modelling of two-stage anaerobic digestion using the iwa anaerobic digestion model no. 1 (adm1)
TRANSCRIPT
ARTICLE IN PRESS
0043-1354/$ - se
doi:10.1016/j.w
�Correspond+49 351 463372
E-mail addr
(F. Blumensaat
Water Research 39 (2005) 171–183
www.elsevier.com/locate/watres
Modelling of two-stage anaerobic digestion using the IWAAnaerobic Digestion Model No. 1 (ADM1)
F. Blumensaata,�, J. Kellerb
aInstitute for Urban Water Management, Technische Universitat Dresden, D-01062 Dresden, GermanybAdvanced Wastewater Management Centre, University of Queensland, QLD 4072, Research Road, St.Lucia, Brisbane, Australia
Received 2 November 2003; received in revised form 4 May 2004; accepted 23 July 2004
Abstract
The aim of the study presented was to implement a process model to simulate the dynamic behaviour of a pilot-scale
process for anaerobic two-stage digestion of sewage sludge. The model implemented was initiated to support
experimental investigations of the anaerobic two-stage digestion process. The model concept implemented in the
simulation software package MATLABTM/Simulinks is a derivative of the IWA Anaerobic Digestion Model No.1
(ADM1) that has been developed by the IWA task group for mathematical modelling of anaerobic processes. In the
present study the original model concept has been adapted and applied to replicate a two-stage digestion process.
Testing procedures, including balance checks and ‘benchmarking’ tests were carried out to verify the accuracy of the
implementation. These combined measures ensured a faultless model implementation without numerical incon-
sistencies. Parameters for both, the thermophilic and the mesophilic process stage, have been estimated successfully
using data from lab-scale experiments described in literature. Due to the high number of parameters in the structured
model, it was necessary to develop a customised procedure that limited the range of parameters to be estimated. The
accuracy of the optimised parameter sets has been assessed against experimental data from pilot-scale experiments.
Under these conditions, the model predicted reasonably well the dynamic behaviour of a two-stage digestion process in
pilot scale.
r 2004 Elsevier Ltd. All rights reserved.
Keywords: Anaerobic two-stage digestion; Sewage sludge; Mathematical modelling; ADM1; Benchmarking; MATLAB/Simulink
1. Introduction
Primary and waste activated sludges are by-products of
municipal wastewater treatment. Prior to their re-use or
disposal, the sludges require further treatment to reduce
pathogen levels and organic content. Anaerobic diges-
e front matter r 2004 Elsevier Ltd. All rights reserve
atres.2004.07.024
ing author. Tel.: +49 351 46336908; fax:
04.
ess: [email protected]
).
tion, particularly the mesophilic single-stage process, is a
well established biological treatment process for the
degradation and stabilisation of municipal sewage sludge,
producing valuable methane gas. Changing standards for
the use and disposal of sewage sludge and the need to
reduce operational costs lead to an increased interest in
processes that treat sewage sludge more effectively. One
way to increase efficiency involves the physical separation
of the single-stage anaerobic digestion process into two
process stages, a thermophilic pre-treatment followed by
a mesophilic main treatment stage.
d.
ARTICLE IN PRESSF. Blumensaat, J. Keller / Water Research 39 (2005) 171–183172
To improve the understanding of this particular
process configuration, a model concept was reviewed,
adapted for a thermophilic/mesophilic digestion process
and implemented into MATLAB/Simulink. The imple-
mentation has been tested, calibrated using lab-scale
data from literature (Siegrist et al., 2002), and finally
validated against data obtained from pilot-scale experi-
ments. The generation of this model implementation was
initiated to support a pilot-scale investigation of the
anaerobic two-stage digestion process for municipal
sewage sludge.
The model presented is based on the International
Water Association (IWA) Anaerobic Digestion Model
No. 1 (ADM1) published in the IWA Scientific and
Technical Report No.9 (STR), (Batstone et al., 2002).
To date, little experience exists in implementing the
ADM1 model concept or derivatives of it. Findings of
this study may contribute further information about
modelling and simulation work with this as yet new
model concept.
2. Theoretical background
2.1. Two-stage anaerobic digestion
The process of anaerobic degradation can be divided
into several reactor processes to optimise operational
conditions. The concept of splitting the conventional
process of anaerobic digestion into two different stages
is known as two-stage, or two-phase anaerobic diges-
tion. These two terms are frequently and equally used to
describe this particular process technology.
The anaerobic two-stage process investigated in this
study is a process that permits selection and enrichment
of different bacteria by independently controlling the
operating conditions in each digesting system. The
original naming ‘‘two-phase digestion’’ is based on
Pohland and Ghosh (1971), who first proposed to
separate the two main groups of microorganisms, acid-
forming bacteria and methanogenic archaea, physically
Feed Sludge
First Stage
Thermophilic Reactor
Biogas
Transfer Sludge
Fig. 1. General process scheme of the tw
into two reactors. Thus, optimal conditions for their
different growth kinetics are provided. The acid form-
ing, or acidogenic bacteria are favoured in the first stage,
producing volatile fatty acids (VFAs) such as acetate
from complex carbon sources. The VFAs are then
utilised by methanogenic archea in the second stage of
the process. A general process scheme of the two-stage
anaerobic process is shown in Fig. 1.
Numerous authors have demonstrated the advantages
of two-stage anaerobic digestion compared to conven-
tional single-stage systems, e.g. Cohen et al. (1979,
1980), Siegrist et al. (1993), Anderson et al. (1994) and
Oles et al. (1997).
On the other hand, few disadvantages such as high
investment costs and operational instability under
certain circumstances are known as drawbacks of the
two-stage process configuration. The benefits however
underline the significance of this particular digestion
technology and hence justify its application and the
development of a process model.
2.2. The ADM1 as basic model concept
The ADM1 is a structured, mathematical model based
on COD as a common base unit in wastewater
characterisation. It describes the anaerobic digestion
process in a continuous-flow stirred tank reactor.
Biochemical processes included in the ADM1 are:
�
o-s
disintegration of composites,
�
hydrolysis of particulate COD, such as carbohy-drates, proteins and lipids,
�
six substrate degradation processes together withtheir six specific biomass growth and decay processes.
The novel aspect, in comparison with other previously
developed anaerobic digestion models (Siegrist et al.,
1993; Vavilin et al., 1994; Angelidaki et al., 1999;
Batstone 2000), is the implementation of the disintegra-
tion step. Cellular solubilisation steps are divided into
disintegration and hydrolysis, of which the first is largely
a non-biological step that converts composite particulate
Digested Sludge
Second Stage
Mesophilic Reactor
Biogas
tage anaerobic digestion process.
ARTICLE IN PRESSF. Blumensaat, J. Keller / Water Research 39 (2005) 171–183 173
substrate to inerts, particulate carbohydrates, proteins
and lipids.
The subsequent phases, microbial-induced enzymatic
hydrolysis of particulates and the stepwise degradation
of the hydrolysed substrates, are represented according
to the generally accepted theory.
Decay processes are described producing inert and
degradable particulate organic matter (composites)
which again undergoes the disintegration step. The
physicochemical process of stripping the gaseous com-
pounds, hydrogen, methane and carbon dioxide, is
included to represent the production of biogas. The
pH calculation is based upon six additional physico-
chemical processes, describing the acid/base equilibria of
CO2/HCO3�, NH4
+/NH3, acetic acid/acetate, propionic
acid/propionate, butyric acid/butyrate and valeric acid/
valerate.
Thus, the reactor state is defined by 26 or 32 state
variables respectively, depending on the implementation
method. Dynamic states for cations and anions are
introduced to address the influence of positively and
negatively charged ions present in the system on the pH.
Using the same approach as in the ASM1 (Henze et al.,
1987), the entire system of dynamic state variables is
expressed in a rate equation matrix.
3. Material and methods
3.1. Model structure
The following section briefly summarises the funda-
mentals of the present model derived from the ADM1
model concept (Batstone et al., 2002).
Dynamic state variables: As a set of ordinary
differential equations (DE system), the model consists
of 32 dynamic state concentration variables and involves
19 biochemical rate processes, three gas–liquid transfer
processes and an additional six acid–base kinetic
processes. In order to characterise the acid–base
dissociation in aquatic systems, the dynamic states for
VFAs, inorganic carbon (IC) and inorganic nitrogen
(IN) are split into two components each.
Rate equation matrix: Table 1 shows the 32 state
variables structured in the modified process rate and
stoichiometric matrix for biochemical and acid/base
equilibria reactions. The original matrix provided in the
STR (Batstone et al., 2002) was modified as suggested by
including physicochemical reactions in order to describe
the effect of physicochemical states, such as pH, upon
the biochemical reactions.
Calculation of the derivatives: As proposed in the
original model concept, it is assumed that there is no
biomass accumulation and hence, the hydraulic reten-
tion time (HRT) equals the solids retention time (SRT).
The calculation of the state variables is shown in Eq. (1),
provided that the bulk volume in the reactor remains
constant.
dSliq;i
dt¼
qSin;i
V liq�
qSliq;i
V liqþ
X
j¼1�19
rjvi;j ; (1)
where the termP
j¼1�19rjvi;j is the sum of the specific
kinetic rates for process j, multiplied by the stoichio-
metric coefficient ni,j (see Table 1). Vliq is the liquid
reactor volume, q the flow into and out of the reactor,
Sin;i the input concentration of the soluble (or X in;i for
the particulate) components and Sliq;i the derivative of
the soluble (or Xi for the particulate) components.
Gas state variables: The integration of the gas state
variables into the system of dynamic state variables was
done to avoid the implementation of an additional gas
vessel, which would have complicated the entire model
structure. The three additional gas state variables were
calculated according to algebraic variables, the overall
gas flow and the gas transfer rate.
The biogas composition was described by the partial
pressures of methane, carbon dioxide and hydrogen as
well as the water vapour pressure.
Acid/Base equilibria: For the integration of the
acid–base rates, a different approach was used to the
one suggested in the STR (Batstone et al., 2002). The
acid–base rates are applied to the differential equations
for the VFAs and IC as follows
dSliq;i
dt¼
qSin;i
V liq�
qSliq;i
V liqþ rA=B;i; (2)
and
dSliq;i�
dt¼
qSin;i�
V liq�
qSliq;i�
V liqþ
X
j¼1�19
rjvi;j � rA=B;i; (3)
where i and i� stand for the free and ionic forms
respectively of organic acids and bicarbonate. On the
other hand, the acid–base equilibrium of inorganic
nitrogen is implemented differently. The biochemical
rate equation termP
j¼1�19rjvi;j is applied to ammonium
rather than to ammonia. The use of this configuration
was motivated by the perception that it is more
appropriate to apply the rate equation term to the
quantitatively dominant form of the split state variables,
rather than to the form which represents only a small
amount of the respective compound. It had been
assumed that this application causes the least numerical
problems. This assumption was proved right by
extensive testing, where the greatest accuracy was
achieved performing simulations with this particular
configuration.
Inhibition: Inhibition kinetics were comprehensively
reviewed by Batstone et al. (2002). To reduce the
complexity of the model it was suggested that only a
few inhibition mechanisms should be addressed. Follow-
ARTICLE IN PRESSTable1
Biochem
icalratecoefficients(n
i;j)andkineticrateequations(r
j)F. Blumensaat, J. Keller / Water Research 39 (2005) 171–183174
ARTICLE IN PRESSF. Blumensaat, J. Keller / Water Research 39 (2005) 171–183 175
ARTICLE IN PRESSF. Blumensaat, J. Keller / Water Research 39 (2005) 171–183176
ing this recommendation, the inhibition forms were
applied as proposed in the STR (Batstone et al., 2002).
Processes excluded: In order to further reduce the
complexity of the model, a number of processes were
excluded. Processes being omitted from the model
concept are:
�
production of lactate from glucose fermentation,�
sulphate reduction and sulphide inhibition,�
nitrate reduction,�
long chain fatty acid (LCFA) inhibition,�
competitive uptake of H2 and CO2 between hydro-genotrophic methanogenic archaea and homoaceto-
genic bacteria (significant in psychrophilic systems),
�
solids precipitation due to high alkalinity or otherchemical precipitation reactions.
The model presented aims to simulate sewage sludge
digestion at typical conditions. Based on this stipulation,
it is presumed that none of the above processes
significantly influence the modelled system.
3.2. Consistency checks
Balance checks for carbon and nitrogen were per-
formed using the stoichiometric coefficients and para-
meters suggested by Batstone et al. (2002). Discrepancies
were found in both balances, and modifications to the
carbon or nitrogen content of particular compounds
were made to eliminate or minimise those. Key
modifications made are outlined in the following section.
COD balance: COD balancing is implied in the rate
equation matrix. No fitted or calculated ‘‘product on
substrate yields’’ are necessary and hence no internal
balance was calculated.
Carbon balance: The rate coefficient for inorganic
carbon is expressed as balancing term (Eq. (4)), as
recommended in the STR (Batstone et al., 2002). The
term is originally applied to the uptake of sugars, amino
acids, propionate, acetate and hydrogen (biochemical
processes j=5, 6, 10, 11, 12) since in these cases
inorganic carbon is the carbon source for, or the
product of, anabolism and catabolism. Due to the
model implementation as ODE system with an expanded
number of state variables, the term was extended for
these additional state variables to give
v15;j ¼ �X
i¼1�13;16�30
Civi;j ; (4)
where Ci is the carbon content (kmol CkgCOD�1) of
the component i and n15,j is the inorganic carbon
stoichiometric coefficient for process j.
Furthermore, an inorganic carbon balance term
for decay processes was introduced. Initial carbon
balance checks revealed that the amount of inorganic
carbon released due to decay of biomass ‘‘is lost’’ in
the system. Assuming that decaying biomass remains
as particulate composites in the system, which is
then partially disintegrated into particulates and
subsequently hydrolysed into soluble substrate, the
carbon cycle must close. However, the carbon
content of biomass (Cbac=0.03125 kmol CkgCOD�1)
and the carbon content of the particulate composites
in the feed (CC=0.0279 kmolCkgCOD�1) differs
significantly. To resolve this discrepancy the following
term (Eq. (5)) was introduced to shift the
inorganic carbon released due to decay of biomass
to the dissolved inorganic carbon state
v15;j ¼ �X
i¼19;23�29
Civi;j : (5)
Nitrogen balance: As with the carbon balance, the
nitrogen balance calculated using the original para-
meters given in the ADM1 concept did not close. This
was again caused by a discrepancy between the nitrogen
content in the biomass and the composite material in the
feed. A balance term was used to close the nitrogen
balance (see Eq. (6)) which related the nitrogen ‘‘losses’’
to the inorganic nitrogen concentration.
The relationship balancing the difference of the
nitrogen content of biomass (Nbac=0.00625 kmol
NkgCOD�1) and the nitrogen content of particulate
composites (NC=0.002 kmolNkgCOD�1) reads as
follows:
v16;j ¼ �X
i¼19;23�29
Nivi;j ; (6)
where Ni is the nitrogen content (kmol NkgCOD�1) of
the component i and n16,j is the nitrogen stoichiometric
coefficient for the respective biochemical process j
(processes j=12–19).
3.3. Benchmarking
The implementation developed was compared with a
‘benchmark implementation’, an independently devel-
oped implementation of the same model concept in a
different simulation system. The aim of this testing
procedure was to prove the absence of numerical
discrepancies. The ‘benchmark implementation’ was
developed in AQUASIM 2.0, a different modelling
platform, by the taskgroup for model testing using the
parameter set in Batstone et al. (2002).
The ‘benchmark simulations’ were performed to
verify the implementation’s numerical accuracy, under
constant as well as under varying input conditions.
Operational conditions, reactor features, input data and
parameter sets used were precisely the same. Solving
algorithms applied in both simulation systems are
similar: ODE 15 s, a variable-step, variable-order solver
based on the numerical differentiation formulas, is used
in MATLAB/Simulink to simulate the implementation
ARTICLE IN PRESS
Integration ofdynamic equations
Initial estimation ofparameter set
Comparison withmeasured data
Manual adjustment of parameter set
Optimisedparameter set Optimal fit
Siegrist, 2002 [1] Lai, 2001 [18]
etc.Romli, 1993 Ramsey, 1997 [17]
ADM1 [2]
Lab-scale experiments
yes
no
Model Validation
ODE solver
Fig. 2. Estimation procedure to identify most sensitive parameters.
F. Blumensaat, J. Keller / Water Research 39 (2005) 171–183 177
generated in this study, whereas the solver used in
AQUASIM 2.0 (‘benchmark implementation’) is based
on a implicit variable-step, variable-order integration
technique known as GEAR’s method.
As a result, a complete verification of the computation
procedure and the model code was expected.
3.4. Parameter estimation
Initial parameter values used have been derived from
a variety of sources (Siegrist et al., 2002; Batstone, 2000;
Costello, 1989; Romli, 1993; Ramsey, 1997; Lai, 2001)
including the STR (Batstone et al., 2002), and other
adequate literature (Stumm and Morgan, 1996; Lide
2001). An iterative procedure (see Fig. 2) was applied to
optimise the a priori as most sensitive identified
parameters used in this model.
Stoichiometric coefficients implied in the rate equation
matrix have been taken from the STR without any
changes. Carbon and nitrogen contents of the system’s
components were recalculated after the implementation
of balance terms into the rate equation matrix in order
to close carbon and nitrogen balance.
Physicochemical parameters, such as equilibrium coef-
ficients and constants, were principally taken from the
STR except for a limited number of physico-chemical
coefficients found in Stumm and Morgan (1996).
Most of the kinetic parameters were considered fixed
since they are generally known to have limited
variability in anaerobic systems (Batstone, 2000). Many
kinetic parameters given in the STR (Batstone et al.,
2002) were reviewed and compared with values found in
related reports (Batstone, 2000; Costello, 1989; Romli,
1993; Ramsey, 1997). Values with a low sensitivity were
preferably taken from these studies without any changes
as this reduces the degrees of freedom and therefore, the
difficulty to estimate the comprehensive parameter set.
Within the given small-scale system, it was considered
reasonable to assign fixed values to physical parameters
such as kLa and total gas pressure in the headspace.
Thus, identical kLa values for all three gases and a
constant headspace gas pressure were assumed.
Ultimately, eight parameters were identified based on
the sensitivity analysis given in the STR (Batstone et al.,
2002) as having the greatest impact upon the model
outputs, and were estimated in comparison to experi-
ARTICLE IN PRESS
Table 2
Initial and estimated parameter values for the parameters that have been optimised manually including literature source of initial
parameter values
Parameter Name Initial value
(meso/thermo)
Estimated value
(meso/thermo)
Unit Source
kdis;35=55 Disintegration constant 0.5/1.0 1.0/0.5 d�1 Batstone et al., (2002)
km;ac Maximum uptake rate for
acetate utilisers
8/16 9/25 Kg
CODkgCOD�1 d�1Batstone et al., (2002)
km;pro Maximum uptake rate for
propionate utilisers
13/20 9/16 Kg
CODkgCOD�1 d�1Batstone et al., (2002)
kS;ac Half-saturation constant for
acetate utilisers
0.15/0.3 0.15/0.4 kg CODm�3 Batstone et al., (2002)
kS;pro Half-saturation constant for
propionate utilisers
0.1/0.3 0.2/0.4 kg CODm�3 Batstone et al., (2002)
kS;H2Half-saturation constant for
hydrogen utilisers
7� 10�6/
5� 10�57� 10�6/
1� 10�5kg CODm�3 Batstone et al., (2002)
kI;H2 ;pro Hydrogen inhibition
constant for propionate
utilisers
3.5� 10�6/
1� 10�53.5� 10�5/
7� 10�6kg COD Batstone et al., (2002)
kI;NH3Ammonia inhibition
constant for acetate utilisers
0.0018/0.011 0.0011/0.011 kmolm�3 Batstone et al., (2002)
F. Blumensaat, J. Keller / Water Research 39 (2005) 171–183178
mental lab-scale data. Data sets were provided by Siegrist
et al. (2002) who performed extensive trials using
single-stage and two-stage digesters in small-scale
(Vmesophil=28 l single-stage and Vthermophil=16 l/
Vmesophil=28 l two-stage respectively).
An iterative, heuristic method (illustrated in Fig. 2)
was applied to estimate the parameters for the thermo-
philic and mesophilic process stage manually. Original
and optimised values for the selected parameters are
given in Table 2.
3.5. Pilot-scale experiments
A set of data for the purpose of model validation was
generated during pilot-scale experiments (Advanced
Wastewater Management Centre and Gold Coast City
Council, 2001). The set-up of the two-stage digestion pilot
plant generally corresponds to the scheme outlined above.
Reactors were completely mixed and had an operating
volume of 160 l for the thermophilic and 800 l for the
mesophilic process stage, respectively. The pilot plant
facility was partially controlled and monitored through a
process logic controller (PLC) connected to a computer.
Temperatures were so kept constant (55 1C for the
thermophilic and 35 1C for mesophilic reactor); biogas
production and operating temperature were continuously
monitored and recorded. Sludge samples were taken
regularly to record quality parameter of sludge and
biogas (VSS, TSS, COD, SCOD, amino acids, acetate,
propionate, TKN, biogas composition, pH). The two-
stage system was semi-continuously fed with 40 l day�1 of
a mixture of primary and secondary sewage sludge from a
BNR (incl. nitrification/denitrification) treatment plant.
This corresponds to a HRT of 4 days for the thermophilic
and 20 days for the mesophilic reactor.
4. Results and discussion
4.1. Benchmarking
Correct benchmarking results were achieved using
steady-state data as well as varying input data, as
exemplary shown in Fig. 3. Both model implementa-
tions, the Simulink implementation and the ‘benchmark
implementation’, were initialised using output data from
steady-state simulations before simulating varying input.
The benchmark simulation with varying input was
performed to verify the model’s numerical accuracy
under dynamic conditions. Similarly to the constant
input simulations, the Simulink model outputs and the
outputs of the ‘benchmark implementation’ do not differ
significantly. The low mean relative error in an order of
magnitude of 0.01% indicates the numerical preciseness
in generating the output for both model implementa-
tions.
4.2. Model calibration
The model has been calibrated for a two-stage
thermophilic/mesophilic process configuration and mu-
nicipal sewage sludge with typical characteristics as
substrate input. Experimental results of Siegrist et al.
(2002) were used to assist the model calibration. Fig. 4
ARTICLE IN PRESS
Fig. 3. Benchmarking results for sugars, amino acids and LCFA including error plot for varying input.
0 10 20 30 40 50 600
0.02
0.04
0.06
0.08
[m3 d
- 1]
Time [days]
Biogas simulatedBiogas measured
0 10 20 30 40 50 600
20
40
60
80
[%]
Time [days]
Methane measuredMethane simulatedCarbon Dioxide simulatedCarbon Dioxide measuredHydrogen simulatedHydrogen measured
first load step second load step
Fig. 4. Biogas production and biogas composition in comparison with experimental data for the thermophilic stage of the two-stage
process after the parameter optimisation.
F. Blumensaat, J. Keller / Water Research 39 (2005) 171–183 179
shows measured and simulated biogas production for
the thermophilic process stage applying the optimised
parameter set.
Both the biogas volume and composition was predicted
accurately for low (10 kg COD (m3d)�1 at HRT=4d)
and medium load (21 kg COD (m3d)�1 at HRT=2d)
ARTICLE IN PRESS
0 10 20 30 40 50 600
0.5
1
1.5
2
2.5
[kg
CO
D m
-3]
Time [days]
Propionate simulatedPropionate measured
0 10 20 30 40 50 600
0.2
0.4
0.6
0.8
1
1.2
1.4
[kg
CO
D m
-3]
Time [days]
Acetate simulatedAcetate measured
Fig. 5. VFA production in comparison with experimental data for the thermophilic stage of the two-stage process after the parameter
optimisation.
F. Blumensaat, J. Keller / Water Research 39 (2005) 171–183180
conditions. The model predicts the decline of the biogas
production as a response of the load increase, as well as
the adaptation of the system after the step changes.
However, at extremely high loading (HRT=1.33 d),
the accuracy of the model prediction decreased. It can be
seen that the model simulates an overload situation, the
biogas production decreases, and the methane propor-
tion in the biogas is reduced. Also, the total COD
removal decreases due to the accumulation of
hydrolysed COD (data are not shown). The VFA
production in the thermophilic reactor, particularly
acetate (lower chart of Fig. 5), also indicates the
difficulty to accurately predict the behaviour under high
loading conditions. It was complicated to further
optimise the parameter set, and a complete match of
simulated and experimental data for all loading condi-
tions could not be obtained. The lab-scale system on the
other hand, responds prompt with a drop in the biogas
production but eventually recovers despite the high
volumetric loading (31.5 kg COD (m3 d)�1).
Initial and estimated parameter values for the a priori
as most sensitive identified parameters are given in Table
2. Many parameters, principally those with low sensi-
tivity on model outputs, have been applied without any
optimisation. Hydrolysis constants for carbohydrates,
proteins and lipids are also among those parameters that
have not been optimised. Even though these parameters
are assumed to be highly sensitive for the degradation of
heterogeneous substrate, the influence is not as decisive
for homogenous substrate, such as the one that was
considered in this study (Batstone et al., 2002). Further
changes to hydrolysis constants or some other, not as
sensitive identified parameters may result in an im-
proved correspondence between model outputs and
experimental data.
4.3. Model validation
To assess the quality of the optimised parameter sets
and their applicability in process models, a validation
study was undertaken. The model outputs were com-
pared with measured data from a two-stage digestion
process in pilot scale. The process was simulated
applying the same implementation as described above
and without changing the previously optimised para-
meter set. The principle of the pilot-scale process is
equivalent to the process scheme that has been outlined
above.
Steady-state conditions were reached after a start-up
phase with rather unstable operation. After stable
functioning had been achieved and data could be seen
as representative the feed regime was changed from
constant to varying input to observe the dynamic
process behaviour.
The comparison of model outputs and experimental
data for the VFA production is exemplary shown in
ARTICLE IN PRESS
0
0.1
0.2
0.3
0.4
[m3 d
-1]
0 20 40 60 80 100 120 1400
0.05
0.1
0.15
0.2
0.25
[m3 d
-1]
Time [days]
0 20 40 60 80 100 120 140
Time [days]
Biogas simulatedBiogas measured
Biogas simulatedBiogas measured
Fig. 7. Validation results with data from pilot-scale experiments for the biogas production in the thermophilic reactor (upper chart)
and the mesophilic reactor (lower chart).
0 50 100 150
0 50 100 150
0
200
400
600
800
1000
0
200
400
600
800
1000
[mg
l -1]
[mg
l -1]
[mg
l -1]
[mg
l -1]
Time [days]
0 50 100 150
Time [days]
Propionate simulatedPropionate measured
Propionate simulatedPropionate measured
Acetate simulatedAcetate measured
0
10
20
30
40
50
60
70
Time [days]
0 50 100 150
Time [days]
0
10
20
30
40
50
60Acetate simulatedAcetate measured
Fig. 6. Validation results with data from pilot-scale experiments for the VFA production in the thermophilic reactor (upper charts)
and the mesophilic reactor (lower charts).
F. Blumensaat, J. Keller / Water Research 39 (2005) 171–183 181
Fig. 6. Particularly, the VFA production in the first
process stage (upper charts in Fig. 6) is well predicted by
the model though smoothed a little.
More detailed data for the mesophilic process stage
would reduce the influence of outliers and hence, give a
better idea of the curve’s fit. Nevertheless, it can be seen
ARTICLE IN PRESSF. Blumensaat, J. Keller / Water Research 39 (2005) 171–183182
that the concentration level of model outputs and
analytical data correspond well.
Some deviations in predicting the biogas production
and quality have been found (see Fig. 7). The differences
can be explained with the non-optimisation of several
parameters, for instance the application of identical and
non-optimised gas transfer coefficients. In fact, gas
transfer coefficients may differ in reality and the
dependence on the specific reactor configuration applied
has been neglected.
5. Conclusions
Several modifications to the original ADM1 model
concept were necessary to allow an application for the
pilot-scale digestion process.
Verification methods, such as consistency checks and
model benchmarking, have been applied to ensure a
completely verified and hence mathematically faultless
implementation of the ADM1 for a thermophilic/
mesophilic digestion process. It became clear that the
current understanding of modelling anaerobic digestion
processes is insufficient, reflected in the necessity to
implement balance terms (Eqs. (4), (5) and (6)). These
account for the inorganic carbon or nitrogen released
from biomass decay, and the implementation of these
‘add-ins’ evidently led to the minimisation of discre-
pancies in the carbon balance and to the closure in the
nitrogen balance. More fundamental research is re-
quired (a) to clarify the disposition of inorganic carbon
and nitrogen from decaying biomass, (b) to close the
carbon balance in the model concept completely and (c)
to verify the application of the implemented balance
terms. The implementation of additional balance terms
for other processes, such as the disintegration of
composite material and the hydrolysis of lipids, should
also be examined when applying the ADM1 in future
studies.
Even though the benchmarking process shed no
information on the ability of the model to predict the
performance of an anaerobic digester, it proved to be a
useful procedure to verify the numerical accuracy of the
model implementation.
In spite of the high-number biochemical parameters in
the model (37 for each process stage) it was possible to
obtain a good fit for most of the important process
parameters. Even though the curve fitting resulted in a
good agreement, principally, a parameter optimisation
for a structural model with such a high number of
parameters requires proof with data from several
dynamic experiments. This becomes evident in case the
model is further adapted, particularly for the parameters
that describe volatile acid production (hydrolysis con-
stants).
The validation results underline the good parameter
estimation and confirm the ability of the model to
adequately predict the behaviour of this particular
digestion process. However, a further assessment of
the model performance it is recommended. Additional
data representing a well-defined dynamic process beha-
viour should be used to confirm a dynamic model
validation. A further validation study is valuable to
show how the model is capable of simulating overload
situations or how the model predicts the degradation
process for different substrates under similar conditions.
References
Advanced Wastewater Management Centre and Gold City
Council, 2001. AWTT Project—Progress Reports 1 and 2
(unpublished).
Anderson, G.K., Kasapgil, B., Ince, O., 1994. Microbiological
study of two stage anaerobic digestion start-up. Water Res
28, 2383–2392.
Angelidaki, I., Ellegaard, L., Ahring, B.K., 1999. A compre-
hensive model of anaerobic bioconversion of complex
substrates to biogas. Biotechnol. Bioeng. 63 (3), 363–372.
Batstone, D. J., 2000. High-rate anaerobic treatment of
complex wastewater. Ph.D. Thesis, University of Queens-
land, Brisbane.
Batstone, D.J., Keller, J., Angelidaki, I., Kalyuzhnyi, S.V.,
Pavlostathis, S.G., Rozzi, A., Sanders, W.T.M., Siegrist, H.,
Vavilin, V.A., 2002. Anaerobic digestion model no.1.
Scientific and Technical Report 9, International Water
Association (IWA), London.
Cohen, A., Zoetemeyer, R.J., van Deusen, A., Andel, J.G.,
1979. Anaerobic digestion of glucose with separated acid
production and methane fermentation. Water Res 13,
571–580.
Cohen, A., Breure, A.M., van Andel, J.G., van Deusen, A.,
1980. Influence of phase separation on the anaerobic
digestion of glucose. I: maximum COD-turnover rate during
continuous operation. Water Res 14, 1439.
Costello, D. J., 1989. Modelling, optimisation and control of
high-rate anaerobic reactors. Ph.D. Thesis, University of
Queensland, Brisbane.
Henze, M., Grady, C. P. L. J., Gujer, W., Marais, G. V. R.,
Matsuo, T., 1987. Activated sludge model no.1. Scientific
and Technical Report 2, IWAQ, London.
Lai, E., 2001. Ph.D. Thesis, University of Queensland,
Brisbane/Australia.
Lide, D., 2001. CRC Handbook of Chemistry and Physics,
Eighty two edn. CRC Press, Boca Raton, FL.
Oles, J., Dichtl, N., Niehoff, H.-H., 1997. Full-scale experience
of two-stage thermophilic/mesophilic sludge digestion.
Water Sci. Technol. 36, 449–456.
Pohland, F.G., Ghosh, S., 1971. Developments in anaerobic
stabilisation of organic wastes: the two phase concept.
Environ. Lett. 1, 255–266.
Ramsey, I., 1997. Modelling and control of high-rate anaerobic
wastewater treatment systems. Ph.D. Thesis, University of
Queensland, Brisbane/Australia.
ARTICLE IN PRESSF. Blumensaat, J. Keller / Water Research 39 (2005) 171–183 183
Romli, M., 1993. Modelling and verification of a
two-stage high-rate anaerobic wastewater treatment system.
Ph.D. Thesis, University of Queensland, Brisbane/
Australia.
Siegrist, H., Renggli, D., Gujer, W., 1993. Mathematical
modelling of anaerobic mesophilic sewage sludge treatment.
Water Sci. Technol. 27, 25–36.
Siegrist, H., Vogt, D., Garcia-Heras, J.L., Gujer, W., 2002.
Mathematical model for meso- and thermophilic anaerobic
sewage sludge digestion. Environ. Sci. Tech. 36 (5),
1113–1123.
Stumm, W., Morgan, J.J., 1996. Aquatic chemistry: Chemical
Equilibria and Rates in Natural Waters, third ed. Wiley,
New York.
Vavilin, V.A., Vasiliev, V.B., Ponomarev, A.V., Rytow,
S.V., 1994. Simulation model ‘methane’ as a tool for
effective biogas production during anaerobic conversion of
complex organic matter. Bioresource Technol. 48 (1), 1–8.