Applied Energy 142 (2015) 352–360

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Parameter estimation and long-term process simulation of a biogas

reactor operated under trace elements limitation
Manfred Lübken a,⇑, Konrad Koch b, Tito Gehring a, Harald Horn c, Marc Wichern a
a
Institute of Urban Water Management and Environmental Engineering, Ruhr-Universität Bochum, Universitätsstraße 150, 44780 Bochum, Germany
b
Chair of Urban Water Systems Engineering, Technische Universität München, Am Coulombwall 8, 85748 Garching, Germany
c
Chair of Water Chemistry and Water Technology, Karlsruhe Institute of Technology, Engler-Bunte-Ring 1, 76131 Karlsruhe, Germany

h i g h l i g h t s

 Estimation of ADM1 parameter uncertainty by nonlinear, correlated parameter analysis.

 Unbounded confidence regions were obtained for single hydrolysis rate constants.
 ADM1 carbohydrates were divided into a slowly and readily degradable part.
 Bioavailability of trace metals explained discrepancies between modeled and measured data.

a r t i c l e i n f o a b s t r a c t

Article history: The Anaerobic Digestion Model No. 1 (ADM1) was modified to describe the long-term process stability of
Received 10 September 2014 a two-stage agricultural biogas system operated for 494 days with a mono-substrate. The ADM1 model
Received in revised form 2 January 2015 fraction for carbohydrates was divided into a slowly and readily degradable part. Significant different
Accepted 4 January 2015
hydrolysis rate constants were found for proteins and single fractions of carbohydrates in batch experi-
Available online 19 January 2015
ments. Degradation of starch, xylan (hemicellulose), cellulose and zein (protein) were modeled with first
order hydrolysis rate coefficients of 1.20 d1, 0.70 d1, 0.18 d1 and 0.30 d1, respectively. While the
Keywords:
hydrolysis rate coefficients found in batch experiments could be used for predicting continuous process
Mathematical modeling
ADM1
data, the statistically calculated confidence regions (nonlinear parameter estimation) showed that the
Parameter estimation upper limits were unbounded. Single discrepancies between measured and modeled process data of
Parameter uncertainty the two-stage pilot system could be explained by the lack of bioavailability of trace elements. Addition
Biogas technology of iron, as Fe(III)Cl3, allowed stable process conditions for an organic loading rate (OLR) up to 2.5 gVS
L1 d1. Additional supplement of trace elements was necessary for process operation at OLRs above
2.5 gVS L1 d1.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction process simulation. While the basic objective of BMP test is to esti-
mate biodegradability of selected substrates, this method is
The ability to produce bioenergy from a broad range of organic beyond that increasingly used to determine the apparent first
substances has stimulated innovations in biogas technology during order hydrolysis rate constant [5,6]. However, the validation of
the last years. The fast progress in both technology development parameters obtained in batch assays on continuous systems is
and substrate usage for anaerobic digestion leads to a demand of often neglected [7].
a general mathematical model, which can be used by scientists The Anaerobic Digestion Model No 1 (ADM1) [8] has been
and practitioners to understand and optimize system’s behavior established in recent years as a common platform for modeling
under dynamic conditions. When the methanization of new sub- anaerobic digestion systems. The model has been used, besides
strates is evaluated, such as microalgae [1,2] or lignocellulosic bio- generally to predict digester performance, to assess reactor failure
mass [3,4], biomechanical methane potential (BMP) tests in batch under external stress conditions [9–11]. One of the strengths of
mode or continuously operated lab-scale studies are commonly structured mathematical models is seen in their ability to assist
performed; and this often without complemental mathematical in controlling the complex and often unstable anaerobic digestion
process [12–14]. This, however, depends on the availability of
⇑ Corresponding author. Tel.: +49 234 32 23114; fax: +49 234 32 14503. reliable kinetic model parameters, at least for the key processes.
E-mail address: [email protected] (M. Lübken). Intensive studies on kinetic parameters estimation with respect

http://dx.doi.org/10.1016/j.apenergy.2015.01.014
0306-2619/Ó 2015 Elsevier Ltd. All rights reserved.
 M. Lübken et al. / Applied Energy 142 (2015) 352–360 353

to agricultural feedstock have been done so far [15–18]. However, 2. Methods

as Donoso-Bravo et al. [19] have pointed out the determination of
parameter variability and uncertainty is often neglected in model 2.1. Chemical analyses
calibration studies. Within this context, the study in [20] has dem-
onstrated that estimation of parameter uncertainty should not be Analytical methods were based on German Standard Methods
done by linear, uncorrelated analysis. Confidence intervals of for the examination of water, wastewater and sludge [26]. Volatile
parameter estimates will only provide approximations of the mod- fatty acids (acetic acid, propionic acid, i-butyric acid, n-butyric acid,
el’s uncertainty. In general, joint confidence areas for two parame- i-valeric acid, n-valeric acid and hexanioc acid) were quantified by
ters, such as the Monod half saturation constants (KS) and the means of a GC/FID (CE Instruments, HRGC 5300). Analysis of the
maximum specific substrate uptake rate constants (km), allow an single organic fractions of the lignocellulosic biomass was similar
estimate about the true parameter value (expressed, e.g., with a as described in [27]. At this, an extended Weender analysis [28] is
probability of 95%). Only few ADM1 simulation studies in which performed to measure organic matter more detailed in terms of
confidence regions for optimized parameters have been deter- raw lipid (RL), raw protein (RP), raw fiber (RF) and N-free extract
mined can be found [20–22,7]. (NfE) as well as in terms of neutral detergent fiber (NDF), acid deter-
Besides parameter uncertainty, there is a lack of experience in gent fiber (ADF) and acid detergent lignin (ADL). All those parame-
model applications for mono-substrates (without manure) fermen- ters are expressed as percentage of TS (% TS). Raw proteins can be
tation. Mono-substrate fermentation is generally considered as less further distinguished into amides, like for instance free amino acids,
stable in long-term operation. Sufficient supply of trace elements is acid amides, or peptides. The residue is consequently pure protein.
regarded as a limitation factor in steadily achieving high biogas The term ‘‘N-free extracts’’ refers to the soluble carbohydrate of the
production rates [23,24]. Especially at high organic loading rates feed and is the only component, which is not determined
the addition of trace elements can avoid process instabilities due analytically, but calculated by difference. Carbohydrates are further
to, e.g., ammonia inhibition [25]. Model calibration upon a system
which is operated with an unrecognized deficit of micronutrients
Table 1
could possibly lead to a wrong estimation of kinetic parameters. Mean substrate characteristics for the maize silage used in this study.
On the other hand, a calibrated model could contribute knowledge
Parameter Unit Maize silage
to the demand of micronutrients, when discrepancies between
model output and system behavior can clearly be explained by TS (%) 36.8
VS (% TS) 97.5
the lack of bioavailability of trace elements.
COD (g kg1) 461
Within this study, the ADM1 was modified to describe mono- Lactic acid (mg kg1) 30,000
fermentation of maize silage, the most commonly used energy crop Propionic acid (mg kg1) 400
for biogas production. Process data were obtained for a two stage Acetic acid (mg kg1) 4,900
pilot reactor consisting of a horizontal digester as the main process pH (–) 3.9
TKN (g kg1) 4.8
step and a second vertical tank reactor, which represents a typical NH4–N (g kg1) 0.78
layout for energy crops digester. Hydrolysis rate constants were RP (raw protein) (% TS) 8.0
estimated on both batch and continuous process data. A two- RF (raw fiber) (% TS) 17.7
parameter analysis method was applied to ADM1 key kinetic RL (raw lipid) (% TS) 2.6
Hemicellulose (% TS) 22.1
parameters, and the values and uncertainties were assessed by non-
Cellulose (% TS) 18.7
linear, correlated parameter estimation. An extended simulation Lignin (% TS) 3.2
period of 494 days was performed to enable insights into the N-free extract (NfE) (% TS) 69.2
model’s capability of simulating long-term process performance. C, H, O, N, S (% VS) 44.1, 6.0, 44.1, 1.3, 0.06

Electric
Biogas
control
analyzer
cabinet
(7)
(6)

(12)
(11) (5)
(5) (5)
Maize silage + (10) (10)
recycled biomass
(1)
(4)

(2)

(8) (9)
(10)
(3)
(3)

(1) Input (4) Gas dome (7) Gas meter (10) Valve
(2) Screw (5) Gas line (8) Paddle mixer (11) Filling level monitor
(3) Sampling points (6) Gas cooling device (9) Propeller mixer (12) Temperature measurement

Fig. 1. Schematic drawing of pilot plant and technical equipment. The horizontal digester was equipped with four sample points at the bottom and four samples points at the
upper part. Bulk level reached shortly into the gas domes to avoid gas mixing between the four compartments.
354 M. Lübken et al. / Applied Energy 142 (2015) 352–360

Table 2
Operational conditions for the investigation period. All parameters were calculated for the horizontal digester only.

Period (d) OLR (gVS L1 d1) HRT (d) Ygas (LN g1
VS )

Period 1: mono-fermentation of maize silage with recycle flow. Ratio maize silage to recycle flow for days 0–102: 1:2 v/v, for days 103–150: 1:8 v/v
(1) 0–16 1.64 62 716
(2) 17–48 2.05 49 703
(3) 49–59 2.33 57 660
(4) 60–102 2.91 46 678
(5) 103–120 3.50 13 571
(6) 121–150 3.33 13 605
Period 2: mono-fermentation of maize silage without recycle flow
(1) 0–15 0.50 800 896
(2) 16–35 1.00 400 627
(3) 36–66 1.50 267 712
(4) 67–148 1.90 200 742
(5) 149–206 2.40 160 660
(6) 207–294 2.50 126 674
(7) 295–315 3.10 104 680
(8) 316–344 3.60 89 663

subdivided into starch (RF + NfE  NDF), cellulose (ADF  ADL) and was operated for 344 days without a recycle flow. The hydraulic
lignin (ADL). Hemicellulose is approximately analyzed by van Soest retention time (HRT) in the first period varied between 13 and
and Wine fractions [29] as the difference between neutral detergent 62 days. HRT in the second period varied between 89 and 800 days.
fiber (NDF) and acid detergent fiber (ADF). As no recycle flow was applied for the second period, HRT
depended only on the daily input of fresh matter (FM) and the
HRT of 800 days corresponded to a low OLR. Gas production of
2.2. Pilot plant operation the second reactor resulted from the non-digested substrate from
the horizontal fermenter. As the first reactor achieved a high gas
The pilot plant (Fig. 1) was a combination of two digesters in production rate, the second fermenter showed only minor dynam-
series: a horizontal fermenter as the main process step and a ver- ics. Experimental data are provided in this study only for the hor-
tical fermenter used for recirculation of active digestate to the first izontal reactor as the second reactor was predominantly used as
reactor. Both fermenters were built as cylindrical steel tanks and storage tank and to provide active biomass for the first stage
operated constantly at 38 ± 1 °C. The horizontal fermenter had a through the recycle stream.
liquid volume of 240 L with an additional reserved headspace
volume of 60 L. The volume of the vertical fermenter was 750 L
2.3. Lab reactor system
partitioned into 500 L liquid phase and 250 L headspace. The
proportions chosen at reduced scale complied with a full-scale
Gas tight glass bottles were used as anaerobic batch fermenters.
biogas plant operated in Rosenheim, Germany. Both gas measure-
Gas production was measured by milligascounter (Ritter, Bochum,
ment and biomass sampling were possible at four points along the
Germany). The milligascounter were connected to a data logger
horizontal fermenter length.
and a PC equipped with LabVIEW. Each batch experiment was per-
Total biogas production of the pilot plant was measured by
formed in quadruplicate. The same amount of blanks were
drum chamber gas meters (TG5, Ritter, Bochum, Germany) after
installed. The gas production of the blanks was subtracted from
passing through condensate separators and a gas-cooling device.
reactors fed during batch experiments. Experiments and determi-
Values for biogas production were normalized. Methane and car-
nation of first order hydrolysis rate coefficients were performed
bon dioxide were quantified by means of the infrared two-beam
similar to [30,31]. Batch reactors were operated separately with
compensation method with pressure compensation (measuring
starch, xylan, zein and cellulose. All chemicals were purchased
error as specified: ±2%). Oxygen, hydrogen and hydrogen sulfide
from commercial supplier (VWR, Darmstadt, Germany).
were measured by electrochemical sensors (measuring error as
specified: ±3%) (Awite, Langenbach, Germany).
Both reactors were inoculated with fermenter content of a full-
scale biogas plant next to Munich, Germany. The full-scale plant
was operated with maize silage in mono-fermentation without
addition of manure. The pilot system was operated quasi continu-
ously and was fed manually once a day with maize silage which
was taken from a farm next to Freising, Germany. The maize silage
used (zea mays) was portioned, stored at 18 °C, gently defrosted
at 4 °C before used and regularly analyzed for total solids (TS), vol-
atile solids (VS), chemical oxygen demand (COD), lactic acid, propi-
onic acid, acetic acid, pH, Kjeldahl nitrogen (TKN), ammonia
nitrogen (NH4–N), total carbon (TC), total organic carbon (TOC)
and the Weender and van Soest fractions. Mean substrate charac-
teristics are shown in Table 1.
The investigation was separated in two periods. In period 1, the
system was operated for 150 days at different organic loading rates
(OLRs), ranging from 1.65 to 3.33 gVS L1 d1 (see Table 2, loading
is calculated only for the horizontal fermenter). A fixed amount
of fermenter content from the vertical fermenter was recycled Fig. 2. Degradation of single particulate components of maize silage over time. First
and mixed with the input maize silage. In period 2, the system order models (dotted lines) were used to reproduce data.
 M. Lübken et al. / Applied Energy 142 (2015) 352–360 355

1200
550

Specific gas yield (LN kgVS-1)

(1) (2) (3) (4) (5) (6)
(1) (2) (3) (4) (5) (6)
Gas production (LNd-1) 1000
500

450 800

400 600

350 400

300 200
measured modeled (a) measured modeled (b)
250 0
0 30 60 90 120 150 0 30 60 90 120 150
Days (d) Days (d)

70 70
(1) (2) (3) (4) (5) (6) (1) (2) (3) (4) (5) (6)

Carbon dioxide (Vol. %)

65 65
Methane (Vol. %)

60 60

55 55

50 50

45 45

40 40

35 measured modeled 35 measured modeled

(c) (d)
30 30
0 30 60 90 120 150 0 30 60 90 120 150
Days (d) Days (d)
100 10.0 8.0
(1) (2) (3) (4) (5) (6) (1) (2) (3) (4) (5) (6)

80 9.0
Hydrogen (ppm.)

6.0

NH4-N (g kg-1)
8.0
60
pH (-)

7.0 4.0
40
6.0
2.0
20 measured pH modeled pH
5.0
measured modeled (e) measured NH4-N modeled NH4-N (f)
0 4.0 0.0
0 30 60 90 120 150 0 30 60 90 120 150
Days (d) Days (d)

5000 1000
(1) (2) (3) (4) (5) (6) (1) (2) (3) (4) (5) (6)
Propionate (mg kg-1)

4000 800
Acetate (mg kg-1)

3000 600
measured modeled
measured modeled
2000 400

1000 200
(g) (h)
0 0
0 30 60 90 120 150 0 30 60 90 120 150
Days (d) Days (d)

Fig. 3. Simulation results for the first operation period used for parameter estimation: comparison of measured and modeled parameters for the gas phase. OLRs applied are
indicated within the figures: (1) 1.64, (2) 2.05, (3) 2.33, (4) 2.91, (5) 3.5 and (6) 3.33. Unit for OLR is gVS L1 d1.

2.4. Model set-up The experimental data obtained for the horizontal fermenter
were modeled with two CSTRs in series, in which the first and
The ADM1 implementation in Matlab7.1/Simba6.5 was used. the last two compartments of the horizontal digester were com-
The suggested values of the ADM1 Scientific and Technical Report bined to one CSTR each. A virtual backflow was computed to real-
[8] for mesophilic solids were set as initial values for kinetic and ize back-mixing between both CSTRs and to give best fit between
stoichiometric parameters. measured and modeled tracer results (pulse injection of LiCl was
356 M. Lübken et al. / Applied Energy 142 (2015) 352–360

performed prior to the simulation study in order to calibrate the q1a ¼ khyd ch s  X ch s ¼ 0:18  X ch s ðkgCOD m3 d1 Þ ð1Þ
hydraulic model, data not shown).
Substrate composition in terms of particulate carbohydrates,
q1b ¼ khyd ch r  X ch r ¼ 0:70  X ch r ðkgCOD m3 d1 Þ ð2Þ
proteins and lipids was defined within the input stream, and a sim- q2 ¼ khyd pr  X pr ¼ 0:30  X pr ðkgCOD m3 d1 Þ ð3Þ
ilar influent characterization approach as described in [27] was
performed. The conversion of the analytical fractions, which are where q1a is the process rate of the slowly degradable carbohydrate
analyzed on a volatile solids (VS) basis, into COD units was per- fraction (kgCOD m3 d1), q1b the process rate of the readily degrad-
formed using COD conversion factors for carbohydrates able carbohydrate fraction (kgCOD m3 d1), q2 the process rate of
(1.19 gCOD gVS1, calculated for the elemental formula C6H10O5), the protein fraction (kgCOD m3 d1), khyd_ch_s the hydrolysis rate
proteins (2.18 gCOD gVS1, calculated for the elemental formula of found for cellulose (d1), khyd_ch_r the hydrolysis rate found for
the dominant maize protein zein C317.02H505.90O95.27N80.12S1.68; xylan (d1), khyd_pr the hydrolysis rate found for zein (d1), Xch_s
the chemical composition of zein was calculated in this study the ADM1 state variable for slowly degradable carbohydrates
based on the elemental analysis given in [32]) and lipids (kgCOD m3), Xch_r the ADM1 state variable for readily degradable
(2.03 gCOD gVS1, calculated for the elemental formula C8H6O2). carbohydrates (kgCOD m3) and Xpr the ADM1 state variable for pro-
The soluble COD of the maize silage was mainly composed of lactic teins (kgCOD m3). Although Xch_r is composed of both hemicellulose
acid. As the anaerobic degradation of lactate follows similar the and starch it was decided to model hydrolysis of this fraction only
stoichiometry of glucose, the influent lactic acid was allocated to with the hydrolysis rate constant found for xylan. This reduces
the model fraction for monosaccharide (Ssu). Measured acetic and model complexity and starch can generally be assumed as being
propionic acids were converted to a COD basis and allocated fully converted in biogas reactors. The concept of considering sub-
directly to the model fractions for acetate (Sac) and propionate strate fractions with different hydrolysis rates is not new. It was,
(Spro) respectively. e.g., likewise considered for activated sludge systems [37].
The estimated hydrolysis parameters were directly applied to
3. Results and discussions the continuous data set (Table 2, period 1). In order to obtain a best
relationship between modeled and measured data, the maximum
3.1. Parameter estimation uptake rate of acetate (km_ac) was adjusted to 9.5 kgCOD kgCOD1 d1,
the free ammonia inhibition constant for acetate uptake (KI_NH3)
Hydrolysis is generally assumed as the rate limiting step in was adjusted to 0.0055 (kmoleN m3), the maximum uptake rate
anaerobic degradation of lignocellulosic biomass [33]. It is usually of propionate (km_pro) was adjusted to 5.6 kgCOD kgCOD1 d1, the
described as a first order process. The gas rate in a batch experi- half saturation coefficient of propionate (KS_pro) was adjusted to
ment follows the hydrolysis rate when no accumulation of VFA 0.12 kgCOD m3, the maximum uptake rate of hydrogen (km_h2)
and no inhibition of the following process steps are ensured. was adjusted to 28 kgCOD kgCOD1 d1 and the decay rate of bacteria
Although several studies have been performed to estimate the (kdec) was adjusted to 0.10 d1. The estimation of the single kinetic
hydrolysis rate constant only from batch tests data [30,34,35], Bat- parameters H was done using the residual sum of squares (RSS)
stone et al. [7] illustrated that the hydrolysis rate determined from objective function J(H):
batch tests was unsuitable to predict continuous process data
X
N
obtained from full-scale digesters. As maize silage, in contrast to, JðHÞ ¼ RSS ¼ ^i Þ2
ðyi  y ð4Þ
e.g., wastewater, is a well-defined mixture of single constituents, i¼1
batch tests were complementally performed to the continuous
experiments in order to obtain an estimation about different where N is the number of measurements, yi the measured data and
hydrolysis kinetics for single organic fractions. ^i the outputs predicted by the model.
y
Main organic components of maize silage are defined as starch, The simulation results are shown in Fig. 3.
hemicellulose, cellulose, protein, lipid and lignin. Lignin was Within the first operation period maize silage was mixed with
assumed not being hydrolysable and the proportion of lipid as recycle content from the storage tank. During the first 100 days
too small to have an influence on the overall hydrolysis rate of the gas production constantly rose corresponding to increasing
maize silage. The remaining components, with xylan as represen- OLRs (see Table 2), while the specific gas yield remained constant.
tative hemicellulose and zein as representative protein, were sep- The decline in biogas production at an OLR of 3.5 gVS L1 d1 was
arately treated in batch systems which were inoculated with the result of an increased recycle flow from the vertical reactor,
fermenter content of the pilot plant. Zein was chosen as represen- by which a proliferated amount of degradable material is displaced
tative protein. Zein is a class of prolamine protein and is the major
protein of maize (Zea mays). The results are shown in Fig. 2. 3.0
Process rates (kgCOD m-3 d-1)

Degradation of starch, xylan and zein was well described by 2.7

first order kinetics. The hydrolysis rate constants found were 2.4
1.20 d1 (starch), 0.70 d1 (xylan) and 0.30 d1 (zein). Degradation 2.1
of cellulose, however, did not follow first order kinetics. As no 1.8
explanation could be derived, first order kinetic was likewise 1.5
applied and the rate coefficient was chosen in the way that the last 1.2
data point was predicted. Hereby an overestimation of cellulose
0.9
degradation was avoided and the deduced dimension of the hydrolysis of readily degradable carbohydrates
0.6
respective coefficient of 0.18 d1 is in the range of published liter- uptake of acetate

ature [36]. 0.3

As significant different hydrolysis rate constants were found for 0.0
0 30 60 90 120 150
the three carbohydrate fractions, the ADM1 carbohydrate fraction
Days (d)
was split into a fraction of readily degradable and a fraction of
slowly degradable carbohydrates. The modified hydrolysis rates Fig. 4. Process rates: acetate uptake and hydrolysis of readily degradable
of ADM1 were calculated as follows: carbohydrates.
 M. Lübken et al. / Applied Energy 142 (2015) 352–360 357

Fig. 5. 95% Confidence regions on two adjusted ADM1 parameters jointly. Left: propionate related; right: acetate related. Black small dots represent parameter combinations
for which SSR is below the defined threshold. Parameters used for process simulation are indicated by arrows.

from the horizontal fermenter into the vertical storage tank. This beginning with an NH4–N content of 4.8 g kg1, which resulted
was reproduced by the model. The measured gas composition fluc- from the inoculum taken from a full-scale biogas plant during
tuated over a broader range compared to the simulation results. start-up. The full-scale plant fermented maize silage in mono-
Gas composition was not measured continuously for the first per- fermentation with an occasional addition of urea to avoid N-limi-
iod, and was conducted at varying time steps during the day. A tation. The NH4–N content in this study leveled off to 3 g kg1.
higher temporal distance of measurement to feeding activity gen- N-limitation was not observed during the whole investigation
erally leads to increased methane values. period.
At an OLR of 3.5 gVS L1 d1 a strong increase of both acetate
and propionate was observed. Recycled biomass from the vertical 3.2. Estimation of parameter uncertainty
tank was, as a counteractive measure, quadrupled by which a
further increase in VFA was avoided. This measure directed a Calculation of confidence regions for the optimized parameters
balance of the actual volumetric COD conversion rates for the was done according to Lobry et al. [38], whereby the sum of the
two processes hydrolysis (readily carbohydrates) and acetate squared residuals (SSR) is limited to the following threshold
uptake (Fig. 4).  
p
The pH remained relatively constant during the whole 150 days SSRðhÞ 6 SSRðh0 Þ 1 þ  Fa ð5Þ
n  p p;np
and it was not affected by the OLR. The system was operated in the
358 M. Lübken et al. / Applied Energy 142 (2015) 352–360

unrealistic high specific gas yield of >900 LN kgVS1. From day

200 on, the gas rate fluctuated strongly. This is explained by tech-
nical shortcomings of the gas measuring devices. The measured
hydrogen gas content was unsteady, but average values were in
the range of the simulation model.
The simulated acetate and propionate content remained con-
stant at a low level for the whole simulation period. However, mea-
surements revealed two distinctive peaks for both acetate and
propionate. The first increase was observed at a relatively low
OLR of 1.9 gVS L1 d1. Iron (as Fe(III)Cl3) was added to the fermen-
ter at day 100 (indicated in Fig. 7g and h), which resulted in an
immediate decline of acetate and propionate concentrations. Avail-
ability of trace elements was assumed as the limiting factor. The
observed benefit of iron addition can be explained by the fact that
sulfide is chemically bounded, by which the bioavailability of other
trace elements within the fermenter is subsequently increased. The
metal binding capacity of sulfide was assumed of being most
important for the present fermenter. Due to the addition of iron
the content of H2S, measured in the gas phase, decreased from
Fig. 6. Joint 95% confidence region on the hydrolysis rate coefficients relating to the
readily (khyd_ch_r) and slowly (khyd_ch_s) degradable carbohydrates. 150 to 20 ppm (data not shown). Iron was added in the following
fermentation progress to limit the H2S gas content to a threshold
of 30 ppm.
where SSR(h0) is the minimum value of SSR, p the number of param- A second increase of acetate and propionate was observed at an
eters within the parameter vector h, n the number of measuring OLR of 2.5 gVS L1 d1. Again, this was not reproduced by the
data and F the value for the F-distribution for a confidence interval model. Release of fermenter acidification was not achieved by an
of a with p and p–n degrees of freedom. intensified addition of iron. A solution of trace elements (composi-
The calculation of confidence regions for the parameters related tion was FeCl24H2O, CoCl26H2O, NiCl26H2O and NaMoO42H2O in
to acetate and propionate degradation were individually com- a concentration of 1.96, 0.23, 0.24 and 0.02 mg gVS_input1 respec-
puted, considering the measured data of each organic acid. Resid- tively) was added to the system from day 250 on (indicated in
uals were normally distributed. For each calculation three kinetic Fig. 7g and h). The effect was directly visible by means of declining
parameters were included, with F-value of 3.0491 and 3.2389 for VFA concentrations to the levels of the simulated values.
acetate and propionate degradation respectively; both for 95% con- The mono-fermentation of maize silage without recycling of
fidence regions. The joint 95% confidence regions are shown in biomass from the vertical tank evolved strongly different com-
Fig. 5. pared to the first period. However, it is not clear that a recycle flow
The number of possible parameter combinations within the has a general positive effect on digester stability, as the vertical
confidence regions is quite high due to the relatively low frequency storage tank was operated in the first period at significant lower
of organic acids measurement. This subject is addressed in the OLRs compared to the horizontal fermenter. The estimated kinetic
study of [20], where the maximum uptake rate and the half satura- parameters could generally be validated on the second data set, as
tion constant could be identified for an anaerobic sequencing batch the lack of biological available trace elements in the second period
system when two cycles with twenty measuring points per cycle explained the discrepancies between measured and modeled VFA
were used for analysis. The parameter values chosen for process content.
simulation are indicated within Fig. 5 by an arrow (SSR(h0)). Knowledge about kinetic parameters is crucial for process sim-
Identification of the hydrolysis rate coefficients on the ulation, and a mathematical model will then allow to identify the
continuous data set was not possible. The 95% confidence spaces most relevant characteristics of the biological processes. The basic
were calculated using the daily measured gas production rate tool for both system design and evaluation of suitable biomass
(N = 55, F-value = 2.7725). Fig. 6 shows the joint confidence regions feedstock is the biochemical methane potential (BMP) test. While
on the readily and slowly carbohydrate hydrolysis rate coefficients. BMP tests are generally performed at lab-scale, the results are
Only lower limits were detected. These were 0.30 d1 and 0.05 d1 often directly transferred to full-scale system layout. The hydroly-
for khyd_ch_r and khyd_ch_s respectively. The upper limits were sis process is assumed as rate limiting for lignocellulosic biomass,
unbounded, which is similar to the study described in [7], where and hence hydrolysis rate constants have been determined upon
no upper limit for the hydrolysis rate constant could be identified. BMP tests in numerous studies. However, the results in this study
The hydrolysis rate constants estimated from batch test data were have shown that in dynamic, continuous systems the hydrolysis
in this study within the lower space of the 95% confidence region rate constant is neither explicitly identifiable nor is the hydrolysis
(indicated by an arrow within Fig. 6). process the rate limiting step for lignocellulosic biomass fermenta-
tion at high OLRs. Model application, although being still an elab-
3.3. Model validation orate task, is recommended as a complementary tool to BMP
tests for design and control of biogas systems. However, model
The model was validated on a second, independent data set assisted controlling of such mono-fermentation systems does not
(Table 2, operation period 2). Measurements and simulation results only depend on reliable kinetic parameters. ADM1 extensions on
for this period are shown in Fig. 7. Both total and specific gas pro- solids precipitation, e.g. metal sulfide precipitation, and influence
duction rate were reproduced by the model over a period of of trace elements availability on microbial activity is likewise
344 days. The gas rate of the first 15 days of the second period essential. At least for the latter, extensive experience is available
was still influenced by the first period, although the feeding was from literature [23,24], but has not integrated into modeling con-
stopped between the two phases for 20 days. This resulted in an cepts so far.
 M. Lübken et al. / Applied Energy 142 (2015) 352–360 359

800 1500
(1) (2) (3) (4) (5) (6) (7) (8) (1) (2) (3) (4) (5) (6) (7) (8)

Specific gas yield (LN kgVS-1)

Gas production (LN d-1) 1200
600

(a) 900
400
600

200
300
measured modeled (b) measured modeled
0 0
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
Days (d) Days (d)

100 100
(1) (2) (3) (4) (5) (6) (7) (8) (1) (2) (3) (4) (5) (6) (7) (8)

Carbon dioxide (Vol.-%)

80 80
Methane (Vol. %)

60 60

40 40

20 20
measured modeled measured modeled
(c) (d)
0 0
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
Days (d) Days (d)

80 10.0
(1) (2) (3) (4) (5) (6) (7) (8) (1) (2) (3) (4) (5) (6) (7) (8)

9.0
60
Hydrogen (ppm.)

8.0
pH (-)

40 7.0

6.0
20 measured modeled
5.0
(e) measured modeled (f)
0 4.0
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
Days (d) Days (d)

4000 3000
(1) (2) (3) (4) (5) (6) (7) (8) (1) (2) (3) (4) (5) (6) (7) (8)

2500
Propionate (mg kg-1)
Acetate (mg kg-1)

3000
2000

2000 1500 measured modeled

measured modeled
(g) 1000 (h) Dosage of Fe(III)Cl3 Dosage of trace elements
1000 Dosage of trace elements
Dosage of Fe(III)Cl3
500

0 0
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
Days (d) Days (d)

Fig. 7. Simulation results for the second operation period used for model validation: comparison of measured and modeled parameters for the gas phase. OLRs applied are
indicated within the figures: (1) 0.5, (2) 1.0, (3) 1.5, (4) 1.9, (5) 2.4, (6) 2.5, (7) 3.1 and (8) 3.6. Unit for OLR is gVS L1 d1.

4. Conclusions data revealed that the upper limits of the single hydrolysis rate
constants were unbounded. The model fraction for carbohydrates
ADM1 kinetic parameters have been estimated to simulate can be divided into a slowly and readily degradable part. The mod-
process performance of a two-stage pilot scale system over a ified ADM1 showed a well agreement with measurements. Single
period of 494 days. Different hydrolysis rate constants were found variations of predicted and measured VFA concentrations were
for proteins and single fractions of carbohydrates in batch experi- observed. The variations occurred when a lack of bioavailability
ments. Estimation of confidence regions using continuous process of trace elements was recognized.
360 M. Lübken et al. / Applied Energy 142 (2015) 352–360

References [20] Batstone DJ, Torrijos M, Ruiz C, Schmidt JE. Use of an anaerobic sequencing
batch reactor for parameter estimation in modelling of anaerobic digestion.
Water Sci Technol 2004;50:295–303.
[1] Mendez L, Mahdy A, Ballesteros M, González-Fernández C. Methane
[21] Batstone DJ, Pind PF, Angelidaki I. Kinetics of thermophilic, anaerobic
production of thermally pretreated Chlorella vulgaris and Scenedesmus sp.
oxidation of straight and branched chain butyrate and valerate. Biotechnol
biomass at increasing biomass loads. Appl Energy 2014;129:238–42.
Bioeng 2003;84:195–204.
[2] Zamalloa C, Boon N, Verstraete W. Anaerobic digestibility of Scenedesmus
[22] Kalfas H, Skiadas IV, Gavala HN, Stamatelatou K, Lyberatos G.
obliquus and Phaeodactylum tricornutum under mesophilic and thermophilic
Application of ADM1 for the simulation of anaerobic digestion of olive pulp
conditions. Appl Energy 2012;92:733–8.
under mesophilic and thermophilic conditions. Water Sci Technol 2006;54:
[3] Quintero M, Castro L, Ortiz C, Guzmán C, Escalante H. Enhancement of starting
149.
up anaerobic digestion of lignocellulosic substrate: fique’s bagasse as an
[23] Ortner M, Rachbauer L, Somitsch W, Fuchs W. Can bioavailability of
example. Bioresour Technol 2012;108:8–13.
trace nutrients be measured in anaerobic digestion? Appl Energy 2014;126:
[4] Kaparaju P, Serrano M, Angelidaki I. Optimization of biogas production from
190–8.
wheat straw stillage in UASB reactor. Appl Energy 2010;87:3779–83.
[24] Lebuhn M, Liu F, Heuwinkel H, Gronauer A. Biogas production from mono-
[5] Koch K, Helmreich B, Drewes JE. Co-digestion of food waste in municipal
digestion of maize silage-long-term process stability and requirements. Water
wastewater treatment plants: effect of different mixtures on methane yield
Sci Technol 2008;58:1645–51.
and hydrolysis rate constant. Appl Energy 2015;137:250–5.
[25] Browne JD, Allen E, Murphy JD. Assessing the variability in biomethane
[6] Zhen G, Lu X, Li YY, Zhao Y. Combined electrical-alkali pretreatment to increase
production from the organic fraction of municipal solid waste in batch and
the anaerobic hydrolysis rate of waste activated sludge during anaerobic
continuous operation. Appl Energy 2014;128:307–14.
digestion. Appl Energy 2014;128:93–102.
[26] DEV. Deutsche Einheitsverfahren zu Wasser-, Abwasser- und
[7] Batstone DJ, Tait S, Starrenburg D. Estimation of hydrolysis parameters in full-
Schlammuntersuchung. German standard methods for examination of water,
scale anerobic digesters. Biotechnol Bioeng 2009;102:1513–20.
wastewater and sludge. Weinheim: Wiley-VCH; 2014.
[8] Batstone DJ, Keller J, Angelidaki I, Kalyuzhnyi S V, Pavlostathis SG, Rozzi A,
[27] Lübken M, Wichern M, Schlattmann M, Gronauer A, Horn H. Modelling the
et al. Anaerobic digestion model no. 1 (ADM1). London, UK: IWA Publishing;
energy balance of an anaerobic digester fed with cattle manure and renewable
2002.
energy crops. Water Res 2007;41:4085–96.
[9] Esposito G, Frunzo L, Panico A, Pirozzi F. Modelling the effect of the OLR and
[28] Naumann C, Bassler R, editors. Handbuch der landwirtschaftlichen Versuchs- und
OFMSW particle size on the performances of an anaerobic co-digestion reactor.
Untersuchungsmethodik: (Methodenbuch). 3. neu bea. Darmstadt: VDLUFA-
Process Biochem 2011;46:557–65.
Verlag; 2004.
[10] Thamsiriroj T, Nizami AS, Murphy JD. Why does mono-digestion of grass silage
[29] Van Soest PJ, Wine RH. Use of detergents in the analysis of fibrous feeds. IV.
fail in long term operation? Appl Energy 2012;95:64–76.
determination of plant cell-wall constituents. J Assoc Off Anal Chem
[11] Lübken M, Wichern M, Letsiou I, Kehl O, Bischof F, Horn H. Thermophilic
1967:50–9.
anaerobic digestion in compact systems: investigations by modern
[30] Gupta A, Chandra R, Subbarao PMV, Vijay VK. Kinetics of batch biomethanation
microbiological techniques and mathematical simulation. Water Sci Technol
process of jatropha and pongamia oil cakes and their co-digested substrates. J
2007;56:19–28.
Sci Ind Res (India) 2009;68:624–9.
[12] Olsson G, Carlsson B, Comas J, Copp J, Gernaey KV, Ingildsen P, et al.
[31] VDI 4630. Fermentation of organic materials – characterisation of the
Instrumentation, control and automation in wastewater – from London 1973
substrate, sampling, collection of material data, fermentation tests.
to Narbonne 2013. Water Sci Technol 2014;69:1373–85.
Düsseldorf: Verein Deutscher Ingenieure. Düsseldorf: 2006.
[13] Gaida D, Wolf C, Meyer C, Stuhlsatz A, Lippel J, Bäck T, et al. State estimation
[32] Dennstedt M, Haßler F. Über den Abbau von Eiweiß. Zeitschrift Für Physiol
for anaerobic digesters using the ADM1. Water Sci Technol 2012;66:1088.
Chemie 1906;48:489–504.
[14] Martinez E, Marcos A, Al-Kassir A, Jaramillo MA, Mohamad AA. Mathematical
[33] Noike T, Endo G, Chang JE, Yaguchi J, Matsumoto J. Characteristics of
model of a laboratory-scale plant for slaughterhouse effluents biodigestion for
carbohydrate degradation and the rate-limiting step in anaerobic digestion.
biogas production. Appl Energy 2012;95:210–9.
Biotechnol Bioeng 1985;27:1482–9.
[15] Zhou H, Löffler D, Kranert M. Model-based predictions of anaerobic digestion
[34] Sambusiti C, Ficara E, Malpei F, Steyer JP, Carrère H. Effect of sodium hydroxide
of agricultural substrates for biogas production. Bioresour Technol
pretreatment on physical, chemical characteristics and methane production of
2011;102:10819–28.
five varieties of sorghum. Energy 2013;55:449–56.
[16] Antonopoulou G, Gavala HN, Skiadas IV, Lyberatos G. ADM1-based modeling of
[35] Koch K, Drewes JE. Alternative approach to estimate the hydrolysis
methane production from acidified sweet sorghum extract in a two stage
rate constant of particulate material from batch data. Appl Energy 2014;120:
process. Bioresour Technol 2012;106:10–9.
11–5.
[17] Wichern M, Lübken M, Horn H, Schlattmann M, Gronauer A. Investigations and
[36] Gunaseelan VN. Predicting ultimate methane yields of Jatropha curcus and
mathematical simulation on decentralized anaerobic treatment of agricultural
Morus indica from their chemical composition. Bioresour Technol 2009;100:
substrate from livestock farming. Water Sci Technol 2008;58:67–72.
3426–9.
[18] Lübken M, Gehring T, Wichern M. Microbiological fermentation of
[37] Orhon D, Çokgör EU, Sözen S. Dual hydrolysis model of the slowly
lignocellulosic biomass: current state and prospects of mathematical
biodegradable substrate in activated sludge systems. Biotechnol Tech 1998;
modeling. Appl Microbiol Biotechnol 2010;85:1643–52.
12:737–41.
[19] Donoso-Bravo A, Mailier J, Martin C, Rodríguez J, Aceves-Lara CA, Wouwer A
[38] Lobry J, Rosso L, Flandrois J. A fortran subroutine for the determination of
Vande. Model selection, identification and validation in anaerobic digestion: a
parameter confidence limits in non-linear models. Binary 1991;3:86–93.
review. Water Res 2011;45:5347–64.