Modeling and Optimization of Fermentativ

Bioresource Technology 102 (2011) 8569–8581
Contents lists available at ScienceDirect
Bioresource Technology
journal homepage: www.elsevier.com/locate/biortech
Modeling and optimization of fermentative hydrogen production

Kaushik Nath a, Debabrata Das b,⇑
a
Department of Chemical Engineering, GH Patel College of Engineering and Technology, Vallabh Vidya Nagar 388 120, Gujarat, India
b
Department of Biotechnology, Indian Institute of Technology, Kharagpur 721 302, India
a r t i c l e i n f o a b s t r a c t
Article history: Biohydrogen is a sustainable energy resource due to its potentially higher efficiency of conversion to
Received 17 December 2010 usable power, non-polluting nature and high energy density. The purpose of modeling and optimization
Received in revised form 30 March 2011 is to improve, analyze and predict biohydrogen production. Biohydrogen production depends on a num-
Accepted 31 March 2011
ber of variables, including pH, temperature, substrate concentration and nutrient availability, among oth-
Available online 9 April 2011
ers. Mathematical modeling of several distinct processes such as kinetics of microbial growth and
products formation, steady state behavior of organic substrate along with its utilization and inhibition
Keywords:
have been presented. Present paper summarizes the experimental design methods used to investigate
Biohydrogen
Optimization
effects of various factors on fermentative hydrogen production, including one-factor-at-a-time design,
Inhibition full factorial and fractional factorial designs. Each design method is briefly outlined, followed by the
Modeling introduction of its analysis. In addition, the applications of artificial neural network, genetic algorithm,
Fermentation principal component analysis and optimization process using desirability function have also been
highlighted.
Ó 2011 Elsevier Ltd. All rights reserved.
1. Introduction concepts led to a proven and long-term functioning industrial scale

bio-hydrogen production system. One of the major deterrents of
In the backdrop of looming energy crisis, as a result of depleting biohydrogen production processes for commercialization is their
fossil fuel reserves coupled with the concerns for greenhouse gas notoriously low hydrogen production rates particularly a low space
emissions, initiatives to develop self-sustaining alternative energy time yield (STY) per unit volume.
sources have assumed paramount importance. Hydrogen is widely
touted as an ultimate clean energy source. It has gained promi-
2. Fermentative hydrogen production
nence over competing technologies such as ethanol and biofuels
because of its ability to burn cleanly without by-product release
Hydrogen evolution by fermentation is common metabolism for
and its efficient convertibility into electrical and thermal energy
bacteria. A large number of microbial species, including signifi-
while supplied to a fuel cell. Biological processes are carried out
cantly different taxonomic and physiological types can produce
largely at ambient temperatures and atmospheric pressures, and
hydrogen by fermentation. Almost 25% of genera listed in the Ber-
hence are less energy intensive than chemical or electrochemical
gey’s Manual of Determinative Bacteriology, 8th edition were rec-
ones (Nath and Das, 2004). Unfortunately, hydrogen is presently
ognized to evolve hydrogen, no matter what the amount of
produced mainly from the reforming of fossil materials (90% of
evolution. These include strict anaerobes such as clostridia (Clos-
worldwide production, representing 45 billion tonnes) with a high
tridium butyricum, Clostridium welchii, Clostridium pasteurianum,
level of pollution generated, i.e., 10 tonnes CO2/tonnes H2 (Maddy
Clostridium beijerincki), Rumen bacteria, hyperthermophilic archae-
et al., 2003).
bacteria, and facultative anaerobes like Enterobacter, Escherichia
Several alternative hydrogen energy generation concepts and
coli, Citrobacter, or even some aerobes like Alcaligenes, Bacillus
technologies are in vogue. The idea to produce hydrogen from
and so on. Apart from pure culture some co and mixed cultures
microorganisms, such as bacteria and algae, via fermentation pro-
have also been found to be efficient producers of molecular hydro-
cesses has re-gained tremendous interest in the recent years (Luo
gen. These have been extensively reviewed by several authors (Das
et al., 2010; Rupprecht, 2009). Biological hydrogen can be pro-
and Veziroglu, 2001; Hallenbeck and Benemann, 2002; Nandi and
duced both by photosynthetic and fermentative microorganisms.
Sengupta, 1998). Anaerobic fermentation enables the mass produc-
However, none of the described microorganisms and proposed
tion of hydrogen via relatively simple processes not only from a
wide spectrum of potentially utilizable substrates, but also from
⇑ Corresponding author. Tel./fax: +91 3222 283758. refuse and waste products (Ntaikou et al., 2009a,b; Kim and Lee,
E-mail address: ddas.iitkgp@gmail.com (D. Das). 2010; Li et al., 2010; Karlsson et al., 2008; Luo et al., 2009).
0960-8524/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.biortech.2011.03.108
8570 K. Nath, D. Das / Bioresource Technology 102 (2011) 8569–8581
Moreover, fermentative hydrogen production generally proceeds estimated following Monod model by regression analysis of Linewe-
at a higher rate and does not rely on the availability of light sources aver–Burk linearized equation.
(Nath and Das, 2004). Carbohydrates, mainly glucose, are the pre- The dependence of H2 production rate on limiting substrate
ferred carbon sources for fermentation process which predomi- (starch) concentration, in presence of mixed anaerobic microflora
nantly give rise to acetic and butyric acids together with was simulated by Lee et al. (2008) using a Monod-type kinetic
hydrogen gas (Classen et al., 1999). model:
Modeling and optimization have been carried out in attempts to
v max;H C
2 starch
improve biohydrogen production. Biological production rate of vH 2
¼ ð2Þ
hydrogen and its molar yield, like many other bioprocesses, are a K s þ C starch
function of several variables, including medium pH, temperature, where v max;H2 is the maximum volumetric H2 production rate (ml/L/
substrate concentration and many more. Modeling and analysis h), Ks is half-saturation (Monod) constant (g COD/L) and Cstarch is
could be used to determine the optimal values of the important starch concentration (g COD/L).
relevant parameters. A variety of modeling methods have been But many of the authors reported that classical Monod model
developed that are broadly applicable to a spectrum of diverse does not hold well for several instances the most dominant of
fields, including engineering, biology, environmental science, food which is inhibition due to high substrate concentration, pH, pres-
processing and industrial processing, and their application in bio- ence of certain salt concentrations, substrate diffusion, mainte-
hydrogen (Hallenbeck and Ghosh, 2009). The first part of the re- nance or effects of cell density on lmax etc. Hence a number of
view gives an overview of the kinetic models as applied in the models are proposed for growth and substrate utilization of hydro-
prediction of microbial growth, product formation and degradation gen producing organisms incorporating the additional parameters
of organic substrate during fermentative hydrogen production. The for substrate inhibition, pH inhibition and biomass decay constant.
second part deals with various experimental design methods used The inclusion of substrate inhibition factor in classical Monod
to investigate the effects of physico-chemical parameters on fer- equation stands out to be a necessity since many authors reported
mentative hydrogen production processes for its optimization. that values of the growth kinetic parameters could not simulate
Several design strategies are briefly exposed, followed by the intro- satisfactorily the experimental data from cultures with high sub-
duction of its analysis and application to the study of fermentative strate concentrations. Table 1 presents kinetic parameters for var-
hydrogen production. ious classical and modified Monod models and M–M equations
used in dark and light fermentation.
3. Modeling of the fermentative process The inhibitory nature of substrates at high concentrations is well
known, and the kinetics of pure and mixed culture hydrogen produc-
Mathematical models can provide valuable information to ana- ing microorganisms on carbon substrates have been described by a
lyze and predict the performance of biological hydrogen production variety of substrate inhibition models (Nath et al., 2008; JianLong
by fermentation. Kinetic equations, which describe the activity of an and Wei, 2008; Ntaikou et al., 2009). Most of these models are
enzyme or a microorganism on a particular substrate, are crucial in empirical. However, they are able to provide satisfactory description
understanding many phenomena in biotechnological processes. It is of substrate utilization data, thus providing a convenient means of
important to gain an understanding of operations where time-vari- modeling microbial growth and substrate utilization.
ant influent concentrations and multiple substrates are encountered Substrate inhibition model following Andrew’s equation was
(Okpokwasili and Nweke, 2005). For an effective design and scale-up found suitable for hydrogen production using glucose as substrate
of the bioreactor, the definition of an appropriate kinetic model of in the first stage of a two stage fermentation process compared to
the process is required. Modeling based on material balances for the classical Monod model (Nath et al., 2008). The model suggests a
the process components of the H2-producing bioreactors requires non-linear relationship between the specific growth rate and sub-
the mathematical description of several distinct processes such as strate concentration.
kinetics of microbial growth (biomass formation) and products
lm S
(hydrogen, VFA, solvents, etc.) formation in the bioreactor, steady l¼ ð3Þ
state behavior of organic substrate along with its utilization and K s þ S þ S2 =K I
inhibition (if any). Additionally a few models are developed to de- In addition, Nath et al. also compared the ability of classical
scribe the effects of several physicochemical parameters like tem- Monod model and an Andrew model to describe the progress of glu-
perature, pH, dilution rate etc. on biohydrogen production using cose degradation and Enterobacter cloacae DM11 growth in batch
either batch or continuous mode of operations. tests and concluded that the latter was the most suitable one. How-
ever, there was little deviation between the experimental data and
3.1. Kinetic models for microbial growth and substrate utilization those predicted by the model which might be due to the assumption
that Yx/s remained constant over the system irrespective of time of
The biomass concentration in the batch hydrogen production fermentation. Also, this was assumed by ignoring the effect of main-
experiments depends on the concentration in limiting substrate. tenance energy on cell growth and metabolism. JianLong and Wei
Classical Monod equation empirically fits a wide range of data sat- (2008) fitted the Andrew’s model to describe the biohydrogen pro-
isfactorily and is the most commonly applied unstructured, non- duction from glucose by mixed microflora using digested sludge
segregated model of microbial growth to describe growth-linked with estimated correlation coefficient (R2) 0.902.
substrate utilization (Shuler et al., 2002).
67:1S
The growth of microorganisms can be modeled by Monod r¼ ð4Þ
equation. 47:7 þ S þ S2 =13:5
lmax S Notwithstanding the fact that the Andrew’s model has been
l¼ ð1Þ widely used to describe the substrate inhibition, it does not ade-
Ks þ S
quately predict the nature of inhibition. A generalized Monod type
where l is the specific growth rate, lm is the maximum specific model (Eq. (5)), originally proposed by Han and Levenspiel (1988)
growth rate, Ks is the saturation constant and S is the limiting could probably better be used to account for substrate stimulation
substrate concentration. The values of lmax and Ks are usually at low concentration and substrate inhibition at high concentration.
K. Nath, D. Das / Bioresource Technology 102 (2011) 8569–8581 8571
Table 1
Kinetic parameters for various classical and modified Monod models and M–M equations used in dark and light fermentation.
Microorganism Substrate and Model used Maximum specific growth Saturation Yield coefficient References
mode of rate(lm) h1 constant (Ks) Yx/s g biomas/g
operation g/L substrate
E. coli BL-21 (recombinant) Glucose, Batch Classical Monod 0.4 11.11 – Chittibabu
et al.
(2006)
Mixed microflora from anaerobic Sucrose, batch Michaelis–Menten – 1.446 – Chen et al.
digestor (2006a,b)
Anaerobic microflora from UASB Sucrose, batch Michaelis–Menten 0.28 g/g VSS h (maximum 13.5 – Mu et al.
reactor treating citrate-producing specific substrate degradation (expressed (2006)
wastewater rate) as km)
Mixed anaerobic microflora from Glucose, CSTR Monod with dilution 0.75 200 mgl1 0.30 Whang
UASB reactor treating food rate (D) term et al.
processing waste water (2006)
Mixed anaerobic microflora from Glucose, CSTR Monod with D and 1.00 178 mgl1 0.45 Do
food processing waste water endogenous rate
constant (ke) term
Do Nonfat dry Do – 6.616 – Do
milk (NFDM),
batch
Do Food waste, Do – 8.692 – Do
Batch
Enterobacter cloacae DM 11 Glucose, Batch Modified Monod using 0.4 5.51 – Nath et al.
Andrew’s equation (2008)
Ruminococcus albus Glucose, batch Modified Monod using 0.603 ± 0.011 276 ± 33.38 0.147 ± 0.01 Nath et al.
pH inhibition and decay (2008)
constant
Ruminococcus Albus DSMZ 20455 Glucose, batch Modified Monod using 0.654 ± 0.039 0.765 ± 0.029 0.139 ± 0.012 Ntaikou
pH and substrate et al.
inhibition (2009a,b)
Mixed microflora from organic farm Glucose, batch Classical Monod 0.03 day1 16.1 – Sharma
soil and Li
(2009)
Mixed anaerobic microflora from Cassava Monod type model 1741 mL/h L (expressed as 16.28 g COD – Lee et al.
sewage sludge starch, batch maximum volumetric H2 L1 (2008)
production rate)
Clostridium acetobutylicum Glucose, batch Monod type model with 0.496 1.0 mmol/L 0.2 (mmol/mmol- Lin et al.
lower pH inhibition glucose) (2007)
Clostridium butyricum Do Do 0.88 (mmol/mol h) 4.34 mmol/L 0.34 (mmol/ Do
mmol-glucose)
Clostridium tyrobutyricum Do Do 0.428 (mmol/mol h) 4.0 mmol/L 0.46 (mmol/ Do
mmol-glucose)
Clostridium beijerinckii Do Do 1.03 (mmol/mol h) 2.6 mmol/L 0.23 (mmol/ Do
mmol-glucose)
Clostridium butyricum CGS5 Xylose Monod type model 0.15 0.67 g COD/L – Lo et al.
(2008)
Clostridium pasteurianum CH4 Sucrose Do 0.31 4.39 g COD/L – Do
Rhodobacter capsulatus strain IR3 Na-lactate, Classical Monod 0.3 10.0 0.1 Obeid
batch et al.
(2009)
Rhodobacter capsulatus strain B10 Na-lactate, Classical Monod 0.4 19.0 0.7 Obeid
batch et al.
(2010)
Thermoanaerobacterium Sucrose, batch Classical Monod 0.31 1.47 – Thong
thermosaccharolyticum PSU-2 et al.
(2008)
S also revealed that inhibition of the fermentative hydrogen

r ¼ kð1 S=Smax Þn ð5Þ
S þ K S ð1 S=Smax Þm production by substrate at higher concentration was uncompeti-
tive in nature as the values of n and m obtained from the fitting
where r (mL/h) is the hydrogen production rate; k (mL/h) is the using Han–Levenspiel model were all positive.
hydrogen production rate constant; S (g/L) is the substrate con- Monod-type kinetic expression, incorporating the empirical
centration; Smax (g/L) is the maximum substrate concentration lower pH inhibition term IpH was put forward by a number of
above which the fermentative hydrogen production stops; KS (g/ authors [Lin et al., 2007; Ntaikou et al., 2008, 2009].
L) is the saturation constant; m and n are the exponent constants
qmax
Glu SGlu X
the specific values of which determine the type of substrate inhi- rGlu ¼ :IpH ð6Þ
K Glu þ SGlu
bition such as non-competitive, competitive, uncompetitive and
mixed inhibition. where X denotes biomass concentration, SGlu denotes residual glu-
JianLong and Wei (2008) compared Andrews model and Han– cose concentration, qmax
Glu denotes maximum specific glucose con-
Levenspiel model to fit the hydrogen production data and consumption rate, and KGlu represents saturation constant. Similar
cluded that the latter could describe better the effect of substrate type of kinetic expression was used by Ntaikou et al. (2008) for
(glucose) concentration on hydrogen production rate. The study the batch production of hydrogen from glucose and sweet sorghum
biomass using Ruminococcus albus. However, together with pH inhi- parameters across different studies. This complicates the applica-
bition coefficient (IpH) Ntaikou et al. (2008) modified the Monod tion of laboratory kinetic information in the design and scale up
equation with one additional term, biomass decay constant kd. of biological hydrogen production via fermentation of either pure
substrate or waste materials.
dx lmax :S
¼ X:IpH kd :X ð7Þ Logistic equation and its modified version have also been used
dt K S þ S to describe the microbial growth during dark as well as light fer-
The growth of fermentative hydrogen producing R. albus on glu- mentation process for biohydrogen production (Nath et al., 2008;
cose was modeled by a modified Monod equation including both Wang and Wan, 2008c,d; Jianlong and Wei, 2008). The logistic
pH inhibition (IpH) and substrate inhibition (IS) factors equation for cell concentration for a batch reactor can be repre-
sented as
ds S
¼ km :X:IpH :IS ð8Þ
dt KS þ S X 0 expðkc tÞ
X¼ ð11Þ
The factor IS of Eq. (8) is a function of S which describes non- 1 XXmax
0
ð1 expðkc tÞÞ
competitive substrate inhibition according to Eq. (9)
where kc is the apparent specific growth, X is the biomass concen-
K:IS tration and Xmax is maximum biomass concentration. Notwith-
IS ¼ ð9Þ
KIS þ S standing the fact that the predictive power of logistic equation
(Eq. (11)) is limited as it does not contain a substrate term, the lo-
where KIS is the value of substrate concentration when the sub-
gistic model is a fair approximation of representing the entire
strate consumption rate is half of the maximum.
growth curve including the lag phase (if present), the exponential
The International Water Association (IWA) task group for math-
growth and the stationery phases (Mu et al., 2006a,b; Nath et al.,
ematical modeling of anaerobic digestion processes developed the
2008). In a study by Nath et al. (2008) hydrogen production by a
anaerobic digestion model No. 1 (ADM1) to describe the stoichiom-
two stage fermentation process, growth of Rhodobater sphaeroides
etry and kinetics of reactions occurred in anaerobic processes
OU 001 on spent medium in the second stage was modeled by Lo-
(Batstone et al., 2002). The empirical lower pH inhibition term,
gistic equation. For initial acetic acid concentration of 2.5 g/L in the
IpH, according to the ADM1, was defined as
spent medium the maximum specific growth rate was found to be
2 ! 0.099 h1 using logistic model in the second stage. Jianlong and
pHmeas pHUL
IpH ¼ exp 3 ð10Þ Wei (2008) used the modified Logistic model (Eq. (12)) to describe
pHUL pHLL
the progress of bacterial growth during fermentative hydrogen pro-
where pHUL and pHLL denote the upper limit at which the acido- duction by mixed cultures in batch tests using glucose as substrate
genic bacteria are not inhibited (IpH = 1 when pH > pHUL), and the Z t
dX Xm
lower limit at which inhibition is complete, respectively. In the dt ¼ ð12Þ
0 dt 1 þ exp½4lm ðk tÞ=X m þ 2
ADM1, the values of pHUL and pHLL are recommended to be 5.5
and 4, respectively, for acidogenic bacteria. The coefficients of determination (R2) of all the fittings were all
Penumathsa et al. (2008) presented a modification of the ADM1 over 0.99, which indicated that the modified Logistic model could
using a variable stoichiometry approach, derived from experimen- describe the progress of cumulative hydrogen production in the
tal information in the continuous production of hydrogen from su- batch tests adequately.
crose as substrate in a CSTR. The authors considered the In order to describe the true behavior of biological hydrogen
concentration of undissociated acidic products as the main govern- production systems, it is imperative to obtain accurate estimates
ing variable for the product yields (including VFAs and hydrogen) of the kinetic parameters in these models. The values of lmax and
to describe fermentation of sugars in non-methanogenic anaerobic KS following Monod model are estimated by regression analysis
systems. The biomass and product yields from glucose degradation of Lineweaver–Burk linearized equation. But the non-linearity of
were assumed to be dynamically depending on the total concen- the modified Monod kinetic models and their integrated forms,
tration of undissociated acids in the reactor. Experimental data as proposed by various workers renders parameter estimation rel-
from an 11 L mesophilic continuous bio-hydrogen reactor fed with atively difficult. However, some of these models can be linearized.
20, 40, 50 and 10 g/L of sucrose, were used to validate the ap- But the linearized expression may transform the error associated
proach. The modified model achieved good predictions of the with the dependent variable making it not to be normally distrib-
experimental data, using the standard ADM1 parameter values, uted, thus making parameter estimate inaccurate. Several authors
without any parameter fitting beyond the implementation of the (Nath et al., 2008; Whang et al., 2006) have used least square
variable stoichiometry (Penumathsa et al., 2008). methods to estimate the kinetic parameters.
Gadhamshetty et al. (2010) compared the batch biohydrogen
production using Gompertz equation and ADM 1 model. Experi- 3.2. Kinetic models for product formation
mental data from eight dark fermentation reactors varying in tem-
perature, pressure release, biomass seed, and substrate were used The products from fermentative biohydrogen processes are
to evaluate the predictive ability of the Gompertz equation and an broadly grouped into two major categories. The first category is
ADM1-based model developed in this study. The authors main- the gaseous products (primarily H2 and CO2), while the second
tained that Gompertz equation was limited by its inability to pre- group includes VFA and solvents. During hydrogen production by
dict volatile fatty acid formation and substrate consumption, anaerobic fermentation the distribution of the acidogenic products
whereas the ADM1-based model was able to predict well not only varies substantially. Moreover, some acidogenic products, e.g., ace-
the hydrogen profile, but also the COD, acetate, and butyrate. In tate, butyrate, hydrogen, carbon dioxide, may form more complex
contrast to the Gompertz equation, only one specific parameter long-chain fatty acids or alcohols as the hydrogen is consumed
per reactor had to be chosen in the case of the ADM1-based model (Mu et al., 2006a,b).
(Gadhamshetty et al., 2010). Modified Gompertz equation has been extensively used to
It is noteworthy to state that since there are a good number of describe the progress of hydrogen production and some soluble
inhibition models to describe experimental data of microbial metabolite production in a batch fermentative hydrogen produc-
hydrogen production, it becomes difficult to compare kinetic tion process. The equation can be expressed as;

Rm :e be modeled by the Luedeking–Piret model (Eq. (15)) or its modified
HðtÞ ¼ P: exp exp ðk tÞ þ 1 ð13Þ
P version (Eq. (16)) (Mu et al., 2006a; Mu et al., 2006b; Lo et al.,
2008; Obeid et al., 2009):
where H(t) represents cumulative volume of hydrogen production
(ml), P the gas production potential (ml), Rm (ml/h) the maximum dpi dx
¼ ai þ bi X ð15Þ
production rate, l k(h) the lag time, t incubation time (h) and e, dt dt
the exp(1) 2:718. The typical cumulative hydrogen production where ai is the growth-associated formation coefficient of product i;
curve can be nonlinearly modeled by the above equation. In a batch bi is non-growth-associated formation coefficient of product i.
test, H increases very slowly with increasing cultivation time from 0 Lo et al. (2008) proposed a kinetic model based on growth asso-
to l, and then increases rapidly almost at the rate of Rm and finally ciated products representing the correlation between hydrogen
reaches an asymptotic value P with further increasing the cultiva- production and biomass gain.
tion time (Wang and Wan, 2009a).
Mu et al. (2006a,b) used the Modified Gompertz equation to 1 dC H2 1 dX
¼ al ¼ a ð16Þ
model the production of hydrogen as well as butyrate and acetate, X dt X dt
during a batch hydrogen production process by mixed anaerobic where C H2 represents H2 concentration (mol), X represents cell con-
culture. centration (g VSS/L), and l represents specific growth rate (h1). By
simplifying Eq. (16), the a value can be easily obtained from the
Rmax;i e
Pi ¼ Pmax;i exp exp ðki tÞ þ 1 ð14Þ slope of specific rate of C H2 versus that of X plot during the exponen-
Pmax;i
tial growth phase. The model fitted the data quite well (with an R2
where i represents hydrogen, butyrate and acetate, respectively; Pi value of over 0.910) and the calculated a value was 0.041 and
is the product i formed per liter of reactor volume at fermentation 0.039 mol/g VSS for xylose and sucrose, respectively during fermen-
time t; Pmax,i is the potential maximum product formed per liter tative hydrogen production using C. butyricum (strains CGS2 and
of reactor volume; Rmax,i is the maximum rate of product formed. CGS5), C. pasteurianum (strains CH1, CH4, CH5, and CH7).
Table 2 presents a summary of several recent studies where modi- The effect of light intensity on hydrogen production by the pho-
fied Gompertz equation has been used to model the cumulative tosynthetic bacterium Rhodobacter capsulatus was modeled by
production of biohydrogen. Obeid et al. (2009) by incorporating an additional factor into the
The relationship between biomass and the products for the Luedeking–Piret equation to take into consideration the light
anaerobic hydrogen production by mixed anaerobic cultures can intensity effect.
Table 2
Some recently used representative Gompertz models and its modified version.
Microorganism Substrate and mode of Maximum rate of production of H2 and Correlation References
operation substrate concentration coefficient
Enterobacter cloacae DM 11 Glucose, batch Rm = 14.14 mL/h 0.992 Nath et al. (2008)
Substrate conc.: 1.4 g/L
Enterobacter cloacae DM 11 Glucose, batch Rm = 13.92 mL/h 0.996 Nath et al. (2006)
Substrate conc.: 1.0 gL1
Rhodobacter sphaeroides OU 001 Spent MYG medium from Rm = 16.15 mL/h 0.996 Nath et al. (2008)
dark fermentation
Mixed anaerobic microflora Cassava starch Rm = 477 mL/h 0.997 Lee et al. (2008)
(T = 37 °C)
Rm = 83 mL/h
(T = 55 °C)
Mixed microflora from digested sludge Glucose Rm = 15.1 ml h1 0.99 JianLong and Wei
Substrate conc.: 25 gl1 (2008)
Mixed microflora from anaerobic digestor Sucrose, batch Rm = 10.6 mL/h 0.999 Chen et al. (2006)
Substrate conc.: 17.9 g COD/L
Do Nonfat dry milk (NFDM), Rm = 22.0 mL/h 0.997 Do
batch Substrate conc.: 96 g COD/L
Do Food waste Rm = 32.3 mL/h 0.998 Do
Substrate conc.: 25 g COD/l
Anaerobic microflora from UASB reactor treating Sucrose, batch Rm = 201 mL/L h 0.999 Mu et al. (2006)
citrate-producing wastewater Substrate conc.: n.a
Anaerobic microflora from UASB Lactose, Rm = 6.8 ± 1.1 mL/h 0.99 Davila-Vazquez et al.
Substrate conc.: 15 g/L (2008)
Do Cheese whey powder (CWP) Rm = 13.1 ± 0.1 mL/h 0.99 Do
Substrate conc.: 15 L1
Do Glucose, Rm = 11.3 ± 0.6 mL/h 0.99 Do
Substrate conc.: 15 gL1
Mixed anaerobic culture Sucrose, Batch Rm = 1511 mL/h – Mu et al. (2006a,b)
Substrate conc.: 24.8 gL1
Mixed anaerobic culture Sweet Sorghum syrup Rm = 51.66 ml H2/L h – Saraphirom and
Substrate conc.: 25 g/L Reungsang (2010)
Mixed anaerobic culture Starch, Batch Rm = 37.9 mL/h 0.999 Lin et al. (2008)
Substrate conc.: 20 g COD/L
Aerobic thermophillic digestion sludge with Poultry slaughterhouse Rm = 4.04 mL/h 0.99 Sittijunda et al.
nutrient sludge (2010)
Seed sludge from anaerobic digestor Coffee drink manufacturing Rm = 12.03 mL/h – Jung et al. (2010)
wastewater Substrate conc.: 20 g COD/L (based on
carbohydrate)
Enterobacter cloacae IIT-BT 08 Glucose, Batch Rm = 72 mLH2 g1 TVS, substrate conc.: 0.99 Khanna et al. (2011)
dV H2 1 equation was applied to estimate the hydrogen production poten-

¼ uðIÞlðSÞX þ bX ð17Þ
dt YXP tial, rate and lag phase time in a batch process for various initial
concentration of glucose, based on the cumulative hydrogen pro-
where V H2 is the volume of hydrogen produced at time and
duction curves. Both the curve fitting and statistical analysis
v arphiðIÞ is the the yield of light energy (lux), YxP the hydrogen pro- showed that the equation was suitable to describe the progress
duction yield, and b the non-growth-associated formation coeffi-
of cumulative hydrogen production (Nath et al., 2008).
cient of product.
Three different models namely the single-substrate model
without endogenous metabolism, the single substrate model with
3.3. Modeling of two-stage fermentation system endogenous metabolism, and the dual substrate model with
endogenous metabolism, were compared to describe the effects
Two-stage fermentation is one among variety of manipulations of dilution rates on the hydrogen production rate and the concen-
that have been proposed by various authors to improve hydrogen trations of glucose, peptone, biomass, ammonium nitrogen, for-
yields. There is a great deal of interest in this area at present mate, acetate, and butyrate in a continuous stirred tank reactor
and the results of many studies are appearing in the literature. for hydrogen production (Whang et al., 2006). The model equations
Hallenbeck and Ghosh (2009) discussed several strategies for com- are stated below:
plete substrate conversion using two-stage systems. These include Single-substrate model without endogenous metabolism:
dark fermentation followed by photofermentation (Nath et al., For glucose and peptone
2008), dark fermentation followed by methane fermentation
(Wang and Zhao, 2009; Cooney et al., 2007; Liu et al., 2006) and DK SG
ð22Þ
dark fermentation followed by electrohydrogenesis. But all these lmax
G D
processes have met with limited success in increasing the overall For biomass, VFAs and alcohols
yield of biohydrogen appreciably.
Wang and Zhao (2009) modeled an integrated two-stage fer- DK SG
Y X=G SG0 ð23Þ
mentation system combining hydrogen and methane production l Dmax
G
for the treatment of food waste by anaerobic mixed microflora.
Single substrate model with endogenous metabolism
The model based on mass balance equation for a two-stage process
For glucose and peptone
combined by a mesophilic hydrogen production reactor and a
methane production reactor. The mass balance equation for each ðD þ ke ÞK SG
ð24Þ
fermentation stage can be presented as: ðlmax
G D ke Þ
dc For biomass, VFAs and alcohols
V ¼ m0 c0 m0 c þ VrðcÞ ð18Þ
dt DY X=G ðSG0 SG Þ
The rate of degradation of the substrates was assumed to follow
ð25Þ
ai X ðDþke Þ
first-order kinetics, thus the following equations were used to de-
scribe the H2/CH4 production and substrate degradation in semi- Dual substrate model with endogenous metabolism
continuous operation For glucose
dc ðD þ ke ÞK SG
¼ rðcÞ ¼ k:c ð19Þ ð26Þ
dt ðlmax
G D ke Þ
The organic loading rate (kg VS/m3 d) For peptone
c0 kct ðD þ ke ÞK SP
h¼ ð20Þ ð27Þ
c 0 ct ðlmax
P D ke Þ
Biogas yield (m3/kgVS) can be expressed as For biomass, VFAs and alcohols
DY X=G ðSG0 SG Þ DY X=P ðSP0 SP Þ
l k ct þ ð28Þ
y¼ ð21Þ ðD þ ke Þ ðD þ ke Þ
h
At steady state, the optimal OLR (organic loading rate) and SRT The simulation study of the above there models suggested that
(solid retention time) for the integrated two-stage process were an increase in D from 0.06 to 0.69 h1 could minimize the adverse
found to be 22.65 kg VS/m3d (160 h) for hydrogen fermentation effects of endogenous and peptone metabolism on net hydrogen
reactor and 4.61 (26.67 d) for methane fermentation reactor, production. For the operational conditions of D > 0.69 h1, how-
respectively. Under the optimum conditions, the maximum yields ever, washout of hydrogen-producing bacteria in the CSTR became
of hydrogen (0.065 m3H2/kg VS) and methane (0.546 m3 CH4/kg substantial. Dual substrate model with endogenous metabolism
VS) were achieved with the hydrogen and methane contents rang- provided a better understanding in anaerobic fermentation of
ing from 29.42% to 30.86%, 64.33% to 71.48%, respectively (Wang glucose and peptone for hydrogen production compared to the
and Zhao, 2009). single-substrate model without endogenous metabolism, the
Nath et al. (2008) reported two-stage fermentation as a combi- single substrate model with endogenous metabolism (Whang
nation of dark and photo fermentation. Spent medium from dark et al., 2006).
fermentation by E. cloacae DM11 was used to photoproduce hydro-
gen by R. sphaeroides O.U. 0 0 1 in a two-stage batch fermentation 4. Optimization of physico-chemical parameters
process. The yield of hydrogen in the first stage was about 3:31 mol
H2/(mol glucose) and that of in the second stage was about 1.5– Various strategies and methods for improving hydrogen pro-
1.72 mol H2/(mol acetic acid). Substrate inhibition model following duction rates and yields have been under investigation for the past
Andrew’s equation was used in the first stage of fermentation and a few decades. Many studies have been conducted on fermentative
model based on logistic equation was used to describe the growth hydrogen production using pure and mixed culture with both pure
of R. sphaeroides O.U 0 0 1 in the second stage. Modified Gompertz substrate and various actual and simulated wastewaters. Results
showed that a number of intervening process parameters, few of principle and relative merits and demerits of various optimization
which are interdependent, play an important role, in the produc- strategies for biohydrogen production are briefly summarized in
tion rates and yields of hydrogen. Most widely studied important Table 3.
key parameters are probably substrate concentration, temperature
and initial medium pH. In addition, there are certain other param- 4.1.1. One-factor-at-a-time optimization
eters like inoculum size, iron concentration, HRT, C/N and C/P ratio The traditional method of process optimization involves varying
and so on. In spite of the fact that fermentative hydrogen produc- the level of one parameter at a time, while keeping the other
tion using carbohydrate substrates in general require similar con- parameters constant. There is large number of works in literature,
ditions, there are significant variations in some of the parameters reporting the effects of various parameters in batch as well as con-
while establishing their suitable values, with respect to maximum tinuous or semicontinuous production of hydrogen. The most use-
hydrogen yield or its production rate, even at times with the same ful application in batch hydrogen production has been to
substrates and same microorganisms. Here comes the importance determine the optimal operating conditions with regard to med-
of optimization of the physico-chemical parameters of a biological ium pH (Hamilton et al., 2010; Wang and Jin, 2009; Chittibabu
hydrogen production, particularly with fermentative production. et al., 2006; Nath et al., 2006), substrate concentration or organic
In addition, these methods could prove helpful in establishing pro- loading rate (OLR) in case of wastewater treatment (Wang and
cess parameters for novel consortia or complex. wan, 2008a; Kim et al., 2006a,b; Wang and Jin, 2009), iron concen-
Initial medium pH is one of the most important factors in tration (Hamilton et al., 2010; Wang and Wan, 2008b; Wang et al.,
hydrogen production, the effect of which has been extensively re- 2007; Nath et al., 2006), temperature (Lin et al., 2008a,b; Mu et al.,
ported in literature. pH has significant effects on metabolic path- 2006a,b; Nath et al., 2006). However, for continuous or semicontin-
ways, thus potentially modulating end product distribution, as uous operations effect of some additional parameters like hydrau-
well as possibly affecting the duration of the lag phase (Bartacek lic retention time, dilution rate, and recycle ratio becomes
et al., 2007; Nath et al., 2006; Davila-Vazquez et al., 2008; Lee significant in the production rate as well as yield of biohydrogen.
et al., 2008; Hallenbeck and Ghosh, 2009; Hawkes et al., 2007; Hydraulic retention time (HRT), which is the residence time of
Saraphirom and Reungsang, 2010). the liquid in a continuously fed reactor, has profound effect on
The poor hydrogen production at low pH, lower than 5.0, is due hydrogen production rate in a continuous system. It controls the
to the increased formation of acidic metabolites, which destroys microbial growth rate and therefore should invariably be greater
the cell’s ability to maintain internal pH (Bowles and Ellefson, than the maximum growth rate of the organism(s), because faster
1985). It results in lowering the intracellular level of ATP, thereby, dilution rates cause washout (Ghosh and Hallenbeck, 2010). Lin
inhibiting glucose uptake. Nath et al. (2006) observed that the opti- et al. (2008a,b) reported hydrogen production from anaerobically
mum pH for maximizing the rate of hydrogen production was enriched paper mill sludge in a continuous fermenter operated at
dependent both on the type of microorganisms and the substrates the HRTs of 12–2 h. The bioactivity was enhanced with peak
used. Initial glucose concentration also has a substantial effect on hydrogen yield (at HRT 12 h), hydrogen production rate (at HRT
the yield and production rate of hydrogen during the course of fer- 4 h), of 1.5 mol-H2/mol-hexose, 450 mmol-H2/L d (18.8 mmol-H2/
mentation. Low initial glucose concentrations result in low rates of L h), respectively. A shift in dominant hydrogen-producing micro-
fermentation, and total fermentation times increase with high ini- bial population along with HRT variation was observed with C.
tial substrate concentrations (Ghosh and Hallenbeck, 2010). The butyricum, C. pasteurianum, K. pneuminiae, Streptococcus sp., and
observation implies that the chances of formation of acid metabo- Pseudomonas sp. being present at efficient hydrogen production
lites increase with high initial glucose concentration, whereby, pH at the HRTs of 4–6 h (Lin et al., 2008a,b). In another study, Hamil-
drops appreciably and results in a lower rate of hydrogen produc- ton et al. (2010) obtained optimal overall hydrogen production rate
tion albeit at the cost of higher percent consumption of glucose of 33.2 mL H2/L h and a yield of 0.83 mol H2/mol glucose in a semi-
(Nath et al., 2006). Moreover, temperature affects the maximum continuous reactor applying the previously optimized parameters
specific growth of the microorganism as well as substrate utiliza- for pH, nitrogen, and iron with a dilution rate of 0.012 h1 and
tion rate during hydrogen fermentation. Thermal deactivation at degassing of biogas by N2 at a 28 mL/min flow rate. Luo et al.
higher temperatures leads to inactivation of the enzymes responsi- (2010) optimized the influent alkalinity (6 g NaHCO3/L) and HRT
ble for controlling metabolic pathways in the fermentative hydro- (24 h) for thermophilic hydrogen production from cassava stillage.
gen production process. They reported maximum hydrogen yield of 76 ml H2/gVS and
hydrogen production rate of 3215 ml H2/L d. Maximum hydrogen
4.1. Various optimization strategies production rate of 66 mmol/(l h) was reported by using glucose
as substrate at a dilution rate of 0.55 h1 in the immobilized whole
Design of experiment (DOE) is a structured method by which cell bioreactor (Chittibabu et al., 2006). However, the production
certain factors are selected and deliberately varied in a controlled rate dropped with further increase in dilution rate. Consumption
manner to obtain their effects on the output of a process, often fol- of substrate (glucose) was found to decrease with increase in dilu-
lowed by the analysis of the experimental results. It should predict tion rate and the maximum utilization was about 70% at a dilution
the important parameters and the relationship between them. rate of 0.25 h1. Similar observations were reported by Hamilton
According to the number of the factors to be investigated at a time, et al. (2010), in their study of biological hydrogen production by
the experimental design can be broadly classified into two catego- Citrobacter freundii CWBI952 in batch, sequenced-batch and semi
ries: one-factor-at-a-time design (single-factor design) and facto- continuous operating mode. They reported maximum H2 yield of
rial design (multiple-factor design) (Wang and Wan, 2009a). 0.95 mol H2/mol glucose with a dilution rate of 0.009 h1 before
Additionally, there are several types of factorial design, namely full growth restriction took place probably as a consequence of
factorial and fractional factorial design that are extensively re- byproduct inhibition. The hydrogen yield then decreased to
viewed by Wang and Wan (2009a). A common and powerful 0.74 mol H2/mol glucose when the dilution rate was increased to
DOE approach is afforded by factorial design (FD) combined with 0.018 h1. Prasertsan et al. (2009) reported the optimum values
response surface methodology (Ghosh and Hallenbeck, 2010). Prin- of 2 d HRT with an OLR of 60 g COD/L d which gave a maximum
cipal component analysis and Genetic algorithm based on a neural hydrogen yield of 0.27 l H2 g/COD with a volumetric hydrogen pro-
network model approaches have also been used, to optimize the duction rate of 9.1 L H2/L d (16.9 mmol/L/h) from palm oil mill
process parameters in fermentative hydrogen production. Basic effluent by thermophilic fermentative process.
Table 3
Summary of various optimization strategies for biohydrogen production adopted in literature.
S Method Principle Relative merits and demerits

No.
1 One factor at a This is a traditional design method investigating one-factor-at-a-time, Merits: Simple, easy to operate and analyze.
time while keeping the levels of other factors constant. The level of the Demerits: Ignores the interactions among different factors.
factor to be investigated is then changed over a desired range to study Involves a relatively large number of experiments.
its effects on a response.
2 Full factorial A statistical method showing the relationship between responses of all Merits: Number of runs increases geometrically as the number of
design experimental factors that are varied simultaneously. factors increases.
Demerits: inability to accurately predict all the positions of the factor
space which are equi-distant from the centre (rotatability)
3 Fractional With a fractional factorial design, the effects of certain factors on a Optimal solution may not be guaranteed to be optimal always due to
factorial response can be studied under an economical and practical condition the poor modeling ability of the second-order polynomial model.
design (FD)
(a) Taguchi Taguchi design is a fractional factorial design using orthogonal array. It Taguchi design may provide a powerful and efficient method to find
design allows the effects of many factors with two or more levels on a an optimal combination of factor levels that may achieve optimum.
response, to be studied in a relatively small number of runs. True optimal factor levels may not be guaranteed using Taguchi
design, because the true optimal factor levels may be different from
the corresponding pre-determined factor levels
(b) Plackett– It is a two-level fractional factorial design developed by Plackett and Based on the Plackette–Burman experimental design, each factor was
Burman Burman, to screen important factors for further investigation. The prepared in two levels: 1 for low level and +1 for high level.
design number of runs for a Plackett–Burman design is equal to a multiple of
four.
(c) Central Central composite design is a five-level fractional factorial design It has a relatively more number of factor levels and contains extreme
composite developed. The design usually consists of a 2n full factorial design, high or extreme low levels.
design (CCD) 2 n axial designs and m central designs.
(d) Box– It is a three level design based upon the combination of two-level It requires less experiments than the FFD or CCD with the same
Benkhen factorial and incomplete block designs. The BBD method employs a number of factors
design (BBD) spherical design with excellent predictability within the design space. BBD technique is rotatable or nearly rotatable regardless of the
number of factors under consideration
4. Artificial A computational statistical modeling process based on the It represents the non-linearities in a much better way.
neural connectivity of biological neurons. It is nonlinear and adaptive. ANN requires much more number of experiments (number of
network patterns) than RSM to build an efficient model.
5. Computational Computational fluid dynamics soft ware can predict fluid flow, heat A powerful tool for the prediction of flow patterns and hydrodynamics
fluid dynamics and mass transfer, chemical reactions, and other related phenomena in the bioreactor.
by solving a set of appropriate mathematical equations.
Effect of iron concentration (mostly Fe2+) is another important theoretical design. Several studies report statistical experimental
parameter, which has been reported in quite a good number of design methods that involve initially identifying the influencing
studies (Hamilton et al., 2010; Wang and Wan, 2008b;Wang parameters by screening method of Plackett–Burman, followed
et al., 2007; Nath et al., 2006). Nath et al. (2006) observed that with by the path of steepest ascent to approach the vicinity of optimal
an iron concentration up to 20 mg/L in the malt extract-yeast ex- region of the parameters and finally central composite or Box–
tract-glucose (MYG) media, the total volume of hydrogen produc- Behnken design for response surface methodology to further inves-
tion had increased marginally. However, higher concentrations of tigate the mutual interaction between the parameters and thereby
iron were found to have some inhibitory effect on hydrogen pro- identifying the optimal values that lead maximum hydrogen pro-
duction. In another study it was observed that within a certain con- duction rate or yield (Pan et al., 2008; Long et al., 2010). Table 4
centration range FeSO4 (0.045 and 0.03 g/L, respectively tested in summarizes some studies using factorial design to study the effects
sequenced-batch and semicontinuous cultures) enhanced the of various factors on fermentative hydrogen production processes.
HPR and hydrogen yield when growing the strain on (NH4)2SO4 A large number of recent studies have examined parameter
(Hamilton et al., 2010). Besides, there are certain other less com- optimization using factorial design methods, and some relevant
mon parameters like concentration of Ni2+,Mg 2+, NO3 ions, which examples are given in Table 3 (others can be found elsewhere)
affects fermentative hydrogen production (Wang and Wan, 2008a; (Wang and Wan, 2009a). Several statistical design methods to opti-
Kim et al., 2006a,b; Wang et al., 2007). mize H2 yields have been reviewed by Wang and Wan (2009a).
Among the different methods, the fractional factorial design ap-
4.1.2. Response surface methodology proach is commonly used by researchers. FFD is not always eco-
Response surface methodology (RSM) is a collection of statisti- nomically and practically feasible due to requirements for a large
cal and mathematical techniques useful for developing, improving number of experiments in order to accurately predict the response.
and optimizing processes. The field of response surface methodol- Response surface designs such as central composite design (CCD)
ogy consists of the experimental strategy for exploring the space of and Box–Benkhen design (BBD) are commonly selected for per-
the process or independent variables, empirical statistical model- forming optimization studies. A first-order polynomial model is
ing to develop an appropriate approximating relationship between usually used to describe the effects of various factors on it based
the yield and the process variables, and optimization methods for on the experimental results from a Plackett–Burman design.
finding the values of the process variables that produce desirable
X
k
values of the response. On the other hand, FD investigates how re- y ¼ b0 bi xi ð29Þ
sponses of multiple factors depend on each other and permits the i¼1
identification of the most important parameters that control the
process and the degree of interaction among them. FD approaches where y is the response, 0 is the constant and i is the linear coeffi-
can be coupled to experimental systems by RSM, which uses a cient, and xi is the coded factor levels. Based on the analysis of var-
response function to fit the obtained experimental data to the iance (ANOVA) of the estimated model, the significant factors can
Table 4
Summary of optimum parameters study of fermentative hydrogen production by factorial design.
Substrate and inoculum Design method Optimum values of the Reported maximum H2 production References
factors studied rate or yield
Glucose, E. coli DJT135 3k full factorial Box–Behnken Substrate conc.: 75 mM 1.69 mol H2/mol glucose Ghosh and
model T: 35 °C, pH: 6.5 Hallenbeck (2010)
Xylose, Enterobacter sp. CN1 Plackette–Burman design Xylose conc.: 16.15 g/l 2.0 ± 0.05 mol H2/mol xylose Long et al. (2010)
FeSO4 concentration:
0.25 g/l
Peptone conc.: 2.54 g/l
Sweet sorghum, anaerobic mixed microflora Plackette–Burman design Total sugar: 25 g/L 6897 ml H2/L Saraphirom and
from sewage sludge Initial pH: 4.75 Reungsang (2010)
FeSO4: 1.45 g/L
Glucose, mixed anaerobic mesophilic culture 3k full factorial Box–Behnken Linoleic acid: 2.0 g/l H2 yield of 3.49 mol H2 mol/glucose Ray et al. (2010)
model initial pH:5.0
Glucose, anaerobic mixed microflora Full factorial central Substrate conc.: 1.75 mol H2/mol glucose Mu et al. (2009)
composite design 9.7 gL1,
T: 38.8 °C,
pH: 5.7
Glucose, Ethanoligenens harbinense B49 3k full factorial Box–Behnken Glucose conc.: 14.57 g/ H2 yield: 2.21 mol/mol glucose Guo et al. (2009)
model L,
Fe2+: 177.28 mg/L
Mg2+: 691.98 mg/L,
Glucose, Enterobacter aerogenes ATCC 29007 3k full factorial Box–Behnken Substrate conc.: 425.8 mL H2/ (g dry cell h) Jo et al. (2008)
model 118.06 mM,
T: 38 °C,
pH: 6.13
Glucose, Clostridium sp. Fanp2 Plackette-Burman design Glucose:23.75 g/l, Maximum specific H2 production Pan et al. (2008)
Phosphate buffer: 0.159 potential: 4165.9 ml H2/l
M
Vitamin solution:
13.3 ml/l.
Glucose, Mixed culture from anaerobic Three factor central T: 37.8 °C, Maximum H2 production rate of Wang and Wan
digester composite design initial pH: 7.2 Substrate 28.2 mL/h (2008c,d)
conc.: 27.6 g/L.
Sucrose RSM with a two factor central Substrate conc.:14.5 g/L H2 yield: 1.62 mol-H2/mol-hexose Zhao et al. (2008)
composite design HRT: 16.3 h
Stillages from distillery, digested sludge RSM with 23 factorial design Temp, cavitation, H2 yield:1.55 mol/mol hexose Espinoza-Escalante
from a mesophillic reactor alkalinization et al. (2008)
Thermoanaerobacterium-rich sludge RSM with a three factor C/N ratio: 74 6:33 ± 0:142 l H2/L effluent O-Thong et al.
acclimated with palm oil mill effluent central composite design C/P ratio: 559 (2008)
Fe2+ conc.: 257 mg/L
be identified (Wang and Wan, 2009b). Pan et al. (2008) used Plack- response in the direction of the maximum change toward the opti-
ett–Burman design to study the effects of eight factors on fermen- mum. However, if minimization of a response is desired, this meth-
tative hydrogen production and then screened three factors od is referred to as the method of steepest descent. The factors
(glucose, phosphate buffer and vitamin solution) that had signifi- screened by the Plackett–Burman design can be further investi-
cant effects on the specific hydrogen production potential for fur- gated using this method. The path of steepest ascent starts from
ther study based on analysis of the experimental results. the design center of the factorial design building the first-order
Saraphirom and Reungsang (2010) reported the use of the Plack- polynomial model and ends until no further improvement can be
ett–Burman design to screen and identify significant variables for achieved in the response, which indicates that the region of opti-
hydrogen production from sweet sorghum syrup by mixed cultures. mal response is in the neighborhood of that condition (Wang and
Culture parameters investigated were total sugar concentration, ini- Wan, 2009b).
tial pH, nutrient addition, iron (II) sulfate (FeSO4), peptone and so-
dium bicarbonate (NaHCO3) concentration. These variables were
optimized by the Box–Benkhen design to maximize hydrogen pro- 4.1.3. Artificial neural network-genetic algorithm
duction. In a recent study, a response surface model, based on the In last two decades, artificial neural network (ANN) has been
BBD technique was developed for computing the H2 yield for glu- applied successfully in multivariate non-linear bioprocesses as a
cose degradation by a mixed batch anaerobic mesophilic culture useful tool to construct models. It has been shown that a neural
under different experimental conditions (Ray et al., 2010). Approx- network is a superior and a more accurate modeling technique
imately 85% of the theoretical H2 yield (4.0 mol H2/mol glucose) compared with the response surface methodology, as it represents
was achieved in experiments conducted near the optimum factor the non-linearities in a much better way. The power of ANN is that
setting identified by the D-optimality analysis. it is generic in structure and possesses the ability to learn from his-
Often, the initial estimates of the optimal conditions for biohy- torical data. The main advantage of ANN compared to RSM are: (i)
drogen production process far from the actual optimum. Therefore, ANN does not require a prior specification of suitable fitting func-
the second step for optimization is to locate the region of factor tion and (ii) ANN has universal approximation capability, i.e., it can
levels that give optimal conditions. In order to approach the area approximate almost all kinds of non-linear functions including
of the optimum, the next experiment is usually carried out along quadratic functions, whereas RSM is useful only for quadratic
the path of steepest ascent. It is a simple and economically efficient approximations (Desai et al., 2008). Genetic Algorithms (GAs), an
procedure developed to move the experimental region of a artificial intelligence-based stochastic non-linear optimization
formalism is used to optimize the input space of ANN model (Da- sets were used in the study. Each data set consisted of three input
vis, 1991). This hybrid methodology is usually referred as ANN–GA variables, i.e., organic loading rate (OLR), hydraulic retention time
modeling. The GA mimics the principles of biological evolution (HRT), and influent bicarbonate alkalinity (BA), and 10 correspond-
namely, ‘‘survival-of-the-fittest’’ and ‘‘random exchange of data ing output variables, i.e., H2 concentration in the biogas, H2 produc-
during propagation’’ followed by biologically evolving species. tion rate, H2 yield, and total organic carbon (TOC), acetate,
ANNs were successfully used to model the results of hydrogen pro- propionate, butyrate, valerate, caporate in the reactor effluent. Of
duction and chemical oxygen demand (COD) removal with an up- these data sets, 100 data sets were used for training the GA–NN
flow anaerobic sludge blanket reactor (UASB) (Mu and Yu, 2007) model, whereas 40 data sets were employed for model verification.
and an expanded granular sludge bed (EGSB) reactor (Guo et al., A fractional factorial design was carried out by Wang and Wan
2008). (2009b) to investigate the effects of temperature, initial pH and
The on-line determination of hydrogen is essential to establish glucose concentration on fermentative hydrogen production by
feedback or feed forward control algorithms. However, in practice mixed cultures in batch tests. The experimental data of substrate
there are few parameters for on-line monitoring in bioreactors. degradation efficiency, hydrogen yield and average hydrogen pro-
These generally include temperature, pH, oxidation–reduction po- duction rate were described by a neural network, based on which
tential, dissolved oxygen and dissolved CO2. The ANNs based math- the simultaneous optimization of the three responses was per-
ematical models are based on the connectivity of biological formed by the method of desirability function. The method of
neurons that have an incredible capability for emulation, analysis, desirability function based on a neural network was a useful tool
prediction, association and adaptation (Hallenbeck and Ghosh, to optimize several responses for fermentative hydrogen produc-
2009; Chen et al., 2004). Therefore, ANN based models can be ap- tion processes simultaneously (Wang and Wan, 2009b).
plied successfully for the online determination and monitoring of
various fermentation products. Rosales-Colunga et al. (2010) 4.1.4. Principal component analysis
developed an artificial neural network (ANN) to estimate the The physico-chemical operating conditions determining the
hydrogen production in fermentative processes. A back propaga- biological pathways are followed inside the cell in order to have
tion neural network (BPNN) of one hidden layer with 12 nodes a good energetic balance (Aceves-Lara et al., 2008). Use of macro-
was selected. The BPNN training was done using the conjugated scopic approach like the pseudo-stoichiometric matrix provides
gradient algorithm and on-line measurements of dissolved CO2, an important method to study the biochemical pathways. The
pH and oxidation–reduction potential during the fermentations pseudo-stoichiometric matrix is the one that contains the minimal
of cheese whey by E. coli WDHL strain with or without pH con- number of reactions capable of describing biochemical pathways
trol.The model was structured as: (Bernard and Bastin, 2005). Aceves-Lara et al. (2008) estimated
the pseudostoichiometric coefficients and reaction rates with an
H2 ¼ FðpH;DCO2 ; ORP;WÞ ð30Þ optimization algorithm coupled to a principal components analysis
where, W is the vector of adjustable parameters of the network or (PCA). The number of reactions was estimated by using the PCA fol-
weight. lowed by coefficient estimation with an optimization based on
The output layer had a node that predicted the value of hydro- algorithms of Sequential Quadratic Programming (SQP). It was as-
gen production whereas the input layer consisted on three nodes sumed that pseudostoichiometric coefficients are constants and
for pH, DCO2 and ORP. All the neurons of hidden layer were non-lin- that the kinetics rates are unknown. Finally, reactions rates were
ear with sigmoid activation function. Minimal squares methodol- expressed as functions of operational conditions (hydraulic reten-
ogy with respect to error function was used for BPNN training. tion time, pH and stirring velocity) in order to compare their role
in the hydrogen production.
X
p The mass balance model of the continuous stirred tank reactor
Error ¼ ð1=ð2PÞÞ ððH2 Þiexp ðHi2 Þ2 Þ ð31Þ fed with sucrose (the main component of molasses) and producing
i
mainly acetate, propionate, butyrate, biomass, carbon inorganic
Where (H2)expi is the experimental value for the i-point, (H2)i is (i.e., CO2, HCO3, etc.) and hydrogen was written according to Eq.
the value estimated by the network, p is the number of data. Wang (32):
and Wan (2009b) compared ANN and RSM methodologies for their 2 3 2 3 2 3
predictive and generalization capabilities, sensitivity analysis and Suc Suc Sucin 0
optimization efficiency in fermentative hydrogen production. 6 Ace 7 6 Ace 7 6 7
6 7 6 7 6 0 7
ANN showed better accuracy and generalization capability than
6 7 6 7 6 7
6 Pro 7 6 Pro 7 6 0 7
RSM even with limited number of experiments. The root mean d6 6
7
7
6
6
7 6
7 6
7
7
6 Bu 7 ¼ K:r D6 Bu 76 0 7 ð32Þ
square error (RMSE) and the standard error of prediction (SEP) dt 6 7 6 7 6 7
6 X 7 6 X 7 6 7
for the RSM and NN models indicates that the RMSE (38.4%) and 6 7 6 7 6 0 7
6 7 6 7 6 7
SEP (16.6%) for the RSM model were both much larger than those 4 Ci 5 4 Ci 5 4 qCi 5
(17.8% and 7.7%, respectively) for the NN model, also indicating H2 H2 qH2
that the NN model had a much higher modeling ability than the
RSM model (Wang and Wan, 2009a). The maximum hydrogen where Suc, Ace, Pro, Bu, X, Ci and H2 represent, respectively, the
yield of 289.8 mL/g glucose identified by response surface method- concentrations in g/L of sucrose, acetate, propionate, butyrate, bio-
ology was a little lower than that of 360.5 mL/g glucose identified mass, inorganic carbon and dissolved hydrogen in the liquid phase.
by the genetic algorithm based on a neural network model, indicat- K represents the matrix of pseudo stoichiometric coefficients (with-
ing that the genetic algorithm based on a neural network model out units and dimension equal to m n), r the vector of the kinetics
had a much higher optimizing ability than the response surface (dimension n 1 in g/L d), D the dilution rate (d1) and qCi and qH2
methodology. the gas flow rates of inorganic carbon and hydrogen expressed in g/
Mu and Yu (2007) reported the simulation of a granule-based L d. Under steady state condition the left hand side term of Eq. (32)
H2-producing upflow anaerobic sludge blanket (UASB) reactor is cancelled and Eq. (33) is obtained defining the vector u
using neural network and genetic algorithm. A model was de- (dim = m 1):
signed, trained and validated to predict the steady-state perfor-
mance of the reactor. A total of 140 experimentally derived data K r ¼u ð33Þ
Where intermediary products. The application of a desirability function

2 3 2 3 combines all the responses into one measurement and gives the
Suc Sucin 0
possibility of predicting the optimum levels for the independent
6 Ace 7 6 0 7
6 7 6 7 variable. The combination of the responses into one desirability
6 7 6 7
6 Pro 7 6 0 7 function requires the calculation of each individual desirability
6 7 6 7
6 7 6 7 function. Responses were maximized and minimized according to
U ¼ D:6 Bu 7þ6 0 7 ð34Þ
6 7 6 7 the Eqs. (36) and (37), respectively.
6 X 7 6 0 7
6 7 6 7
6 7 6 7
4 Ci 5 4 qCi 5 Y i Y min
di ¼ ð36Þ
H2 qH2 Y max Y min
Aceves-Lara et al. (2008) carried out the experiments at 37 °C
Y max Y i
using a continuous stirred tank reactor fed with molasses. Principal di ¼ ð37Þ
component analysis method was used to determine the main reac- Y max Y min
tions implied in the degradation of sucrose into acetate, propio- The overall desirability values were calculated from the individ-
nate, butyrate, hydrogen, inorganic carbon and biomass. Two ual values by using the following equation for n responses:
reactions explained 89% of the experimental variance. The estima-
tion of the pseudo-stoichiometric coefficients, using SQP optimiza- D ¼ ðd1 d2 :::::dn Þ1=n ð38Þ
tion, showed that the first reaction mainly produced butyrate, 2
A 3 central composite design was used to determine the effect
hydrogen and biomass whereas the second reaction mainly pro-
of the solid concentration of the substrate, the proportion of waste
duced acetate and biomass. HRT ranged from 6 to 14 h, pH from
in the feed and the hydraulic retention over the responses selected
5.5 to 6 and the stirring velocity from 150 to 300 rpm. The maxi-
(Cuetos et al., 2007). Optimization of the process using a desirabil-
mum hydrogen production of 5.4 L/d L was obtained for a pH of
ity function for dark fermentation phase coupled with a second
5.5, a HRT of 6 h and a stirring velocity of 300 rpm, at a loading rate
methanogenic phase resulted in a reduction in the production of
of 37.12 g COD/d L, which corresponds to 2.5 mol H2/mol sucrose.
hydrogen when the generation of intermediary products was max-
Only linear effects of pH, HRT and stirring velocity were significant
imized. Hydrogen production fell from 0.97 L/L d when the opti-
and not their interactions or quadratic effects. Aceves-Lara et al.,
mum point was based on hydrogen maximization, to 0.25 L/L d
2010 presented an approach of dynamic optimization of hydrogen
when maximization of soluble products was also considered. Ding
production from an agricultural effluent by combining an optimal
et al. (2010) reported three-dimensional computational fluid
closed-loop control together with state and input variable estima-
dynamics (CFD) simulations of gas–liquid flow in a laboratory scale
tions by an asymptotic observer. A general mass balance model of a
continuous stirred tank reactor for biohydrogen production. Here,
continuous stirred tank reactor fed with carbohydrates was used in
the CFD modeling predicted the hydrodynamics, heat and mass
their study. The glucose metabolism performed by a single micro-
transfer and chemical reactions by solving a set of mathematical
organism producing acetate, propionate, butyrate, biomass, carbon
equations, describing these processes as mass, momentum, energy
inorganic (i.e., CO2, HCO
3 etc.) and hydrogen was written according
and species balances. The CFD simulation was used to portray
to the Eq. (35)
hydrodynamics behavior in the reactor, including the velocity field,
2 3 2 3 2 3 2 3
Glu Glu Gluin 0 biogas volume fraction, turbulence kinetic energy and shear strain
6 7 6 7 6 7 6 0 7 rate. The turbulence equations were solved in conjunction with the
6 Ace 7 6 Ace 7 6 0 7 6 7
6 7 6 7 6 7 6 7 continuity equation, the Navier–Stokes equation and the energy
6 Pro 7 6 Pro 7 6 0 7 6 0 7
d6 6
7
7
6
6
7
7
6
6
7 6
7 6 0 7
7 equation (Ding et al., 2010). The inhomogeneous particle model
6 Bu 7 ¼ Y:r D6 Bu 7 þ D6 0 7 6 7 ð35Þ used in the simulation assumed that the mixture was continuous
dt 6 7 6 7 6 7 6 7
6 X 7 6 X 7 6 0 7 6 0 7 phase whereas biogas was dispersed phase. The study revealed
6 7 6 7 6 7 6 7
6 7 6 7 6 7 6 7 that impeller type and speed significantly affected flow patters in
4 CO2 5 4 CO2 5 4 0 5 4 qCO2 5
H2 H2 0 qH2 CSTR, thereby offering different optimal efficiencies for biohydro-
gen production.
where the state variables Glu, Ace, Pro, Bu, X, CO2 and H2 represent,
respectively, the concentrations in g L1of glucose, acetate, propio-
5. Conclusion
nate, butyrate, biomass, carbon dioxide (in mol L1) and dissolved
hydrogen in the liquid phase. The vector r = [r1 r2]t describes
This review highlights the recent studies on modeling of micro-
the kinetics of the involved biological reactions (in g L1 h1), D is
bial growth, substrate utilization and product formation along with
the dilution rate (h1) and qCO2 and qH2 the gas flow rates of carbon
various optimization strategies. It is worthwhile to note that any
dioxide and hydrogen expressed in g L1 h1. Y represents the ma-
biohydrogen process depends largely on optimization of several
trix of pseudo-stoichiometric coefficients. The process was de-
controlling factors. This is only possible through modeling of the
scribed by a dynamic nonlinear model. The influent concentration
factors that determine process rate. Notwithstanding the fact that
of molasses together with the effluent substrate and product con-
all mesophilic fermentations require similar conditions, there are
centrations of acetate, propionate, butyrate and biomass were esti-
sufficient differences in some of the parameters. This substantiates
mated by an asymptotic online observer from measurements of gas
the need for optimization of a particular system. In addition, these
composition in H2 and CO2 and gas flow rate. The observer was
methods could prove helpful in establishing process parameters for
tested experimentally before applying model predictive control
novel consortia or complex substrates like effluent streams.
strategy (Aceves-Lara et al., 2010).
Cuetos et al. (2007) employed a desirability function in order to
optimize the fermentative biohydrogen from municipal solid References
wastes and blood from a poultry slaughterhouse when a second,
methanogenic, phase is coupled with it. In this last case, the opti- Aceves-Lara, C.A., Latrille, E., Buffiere, P., Bernet, N., Steyer, J.-P., 2008. Experimental
determination by principal component analysis of a reaction pathway of
mum conditions lead to a reduction in the production of hydrogen biohydrogen production by anaerobic fermentation. Chem. Eng. Proc. 47, 1968–
when the optimization process involves the maximization of 1975.
Aceves-Lara, C.A., Latrille, E., Buffiere, P., Bernet, N., Steyer, J.-P., 2010. Optimal Karlsson, A., Vallin, L., Ejlersson, J., 2008. Effects of temperature, hydraulic retention
control of hydrogen production in a continuous anaerobic fermentation time and hydrogen extraction rate on hydrogen production from the
bioreactor. Int. J. Hydrogen Energy 35, 10710–10718. fermentation of food industry residues and manure. Int. J. Hydrogen Energy
Bartacek, J., Zabranska, J., Lens, P.N.L., 2007. Developments and constraints in 33, 953–962.
fermentative hydrogen production. Biofuels, Bioprod. Bioref. 1, 201–214. Khanna, N., Kotay, S.M., Gilbert, J.J., Das, D., 2011. Improvement of biohydrogen
Batstone, D.J., Keller, J., Angelidaki, I., Kalyuzhnyi, S.V., Pavlostathis, S.G., Rozzi, A., production by Enterobacter cloace IIT-BT08 under regulated pH. J. Biotechnol.
2002. Anaerobic digestion model No. 1. IWA Scientific and Technical Report No. 152, 9–15.
13, London. Kim, M.-S., Lee, D.-Y., 2010. Fermentative hydrogen production from tofu-
Bernard, O., Bastin, G., 2005. On the estimation of the pseudo-stoichiometric matrix processing waste and anaerobic digester sludge using microbial consortium.
for macroscopic mass balance modeling of biotechnological processes. Math. Bioresour. Technol. 101, S48–S52.
Biosci. 193, 51–77. Kim, S.H., Han, S.K., Shin, H.S., 2006a. Effect of substrate concentration on hydrogen
Bowles, L.K., Ellefson, W.L., 1985. Effects of butanol on Clostridium acetobutylicum. production and16S rDNA-based analysis of the microbial community in a
Appl. Environ. Microbiol. 50, 1165–1170. continuous fermenter. Process Biochem. 41, 199–207.
Chen, L., Nguang, S., Li, X., Chen, X., 2004. Soft sensors for on-line biomass Kim, J.O., Kim, Y.H., Yeom, S.H., Song, B.K., Kim, I.H., 2006b. Enhancing continuous
measurements. Bioprocess Biosyst. Eng. 26, 191–195. hydrogen gas production by the addition of nitrate into an anaerobic reactor.
Chen, W.H., Chen, S.Y., Khanal, S.K., Sung, S., 2006a. Kinetic study of biological Process Biochem. 41, 1208–1212.
hydrogen production by anaerobic fermentation. Int. J. Hydrogen Energy 31 Lee, K.-S., Hsu, Y.-F., Lo, Y.-C., Lin, P.-J., Lin, C.-Y., Chang, J.-S., 2008. Exploring optimal
(15), 2170–2178. environmental factors for fermentative hydrogen production from starch using
Chen, W.-S., Chen, S.-Y., Khanal, S.K., Sung, S., 2006b. Kinetic study of biological mixed anaerobic microflora. Int. J. Hydrogen Energy 33, 1565–1572.
1053 hydrogen production by anaerobic fermentation. Int. J. Hydrogen Energy Li, Y., Zhu, J., Wu, X., Miller, C., Wang, L., 2010. The effect of pH on continuous
31, 1054 2170-2178. biohydrogen production from swine waste water supplemented with glucose.
Chittibabu, G., Nath, K., Das, D., 2006. Feasibility studies on the fermentative Appl. Biochem. Biotechnol. 162, 1286–1296.
hydrogen production by recombinant Escherichia coli BL-21. Proc. Biochem. 41, Lin, P.Y., Whang, L.M., Wu, Y.R., Ren, W.J., Hsiao, C.J., Li, S.L., 2007. Biological
682–688. hydrogen production of the genus Clostridium: metabolic study and
Classen, P.A.M., van Lier, J.B., Lopez Contreras, A.M., van Niel, E.W.J., Sijtsma, L., mathematical model simulation. Int. J. Hydrogen Energy 32 (12), 1728–1735.
Stams, A.J.M., de Vries, S.S., Weusthuis, R.A., 1999. Utilisation of biomass for the Lin, C.-Y., Chang, C.-C., Hung, C.-H., 2008a. Fermentative hydrogen production from
supply of energy carriers. Appl. Microbiol. Biotechnol. 52, 741–755. starch using natural mixed cultures. Int. J. Hydrogen Energy 33, 2445–2453.
Cooney, M., Maynard, N., Cannizzaro, C., Benemann, J., 2007. Two phase anaerobic Lin, C.Y., Wu, C.C., Hung, C.H., 2008b. Temperature effects on fermentative hydrogen
digestion for production of hydrogen methane mixtures. Bioresour. Technol. 98, production from xylose using mixed anaerobic cultures. Int. J. Hydrogen Energy
2641–2651. 33, 43–50.
Cuetos, M.J., Ǵomez, X., Escapa, A., Moŕan, A., 2007. Evaluation and simultaneous Liu, D.W., Liu, D.P., Zeng, R.J., Angelidaki, I., 2006. Hydrogen and methane
optimization of bio-hydrogen production using 32 factorial design and the production from household solid waste in the two-stage fermentation
desirability function. J. Power Sour. 169, 131–139. process. Water Res. 40, 2230–2236.
Das, D., Veziroglu, T.N., 2001. Hydrogen production by biological processes: a Lo, Y.C., Chen, W.M., Hung, C.H., Chen, S.D., Chang, J.S., 2008. Dark H2 fermentation
survey of literature. Int. J. Hydrogen Energy 26, 13–28. from sucrose and xylose using H2-producing indigenous bacteria: feasibility
Davila-Vazquez, G., Alatriste-Mondragón, F., León-Rodrı́guez, A., Razo-Flores, E., and kinetic studies. Water Res. 42, 827–842.
2008. Fermentative hydrogen production in batch experiments using lactose, Long, C., Cui, J., Liu, Z., Liu, Y., Long, M., Hu, Z., 2010. Statistical optimization of
cheese whey and glucose: Influence of initial substrate concentration and pH. fermentative hydrogen production from xylose by newly isolated Enterobacter
Int. J. Hydrogen Energy 33, 4989–4997. sp. CN1. Int. J. Hydrogen Energy 35, 6657–6664.
Davis, L., 1991. Handbook of Genetic Algorithms. Van Nostrand Reinhold, New York. Luo, G., Xie, L., Zou, Z., Wang, W., Zhou, Q., 2009. Evaluation of pretreatment
Desai, K.M., Survase, S.A., Saudagar, P.A., Lele, S.S., Singhal, R.S., 2008. Comparison of methods on mixed inoculum for both batch and continuous thermophilic
artificial neural network (ANN) and response surface methodology (RSM) in biohydrogen production from cassava stillage. Bioresour. Technol. 101, 959–
fermentation media optimization: case study of fermentative production of 964.
scleroglucan. Biochem. Eng. J. 41, 266–273. Luo, G., Xie, L., Zou, Z., Wang, W., Zhou, Q., 2010. Exploring optimal conditions for
Ding, J., Wang, X., Zhou, X.-F., Ren, N.-Q., Guo, W.-Q., 2010. CFD optimization of thermophilic fermentative hydrogen production from cassava stillage. Int. J.
continuous stirred-tank (CSTR) reactor for biohydrogen production. Bioresour. Hydrogen Energy 35, 6161–6169.
Technol. 101, 7005–7013. Maddy, J., Cherryman, S., Hawkes, F.R., Hawkes, D.L., Dinsdale, R.M., Guwy, A.J.,
Espinoza-Escalante, F.M., Pelayo-Ortiz, C., Humberto, G.P., Vıćtor, G.A., Vıćtor, A.G., Premier, G.C., Cole, S., 2003. HYDROGEN 2003 Report Number 1 ERDF part-
Bories, A., 2008. Multiple response optimization analysis for pretreatments of funded project entitled: ‘‘A Sustainable Energy Supply for Wales: Towards the
Tequila’s stillages for VFAs and hydrogen production. Bioresour. Technol. 99, Hydrogen Economy’’, University of Glamorgan.
5822–5829. Mu, Y., Yu, H.-Q., 2007. Simulation of biological hydrogen production in a UASB
Gadhamshetty, V., Arudchelvam, Y., Nirmalakhandan, N., Johnson, D.C., 2010. reactor using neural network and genetic algorithm, Int. J Hydrogen Energy. 32,
Modeling dark fermentation for biohydrogen production: ADM1-based model 3308–3314.
vs. Gompertz model. Int. J. Hydrogen Energy 35, 479–490. Mu, Y., Wang, G., Yu, H.-Q., 2006a. Kinetic modeling of batch hydrogen production
Ghosh, D., Hallenbeck, P.C., 2010. Response surface methodology for process process by mixed anaerobic cultures. Bioresour. Technol. 97, 1302–1307.
parameter optimization of hydrogen yield by the metabolically engineered Mu, Y., Zheng, X.J., Yu, H.Q., Zhu, R.F., 2006b. Biological hydrogen production by
strain Escherichia coli DJT135. Bioresour. Technol. 101, 1820–1825. anaerobic sludge at various temperatures. Int. J. Hydrogen Energy 31, 780–785.
Guo, W.-Q., Ren, N.-Q., Chen, Z.-B., Liu, B.-F., Wang, X.-J., Xiang, W.-S., Ding, J., 2008. Mu, Y., Zheng, X.-J., Yu, H.-Q., 2009. Determining optimum conditions for hydrogen
Simultaneous biohydrogen production and starch wastewater treatment in an production from glucose by an anaerobic culture using response surface
acidogenic expanded granular sludge bed reactor by mixed culture for long- methodology (RSM). Int. J. Hydrogen Energy 34, 7959–7963.
term operation. Int. J. Hydrogen Energy 33, 7397–7404. Nandi, R., Sengupta, S., 1998. Microbial production of hydrogen: an overview. Crit.
Guo, W.-Q., Ren, N.-Q., Wang, X.-J., Xiang, W.-S., Ding, J., You, Y., Liu, B.-F., 2009. Rev. Microbiol. 24, 61–84.
Optimization of culture conditions for hydrogen production by Ethanoligenens Nath, K., Das, D., 2004. Improvement of fermentative hydrogen production various
harbinense B49 using response surface methodology. Bioresour. Technol. 100, approaches. Appl. Microbiol. Biotechnol. 65, 520–529.
1192–1196. Nath, K., Kumar, A., Das, D., 2006. Effect of some environmental parameters on
Hallenbeck, P.C., Benemann, J.R., 2002. Biological hydrogen production; fermentative hydrogen production by Enterobacter cloacae DM11. Can. J.
fundamentals and limiting processes. Int. J. Hydrogen Energy 27, 1185–1193. Microbiol. 52, 525–532.
Hallenbeck, P.C., Ghosh, D., 2009. Advances in fermentative biohydrogen Nath, K., Muthukumar, M., Kumar, A., Das, D., 2008. Kinetics of two-stage
production: the way forward? Trends Biotechnol. 27, 287–297. fermentation process for the production of hydrogen. Int. J. Hydrogen Energy
Hamilton, C., Hiligsmann, S., Beckers, L., Masset, J., Wilmotte, A., Thonart, P., 2010. 33, 1195–1203.
Optimization of culture conditions for biological hydrogen production by Ntaikou, I., Gavala, H.N., Kornaros, M., Lyberatos, G., 2008. Hydrogen production
Citrobacter freundii CWBI952 in batch, sequenced-batch and semicontinuous from sugars and sweet sorghum biomass using Ruminococcus albus. Int. J.
operating mode. Int. J. Hydrogen Energy 35, 1089–1098. Hydrogen Energy 33, 1153–1163.
Han, K., Levenspiel, O., 1988. Extended monod kinetics for substrate, product, and Ntaikou, I., Gavala, H.N., Lyberatos, G., 2009a. Modeling of fermentative hydrogen
cell inhibition. Biotechnol. Bioeng. 32 (4), 430–437. production from the bacterium Ruminococcus albus: definition of metabolism
Hawkes, F.R., Hussy, I., Kyazze, G., Dinsdale, R., Hawkes, D.L., 2007. Continuous dark and kinetics during growth on glucose. Int. J. Hydrogen Energy 34 (4), 3697–
fermentative hydrogen production by mesophilic microflora: principles and 3709.
progress. Int. J. Hydrogen Energy 32 (2), 172–184. Ntaikou, I., Koutros, E., Kornaros, M., 2009b. Valorisation of waste paper using the
JianLong, W., Wei, W., 2008. The effect of substrate concentration on biohydrogen fibrolytic/hydrogen producing bacterium Ruminococcus albus. Bioresour.
production by using kinetic models. Sci. China Ser. B: Chem. 51, 1110–1117. Technol. 100, 5928–5933.
Jo, J.H., Lee, D.S., Park, D., Choe, W.-S., Park, J.M., 2008. Optimization of key process Obeid, J., Magnin, J.P., Flaus, J.M., Adrot, O., Willison, J.C., Zlatev, R., 2009. Modeling
variables for enhanced hydrogen production by Enterobacter aerogenes using of hydrogen production in batch cultures of the photosynthetic bacterium
statistical methods. Bioresour. Technol. 99, 2061–2066. Rhodobacter capsulatus. Int. J. Hydrogen Energy 34, 180–185.
Jung, K.-W., Kim, D.-H., Shin, H.-S., 2010. Continuous fermentative hydrogen Obeid, J., Flaus, J.-M., Adrot, O., Magnin, J.-P., Willison, J.-C., 2010. State estimation of
production from coffee drink manufacturing wastewater by applying UASB a batch hydrogen production process using the photosynthetic bacteria
reactor. Int. J. Hydrogen Energy 35, 13370–13378. Rhodobacter capsulatus. Int. J. Hydrogen Energy 35, 10719–10724.
Okpokwasili, G.C., Nweke, C.O., 2005. Microbial growth and substrate utilization Sittijunda, S., Reungsang, A., O-Thong, A., 2010. Biohydrogen production from dual
kinetics. Afr. J. Biotechnol. 5, 305–317. digestion pretreatment of poultry slaughterhouse sludge by anaerobic self-
O-Thong, S., Prasertsan, P., Karakashev, D., Angelidaki, I., 2008. Thermophilic fermentation. Int. J. Hydrogen Energy 35, 13427–13434.
fermentative hydrogen production by the newly isolated Wang, X., Jin, B., 2009. Process optimization of biological hydrogen production from
Thermoanaerobacterium thermosaccharolyticum PSU-2. Int. J. Hydrogen Energy molasses by a newly isolated Clostridium butyricum W5. J. Biosci. Bioeng. 107
33 (4), 1204–1214. (2), 138–144.
Pan, C.M., Fan, Y.T., Xing, Y., Hou, H.W., Zhang, M.L., 2008. Statistical optimization of Wang, J.L., Wan, W., 2008a. Influence of Ni2+ concentration on biohydrogen
process parameters on biohydrogen production from glucose by Clostridium sp. production. Bioresour. Technol. 99, 8864–8868.
Fanp2. Bioresour. Technol. 99, 3146–3154. Wang, J.L., Wan, W., 2008b. Effect of Fe2+ concentrations on fermentative hydrogen
Penumathsa, B.K.V., Premier, G.C., Kyazze, G., Dinsdale, R., Guwy, A.J., Esteves, S., production by mixed cultures. Int. J. Hydrogen Energy 33, 1215–1220.
Rodrı́guez, J., 2008. ADM1 can be applied to continuous bio-hydrogen Wang, J.L., Wan, W., 2008c. The effect of substrate concentration on biohydrogen
production using a variable stoichiometry approach. Water Res. 42, 4379– production by using kinetic models. Sci. China Ser. B 51, 1110–1117.
4385. Wang, J., Wan, W., 2008d. Optimization of fermentative hydrogen production
Prasertsan, P., O-Thong, S., Birkeland, N.-K., 2009. Optimization and microbial process by response surface methodology. Int. J. Hydrogen Energy 33, 6976–
community analysis for production of biohydrogen from palm oil mill 6984.
effluent by thermophilic fermentative process. Int. J. Hydrogen Energy 34, Wang, J., Wan, W., 2009a. Optimization of fermentative hydrogen production
7448–7459. process using genetic algorithm based on neural network and response surface
Ray, S., Reaume, S.J., Lalman, J.A., 2010. Developing a statistical model to predict methodology. Int. J. Hydrogen Energy 34, 255–261.
hydrogen production by a mixed anaerobic mesophilic culture. Int. J. Hydrogen Wang, J., Wan, W., 2009b. Application of desirability function based on neural
Energy 35, 5332–5342. network for optimizing biohydrogen production process. Int. J. Hydrogen
Rosales-Colunga, L.M., Garcia, R.G., Rodriguez, A.D.L., 2010. Estimation of hydrogen Energy 34, 1253–1259.
production in genetically modified E. coli fermentations using an artificial Wang, X., Zhao, Y.-C., 2009. A bench scale study of fermentative hydrogen and
neural network. Int. J. Hydrogen Energy 35, 13186–13192. methane production from food waste in integrated two-stage process. Int. J.
Rupprecht, J., 2009. From systems biology to fuel—Chlamydomonas reinhardtii as a Hydrogen Energy 34, 245–254.
model for a systems biology approach to improve biohydrogen production. J. Wang, X.J., Ren, N.Q., Xiang, W.S., Guo, W.Q., 2007. Influence of gaseous end-
Biotechnol. 142, 10–20. products inhibition and nutrient limitations on the growth and hydrogen
Saraphirom, P., Reungsang, A., 2010. Optimization of biohydrogen production from production by hydrogen-producing fermentative bacterial B49. Int. J. Hydrogen
sweet sorghum syrup using statistical methods. Int. J. Hydrogen Energy 35, Energy 32, 748–754.
13435–13444. Whang, L.-M., Hsiao, C.-J., Cheng, S.-S., 2006. A dual-substrate steady-state model
Sharma, Y., Li, B., 2009. Optimizing hydrogen production from organic wastewater for biological hydrogen production in an anaerobic hydrogen fermentation
treatment in batch reactors through experimental and kinetic analysis. Int. J. process. Biotech. Bioeng. 95, 492–500.
Hydrogen Energy 34, 6171–6180. Zhao, B.-H., Yue, Z.-B., Zhao, Q.-B., Mu, Y., Yu, H.-Q., Harada, H., Li, Y.-Y., 2008.
Shuler, L.M., Kargi, F., 2002. Bioprocess engineering—basic concepts, second ed. New Optimization of hydrogen production in a granule-based UASB reactor. Int. J.
Delhi, Prentice-Hall of India, pp. 161–79. Hydrogen Energy 33, 2454–2461.

Modeling and Optimization of Fermentativ

Uploaded by

Copyright:

Available Formats

Modeling and Optimization of Fermentativ

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Modeling and Optimization of Fermentativ

Uploaded by

Copyright:

Available Formats

Bioresource Technology 102 (2011) 8569–8581

Contents lists available at ScienceDirect

Modeling and optimization of fermentative hydrogen production

1. Introduction concepts led to a proven and long-term functioning industrial scale

S also revealed that inhibition of the fermentative hydrogen

dV H2 1 equation was applied to estimate the hydrogen production poten-

The organic loading rate (kg VS/m3 d) For peptone

S Method Principle Relative merits and demerits

Where intermediary products. The application of a desirability function

You might also like