, 146, 149, 152, 153, 154, 156, 163, 166, 167, 169, 173, 174, 183, 184,
185,186,191,192,214,215
decayrate .................. 62, 67,108,110,117,149,153,156,166,169,184,191,192,215
denitrification .. 13,18,19,20,21,22,24,25,60,75,107,109,110,112,114,118,121,
129,134,135,140,143,161,162,170,181,182,183,185,188, 190, 191, 192, 193,
194,195,196,197, 198, 199,200,201,202,203,204,205,206,207,208,209,210,
211,217,218,219,220,221,223
endogenous respiration .. 83, 108, 110, 112,114,115,137,152,153,157,163,169,173,
174,175,176,181,185,186,192,211,219
enzymology ..................................................................................................................... 1
evaluation of control strategies ...................................................................................... 73
experimental design ............................................................. 101,137,158,159,179,184
external carbon addition ............................................................................ 17,18,60,193
fault detection ...................................................................................... 88, 89, 93, 99, 100
fuzzy logic ................................................................................................... 87, 88, 93, 99
growth .... 17, 19,22,23,25,29,43,58,62,67,68,74,79,102,103,104,106, 107, 108,
109,110,111,112,113,114,117,120,121,132,134,137,139,147,151, 152,154,
156, 157, 158, 161, 162, 163, 166, 167, 168, 169, 170, 172, 173, 174, 175, 176, 178,
179,182,183,184,186, 187, 188,212,217,218,219,220,221,222,232,233,234,
241,252,262,264
half-saturation concentration ............................................................... 154, 156, 157, 161
heterotrophic biomass .... 20, 63, 66, 67, 68,103,104,107, 108, 109, 114, 132, 138, 139,
140,152,153, 154, 157, 172, 182, 183, 192
hydrolysis .... 18, 20, 21, 63, 106, 108, 109, 110, 111, 115, 124, 125, 131, 132, 137, 138,
139,147,157, 159, 161, 163, 164, 169, 170, 171, 172
hydrolysis rate .............................................................. 108,125,137, 138, 139, 157, 164
identifiability ................................ 12,17,21,60,115,116, 117, 173, 181, 183, 184, 186
kinetic parameters .. 60, 109, 110, 121, 150, 152, 155, 157, 158, 159, 160, 161, 162, 163,
165, 170, 171, 172, 173, 180, 182, 183, 189, 199
mathematical modeL ................................................................... 61, 72, 88, 99, 182, 184
maximum specific autotrophic growth rate ......................................................... 157, 167
maximum specific heterotrophic growth rate .............................................................. 167
membrane ..................................................................... 125,138,147,231,233,236,257
model reduction .. 11, 12, 13, 14, 15, 16, 17,20,25,28,30,32,37,51,57,58,59,61,62
modelling11, 12, 15, 16, 17,25,31,61,62,74,75,83,99,108,117,118,120,172,173,
182, 183, 184, 185, 186
Monod ............................ 17,19,21,22,25,43,44, 45, 50, 52, 57, 59, 60,107,154,252
nitrate utilisation rate ............................ 129, 130, 140, 145, 150, 160, 161, 162, 170, 185
nitrification ...... 15, 17, 18, 19,21,22,60,82,92,99,107,109,110,114,118,121,129,
134,135,138,139,140,141,142,143,144,146,151,152,157, 158, 159, 160, 161,
162,167,180,181,182,187,188,189,190,191,193,194, 195, 196, 197, 198, 199,
200,201,203,204,211,212,213,214,215,216,217,218,219,220,221,273
nitrogen removal.. 13, 17,20,61, 102,121,183,185, 186, 187, 188, 189, 193, 194,195,
196,197,198,199,204,206,211,213,216,219,221,222
optimisation .... 12, 13,20,25,59,116,117,121,133,157,186,187,195,204,211,213
order reduction .......................................................... 11,14,16,26,28,31,37,38,58,59
276
parameter estimation ............................................................................ 115, 116, 173,209
performance assessment ................................................................................................ 82
process design ............................................... 118,187,188,189,194,195, 2l3, 221, 222
quasi steady state assumption .................................................................................. 16,58
respiration rate ..... 73, 74, 78, 79, 80, 82, 83, 85, 114, 129, 131, l33, l36, l37, l38, l39,
140,150,152,153,154,155,157,158,159,160,173,183,184
respirometric 74,78,81,83, 127, l33, l36, l38, l39, 144, 146, 148, 149, 150, 151, 152,
153,155,158,159,161,162,163,170,173,180,181,184,185
respirometry ... 73,74,75,78,81,82,83,84,85,126,129, l30, l31, l39, 140, 143, 145,
146, 149, 150, 152, 161, 162, 164, 182, 184, 185, 199,222
simulation 11, 12,53,54,55,69, 70, 71, 73, 74, 75, 76, 77, 78, 83, 84,85,95, 109, 148,
181,203,204,208,209,210,213,215,237
singular perturbation ....... 11, 14, 16,24,26,27,28,30,31,32,36,37,38,51,59,61,62
sludge composition .............................................................................................. 148, 180
SolXo ................................................................................................................... 181, 183
SRT .... 79, 80, 81, 109, 110, 188, 189, 190, 191, 192, 193, 194, 195,211,212, 2l3, 214,
215,216,217,218,221,223,224
storage ... 90,110,111,112,113,114,115,147,151,162,163,169,170, 176, 178, 179,
182,183,192,214,215,220,262
switching functions ...................................................................................................... 158
titrimetric ...................................................... 129, 141, 143, 161, 162, 180, 182, 183, 186
titrimetry ...................................................................................... 130, 146, 150, 161, 162
transferability ....................................................................... 101,166,167,168,173,181
trickling filters ............................................................................................. 217,241,242
wastewater characterisation .111, 118, 121, 128, 129, l30, l33, 140, 141, 143, 144, 146,
147, 150, 163, 171, 180, 185, 186
wastewater treatment 11 , 12, 19,25,38,61,65,66,67,69,70,72,73,74,84,85,87,99,
101,125,131,181,182,183, 184, 197,201,214,246,247,251,252,259,262,270
yield coefficient.. .... 67, 109, l36, l39, 146, 148, 149, 150, 151, 152, 154, 157, 171, 172
277

(Focus on Biotechnology 3C) S. R. Weijers (auth.), Spiros N. Agathos, Walter Reineke (eds.) - Biotechnology for the Environment_ Wastewater Treatment and Modeling, Waste Gas Handling-Springer Netherla

Uploaded by

Copyright:

Available Formats

You might also like

(Focus on Biotechnology 3C) S. R. Weijers (auth.), Spiros N. Agathos, Walter Reineke (eds.) - Biotechnology for the Environment_ Wastewater Treatment and Modeling, Waste Gas Handling-Springer Netherla

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Focus on Biotechnology 3C) S. R. Weijers (auth.), Spiros N. Agathos, Walter Reineke (eds.) - Biotechnology for the Environment_ Wastewater Treatment and Modeling, Waste Gas Handling-Springer Netherla

Uploaded by

Copyright:

Available Formats

BIOTECHNOLOGY FOR THE ENVIRONMENT:

WASTEWATER TREATMENT AND MODELING, WASTE GAS HANDLING

SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

ISBN 978-90-481-6224-6 ISBN 978-94-017-0932-3 (eBook)

Printed on acid-free papa

AlI Rights Reserved

Spiros N. Agathos Walter Reineke

1.1. NEED FOR REDUCTION OF RIGOROUS, MECHANISTIC MODELS

Dynamic simulation based on rigorous, physics-based mechanistic modelling has

analysis in electrical engineering and dynamic flowsheeting in chemical engineering. In

meeting more stringent treatment requirements that are imposed on wastewater

1.2. PROBLEM STATEMENT AND METHODOLOGY

partitioning. Therefore, a method for this task is developed. A scaling procedure

1.3. REDUCTION APPROACHES FOR PROCESS ENGINEERING SYSTEMS

1.3.1. New model building from 'scratch'

1.3.2. Simplifying assumptions

• Components; e.g. aggregation of variables (for example COD and BOD as

1.3.4. Order reduction methods

! With 'model reduction', we will also denote 'model order reduction'.

1.3.5. Black-box identification

2. Reduction approaches for ASMI

2.1. NEW MODEL BUILDING FROM 'SCRATCH'

Marsili-Libelli (1989) developed a low-order model for conventional activated sludge

This relational model is estimated from measured DO and nitrification rates in

2.2. SIMPLIFYING ASSUMPTIONS

Several simplifying assumptions are applied in ASMI reduction, which can be

2.2.1. Simplification with respect to components

• No suspended organic bound entrapped N (No XND )

• Do not consider alkalinity

2.2.2. Simplification due to separation of aerobic and anoxic conditions

• No denitrification under aerobic conditions

2.2.3. Simplifying assumptions with respect to kinetics

• Replace terms with Monod kinetics by fITst-order or zero-order kinetics (the

2.2.4. Simplification with respect to dynamics

2.2.5 Reduced order models

• No method for measuring the concentration of the autotrophic biomass exists,

• One overall maximal denitrification rate rdenit,max is used.

• For nonlinear (open-loop) optimisation (optimisation horizon 1 day, Lukasse et

timescale of interest of hours. Assumptions were stated in Lukasse et al.

• Do not consider alkalinity

• Approximate Monod expressions by Blackman kinetics (it is assumed that

rx BH =rH ,XCODXBH -bHX BH (2.16)

2.3. DYNAMICS OF VARIABLES

2.4. ORDER REDUCTION METHODS

2.5. BLACK-BOX IDENTIFICATION

Lindberg (1997) applied subspace state space identification to identify a fourth-order

2.6. DISCUSSION AND CONCLUSIONS

be selected. A more systematic investigation of the validity and implications of the

3. Singular perturbation of bioprocess systems: theory and review

Singular perturbation application to bioprocess systems is reviewed in this section. This

3.1. SINGULAR PERTURB AnON THEORY

Let the system be described by n+m equations in state-space notation

£1: = g(x, Z, u, t, £), z(to) = zo,z E R rn • (3.2)

~ == f(X,c1>i (X, U, t), U, t,O) == r(X, U, t), x(tO) == xO' (3.3)

dZ == g(xo, Z(tf + z(to),O, to), z(O) == Zo - z(t o ) (3.4)

Condition 3.2: The equilibrium z(tf) == 0 of the boundary system (3.4) IS

3.2. REVIEW OF MODEL REDUCTION OF (BIO)PROCESS SYSTEMS USING

3.2.1. General rule for order reduction

In this case, the rule holds (for Psat sufficiently small).