Modelling of wastewater treatment plants – how far

Water Science & Technology Vol 53 No 3 pp 79–89 Q IWA Publishing 2006

shall we go with sophisticated modelling tools?
G.C. Glover, C. Printemps, K. Essemiani and J. Meinhold
Veolia Water – Anjou Recherche, Chemin de la Digue, BP 76, 78603 Maisons Laffitte, France

Abstract Several levels of complexity are available for modelling of wastewater treatment plants. Modelling
local effects rely on computational fluid dynamics (CFD) approaches whereas activated sludge models
(ASM) represent the global methodology. By applying both modelling approaches to pilot plant and full
scale systems, this paper evaluates the value of each method and especially their potential combination.
Model structure identification for ASM is discussed based on a full-scale closed loop oxidation ditch
modelling. It is illustrated how and for what circumstances information obtained via CFD (computational fluid
dynamics) analysis, residence time distribution (RTD) and other experimental means can be used.
Furthermore, CFD analysis of the multiphase flow mechanisms is employed to obtain a correct description of
the oxygenation capacity of the system studied, including an easy implementation of this information in the
classical ASM modelling (e.g. oxygen transfer).
The combination of CFD and activated sludge modelling of wastewater treatment processes is applied to
three reactor configurations, a perfectly mixed reactor, a pilot scale activated sludge basin (ASB) and a real
scale ASB. The application of the biological models to the CFD model is validated against experimentation
for the pilot scale ASB and against a classical global ASM model response. A first step in the evaluation of
the potential of the combined CFD-ASM model is performed using a full scale oxidation ditch system as
testing scenario.
Keywords Activated sludge modelling; computational fluid dynamics (CFD); wastewater treatment plant

Introduction
The use of mathematical models in wastewater industry has become important for several
goals such as design, optimisation, diagnostics and the elaboration of optimised manage-
ment strategies. According to the goal, the modelling is performed at different levels.
Dynamic simulation approaches can be subdivided into two levels:
(1) The “global” level relying on well established models, such as the activated sludge
models (ASM) for simulating the nutrient removal (Printemps et al., 2004);
(2) The “local” process level focusing on the hydrodynamic behaviour using compu-
tational fluid dynamics (CFD) or residence time distribution analysis (RTD)
(Essemiani et al., 2004).
For the evaluation of treatment processes via the global approach, the activated sludge
basin is described as a perfectly mixed tank-in-series model, to mimic the different
anoxic and aerobic zones in the basin. The actual modelling work focuses on the biologi-
cal conversion reactions. The use of the CFD techniques allows detailed study of the
transport phenomena taking place and hence provides insight in mixing efficiency, spatial
distribution of particles and oxygen concentrations. Within wastewater industry, CFD has
found increasing applications (Essemiani et al., 2004). Design and diagnostic studies, for
example, deal with mixing efficiency and flow characteristics, addressing such questions
as agitator positions for minimum velocity to avoid sludge settling and oxygen transfer
optimisation.
doi: 10.2166/wst.2006.078 79
 Currently, these approaches are performed more or less isolated from each other. It is,
however, of interest to investigate how the information obtained from both approaches
can be combined in a smart way. The objective is to take advantage of the local hydraulic
analysis to gain a deeper insight on the biological removal processes. This translates, for
example, in predicting the impact of local anaerobic or anoxic macro-environments in an
oxidation ditch on the response of the biological system (e.g. Littleton et al., 2001;
2003). Potential approaches are (a) incorporating biological model equations in the CFD
G.C. Glover et al.

code, and (b) linking the information obtained from CFD analysis to the biological
model, keeping the two models as a separate unit.
The objective of this paper is to present the approaches mentioned above and to dis-
cuss their feasibility and their draw-backs for future applications. It is intended to initiate
the discussion for such approaches within the community of WW treatment.

Results and discussion

Local approach: the local hydrodynamic modelling of ASB units
To model the effect of local hydrodynamic phenomena on biochemical reactions, the
description of the reactor has a critical importance to obtain representative process
dynamics. Developing a numerical model of a reactor needs careful consideration of the
number of phases and their description, the reactor geometry, the spatial resolution of the
numerical domain, the conditions applied (i.e. gas and liquid flow rates), plus the discreti-
zation and solution techniques used to characterise the local hydrodynamics.
The models of the reactors created within computational fluid dynamics (CFD) must
account for the interactions between the gas and liquid phases, any energy input (either
by aeration or by agitation) and the turbulence that is generated by these inputs. For the
underlying study, this is achieved using a Euler –Euler representation of the gas–liquid
mixture that treated both the gas and liquid phases as two continuous phases (i.e. no indi-
vidual bubbles are modelled). The gas phase is treated as a bubbly phase with spherical
bubbles of uniform diameter with the liquid phase being treated as water. A dispersed
phase treatment of the k-1 turbulence model is used to account for the turbulent flow
energy generated by aeration and agitation, and the dissipation of this energy throughout
the reactor with standard wall conditions and functions. The oxygen transfer is modelled
based on the bubble diameter, the local gas hold-up, the local slip velocity through differ-
ent correlations on the film transfer coefficient (Cartland Glover et al., 2004).

Application of a biochemical model to a local hydrodynamic model. For the

incorporation of the biological model equations into the hydraulic model, the Activated
Sludge Model 1 (ASM1) (Henze et al., 2000) was selected as it represents a well-known
model for carbon and nitrogen removal. All thirteen components of the ASM1 were
implemented into the CFD code, FLUENT (Fluent Inc., 2003) through the use of
classical convective scalar transport equations. The obtained model, subsequently called
CFD-ASM1, is then analysed at different levels. The purpose is to verify correct
implementation of the equations, to discuss the different degree of information obtained
and to run test cases for real scale wastewater systems. As the precision of the final
result, as well as the amount of computational time required, are important issues, they
are also included in the evaluation. Of course, it is known that these aspects are
dependent on the volume of the reactor, the number of transport equations used and the
spatial resolution of the domain.

Configurations studied by local modelling approach. Different systems were used to

80 investigate the CFD-ASM1 model. These systems are as follows:
(1) a pilot plant scale oxidation ditch (closed loop reactor, CLR), being operated as a
sequential batch reactor (SBR). This configuration was used to validate the model.
(2) a rectangular aerated batch reactor section; used to study the behaviour of a per-
fectly mixed reactor.
(3) a real scale oxidation ditch system (also CLR), used to apply the model on a real
scale.
Details and results for each configuration are given below.

G.C. Glover et al.

Configuration no. 1: validation of CFD-ASM1 on a SBR. To validate the CFD-ASM1,
an experimental study and two numerical studies of an SBR-oxidation ditch were
performed (Figure 1). Once the experimental data had been obtained, the ASM1 is
calibrated on the SBR model unit in WESTw (Vanhooren et al., 2002), and it is
further used to estimate component concentrations at the start of the feed phase
(SBR-ASM1). The values of the parameters, e.g. mA and YA (0.54 d21 and 0.32 gCOD
gN21, respectively), obtained by calibration of the SBR-ASM1, are then applied to the
CFD-ASM1, which is applied to aerated and non-aerated conditions.
Figure 1 depicts the change in concentration of the soluble COD, the ammonia nitro-
gen and nitrate nitrogen during the non-aerated feed (experimental and SBR-ASM1
studies only), the aerated and non-aerated react phases. In the feed phase, the concen-
trations of soluble COD and ammonia nitrogen increase as wastewater is fed into the
SBR pilot, while nitrate nitrogen is consumed under anoxic conditions. The ammonia
nitrogen and soluble COD are consumed during the aeration period with a parallel
increase of the nitrate. The dissolved oxygen is quickly consumed at the start the non-aer-
ated phase, and once all oxygen is consumed, denitrification can be observed.
Good agreement is observed between the two numerical and the experimental cases
(e.g. changes in the concentrations of ammonia and nitrate-nitrogen in the reactor). How-
ever, there is a constant difference of 30% between both the numerical profiles of soluble
COD concentration and the experimental profile. This difference is due to not taking into
account the addition of anti-foam agent, which caused a constant increase of the non-bio-
degradable COD fraction.
The significant conclusions to be drawn from the results obtained are: (a) the correct
implementation of the equations in CFD-ASM1 is verified; (b) due to calibration, the two
numerical models (CFD-ASM1 and SBR-ASM1) can be used for further comparative
examination.

Figure 1 Profiles of the concentrations of COD soluble, dissolved oxygen, ammonia and nitrate-nitrogen of
the combined numerical and experimental study of a pilot scale sequential batch reactor (Vermande, 2005) 81
 Having achieved this first step in validation, the CFD-ASM1 can now be used for simu-
lation studies testing and analysing biological and hydrodynamic behaviour of various
processes.

Configuration no. 2: analysis of CFD-ASM1 of a perfectly mixed reactor. A small

aerated batch reactor (2 m3) section is modelled with the assumption of perfectly mixed
reactor hydrodynamics to test the response of CFD-ASM1 to different aeration
G.C. Glover et al.

conditions. This reactor consists of a rectangular cross section of 1 m2 and 2 m long that
is aerated with a 0.25 m2 diffuser zone at the base of the reactor at a gas flow rate of
15 Nm3 h21 (Figure 3).
A critical part of combining the local hydraulic model and ASM1 is precisely predict-
ing the transfer of oxygen across the gas– liquid interface for submerged aerators. The
selection of the kLa is dependent on the liquid side film mass transfer coefficient selected
(in this case either the Higbie Isolated Bubble or the Lamont coefficient) and the charac-
teristic size of the bubbly phase that influences the volume fractions predicted and the
interfacial surface area (Cartland Glover et al., 2004). Therefore, parallel simulations
between PMR-ASM1 performed using the ASM1 in WESTw and CFD-ASM1 were used
to test the hypothesis of a perfectly mixed reactor for configuration no. 2 and how CFD-
ASM1 responded to the change in the kLa.
The initial conditions applied to both the CFD-ASM1 and PMR-ASM1 cases were
obtained from quasi steady state conditions for the biomass, particulate inerts concen-
trations and nutrient removal performance at the end of the SBR cycle. The initial con-
ditions mimic the instantaneous charging of the reactor with a solution containing
ammonia, organic nitrogen and organic substrates. The change in the concentration of all
the components is then monitored for more than 4 h of simulated time.
Examination of the CFD-ASM1 revealed a good comparison for two different specifica-
tions (Figure 2) of the gas phase condition and the liquid side film transfer coefficient used
(Bubble modelling impact, A, 1 mm spherical bubble with the Lamont coefficient; B, 3 mm
spherical bubble with the Higbie isolated coefficient). The kLa of the oxidation ditch calcu-
lated in CFD-ASM1 was then used to define the kLa in the PMR-ASM1 model. As can be

Figure 2 Oxygen concentrations profiles for perfectly mixed batch reactor; A: db ¼ 1 mm; kL ¼ Lamont;
82 kLa; B: db ¼ 3 mm; kL ¼ higbie isolated bubble
seen in Figure 2, B the selection of the kLa has a significant effect on the change in the oxy-
gen concentration profile, where a point and a global value of the kLa were used in defining
the PMR-ASM1 model. The global definition of the kLa resulted in a profile that was more
representative of the CFD-ASM1. Therefore, it is especially important when calibrating
model parameters to have a correct description of the oxygenation capacity of the reactor
studied, as the kLa could have a strong impact on the parameterisation process.
Examining the contour plots of the oxygen concentration (Figure 3), there is a small

G.C. Glover et al.

measurable change in the dissolved oxygen concentration from the centre of the reactor,
where the bubble plume from the aerator is found to the reactor walls. The concentration
of oxygen is sufficiently high enough to not affect the growth and consumptions rates of
the biomass components. Thus, negligible differences in the concentration gradients of
other components were observed. Therefore, the mixing characteristics of the other com-
ponents (ammonia, nitrate, organic substrates and organic nitrogen) can still be con-
sidered to be perfectly mixed.

Configuration no. 3: effect of complex hydrodynamics on CFD-ASM1. For observing

the effect of an non-perfectly mixed reactor on the local distribution of the ASM1
components, a real scale oblong oxidation ditch (6200 m3), being 28 m long, 17 m wide
and 9 m high was then selected. Five racks of membrane type aerators were used to
aerate the basin at a rate of 2000 Nm3 h21, to stabilise the bubble plumes and improve
the aeration efficiency. Liquid circulation in the horizontal direction was applied by three
marine type agitators rotating at 40 rpm.
As this case was used as a testing scenario to evaluate first tendencies, the initial con-
ditions and parameters values from case no. 1 (pilot plant oxidation ditch) were used.
The model of the reactor was also treated as a batch process. This significantly differs
from the actual process where the ditch is operated in a continuous mode. Thus, the con-
centrations of the components may be significantly different from those observed at the
plant due to the effect of continuous influent dilution with the mixed liquors in the
reactor.
Figure 4 shows the change in the concentration of soluble COD with time and the con-
centration gradient observed after 1 h of simulation time. The 3 CFD-ASM1 curves rep-
resent the numerical value in two different points (in black on the figure) and the volume
averaged concentration for the whole of the reactor.
Examining the profiles and the range in concentrations observed over the whole reac-
tor, the soluble COD is homogeneous, however, this is not observed to the same degree
in other components such as nitrate nitrogen and oxygen (Figure 5). The heterogeneities
arise from the insufficient injection of air into the reactor, resulting in a low oxygen con-
centration outside of the aeration regions. There was also a small gradient in the nitrate
nitrogen concentration with higher concentrations found on the aerator side of the reactor

Figure 3 Oxygen concentrations contours for the perfectly mixed batch reactor after 250 min of simulation
time; (a): db ¼ 1 mm; kL ¼ Lamont; kLa; (b): db ¼ 3 mm; kL ¼ Higbie isolated bubble 83
 50
CFD-ASM1 aeration side point 31.08
Soluble COD (mg l–1)

45 CFD-ASM1 agitation side point

CFD-ASM1 volume average 31.94SI+SS
PMR-ASM1
40
31.81
35
31.67
30

25 30.53
G.C. Glover et al.

0 10 20 30 40 50 60
y
Time (min) 30.40 z
x

Figure 4 Soluble COD concentration profiles and contours for a real plant ASB, where db ¼ 3.5 mm; kL
¼ Higbie isolated bubble; a ¼ 0.6 to give kLa0 ¼ 1.96 h21

1.27 0.45

1.02 SO SNO
0.39

0.75 0.32

0.51 0.26

0.25 0.20
y y
z
0.00 x 0.13 z
x

Figure 5 Dissolved oxygen and nitrate nitrogen for a real plant ASB, where db ¼ 3.5 mm; kL ¼ Higbie
isolated bubble; a ¼ 0.6 to give kLa0 ¼ 1.96 h21

(Figure 5). A comparative PMR-ASM1 case was also performed with the same initial
conditions and parameters, to observe the response of the components to perfectly mixed
reactor hydrodynamics. The average profile given for soluble COD is near to the CFD-
ASM1 case.
It must be stated that the heterogeneity is not as expressed as assumed at the begin-
ning for this system. This indicates that this real scale system is close to a homogeneous
or well-mixed system. Hence, a different system (e.g. channel) or different scenarios (e.g.
aerator failures) shall be simulated and studied in order to analyse the potential value of
such a CFD-ASM1 approach. Nevertheless, valuable insight of the potential has been
attained so far. Optimisation of the step-size during simulation, still respecting adequate
precision, makes the CFD-ASM1 approach feasible – at least for detailed diagnostic
work.
The biological model equations (ASM1) have been successfully implemented in the
CFD code. This model has been validated on two levels. Further work is necessary to
gain more experience with respect to type of activated sludge systems studied, as well as
methods for validations of local gradients and variations.

Global approach: systemic method in WWTP modelling

When modelling full-scale activated sludge plants, the question on the adequate model
structure arises, i.e. number and position of the PMR units used in common modelling
software. Often, the choice relies on the experience of the person involved and any inac-
curacy is shifted to the model calibration procedure (parameter estimation). The follow-
ing section presents the work performed with respect to the model structure identification
for ASM type models and discusses how information obtained via CFD analysis, resi-
dence time distribution (RTD) and other experimental means can be used. A closed loop
84 oxidation ditch has been chosen as base for discussion and a case study has been
performed on a 10,450 m3 AS unit (18,500 m3 d21, gas flowrate: 8,000 Nm3 h21 and six
impellers). The CFD results and the geometry of the ASB are presented in Figure 6.
The gas volume fraction depicted Figure 6 should not be mixed up with the dissolved
oxygen concentration, which, of course, can be differently distributed throughout the
reactor system.
The closed loop oxidation ditches are specific systems characterised by an internal
recycling rate. Hence, we study here the impact of the number of PMR in series and of

G.C. Glover et al.

the recycling rate on the effluent composition.
Several CFD simulations (variation of the inlet flow rate and the number of running
impellers) allowed the importance of the pumping flowrate delivered by the impellers on
the mean recirculation velocity (with negligible effect of the inlet flow rate) to be seen.
Therefore, the recycling flow rate should, so it seems, be considered more as a fixed
value for a specific system other than an adjustable parameter.
For identifying the model structure, i.e. number of PMR in series or other combi-
nation, the RTD curves can be used. The RTD curves calculated by CFD and reproduced
under WESTw are shown in Figure 7. Astonishingly, high numbers of PMR in series
were determined: approximately 90 CSTR in series in non-aerated conditions and
20 PMR in aerated conditions (Figure 7, top); with a recycling flow rate “Qrec” of
97650 m3 h21 for non aerated conditions; and only of 88,200 m3 h21 for aerated con-
ditions. These values were calculated using the CFD approach.
We can observe that, with an increasing number of PMR, the number and amplitude
of the oscillations of the RTD signal are increasing, approaching plug flow behaviour.
Generating a suitable model structure (i.e. number of PMR and flow pattern between
the tanks) from CFD results by making use of a transfer function, which results in a con-
figuration with 20% of the reactors in the recycle flow. The approach chosen was similar
to the one presented by Alex et al. (2002). This result could be due to the fact that the
inlet and the outlet to the basin are not at the same level. However, putting 20% of the
PMR in the recycle stream does not change the RTD signal significantly at this high
number of PMR (Figure 8).
The difference in number and size of the amplitudes during the first hour indicates a
more important mixing during the aerated period. The shape of the signal after the first
hour also illustrates that the hydraulic behaviour during the rest of the period reflects one
of a completely mixed system, which is in line with the results of De Clercq et al.
(1999). As a consequence, it is questionable that, although there are almost 100 PMR in
series needed to reproduce the RTD signal, this high number of PMR is needed to simu-
late/predict the effluent components of this specific activated sludge system. Indeed, no
significant differences in effluent components were detected when simulating the system
with different numbers of PMR at high recycling rates (94,500 m3 h21). This is certainly
due to the high recycling rate. It must be stated here, however, that due to numerical pro-
blems, the amount of PMR in series simulated remained restrained to the order of 10.

25 m 2*3 mixers

7m
Input

6.1 m 12.5 m
Aer
atio
n ra
mps
Output
66 m

Figure 6 ASB geometry (left) and gas volume fraction distribution (right) 85
G.C. Glover et al.

Figure 7 Comparison of RTD (top), impact of the recycling rate on the RTD and E(t) obtained with WESTw
for one PMR (bottom)

Figure 9 displays simulation results for different recycling rates and 5 PMR in series.
No phase shift can be detected, i.e. the different simulations exhibit the same variation in
time. With decreasing recycling rates, higher removal performances are obtained, i.e.
lower effluent concentrations. This is not a surprise as we are approaching plug flow
character, being known for better performances.
This result might suggest that lowering the internal recycling rate for this closed loop
oxidation ditch represents a method for improving its performance. One has to be
reminded, however, that the horizontal velocity, and hence the internal recycling rate, has
to be kept at above a minimum value in order to avoid sludge settling and decreased
aeration efficiency.
Overall, the investigations have underlined that closed loop oxidation ditches with
well functioning hydraulics can be modelled by a simple PMR unit. This is due to the
high internal recycling rate, specific to these systems, inducing homogenous conditions.
Hence, under these circumstances, there is no need for hydraulic pre-investigations via
CFD. However, in the case of malfunctioning due to erroneous design or equipment fail-
ure, a reduced internal recycle and hence changes in hydraulic conditions are likely to
occur. This can cause revisions to the model structure. Although errors related to the
86 choice of the model structure can be compensated partly by the calibration of the ASM
 G.C. Glover et al.

Figure 8 Two configurations for model structure (top) and comparison of RTD signal (bottom)

Figure 9 Impact of the internal recycling rate (N ¼ 5) on the soluble substrate SS in the effluent

models (Alex et al., 1999), hydraulic investigations are of great help under these con-
ditions. Of course, this is true also for other activated sludge systems that do not have
such an expressed internal recycle, e.g. channel systems, especially as transposing the
correct hydraulics into the activated sludge modelling avoids compensating errors related 87
 to the choice of the model structure during calibration. Furthermore, CFD analysis of the
multiphase flow mechanisms can be used to obtain a correct description of the oxygen-
ation capacity of the system studied. This information can then easily be implemented in
the classical ASM modelling (e.g. oxygen transfer).
With respect to the question as to whether there is a general method in model structure
selection when modelling activated sludge plants, the following approaches are
suggested. Any system that is characterised by a high internal recycling rate (oblong or
G.C. Glover et al.

annular AS units) can be modelled by one PMR unit. If one is unsure, if really hom-
ogenous conditions are prevailing, measurements of the dissolved oxygen profile along
the reactor can be a simple way to gain deeper insight. If no homogeneity is observed,
one might roughly derive the model structure for a chosen/fixed internal recycling rate by
reproducing the corresponding profile, as applied by Printemps et al. (2004). Careful
attention should be paid to systems without internal recycle flow rate and hydraulics that
are expected to approach plug flow behaviour.
More precisely, but also more expensive, would be to perform a tracer test to establish
the RTD signal for the system and use this signal for model structure determination.
A precise and time consuming approach consists of performing a detailed hydraulic anal-
ysis via CFD, allowing a deeper insight into complicated hydraulic conditions. This last
approach is favoured if detailed diagnostics are to be carried out in cases of malfunction-
ing or new design issues, accessing local hydrodynamics conditions induced in the unit
by factors such as the geometry of the ASB, and number, position, type, diameter and
rotating speed of the mixing systems.

Conclusions
Several approaches addressing the coupling of hydraulic modelling to biological model-
ling have been presented here. In general, it can be stated that activated sludge modelling
is enriched by CFD analysis and by incorporating a biological –chemical model equation
in the hydraulic model. Applying these CFD-bio models has, definitively, another objec-
tive than classical AS modelling. Currently, CFD-bio models are still limited to model-
ling several hours up to one day, due to the computational time required. This time
horizon already offers, however, the possibility for short term diagnostic work. Future
work will have to include their application to a variety of systems, in order to increase
the confidence of their prediction power.
Furthermore, CFD analysis can be useful to deduce the model structure of an ASM
type of model. It was shown, however, that closed loop oxidation ditches or race track
systems can sufficiently be modelled by a PMR approach. If homogenous conditions are
not evident, hydraulic analysis is necessary. Hence, classical biological modelling can
take full advantage of indirectly using the CFD results in order to obtain the local oxygen
capacity and the hydraulic structure (recycling rate and number of PMR) of the system,
especially in troubleshooting configurations.

References
Alex, J., Kolisch, G. and Krause, K. (2002). Model structure identification for wastewater treatment
simulation based on computational fluid dynamics. Wat. Sci. Tech., 45(4 –5), 325 – 334.
Alex, J., Tschepetzki, R., Jumar, U., Obenaus, F. and Rosenwinkel, K.-H. (1999). Analysis and design of
suitable model structures for activated sludge tanks with circulating flow. Wat. Sci. Tech., 39(4), 55 – 60.
De Clercq, B., Coen, F., Vanderhaegen, B. and Vanrolleghem, P.A. (1999). Calibrating simple models for
88 mixing and flow propagation in waste water treatment plants. Wat. Sci. Tech., 39(4), 61 –69.
Cartland Glover, G., Dhanasekharan, K., Vermande, S., Essemiani, K. and Meinhold, J (2004). Oxygen
transfer prediction in external loop airlift reactors. Impact of bubble modelling. JIE 2004, Poitiers,
France.
Essemiani, K., Vermande, S., Marsal, S., Phan, L. and Meinhold, J. (2004). Optimisation of WWTP Units
Using CFD – A Tool Grown For Real Scale Application. IWA Leading Edge Conference 2004, Prague.
Fluent Incorporated (2003). User Guide, Fluent Incorporated, Lebanon, New Hampshire, USA.
Henze, M., Gujer, W., Mino, T. and van Loosdrecht, M. (2000). Activated sludge models ASM1, ASM2,

G.C. Glover et al.

ASM2d ASM3. Scientific and Technical Report No. 9. IWA Publishing, London, UK.
Littleton, H.X., Daigger, G.T. and Strom, P.F. (2001). Application of computational fluid dynamics to closed
loop bioreactors. In Proceedings of WEFTEC 2001, Atlanta, Georgia, US.
Littleton, H.X., Daigger, G.T., Strom, P.F. and Cowan, R.A. (2003). Simultaneous biological nutrient
removal: evaluation of autotrophic denitrification, heterotrophic nitrification and biological removal in
full-scale systems. Wat. Env. Res., 75(2), 138 – 150.
Printemps, C., Baudin, A., Dormoy, T., Vanrolleghem, P.A. and Zug, M. (2004). Optimisation of a large
WWTP thanks to mathematical modelling. Wat. Sci. Tech, 50(7), 113 – 122.
Vanhooren, H., Meirlaen, J., Amerlinck, Y., Claeys, F., Vangheluwe, H. and Vanrolleghem, P.A. (2002).
WEST: Modelling biological wastewater treatment. J. Hydroinform., 5, 27 – 50.
Vermande, S. (2005). Hydraulics and biological modelling of activated sludge basins. PhD thesis, INSA
Toulouse France.

89