Professional Documents
Culture Documents
Ficara (2010) Anaerobic Digestion Models - A Comparative Study
Ficara (2010) Anaerobic Digestion Models - A Comparative Study
a Comparative Study ⋆
Elena Ficara ∗ Sonia Hassam ∗ Andrea Allegrini ∗
Alberto Leva ∗ Francesca Malpei ∗ Gianni Ferretti ∗
∗
Politecnico di Milano, 20133 Milan, Piazza Leonardo da Vinci 32,
Italy (e-mail: ferretti@elet.polimi.it).
Abstract: Based on the adoption of object-oriented modelling and simulation tools, in this
paper a comparison between the ADM1 and the AMOCO models is investigated, mainly in
order to assess the performance of AMOCO for control design. The ADM1 model has been
calibrated with reference to the degradation of waste activated sludge, based on literature data,
and assumed as reference model. Then, in order to compare the outputs of the two models, some
variables of ADM1 have been lumped to match the relevant aggregated AMOCO variables. Since
AMOCO failed in predicting the steady state values relevant to the inorganic carbon species and
alkalinity, a new version of it has been developed, accounting for the dynamics of the inorganic
nitrogen concentration. Steady-state and dynamic simulations based on this new version showed
an improvement with reference to simulations obtained with ADM1.
l
dC10
19 R being the gas constant.
qin l l
X ′
= C10,in − C10 + ρj υ10,j − ρT,10 (2)
dt Vl
j=1 3. THE AMOCO MODEL
19
l
dC11 qin l l
X
= C11,in − C11 + ′
ρj υ11,j (3) In the literature several anaerobic digestion models are re-
dt Vl
j=1 ported which are much simpler than ADM1. Among them,
′
where υ10,j ′
and υ11,j account respectively for the carbon the most important is AMOCO Bernard et al. (2001), de-
and nitrogen content of the components and ρT,10 is the veloped to support monitoring and control system design
gas transfer rate relevant to component C10 g
. For the sake for anaerobic digestion processes, rather than as a tool for
of brevity the Petersen matrix is not reported here, the numerical simulation of the process behaviour, and mainly
reader is deferred to Blumensaat and Keller (2005) or focused on the description of the anaerobic digestion of sol-
Allegrini (2010). Anions and cations dynamics is simply uble substrates or with a negligible particulate content. In
defined by the dilution effect of the reactor, since they are contrast to ADM1, AMOCO has not been largely applied
non reactive species, yielding in the research field, probably because of its limited range
of applicability.
dCil qin l l
= Ci,in − Ci , i = 25, 26 (4) Only two bacterial populations are considered (instead of
dt Vl
the seven considered by ADM1), in particular acidogenic
while the dynamics of the dissociated acid and free ammo- and methanogenic, while the hydrolysis and acetogenic
nia are given by phases are no longer considered explicitly. In the first step,
the acidogenic bacteria X1 (gVS L−1 ) consume the organic
dCil
= −ρA,i , i = 27, . . . , 32 (5) substrate S1 (gCOD L−1 ) and produce CO2 (mmol L−1 )
dt and volatile fatty acids S2 (mmol L−1 ). The population of
methanogenic bacteria X2 (gVS L−1 ) uses, in the second
where ρA,i (kgCOD m−3 d−1 ) is the acid-base kinetic
step, the VFAs as substrate for growth and produces CO2
rate relevant to component Cil Rosen et al. (2006). The and methane (the gaseous species). The model is thus
dynamic equations for the gas phase are defined by 6 differential equations: 2 for the mass-balances
dCig qg g Vl of the bacterial populations X1 and X2 , 2 for the organic
=− C + ρT,i , i = 1, 2, 3 (6)
dt Vg i Vg substrate S1 and the VFAs S2 and, finally, 2 for alkalinity
Z and inorganic carbon C. This provides
where Vg (m3 ) is the total gas volume inside the reactor
and ρT,i (kgCOD m−3 d−1 ) is the gas transfer rate relevant dS1 qin S1
= (S1,in − S1 ) − k1 µ1,max X1 (11)
to component Cig Rosen et al. (2006). The only algebraic dt VL S1 + KS1
equation, defining the charge balance dS2 qin S1
= (S2,in − S2 ) + k2 µ1,max X1
dt VL S1 + KS1
Sacm Sprom Sbum
Scat + SNH4 − SHCO3 − − − S2
64 112 160 (7) − k3 µ2,max X2 (12)
Svam Kw S2 + KS2 + S22 /KI2
− − San + Sh − =0
208 Sh dX1 qin S1
=− X1 + µ1,max X1 (13)
dt VL S1 + KS1
where SN H4 = SIN − SN H3 is the concentration of am- dX2 qin S2
monium and KW (kmol2 L−2 ) is the dissociation constant dt
=−
VL
X2 + µ2,max
S2 + KS2 + S22 /KI2
X2 (14)
of water, determines the molar concentration of hydrogen
dZ qin
ion SH , which in turn determines the pH = − log10 SH , = (Zin − Z) (15)
dt VL
which affects the dissociation of VFAs and ammonia. Two
dC qin S1
other main straightforward modifications were introduced. = (Cin − C) + k4 µ1,max X1
dt VL S1 + KS1
Firstly, in order to have a single variable describing the
S2
neutralization capacity of the solution inside the reactor, + k5 µ2,max X2 − r C (16)
S2 + KS2 + S22 /KI2
its alkalinity was calculated by
ϕ = C + S2 − Z + K H P T
A = Scat − San + SIN (8)
k6 S2
+ µ2,max X2 (17)
kLa S2 + KS2 + S22 /KI2
Secondly, to derive the pH of the influent substrate, the
rC = kLa (C + S2 − Z)
charge balance in (7) was calculated with reference to the p
influent composition. Finally, the biogas output flowrate ϕ− ϕ2 − 4KH PT (C + S2 − Z)
− kLa (18)
qg can be computed as 2
pg S2
qg = kp (pg − patm ) (9) rCH4 = k6 µ2,max X2 (19)
patm S2 + KS2 + S22 /KI2
where ki , (i = 1, . . . , 6) are stoichiometric coefficients, X̃2 = (Xac + XH2 )/1.55 (23)
µi,max , (i = 1, 2) (d−1 ) are the maximum specific growth Z̃ = 1000 (Sva /208 + Sbu /160
rates, KS1 (gVS L−1 ) and KS2 (mmol L−1 ) are the half-
saturation constants, KI2 (mmol L−1 ) is the inibition con- + Spro /112 + Sac /64 + SHCO3 ) (24)
stant, KH is Henry’s constant for CO2 (mmol L−1 atm−1 ), C̃ = 1000 Sic (25)
PT = 1 (atm) is the atmospheric pressure, kLa (d−1 ) is
the liquid-gas transfer coefficient, rC (mmol L−1 d−1 ) is q̃c = 1000 ρT,10 (26)
the carbon dioxide production rate and rCH4 (mmol L−1 q̃CH4 = 1000 ρT,9 /64 (27)
d−1 ) is the methane production rate. The inorganic carbon
is assumed mainly composed of dissolved carbon dioxide
CO2 (mmol L−1 ) and bicarbonate B (mmol L−1 ), ne- The AMOCO parameters can be estimated based on
glecting the amount of carbonate in the normal operating steady state measurements in different operating condi-
conditions, while the total alkalinity Z is defined as the tions Bernard et al. (2001), replaced in our case by steady
sum of dissociated acids in the liquid phase (bicarbonate state values obtained from ADM1, fed with the benchmark
substrate Rosen et al. (2006) at varying hydraulic retention
and VFAs, assumed as completely dissociated in the pH time. However, the simple linear least-square regression
range of interest). proposed in Bernard et al. (2001) failed at high tHR due
Of course, AMOCO introduces a number of simplifica- to the absence of a decay term in the growth rate of
tions with respect to ADM1. The influent substrate is the biomasses. A first modification of AMOCO has been
therefore introduced, by adding a decay term for both
considered to be already fully hydrolised, no particulate biomasses, estimated as the 10% of the maximum bacterial
compound is therefore taken into account. Inert fractions growth rates µ1,max and µ2,max as
(Xi and Si ) are not considered because the model does
not consider non biodegradable fractions. The nitrogen dX1 qin
S1
=− X1 + µ1,max − k d X1 (28)
balance is not taken into account nor it is included in dt VL S1 + KS1
the computation of the alkalinity. The pH calculation is dX2 qin
based on the inorganic carbon equilibrium only and it has =− X2
dt VL
no influence in any process: it is just a rough estimation
S2
of the pH based on the bicarbonate chemical equilibrium + µ2,max − kd X2 (29)
S2 + KS2 + S22 /KI2
only, Finally, relevant inhibition effects are neglected, such
as excess of free ammonia or H2 inside the reactor. where kd is a decay constant. As a consequence, the
linearity of the regression used for parameters tuning was
4. MODEL TUNING lost, while linear least-squares regression was still applied
to the estimation of the yield coefficients (ki , i = 1, . . . , 6).
ADM1 requires a detailed substrate definition but, on the Anyway, the steady state values computed by AMOCO
other hand, the parameters monitored during the full-scale with the newly identified parameters were different from
those computed by ADM1. Indeed, the steady-state con-
process are limited (just a few online) and, usually, they
are macro-parameters describing aggregated information. centrations of substrates and biomasses where sufficiently
Substrate characterization methods have been studied closed to the equivalent ADM1 values, whereas all vari-
Zaher et al. (2003); Kleerebezem and Van Loosdrecht ables concerning inorganic carbon (C, B, Z, CO2 , qc )
(2006); Huete et al. (2006) but, in general, the complete assumed much more different values. Note that also in
characterization of the substrate for ADM1 must be dealt Bernard et al. (2001), comparing the modelling results
with case by case. The work by Rosen and Jeppsson with experimental data, a bias was observed in total inor-
Rosen et al. (2006) reported an example of waste activated ganic carbon C and alkalinity Z, and it was ascribed to
sludge characterization used for steady-state simulation. the uncertainty in measurements of influent alkalinity. In
According to the authors, the input values may not be this work, to improve the description of the total inorganic
completely realistic for all variables but they have been carbon inside the fermenter, a modification of AMOCO is
chosen such that every input is active (i.e. non-zero) and proposed and described in the next section.
able to excite all internal modes of ADM1. For this reason
it was also chosen to be the benchmark substrate in this 5. AMOCO MODIFICATION ACCOUNTING FOR
work. All parameters of ADM1 were therefore set to the THE ROLE OF NITROGEN
values suggested in Rosen et al. (2006).
As a consequence of having neglected the nitrogen species,
A major issue in the comparison between the outputs of AMOCO considers alkalinity Z as non-reactive and, conse-
the two models is related to the need of lumping sev- quently, its dynamics is just described by the dilution effect
eral variables of ADM1 into single variables of AMOCO. of the reactor (15). On the contrary, in ADM1 alkalinity
Omitting details for space reasons, and denoting with is given by eq. (8) and its dynamics reflects the sum of
the tilde the equivalent AMOCO variable obtained by the dynamics of the alkalinity constituents: bicarbonates,
lumping ADM1 variables, the conversion formulas can be VFAs, hydroxide ions and, above all, free ammonia. In fact,
summarized as from eqs. (7) and (8) one obtains
A = Scat − San + SIN (30)
S̃1 = Ssu + Saa + Sf a + Xc + Xch + Xpr + Xli (20)
so that if the inorganic nitrogen is neglected alkalinity
S̃2 = 1000 (Sva /208 + Sbu /160 + Spro /112 + Sac /64) (21)
is given by the difference between cations and anions
X̃1 = (Xsu + Xaa + Xf a + XC4 + Xpro )/1.55 (22) concentrations in the solution only, which are actually
non-reactive species in ADM1. Neglecting the contribution 1 .6
1 .4
1 .6
1 .4
of the nitrogen species and, in particular, their interac-
0
1
0
2
1 .2 1 .2
/X
/X
tion with the bicarbonate equilibrium, has also a direct
2
X
X
1 1
0
1
0
2
/S
/S
2
1
monium (N H4+ ) released upon organic protein hydrolisis,
1 2
S
S
0
to form ammonium bicarbonate (N H4 HCO3 ). AMOCO 0 .5
1 .6
0 5 0 1 0 0 1 5 0 2 0 0 2 5 0 3 0 0 3 5 0
1 .8
0 5 0 1 0 0 1 5 0 2 0 0 2 5 0 3 0 0 3 5 0
4
1 .4
H
1 .2
0
/rC
0
rc /rc
4
1 .2
H
rC
the introduction of new parameters describing the nitrogen 1
1
0 .8
content of the organic substrate S1 and of the biomasses, 0 5 0 1 0 0 1 5 0 2 0 0
T im e ( d a y s )
2 5 0 3 0 0 3 5 0
0 .8
0 5 0 1 0 0 1 5 0 2 0 0
T im e ( d a y s )
2 5 0 3 0 0 3 5 0
hence
Fig. 2. Responses of X1 /X10 a) X2 /X20 b) S1 /S10 c) S2 /S20
0
dN qin S1 d) qCH4 /qCH e) qc /qc0 f) to 50% perturbations to
= (Nin − N ) + [k1 NS1 − Nbac ] µ1,max X1 4
dt VL S1 + KS1 S1,IN . Solid line: ADM1 model, dotted line: AMOCO
S2 model, dashed line: AMOCO N model.
− Nbac µ2,max X2
S2 + KS2 + S22 /KI2 1 .6 1 .6
1 .4 1 .4
+ kd Nbac µ1,max X1 + kd Nbac µ2,max X2 (31)
0
1 .2 1 .2
0
Z /Z
B /B
where NS1 is represented by the nitrogen content of 0 .8
1
0 .8
1
0
2
acidogenic process; Nbac is the nitrogen content in the
0
/C O
1 .2 1 .1
C /C
2
C O
1
biomass, respectively uptaken from or released into the 0 .8 0 .9
1
1 .4 1 .0 5
0
0
1 .2
p H /p H
1
N /N
eqs. (15) and (31) while still denoting with Z the alkalinity,
AMOCO may be simply modified, in order to account for Fig. 3. Responses of Z/Z 0 a) B/B 0 b) C/C 0 c) CO2 /CO20
the contribution of the nitrogen species, by substituting d) N/N 0 e) pH/pH0 f) to 50% perturbations to S1,IN .
eq. (15) with Solid line: ADM1 model, dotted line: AMOCO model,
dashed line: AMOCO N model.
dZ S1
= D(Zin − Z) + [k1 NS1 − Nbac ] µ1,max X1 6.2 Dynamic simulation results
dt S1 + KS1
S2
− Nbac µ2,max X2 More than in the steady state, we are interested in the pre-
S2 + KS2 + S22 /KI2 diction of the dynamic response by the simplified AMOCO
+ kd Nbac µ1,max X1 + kd Nbac µ2,max X2 (32) models. Therefore, dynamic simulations of the digester
response to a 50% step increase followed by an equiva-
lent step decrease in the influent S1 concentration were
The model thus maintains the same number of state performed with all models. For comparison purposes, the
variables of the original version, eq. (31) may be considered dynamic value of each state variable was referred to its
just to evaluate the dynamics of the variable N , if of steady state one, thus computing a percentage variation
interest. Y /Y 0 , where Y 0 is the steady-state value of variable Y .
This was meant to make it easier to compare dynamic
responses of variables having different, sometime remark-
6. COMPARISON BETWEEN ADM1 AND AMOCO ably, steady state values. Results are plotted in Figs. (2)
and (3).
6.1 Steady-state analysis For all models, biomasses concentration (X1 and X2 ) dy-
namics shows the typical response of first order systems,
After the ADM1 variables lumping procedure and the coherently with the fact that unlimited growth of microor-
parameter identification, steady-state computations were ganisms is described by a first order reaction. Slightly
run. Details are omitted for space reasons; suffice to say faster responses are simulated by the two AMOCO models,
that both models closely predict the behaviour of the indicating that the calibration process has slightly over-
organic substrate, the biomass concentrations , the pH, estimated the growth rate of these two lumped biomass
and the methane flow rate, with percentage differences populations. Growing faster, those biomasses also reach
always below 15%, while the performance of both AMOCO a slightly higher asymptotic concentration. More relevant
models with respect to the inorganic species is far less differences can be observed when considering the S1 and S2
satisfactory, owing to the simplifications adopted in the dynamics. As for S1 , the ADM1 response is again that one
description of the complex CO2 -producing fermentation of a first order system, since X1 includes particulate com-
processes. ponents whose hydrolysis to soluble organic matter is de-
scribed by ADM1 as a first order reaction. On the contrary, the description of the physico-chemical processes, such as
this first order process is missing in both AMOCO ver- the hypothesis of a non reactive alkalinity, the inorganic
sions, since the model was originally proposed to describe species were also poorly predicted. As a first attempt to
the anaerobic degradation of soluble organic compounds, improve the AMOCO performance, the release of ammo-
as those taking place in UASB (Upflow Anaerobic Sludge nium during the first degradation step was implemented,
Blanked) reactors. In AMOCO, S1 is used as the growth leading to a remarkable improvement in the simulation
substrate for X1 , therefore it initially increases in response of the inorganic species. To further improve the AMOCO
to the step increase in its influent concentration and is performance and applicability to complex substrates, one
later reduced thanks to the increased X1 concentration. further first order hydrolysis step should be included,
Generally speaking, the S1 concentration in both AMOCO which requires the implementation of one more state vari-
models follows the typical response of a soluble substrate able.Also, the ammonia inhibition term may be added, as
degraded biologically inside a CSTR. As for S2 , a major well as the effect of the process pH on the methanogenic
difference can be observed between AMOCO models and process in order to couple the behaviour of the inorganic
ADM1. This is related to the free ammonia inhibition on species to that one of the biological system. These modi-
the growth of methanogens (the most relevant X2 biomass) fications are expected to improve the model performance
that is taken into account in ADM1, but that is neglected without loosing its simple structure.
in the AMOCO ones. Being methanogens growth inhib-
ited, ADM1 predicts both a slower increase in X2 , and an REFERENCES
increase in its substrate S2 . Due to the commented differ-
ence in the methanogenic process, the methane production Allegrini, A. (2010). Anaerobic Digestion Modelling: a
dynamics is accordingly different, with less methane pro- Comparison between ADM1 and AMOCO. Master’s
duced according to ADM1. When considering inorganic thesis, Politecnico di Milano.
components, differences are observed between the original Batstone, D.J., Keller, J., Angelidaki, I., Kalyuzhnyi, S.V.,
AMOCO and the modified one, the major one concerning Pavlostathis, S.G., Rozzi, A., Sanders, W.T.M., Siegrist,
the alkalinity dynamics. AMOCO N correctly predicts the H., and Vavilin, V.A. (2002). The IWA Anaerobic
alkalinity dynamic response, while the original AMOCO Digestion Model No. 1 (ADM1). Water Science and
does not consider alkalinity as a reactive species. As a Technology, 45(10), 65–73.
consequence, dynamic response for all inorganic carbon Bernard, O., Hadj-Sadok, Z., Dochain, D., Genovesi, A.,
species are better predicted by AMOCO N, although time and Steyer, J. (2001). Dynamical model development
constants for AMOCO N are overestimated, possibly be- and parameter identification for an anaerobic wastewa-
cause of the already commented difficulty in estimating the ter treatment process. Biotechnology and Bioengineer-
CO2 release by the acidogenic process. Having improved ing, 75(4), 424–438.
the description of the alkalinity behaviour, also pH pre- Blumensaat, F. and Keller, J. (2005). Modelling of two-
dictions are remarkably improved. Finally, the ammonium stage anaerobic digestion using the IWA Anaerobic
dynamics is sufficiently well predicted by the AMOCO N Digestion Model No. 1 (ADM1). Water Research, 39(1),
process, with a slightly faster release, that also causes the 171–83.
higher steady state nitrogen concentration. Gal, A., Benabdallah, T., Astals, S., and Mata-Alvarez, J.
(2009). Modified version of adm1 model for agro-waste
7. CONCLUSIONS application. Bioresource Technology, 100(11), 2783–
2790.
The simple AMOCO process, originally proposed to de- Graef, S.P. and Andrews, J.F. (1974). Stability and control
scribe the anaerobic degradation of soluble organic matter, of anaerobic digestion. Journal of the Water Pollution
was applied to describe the anaerobic degradation of the Control Federation, 46(4), 666–683.
more complex substrate that is waste activated sludge, Huete, E., de Gracia, M., Ayesa, E., and Garcia-Heras,
taking place in a CSTR. Its applicability to this case study J.L. (2006). ADM1-based methodology for the charac-
was checked by using ADM1 (Anaerobic Digestion Model terisation of the influent sludge in anaerobic reactors.
n. 1) that is, at present, the most comprehensive model for Water Science and Technology, 54(4), 157–166.
the description of anaerobic digestion. To do that, a lump- Kleerebezem, R. and Van Loosdrecht, M. (2006). Waste
ing procedure to compare the many ADM1 variables to characterization for implementation in ADM1. Water
the few AMOCO ones was proposed. Then, the AMOCO Science and Technology, 54(4), 167–74.
parameters were calibrated according to synthetic data Rosen, C., Vrecko, D., Gernaey, K., Pons, M., and Jepps-
series obtained by ADM1 simulations. Results show that, son, U. (2006). Implementing ADM1 for plant-wide
despite the much simpler model structure, AMOCO was benchmark simulations in Matlab/Simulink. Water Sci-
able to well predict steady state values of biomasses and ence and Technology, 54(4), 11–19.
methane production rates, while being less performing in Stemann, S.W., Ristow, N.E., Wentzel, M.C., and Ekama,
predicting the inorganic carbon species and pH values. Its G.A. (2005). A steady state model for anaerobic diges-
dynamic response to a step modification in the influent tion of sewage sludges. Water SA, 31(4), 511–528.
concentration highlighted the major limits of this model. Zaher, U., Rodrguez, J., Franco, A., and Vanrolleghem,
First, it does not include a first order hydrolysis step, that P.A. (2003). Application of the IWA ADM1 model to
has a relevant impact in the simulation of the AMOCO simulate anaerobic digester dynamics using a concise set
S1 variable. Moreover, it does not consider the inhibition of practical measurements. In Proceedings IWA Confer-
effects that some operative conditions may cause on the ence on Environmental Biotechnology, Advancement on
degradation process, such as the free ammonia inhibition Water and Wastewater Applications in the Tropics, 12.
on methanogens growth. Finally, due to simplifications in Kuala Lumpur, Malaysia.