Bioresource Technology 118 (2012) 89101

Contents lists available at SciVerse ScienceDirect

Bioresource Technology
journal homepage: www.elsevier.com/locate/biortech

Integration of kinetic modeling and desirability function approach

for multi-objective optimization of UASB reactor treating poultry
manure wastewater
Kaan Yetilmezsoy
Yildiz Technical University, Faculty of Civil Engineering, Department of Environmental Engineering, 34220 Davutpasa, Esenler, Istanbul, Turkey

h i g h l i g h t s g r a p h i c a l a b s t r a c t

" A holistic approach is proposed for

multi-objective optimization of
mesophilic UASB.
" ChenHashimoto and Stover
Kincannon models are integrated on
a composite surface.
" A bio-economic model including
several technical and
thermophysical data is derived.
" Inhibition parameter is derived
empirically for the rst time for
poultry manure.
" The methodology demonstrated a
useful tool with an overall
desirability of 0.896.

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

Article history: An integrated multi-objective optimization approach within the framework of nonlinear regression-
Received 7 March 2012 based kinetic modeling and desirability function was proposed to optimize an up-ow anaerobic sludge
Received in revised form 17 May 2012 blanket (UASB) reactor treating poultry manure wastewater (PMW). ChenHashimoto and modied
Accepted 18 May 2012
StoverKincannon models were applied to the UASB reactor for determination of bio-kinetic coefcients.
Available online 26 May 2012
A new empirical formulation of volumetric organic loading rate was derived for the rst time for PMW to
estimate the dimensionless kinetic parameter (K) in the ChenHashimoto model. Maximum substrate
Keywords:
utilization rate constant and saturation constant were predicted as 11.83 g COD/L/day and 13.02 g
Desirability function
Kinetic modeling
COD/L/day, respectively, for the modied StoverKincannon model. Based on four process-related vari-
Multi-objective optimization ables, three objective functions including a detailed bio-economic model were derived and optimized
Poultry manure wastewater by using a LOQO/AMPL algorithm, with a maximum overall desirability of 0.896. The proposed optimiza-
Up-ow anaerobic sludge blanket reactor tion scheme demonstrated a useful tool for the UASB reactor to optimize several responses
simultaneously.
2012 Elsevier Ltd. All rights reserved.

1. Introduction eutrophication of surface water resources, groundwater contami-

Poultry manure-related problems are one of the potential
sources of many major environmental problems resulting several
nuisance consequences (i.e. odor problems, release of pathogens,
Tel.: +90 212 383 5376; fax: +90 212 383 5358.
E-mail addresses: yetilmez@yildiz.edu.tr, kyetilmezsoy@gmail.com
nations, surface water runoff, deterioration of biological structure
of the earth, etc.) to the environment and human health. The pro-
duction of a huge amount of poultry wastes, particularly in concen-
trated areas, has become one of the most critical environmental
concerns in recent years (Yetilmezsoy, 2008; Yetilmezsoy and
Sakar, 2008a, 2008b). For this reason, urgent treatment and dis-
posal solutions have been investigated to manage improperly

0960-8524/$ - see front matter 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.biortech.2012.05.088
90 K. Yetilmezsoy / Bioresource Technology 118 (2012) 89101
managed poultry and livestock wastes/wastewaters using different
process congurations (Sakar et al., 2009).
Anaerobic digestion has been reported as a benecial, feasible
and advantageous process in treatment and disposal of poultry
and livestock wastes (Sakar et al., 2009). In particular, this technol-
ogy have been successfully applied to treat a variety of solid,
semi-solid and liquid poultry waste with the use of different reac-
tor congurations, such as mesophilic digestion of poultry manure
slurry using up-ow anaerobic sludge blanket (UASB) reactor
(Yetilmezsoy, 2008) and dry fermentation of chicken manure using
batch type digester (Abouelenien et al., 2009). Although anaerobic
digestion has been proven as an established biotechnological
method for degrading a variety of polluting organic wastes, how-
ever, it is a complex process requiring careful operation and mon-
itoring the several parameters, such as pH, alkalinity, temperature,
volatile fatty acids, organic loading rate, hydraulic retention times,
biogas composition, etc. For this purpose, development of inte-
grated solutions is worthy of investigation for the optimization
and simulation of complex anaerobic digestion processes in a
straightforward manner.
Kinetic models can be used to control and predict the treatment
plant performance, to optimize the plant design and to determine
the importance of relationships between variables in fundamental
research of anaerobic processes (Isik and Sponza, 2005). These
models can also provide a good insight into reaction mechanisms
by reducing laborious and complex experimental data to simple
and convenient mathematical expressions (Turkdogan-Aydinol
et al., 2010). Therefore, a growing number of publications have re-
ported the application of several kinetic models in the anaerobic
digestion of various waste streams which have different character-
istics and higher contents of solids. So far, numerous kinetic mod-
els (i.e. Monod-type kinetic, Grau second-order multicomponent,
StoverKincannon, ChenHashimoto, Contois, Haldane, pseudo-
mechanistic, LineweaverBurk (Sundstrom model), Barthakur,
Dewalle and Chian, Young and McCarty, Benjes, etc.) have been
introduced in anaerobic digestion to explore biokinetics and to
determine the importance of relationships between the design
data and experimental results (Yetilmezsoy, 2008).
Several factors such as operating temperature, initial pH, organ-
ic and hydraulic loading rates and substrate concentration can sig-
nicantly inuence the stability of anaerobic digestion process,
because they can directly affect the activity of microbial consor-
tium. For this reason, optimal ranges of process-related parameters
should be chosen in design and operation of anaerobic digesters to
increase the ability of bacterial population and to produce methane
as a valuable source of renewable energy. Therefore, the multi-var-
iate optimization of complicated biosystems has been received
increasing attention in recent years (Wang and Wan, 2009).
The performance of a biosystem is associated with a set of sev-
eral process variables, which are dened operationally and exper-
imentally, often interrelated with different signicance and
measured in different scale of units. At this point, the concept of
desirability scale arises when there is a need to combine the mag-
nitude of several characteristics to a dimensionless scale (Das et al.,
2009). The method of desirability function, as one of the most
important multi-response optimization techniques, can be used
as a useful tool to optimize several responses simultaneously when
there are many responses to be optimized for a bioprocess (Wang
and Wan, 2009). It must also be remembered that an appropriate
modeling of the process data often becomes a prerequisite for
the simultaneous optimization of multiple responses by the meth-
od of desirability function (Wang and Wan, 2009; Das et al., 2009).
Recently, this methodology has been successfully utilized to
explore the optimal process conditions for several challenging
problems, such as simultaneous analysis of neurotoxicological
endpoints (Coffey et al., 2007), optimization of chemical admixture
for concrete design (Akalin et al., 2010), and optimization of a
microemulsion liquid chromatographic system (Malenovic et al.,
2011).
To the best of the authors knowledge, however, there are al-
most no papers in the scientic literature specically devoted to
a study of an integrated multi-objective optimization approach
based on nonlinear regression-based kinetic modeling and desir-
ability function for the optimization of UASB reactor treating poul-
try manure wastewater (PMW). Thus, in this study, bio-kinetics of
the present digestion process were rst explored by using two of
the most widely used kinetic models (ChenHashimoto and mod-
ied StoverKincannon), and then the several responses were opti-
mized simultaneously by the method of desirability function, with
the purpose of obtaining the optimal conditions for the methane
production rate, the efuent substrate concentration, and the net
operating cost.
2. Methods
2.1. Summary of previous ndings
The experimental methodology including acclimation step, feed
preparation and characteristics, seed sludge, basal medium, re-
agents, equipments, sample collection, chemical analyses, reactor
conguration, gas collection system, and other technical and
operational details was fully documented in the authors previous
studies on the anaerobic treatability of high strength PMW
(Yetilmezsoy, 2008; Yetilmezsoy and Sakar, 2008a, 2008b;
Yetilmezsoy et al., 2009). For this numerical study, technical and
operational data about the UASB system treating PMW are summa-
rized in Table 1.
On the basis of the steady-state data obtained from the previous
work (Yetilmezsoy, 2008), in the present study, an integrated mul-
ti-objective optimization using kinetic modeling, nonlinear regres-
sion methodology, cost analysis and desirability function method
was proposed as a straightforward approach to optimize a specic
digestion process with multiple responses.
2.2. Kinetic modeling
In the present study, ChenHashimoto and modied Stover
Kincannon models were implemented to the UASB reactor for
determination of bio-kinetic coefcients. The proposed models
were formulated based on the following hypothesis and consider-
ations: (i) the reactor was completely mixed during the feeding
and operated in a daily-continuous mode without solids recycle,
(ii) no microbial solids were contained in the inuent substrate,
(iii) the effective volume of the reactor must constant, (iv) hydrau-
lic retention time hh is dened as the volume of the reactor divided
by the ow rate of the inuent and (v) steady-state conditions
prevail throughout the system. Implementation of the above-men-
tioned kinetic models in the proposed multi-objective optimiza-
tion scheme is discussed in brief in the following sections.
2.3. ChenHashimoto model
For a completely mixed reactor without solids recirculating,
Chens and Hashimotos equation is expressed as follows (Videla
et al., 1990; Yetilmezsoy, 2008):
   
1  RY m Si K
YV 1 1
hh hh lm  1 K
where YV is the volume of methane produced per unit volume of di-
gester per unit time (L CH4/L/day), R is the fraction of refractory
(non-biodegradable) COD, Ym is the volume of methane produced
 K. Yetilmezsoy / Bioresource Technology 118 (2012) 89101 91

Table 1 K cattle 0:8 0:016 exp0:060 VS0 3

Technical and operational data about the UASB system treating poultry manure
wastewater (Yetilmezsoy, 2008).
Technical and operational
constituents
Material of the reactor
Total tank capacity and working
volume
Internal diameter and total height
Cross-sectional area of the reactor
Location of sampling ports from the
bottom
Diameter of sampling ports and GSL
separator
Operating temperature
Feeding mode
Total period of operation
Volume of seed sludge in the reactor
Inuent COD concentration
Volumetric organic loading rate
(OLR)
Hydraulic loading rates
Volumetric COD removal rate
Hydraulic retention time (HRT)
Daily biogas production rate
Daily methane (CH4) production rate
pH and buffering capacity (referred
as alkalinity)
Value
5.0-mm transparent Plexiglas
19.85 and 15.7 L
0.11 and 1.60 m
0.0095 m2
0.35, 0.50, 0.65, 0.80, 0.95 and 1.10 m
0.015 and 0.13 m
3034.5 C (32.8 1.3 C)
Daily-continuous mode with no
recirculation
Up-ow velocity = 0.70 m/h
Pumping ow rate = 133 mL/min
Pumping time = 7.515 min (for 12 L
of PMW)
Pumping rate = 50 rpm
356 days including start-up period
About 28.6% of the working volume
(4.5 L)
11.7819.62 g COD/L
(15.2 3.55 g COD/L)
0.7521.635 g COD/L/day
(1.096 0.294 g COD/L/day)
0.1050.138 m3/m2/day (for 12 L of
PMW)
(0.119 0.016 m3/m2/day)
0.6411.319 g CODremoved/L/day
(0.913 0.225 g CODremoved/L/day)
1215.7 days (14.1 1.86 days)
5.210.6 L/day (7.3 1.8 L/day)
3.87.7 L/day (5.4 1.3 L/day)
7.017.56 (7.27 0.11)
28105730 mg CaCO3/L
(4180 1090 mg CaCO3/L)
K swine 0:5 0:0043 exp0:091 VS0 4
Eqs. (3) and (4) indicate that the overloading of the digester
with substrate concentration, heavy metals, salts, or ammonia will
inhibit the digestion process (Nijaguna, 2002). It is interesting to
note that, to the best of the authors knowledge, a similar empirical
relationship K a b expck between K and Si for poultry man-
ure wastewater has not been derived in a multi-objective optimi-
zation-based modeling study so far. From this point of view, the
ChenHashimoto model was also implemented as an important
objective in the scope of the present study to fulll this gap in
the relevant literature. Considering the above-mentioned facts,
the ChenHashimoto model is expressed as follows:
   
1  RY m Si a bexpck
YV 1
hh hh 0:013T  0:129  1 a bexpck
5
where a, b and c are the empirical constants, k is the model variable
representing substrate loading rate (g COD/L/day) or substrate con-
centration (g COD/L). Because both inuent substrate concentration
(Si) and hydraulic retention time (hh) have signicant inuence on
the methane fermentation process (Yetilmezsoy, 2008; Yetilmezsoy
and Sakar, 2008b; Turkdogan-Aydinol et al., 2010), k is introduced
into Eq. (5) as the volumetric organic loading rate LO Si =hh . Then,
the volume of biogas produced per unit time (Qg, L/day) is predicted
from the following equation by substituting the effective volume of
the reactor (VR) and the steady-state methane content (am) into
Eq. (5):
   
1  RY m Si V R a bexpcSi =hh
Qg 1
hh am hh 0:013T  0:129  1 a bexpcSi =hh
6

In order to determine empirical constants (a, b and c); the vol-

per gram of CODremoved (L CH4/g CODremoved), Si is the concentration
of substrate in the inow (g COD/L), hh is the hydraulic retention
time (day), K is a dimensionless kinetic constant serving as an indi-
cator of the level of microbial growth inhibition (Nijaguna, 2002),
and lm is the maximum specic microbial growth rate (day1).
Chen and Hashimoto (1978) dened an empirical relationship
between lm and temperature (T) for temperatures between 20
and 60 C based on the analysis of data from several sources as fol-
lows (Yetilmezsoy, 2008; Turkdogan-Aydinol et al., 2010):
lm 0:013T  0:129
Eq. (2) was developed based upon a t of the data between
these two points and could likely be extrapolated down to 15 C
safely (Nijaguna, 2002). More importantly, this expression also
demonstrates that activity of microorganisms in the reactor essen-
tially stops (lm(10 C) = 0.001 day1) below 10 C (Yetilmezsoy,
2008). According to Eq. (2), Yetilmezsoy (2008) has reported that
the maximum specic microbial growth rate values are deter-
mined as lm(35 C) = 0.326 day1 and lm(25 C) = 0.196 day1 for
operating temperatures of 35 and 25 C, respectively. This indi-
cates that the maximum specic microbial growth rate decreases
about 40% when the operating temperature is decreased from 35
to 25 C. Likewise, decreasing the temperature from 35 to 30 C re-
sults in a 20% decrease in the maximum specic microbial growth
rate (lm(30 C) = 0.261 day1).
Hashimoto proposed the following equations to describe
the relationships between K and inuent total volatile solids
concentration per inuent volume (VS0, kg/m3) for cattle manure
(Hashimoto, 1982) and swine manure (Hashimoto, 1984):
ume of methane produced per gram of CODremoved, inuent sub-
strate concentration, hydraulic retention time and operating
temperature were respectively selected as the model inputs
(x1 = Ym, x2 = Si, x3 = hh, x4 = T). The volume of methane produced
per unit volume of digester per unit time was considered as the
model output (/1 = YV) for the nonlinear regression-based analysis.
Considering the above-mentioned facts, the rst objective function
(/1) to be maximized in the multi-objective optimization is given
by means of model variables and empirical constants:
   
1  Rx1 x2 a bexpcx2 =x3
/1 1
x3 x3 0:013x4  0:129  1 a bexpcx2 =x3
2
7
2.4. Modied StoverKincannon model
The StoverKincannon model is one of the most widely used
mathematical models to determine the kinetic constants for immo-
bilized systems (Kapdan and Erten, 2007). However, due to exper-
imental difculties in measuring the active surface area that
supports the biolm growth for the moving bed reactor, a simple
modication has been applied to the original model by introducing
the effective volume of the reactor (Debik and Coskun, 2009). The
modied form of this model has been successfully utilized in
anaerobic treatment of various types of wastewater, such as starch
wastewater (Ahn and Forster, 2000), synthetic wastewater
(Buyukkamaci and Filibeli, 2002; Kapdan and Erten, 2007; Sponza
and Ulukoy, 2008), poultry slaughterhouse wastewater (Debik and
Coskun, 2009), textile wastewater (Isik and Sponza, 2005), and mu-
nicipal wastewater (Turkdogan-Aydinol et al., 2010).
92 K. Yetilmezsoy / Bioresource Technology 118 (2012) 89101

According the modied StoverKincannon model, the removal has become a matter of increasing concern to the poultry industry
of organic substrate in the UASB process can be expressed by (Yetilmezsoy et al., 2009). For this reason, Yetilmezsoy and Sakar
means of the rate of substrate consumption in two different func- (2008b) reported that pollutant loads discharged from poultry
tions as follows (Yetilmezsoy, 2008; Sponza and Ulukoy, 2008; farms should be rst reduced to a certain extent, and then an effec-
Debik and Coskun, 2009): tive post-treatment unit should be installed for the anaerobically
pretreated PMW to provide the requirements of environmental
Q Si  Se Si  Se
rS 8 protection laws. Considering the worldwide increase in stringent
VR hh
discharge standards and regulatory compliance, several physico-
chemical processes, such as Fentons oxidation (Yetilmezsoy and
U max QSi =V R U max LO
rS 9 Sakar, 2008a), electrocoagulation process (Yetilmezsoy et al.,
K B QSi =V R K B LO
2009), struvite precipitation (Yetilmezsoy and Sapci-Zengin,
where rS is the rate of substrate consumption (g COD/L/day), Umax is 2009), have recently been proposed for the post-treatment of UASB
the maximum utilization rate constant (g COD/L/day), KB is the sat- pretreated PMW to improve the quality of the nal discharge. On
uration constant (g COD/L/day), Q is the ow rate (L/day), VR is the the basis of the above-mentioned facts, post-treatment cost was
effective volume of the reactor (L), LO is the organic loading rate (g considered as an important factor in the present calculations.
COD/L/day), hh is the hydraulic retention time (day), and Si and Se Temperature is one of the most important process-related fac-
are the inuent and efuent substrate concentrations (g COD/L), tors because it affects microbial activity, and a stable and efcient
respectively. operation must therefore be conducted to maintain a constant
From the mathematical point of view, combining the inverses of digestion temperature (Basrawi et al., 2012). Heat is necessary to
Eqs. (8) and (9) gives the linearized form of the modied StoverKin- raise the temperature of the feed sludge to the operating temper-
cannon model for the steady-state conditions (Yetilmezsoy, 2008): ature of digester, and to compensate for heat loss through walls,
 1 oor and roof of the digester (Videla et al., 1990). For this reason,
Q Si  Se VR KB VR 1 heating cost was taken into account as another important factor
r S 1
VR Q Si  Se U max QSi U max in the present analysis.
VR KB VR 1 It has been reported that the energy equivalent per kg of
Y bX b0 ; Y ! ; b! ; X! ; b0 !
Q Si  Se U max QSi U max CODremoved under mesophilic anaerobic conditions can be deter-
10 mined by taking into account some factors, such as biogas produc-
tion (0.20.5 m3 of biogas/kg CODremoved), energy value of produced
Based on the mass balance of substrate into and out of the con- biogas (Eb = 6.58 kWh/m3 of biogas) and efciency of combustion
trol volume, the above expression can be solved for either the efu- process (gc). For instance, this corresponds an energy value of about
ent substrate concentration [Eq. (11)] or the required volume of the 1.561.92 kWh/kg CODremoved for a biogas production of 0.3 m3/kg
UASB reactor [Eq. (12)] by substituting the kinetic constants, Umax CODremoved and an average combustion efciency of 80% under
and KB (Yetilmezsoy, 2008): mesophilic anaerobic conditions (Yetilmezsoy, 2008). Considering
U max Si U max Si a heat-to-electricity conversion efciency range from 20% to 25%,
Se Si  Si  11 the electricity output replaced by the thermal heat output can be
K B QSi =V R K B LO
determined (Thiers et al., 2010). According to this conversion, in-
come generated per kWh of electricity produced was calculated
QSi
VR 12 by introducing a coefcient as a revenue factor (C1: /kWh of
U max Si =Si  Se   K B
electricity produced) in the cost function. Moreover, to simulate
Eq. (11) can be employed to determine the efuent organic con- the economic aspect of the post-treatment (i.e. electrocoagulation,
centration at a given organic loading rate and inuent concentra- Fentons oxidation, etc.) of UASB pretreated poultry manure
tion for the anaerobic reactor. Therefore, in order to determine wastewater, a reasonable cost coefcient (C2: /kg CODremoved)
kinetic constants (Umax and KB), the inuent substrate concentra- representing the total post-treatment cost (including direct and
tion and the organic loading rate were respectively selected as indirect cost items such as chemicals, labor, maintenance, etc.) to
the model inputs (x2 = Si, x2/x3 = LO = Si/hh), and the efuent sub- remove per gram of COD was also included in this work.
strate concentration was considered as the model output (/2 = Se) Based on the above-mentioned facts, following economic fac-
for the multivariate regression-based analysis. To the best of the tors were taken into consideration in the present analysis: (i) total
authors knowledge, the present work is the rst multi-objective heating cost (Ch) for heat losses plus feed-warming energy require-
optimization-based report about the application of the modied ment, (ii) total post-treatment cost (Cpt) for UASB pretreated efu-
StoverKincannon model to the UASB reactor treating poultry ent and (iii) income (Cinc) generated per kWh of electricity replaced
manure wastewater. Based on the selected model variables and ki- by the biogas produced. From the parametric point of view, the
netic constants (c and d represent Umax and KB, respectively), the cost analysis was conducted based on the rst and the second
second objective function (/2) to be minimized in the multi-objec- objective functions (/1 and /2), as well as on the maximum allow-
tive optimization is expressed as follows: able concentration of COD in the post-treated efuent (SL), effective
cx2 volume of the reactor (VR), steady-state methane content (am), sev-
/2 x2  13 eral economic coefcients (C1, C2 and Cf), energy value of biogas
d x2 =x3
(Eb: kWh/m3 of produced biogas), average combustion efciency
(gc), heat-to-electricity conversion efciency (ge), thermal conduc-
2.5. Cost analysis tivities (k1 and k2), convection coefcients (hb and hs), other heat
transfer coefcients (Cp and k), thickness of the construction mate-
Considering the dual benets of environmental pollution con- rial (l1), thickness of the insulation material (l2), inuent ow rate
trol and meeting national energy needs, anaerobic digestion of (Qi), inuent density (qi), ambient temperature (Ta), temperature of
PMW has been proposed as an attractive treatment option in re- feed sludge (Ti), lower heating value of fuel (LHV) and efciency of
cent years (Yetilmezsoy, 2008; Yetilmezsoy and Sakar, 2008a, heating system (gh).
2008b; Yetilmezsoy and Sapci-Zengin, 2009; Yetilmezsoy et al., In the cost analysis, the maximum permissible COD level (for
2009). However, the necessity to comply with discharge limits composite samples taken in both 24 h) for the present type of
 K. Yetilmezsoy / Bioresource Technology 118 (2012) 89101 93

discharge (poultry sector) was set at 400 mg/L according to the design objectives (/i). Considering the simultaneous effects of mul-
Turkish Water Pollution Control Regulation (2004). For the rst tiple responses or individual objective functions, a systematic
quarter of 2012 in Turkey, price of electricity was taken as 0.07 / transformation scheme (from /i to di) can be utilized to aggregate
kWh in the present analysis. Average values of methane content every value of individual desirability into a single global desirabil-
(am), combustion efciency (gc) and heat-to-electricity conversion ity index through a weighted geometric mean. From this principle,
efciency (ge) were taken as 73%, 80% and 25%, respectively. this methodology has been successfully conducted to develop a
For a cylindrical and vertical tank, the amount of heat required composite response surface, referring as composite desirability
to compensate heat losses through the walls of the digester is given function, in various multi-objective optimization-based studies
by the following equation (Videla et al., 1990): (Wang and Wan, 2009; Coffey et al., 2007; Malenovic et al., 2011).
  In order to discover the global optimal conditions, each pre-
2pRd H1 T  T a 2pRd H2 T  T a
E1 3600 dicted ith response /i (i = 1, 2,. . ., n) is transformed to a scale free
l1 =k1 l2 =k2 l1 =k1 l2 =k2 1=hb 1=hs
desirability value di 0 6 di 6 1, where di 0 represents com-
 24 14 pletely undesirable response and di = 1 represents completely
where E1 is the heat lost through the walls of the digester (J/day), Rd desirable or ideal response. Different desirability functions are de-
is the radius of the digester (m), H1 is the height of sludge bed in the ned depending on whether the objective is for response variable
digester (m), H2 is the height of sludge blanket allowed above the to be maximized, minimized or assigned at a target value (Akalin
sludge for degassing (m), T is the operating temperature (in et al., 2010; Malenovic et al., 2011). If a response variable is to be
the present analysis, T = x4) (C), Ta is the ambient temperature (out- maximized (larger-the-better (LTB) type), its individual desirability
side the digester) (C), l1 is the thickness of the construction mate- function (di) is dened by
8
rial and insulation material (m), and l2 is the thickness of the > 0 /i < /i min
>
>
insulation material and insulation material (m), k1 is the thermal < wi
wi /i /i min
conductivity of the construction material (W/m/K), k2 is the thermal di f /i /i min 6 /i 6 /i max ; wi > 0
>
>
/i max /i min
conductivity of the insulation material (W/m/K), hb is the convec- >
:
tion coefcient of biogas (W/m2/K) and hs is the convection coef- 1 /i > /max
cient of sludge blanket (W/m2/K). /i 2 R; di f/i wi
: /i min ; /i max   R ! 0; 1  R;
The amount of heat necessary to heat the sludge to the temper-
ature of the digester is given by the following equation (Videla i 1; 2; . . . ; n 18
et al., 1990; Basrawi et al., 2012): where n denotes the number of responses, (/i)min and (/i)max are
E2 qi Q i C p T  T i 15 the minimum and the maximum acceptable values of /i, respec-
tively. In Eq. (18), wi is the user-specied exponential parameter
where E2 is the heat used to raise temperature of the feed sludge to or the weight factor that determines the shape (convex for wi < 1
the operating temperature (T = x4) of digester (J/day), qi is the inu- or concave for wi > 1) of desirability function. When the weight
ent density (kg/m3), Qi is the inuent ow rate (m3/day), Cp is the wi = 1, the desirability function increases linearly. Choosing wi > 1
specic heat of the inuent sludge (J/kg/K) and Ti is the temperature places more emphasis on being close to the target value, and choos-
of feed sludge (C). In the present analysis, the inuent density (qi), ing 0 < wi < 1 makes this less important (Akalin et al., 2010). The
the specic heat of the feed sludge (Cp) and the convection coef- weight factor can be specied by the user based on technical, eco-
cient of sludge blanket (hs) were assumed to be equal to that of nomical and other considerations. If the response of interest is min-
water because of their high water content (Videla et al., 1990; imization (smaller-the-better (STB) type), the individual desirability
Basrawi et al., 2012). Based on Eqs. (14) and (15), total heating cost function (di) is dened as follows:
(Ch) for heat losses plus feed-warming energy requirement is deter- 8
mined from the following equation: >
> 1 /i < /i min
>
< wi
wi /i /i max
E1 E2 C f di f /i /i min 6 /i 6 /i max ; wi > 0
Ch 16 >
>
/i min /i max
LHVgh >
:
0 /i > /i max
where Cf is the fuel cost in /kg, /m3 or /kWh depending on the wi
/i 2 R; di f/i : /i min ; /i max   R ! 0; 1  R;
fuel type (i.e. natural gas, coal, fuel oil, LPG, electricity), LHV is the
lower heating value of fuel in J/kg, J/m3 or J/kWh depending on i 1; 2; . . . ; n 19
the fuel type and gh is the efciency of heating system (%) (Kaynakli,
Consequently, the desirability approach can convert a multi-re-
2008). Considering all the above-mentioned economic factors, the
sponse problem into a single-response one. Then, the optimal con-
net cost function (Cnc: /m3) dened as the third objective function
ditions for several responses is obtained by maximizing the overall
(/3) to be minimized in the multi-objective optimization is given by
(global) desirability D, which is the weighted geometric mean of all
the following equation:
the individual desirability functions 0 6 D 6 1 (Coffey et al.,
2007; Wang and Wan, 2009; Akalin et al., 2010):
/1 f x1 ;x2 ;x3 ;x4 ; /2 f x2 ;x3 ; E1 ;E2 f x4
!1=I1 I2 ...In
C h f E1 ;E2 ; C pt f /2 ; C inc f /1 Y
n

C nc C h C pt  C inc ! /3 D f /i wi Ii ; f /i wi di ; wi > 0;
     i1
E1 E2 C f / VR 1
/3 C 2 /2  SL Q i  C 1 1 Eb gc ge i 1; 2; . . . ; n 20
LHVgf am Qi
X
n
17 1= Ii
I I I
Y
n
I
D d11  d22  . . .  dnn 1=I1 I2 ...In dii i1

2.6. Desirability function approach
In a specic system, every performance variable (xi), and there-
fore every desirability (di) function, may be related to particular
i1
where D is the overall desirability, n is the number of responses, Ii is
the relative importance that varies from the least important a value
of 1, to the most important a value of 5 (Akalin et al., 2010). The
94 K. Yetilmezsoy / Bioresource Technology 118 (2012) 89101
Fig. 1. A schematic of the proposed multi-objective optimization procedure.
simultaneous optimization of several responses can be achieved by
using various numerical techniques (i.e. response surface design
(RSM), linear or nonlinear programming (LP or NLP), genetic
algorithm (GA), etc.) to nd a global point that maximizes the over-
all (composite) desirability (Wang and Wan, 2009; Akalin et al.,
2010).
In this study, the method of desirability function was used (i) to
maximize the volume of methane produced per unit volume of di-
gester per unit time (Y V ! /1 : L CH4/L/day), (ii) to minimize efu-
ent substrate concentration (Se ! /2 : g COD/L) and (iii) to
minimize the net cost (C nc ! /3 : /m3). For the application of the
method, converted responses /i ! di were then combined on a
single overall (global) desirability (D), which is maximized to nd
the optimum set of input variables.
It is apparent from the literature that there are almost no papers
investigating the effect of weight factor (wi) on individual (di) and
composite (D) desirability values in the optimization of a specic
biosystem. Considering this gap in the literature, this study also in-
tends to explore the global optimal conditions by shifting the
shape of each individual desirability function by means of the
weight factor. In the present study, a total of 19 different weight
factor scenarios were implemented computationally for a number
of weight factors ranging from 0.1 to 8. For computational conve-
nience and comparative purposes, the numerical solution of the
overall desirability function was conducted on an equally impor-
tant response surface (I1 = I2 = I3) to avoid possible knowledge bias.
A schematic of the proposed multi-objective optimization proce-
dure is depicted in Fig. 1.
2.7. Statistical analysis
In derivation of objective functions (/i), the experimental data
was evaluated by a multiple regression software package (DataFit
V8.1.69, Copyright 19952005 Oakdale Engineering, PA, RC167),
containing 298 two-dimensional (2D) and 242 three-dimensional
(3D) nonlinear regression models. The regression analysis was con-
ducted according to the LevenbergMarquardt method with dou-
ble precision, as similarly done in the authors previous studies
(Yetilmezsoy and Sakar, 2008b; Yetilmezsoy and Sapci-Zengin,
2009).
The experimental data was imported directly from Microsoft
Excel used as an open database connectivity data source, and then
the regression analysis was performed. As regression models were
solved, they were automatically sorted according to the goodness-
of-t criteria into a graphical interface on the DataFit numeric
computing environment. Moreover, t-ratios and the corresponding
p-values were determined to appraise the signicance of the
regression coefcients. Descriptive statistics of the residual errors
were also provided for the appraisal of the multiple regression
model performance. An alpha (a) level of 0.05 (or 95% condence)
was used to determine the statistical signicance of the predicted
values. The overall desirability function (D) was optimized by using
 K. Yetilmezsoy / Bioresource Technology 118 (2012) 89101 95

Table 2
Summary of nonlinear regression-based results for ChenHashimoto and modied StoverKincannon models.

Kinetic models Descriptive statistics

SEE SR RA RSS R2 R2a NNI

ChenHashimoto model (prediction of YV) (L CH4/L/day)

K a b expck a b expcx2 =x3 0.0025 0.009 0.0002 0.0002 0.999 0.999 70
 h  i
1 x2
/1 1Rx
x3 1  K
x3 0:013x4 0:1291K

Modied StoverKincannon model (prediction of Se) (g COD/L)

/2 x2  dxcx22=x3 0.0831 0.2665 0.0072 0.242 0.991 0.989 6

2
SEE, standard error of the estimate; SR, sum of residuals; RA, residual average; RSS, residual sum of squares; R , determination coefcient; R2a , adjusted determination
coefcient and NNI, number of nonlinear iterations.

a LOQO/AMPL multi-objective optimization algorithm (by Prof.
Robert J. Vanderbei from Princeton University, NJ, USA), running
on a Pentium 4 CPU (Intel Atom Processor 3.00 GHz, 480 MB
of RAM) PC. SigmaPlot software (V10.0.0.54, Copyright 2006
Systat Software, Inc., GmbH, Germany) was used to create graphs
for the computational results.
ship between the observed methane production rate and efuent
COD concentration from the experiments carried out under
steady-state conditions and predicted methane production rate
and efuent COD concentration calculated by using Eqs. (21) and
(22) based on ChenHashimoto and modied StoverKincannon
kinetics.

K poultry 0:1222 0:0157 exp1:5454Si =hh  0:1222 0:0157 exp1:5454 LO

  
0:1222 0:0157 exp1:5454 LO  21
/1 ! Y V 1  RY m LO 1 
hh 0:013T  0:129  1 0:1222 0:0157 exp1:5454 LO 

3. Results and discussion
3.1. Kinetic modeling
Nonlinear regression-based results including several descriptive
statistics such as standard error of the estimate (SEE), sum of resid-
uals (SR) residual average (RA), residual sum of squares (RSS),
determination coefcient (R2), adjusted determination coefcient
(R2a ) and number of nonlinear iterations (NNI) for ChenHashimoto
and modied StoverKincannon models are summarized in Table
2. Moreover, regression variable results including standard error,
t-ratios and the corresponding p-values for the implemented ki-
netic models are given in Table 3.
The dimensionless parameter (K) in the ChenHashimoto
model was proposed for the rst time for PMW as a new empirical
formulation of volumetric organic loading rate in Eq. (21). The
efuent substrate concentration (Se) was predicted as a function
of inuent substrate concentration, volumetric organic loading
rate, maximum utilization rate constant and saturation constant
[Se = f (Si, LO, Umax, KB)] in Eq. (22). Fig. 2a and b shows the relation-
11:83Si 11:83Si
/2 ! Se Si  Si  22
13:02 Si =hh 13:02 LO
For the ChenHashimoto model, the value of determination
coefcient (R2 = 0.999) indicated that only 0.1% of the total varia-
tions were not explained by the regression model. In addition,
the value of adjusted determination coefcient (R2a 0:999) was
also very high, showing a high signicance of the model, as simi-
larly reported by Yetilmezsoy et al. (2009). In experimental terms
(Yetilmezsoy, 2008), the effectiveness of the present process in
terms of COD removal was above 83% on average, meaning that
most of the CODremoved was employed for a substantial quantity
of methane production in the volumetric organic loading rate
range studied. It is noted that the maximum volume of methane
produced per gram of CODremoved ((Ym)max, L CH4/g CODremoved) is
determined according to the criterion that 0.35 L of methane is
theoretically produced (at 1 atm, 0 C) per gram of CODremoved
(Martn Santos et al., 2010). Therefore, considering the biomass
growth and cell maintenance stable, 0.392 L of methane is theoret-
ically produced (at 1 atm, 32.5 C) per gram of CODremoved. The
experimental results (Yetilmezsoy, 2008) indicated that about

Table 3
Regression variable results for ChenHashimoto and modied StoverKincannon models.

Kinetic constants and coefcients Value Standart error t-Ratio p-Valuea

ChenHashimoto model (prediction of YV) (L CH4/L/day)
K a b expck a b expcx2 =x3
 h  i
1 x2 ab expcx2 =x3
/1 1Rx
x3 1  x3 0:013x4 0:1291ab expcx2 =x3
a (constant term in formulation of K) 0.1222 0.05829 2.0971 0.0435
b (coefcient of exponential term) 0.0157 0.02509 0.6275 0.5345
c (coefcient of organic loading rate) 1.5454 0.80846 1.9115 0.0644
Modied StoverKincannon model (prediction of Se) (g COD/L)
/2 x2  dxcx22=x3
c = Umax (g COD/L/day) 11.83 0.6187 19.1171 0.000
d = KB (g COD/L/day) 13.02 0.7431 17.5206 0.000
a
p values <0.05 were considered signicant in the regression model.
96 K. Yetilmezsoy / Bioresource Technology 118 (2012) 89101

0.55 For a randomly selected operating scenario (Si = 12 g COD/L (as

Kpoultry = 0.1222 + 0.0157 exp (1.5454 LO ) initial substrate concentration), hh = 15.7 days, T = 32.5 C, R = 0.1,
Methane production rate (YV, L CH4/L/day)

0.50 Ym/(Ym)max = 0.95), the proposed empirical formulation (Eq. (21))

(1 R )Ym Si Kpoultry
YV = 1

demonstrates that increasing the volumetric organic loading rate
0.45 h
h m 1 + Kpoultry up to a certain level (i.e. LO = 2.5 g COD/L/day) may stimulate the
(a) methane yield (corresponding to the maximum methane produc-
0.40 tion rate of YV = 0.675 L CH4/L/day). However, overloading of the
(Chen-Hashimoto model)

digester with high substrate concentration may cause a noticeable

0.35
decrease in the volume of methane produced per unit volume of
digester per unit time, thus suggesting the occurrence of a poten-
0.30
tial inhibition process. According to Eq. (22), the COD removal ef-
0.25
(YV )model = 0.9983 (YV )exp + 0.0008 ciency is more than 75% at the volumetric organic loading rate
from 0.76 to 2.5 g COD/L/day. Fig. 3 shows the variations of meth-
R 2 = 0.999
ane production rate (YV), efuent substrate concentration (Se) and
0.20
0.20 0.25 0.30 0.35 0.40 0.45 0.50 COD removal efciency (%) as simulated by the present formula-
Methane production rate (YV, L CH4/L/day) (Experimental) tions (Eq. (21) and (22)) with the increasing volumetric organic
loading rate (LO) in the UASB reactor.
The above-mentioned analysis indicates that although the
(Se, g COD/L) (Modified Stover-Kincannon model)

4 .0
methane production is stimulated by the volumetric organic load-
11.83Si
Se = Si ing rate up to a certain level, at the same time, the substrate con-
13.02 + LO
3 .5 centration in the efuent will increase, resulting a noticeable
decrease in the treatment efciency and increase in the cost of
any post treatment of the UASB-pretreated efuent. Therefore, it
(b)
Effluent substrate concentration

3 .0 should be noted that optimal solution attempts to the original mul-

ti-objective problem is important to transfer non-homogenous
2 .5
objectives (methane yield, efuent substrate concentration, and
net operating cost in the present case) into a particular homoge-
neous optimization (maximization or minimization) criterion, as
2 .0 (Se )model = 1.0178 (Se )exp 0.052 similarly stated by Videla et al. (1990).
Results of the regression analysis incidated that both determina-
R 2 = 0.991
tion coefcient (R2 = 0.991) and adjusted determination coefcient
1 .5 (R2a 0:989) values showed that the modied StoverKincannon
1 .5 2. 0 2 .5 3.0 3.5 4. 0
model signied an excellent t between the experimental data
Effluent substrate concentration (Se, g COD/L) (Experimental)
and the predicted values (responses). Table 4 summarizes a com-
Fig. 2. Relationships between observed and predicted methane production rate for
parison of several different process typologies on anaerobic pro-
the ChenHashimoto model (a) and efuent COD concentration for the modied cessing of various types of wastewater by means of the modied
StoverKincannon model (b). StoverKincannon model.
As seen in Table 4, the kinetic data reveals that a wide range of
95.2 (0.02)% of the maximum volume of methane theoretically kinetic constants (Umax and KB) calculated have been predicted
produced per gram of CODremoved was achieved in the volumetric depending on the characteristics of studied wastewaters and
organic loading rate range (0.7521.635 g COD/L/day) conducted experimental conditions. For the present application, the maxi-
in the steady-state conditions. mum utilization rate constant (Umax) is higher than the values

100 0.8

YV
Methane production rate (L CH4 /L/day)
Effluent COD concentration (g COD/L)

80
0.6
CODrem
COD removal efficiency (%)

60

0.4
Methane production rate (Model)
40 Effluent COD concentration (Model)
COD removal efficiency (Model)
Methane production rate (Yetilmezsoy, 2008)
Effluent COD concentration (Yetilmezsoy, 2008)
0.2
COD removal efficiency (Yetilmezsoy, 2008)
20
Se

0 0.0
0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4
Organic loading rate (g COD/L/day)

Fig. 3. Variations of predicted methane production rate (YV), efuent substrate concentration (Se) and COD removal efciency (%) with the increasing volumetric organic
loading rate (LO).
 K. Yetilmezsoy / Bioresource Technology 118 (2012) 89101 97

reported as 1.996 and 7.05 g/L/day by Turkdogan-Aydinol et al. Table 5

(2010) and Kapdan and Erten (2007), but lower than those ob- Technical, thermophysical and economic parameters used in the integrated multi-
objective optimization approach.
tained as 49.8 and 83.3 g/L/day by Ahn and Forster (2000) and
Buyukkamaci and Filibeli (2002), respectively. The present start- Constituents Value
up kinetics (Umax = 11.83 g COD/L/day, KB = 13.02 g COD/L/day) Anaerobic reactor (UASB) Cylindrical and vertical tank,
seem to be lower than kinetic constants obtained for some syn- Rd = 0.055 m
thetic-based wastewaters containing readily biodegradable organ- Height of sludge bed, H1 = 0.47 m
Height of sludge blanket, H2 = 0.83 m
ic matters. This may be expected because a stable synthetic Construction materials Walls: transparent Plexiglas,
solution medium will be much more efcient for a controlled treat- l1 = 5.0 mm
ment process and bacterial activity, as similarly reported by Turk- Insulation material: cotton wool,
dogan-Aydinol et al. (2010). Results clearly indicated that higher l2 = 10 mm
Thermal conductivities Transparent plexiglas, k1 = 0.19 W/m/
kinetic constants have been obtained for some mesophilic reactors
K
(2535 C) treating readily biodegradable substrates, such as Cotton wool, k2 = 0.029 W/m/K
starch and molasses (Ahn and Forster, 2000; Buyukkamaci and Fil- Convection coefcients Biogas, hb = 9.6 W/m2/K (Videla et al.,
ibeli, 2002). Lower kinetic constants (Sponza and Ulukoy, 2008; 1990)
Turkdogan-Aydinol et al., 2010) may be ascribed to the presence Sludge blanket, hs = 20 W/m2/K
Specic heat of the inuent sludge Cp = 4184 J/kg/K (Videla et al., 1990)
of several recalcitrant and complex compounds, and other undesir-
Inuent density qi = 1000 kg/m3
able impurities in the wastewaters, as well as to experimental con- Operating temperature (T = x4) T = x4 = 30 34.5 C (Yetilmezsoy,
ditions within low temperature range and low activity of acidogens 2008)
and methanogens in the start-up stage. Ambient temperature (air or feed Ta = Ti = 20 C
sludge)
Lower heating value (LHV) of fuel LHV = 3.599  106 J/kWh (Kaynakli,
3.2. Multi-objective optimization and decision phase (electricity) 2008)
Energy value of biogas 7.0 kWh/m3 of produced biogas
Average methane content in the am = 73% (Yetilmezsoy, 2008)
The method of desirability function was used to obtain maxi- produced biogas
mum methane production rate (YV), the minimum efuent sub- Average combustion efciency gc = 80% (Yetilmezsoy, 2008)
strate concentration (Se) and the minimum net operating cost Heat-to-electricity conversion ge = 25% (Thiers et al., 2010)
(Cnc) simultaneously. The engineering parameters considered in efciency
Efciency of heating system gh = 99% (Kaynakli, 2008)
optimizing the operation of the UASB system treating PMW are gi-
(electricity)
ven in Table 5. The constraints imposed on the design variables (xi) Maximum allowable concentration SL = 400 mg/L (Turkish Water
and the individual objective functions (/i) are listed in Table 6. of COD in the post-treated efuent Pollution Control Regulation, 2004)
The overall desirability function (D) was optimized by using a for poultry sector
LOQO/AMPL multi-objective optimization algorithm to explore Cost of electricity C1 = Cf = 0.07 /kWh of electricity
produced or used (for the rst
the effect of weight factor (wi) on individual (di) and composite quarter of 2012 in Turkey)
(D) desirability values. Fig. 4 shows a graphical representation of Cost of post-treatment (i.e. C2 = 0.0114 /kg CODremoved
desirability values with respect to the different weighting factors electrocoagulation with iron (Bayramoglu et al., 2006)
(wi) ranging from 0.1 to 8. Computational results indicated that electrodes)
the highest desirabilities of the individual objectives (d1 = 0.874,
d2 = 0.822 and d3 = 1.000) were obtained for the weight factors of
w1 = 1, w2 = 0.1 and w3 = 1. This revealed that minimizing the efu- (Yetilmezsoy et al., 2009). Moreover, this process is characterized
ent substrate concentration could be appreciated as a less urgent by simple equipment, easy operation, a shortened reactive reten-
objective than those of maximizing the methane production rate tion period and decreased amount of sludge (Bayramoglu et al.,
and minimizing of the net operating cost. It is apparent from the lit- 2006). Furthermore, there is also an advantage for applying struvite
erature that several attractive and suitable methods have been pro- precipitation process to anaerobically pretreated efuents due to
posed for the post-treatment of UASB pretreated PMW wastewater predominance of struvite constituent ions, NH4+N and PO43P,
(Yetilmezsoy and Sakar, 2008a; Yetilmezsoy and Sapci-Zengin, in the methane fermentation efuent, thereby minimizing the need
2009; Yetilmezsoy et al., 2009). Among them, electrocoagulation to add chemicals (Yetilmezsoy and Sapci-Zengin, 2009). More
has been proven to be highly effective and low cost process importantly, it has been reported that the struvite precipitation

Table 4
Comparison of different process typologies on anaerobic processing of various types of wastewater by means of the modied StoverKincannon model.

Type of substrate Type of digester Operating Modied StoverKincannon kinetics and References
temperature (C) determination coefcients
Umax (g /L/ KB (g /L/day) R2
day)
Poultry manure wastewater UASB 3034.5 11.83 13.02 0.991 Present study
Municipal wastewater UASB 17.121 1.996 1.536 0.972 Turkdogan-Aydinol et al.
(2010)
Synthetic wastewater (saline) Anaerobic lter 37 7.05 5.3 0.910 Kapdan and Erten (2007)
Synthetic wastewater (starch) Mesophilic 35 49.8 50.6 0.998 Ahn and Forster (2000)
anaerobic lter
Synthetic wastewater (10 mg/L 2,4- UASB 0.01 (mg/L/ 0.0098 (mg/L/ 0.992 Sponza and Ulukoy
dichlorophenol) day) day) (2008)
Synthetic wastewater (molasses, urea, Anaerobic hybrid 83.3 186.23 0.990 Buyukkamaci and Filibeli
K2HPO4) reactor (2002)
Poultry slaughterhouse wastewater SASBR and SGBR 22 121.71 and 130.28 and 0.999 and Debik and Coskun (2009)
164.48 177.21 0.996
98 K. Yetilmezsoy / Bioresource Technology 118 (2012) 89101

Table 6
Model components and constraints for the integrated multi-objective approach.

Model components Constraints Units

Objective functions (/i) /i 2 R : /i min ; /i max   R i = (1, 2,. . ., n)
Maximize (/1) (Eq. (7)) 0:241 6 /1 Y V 6 0:491 (L CH4/L/day)
Minimize (/2) (Eq. (13)) 1:67 6 /2 Se 6 3:78 (g COD/L)
Minimize (/3) (Eq. (17)) 16:24 6 /3 C nc 6 24:91 (/m3)
Design variables (xi) xi 2 R : xi min ; xi max   R i = (1, 2,. . ., n)
Ym: Volume of methane produced per gram of CODremoved 0:354 6 Y m x1 6 0:386 (L CH4/g CODremoved)
Si: Inuent substrate concentration 11:78 6 Si x2 6 19:62 (g COD/L)
hh : Hydraulic retention time 12 6 hh x3 6 15:7 (day)
T: Operating temperature 30 6 T x4 6 34:5 (C)
lm : Maximum specic microbial growth rate lm 0:013T  0:129 for 20  C 6 T x4 6 60  C (day1)
Faction of refractory (non-biodegradable) COD R = 0.1 (Videla et al., 1990)
Individual desirability functions (di) 0 6 di f /i wi 6 1 i = (1, 2,. . ., n)
Weight factor (wi) wi P 0 (herein wi = 0.18.0) i = (1, 2,. . ., n)
Relative importance (Ii) 1 6 Ii 6 5 (herein I1 = I2=I3) i = (1, 2,. . ., n)
Composite desirability function (D) 0 6 D 6 1 ! Maximize D (Eq. (20))
Verication of Ym /1 x3 =x2  /2 x1 (L CH4/g CODremoved)
Limit of theoretical methane production (Ym/(Ym)max < 1) (x1)/[(x4+273)(0.35/273)] < 1

1.0
Desirability values (d1, d2, d3, D)

0.8

0.6

0.4
Global point: d1
d1= 0.874
d2 = 0.822 d2
0.2
d3 = 1.000 d3
D = 0.896
D
0.0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Number of weight factor (wi) scenario (i th)

A B
No w1 w2 w3 No w1 w2 w3
w i : Weight factor
1 0.1 1 1 11 1 4 1 1.0 1.0
0.1
2 0.2 1 1 12 1 8 1 w=
.2
=0
w
3 0.4 1 1 13 1 1 0.1
4
Desirability (d )

Desirability (d )
0.
=
w

4 2 1 1 14 1 1 0.2
1
=
w

5 4 1 1 15 1 1 0.4
2
=
w

6 8 1 1 16 1 1 2 4
=
w
8

7 1 0.1 1 17 1 1 4
=
w

8 1 0.2 1 18 1 1 8
0.0 0.0
di = f (i )
wi
9 1 0.4 1 19 1 1 1
10 1 2 1 No : Scenario number A B
Response value

Fig. 4. Effect of weight factor (wi) on individual (di) and composite (D) desirability values.

can be reliable, viable and cost effective when the commercial value The experimental studies indicated that the operating tempera-
of obtained product is taken into account (Yetilmezsoy et al., 2011). ture has a crucial effect on the growth, activity and survival of
Considering both computational results and the above-mentioned microorganisms, as well as on the biogas production (Yetilmezsoy,
cost-effective post-treatment solutions, it can be concluded that 2008). It is reported that a change from mesophilic to thermophilic
the individual objective of efuent substrate concentration may temperatures (or vice versa) can result in a sharp decrease in biogas
be relaxed ensuring a removal of at least 80% of COD for the present production until the necessary microbial populations have in-
application. However, it should be noted that the relative weighting creased in number. Moreover, even small changes in temperature,
can be made more stringent depending on the other specic consid- from 35 to 30 C and from 30 to 32 C have been shown to reduce
erations to meet the compliance with the desired efuent quality. biogas production rate (Ward et al., 2008). Therefore, optimal
 K. Yetilmezsoy / Bioresource Technology 118 (2012) 89101 99

YV 1 d1 d1 = 0.874 Se 2 d 2 d2 = 0.822

(a) (b)

1 .0 1 .0

y of S e (d 2)
)
ability of YV (d 1

0 .8 0 .8

0 .6 0 .6

abilit
Individual desir

Individual desir
0 .4 0 .4

)
ay
1 .5

)
1 .5

ay

/d
0 .2

L
0 .2

/d
1 .4

D/
1 .4

CO
D/
1 .3 1 .3

CO

(g
0 .0 1 .2 0,0 1 .2

(g

t e
34

ra
34
e
1 .1 1 .1
t
ra 33

g
33

din
g
32 1 .0 32 1 .0
din

loa
Temp 31 Temp 31
loa

e rature e rature

nic
(C) 30 (C) 30
nic

ga
ga

Or
Or

D = 0.896
Cnc 3 d3 d3 = 1.000

(c) (d)
1 .0

1 .0
)
ability of Cnc (d 3

ability (D)

0 .8

0 .8
0 .6
Composite desir

0 .6
Individual desir

0 .4
0 .4

y)
1 .5

da
1 .5 0 .2

L/
1 .4
)
ay

D/
0 .2 1 .4
/d

1 .3

CO
L
D/

1 .3 0 .0 1 .2

(g
CO

a te
0 .0 1 .2 34 1 .1
(g

34 33
gr
1 .1
e

33 32
t

1 .0 din
ra

32 Temp 31
loa

1 .0
g

e
din

Temp 31 rature 30
(C)
nic

e
loa

rature 30
(C)
ga
nic

Or
ga
Or

Fig. 5. 3D plots showing the optimal desirability values (d1 = 0.874, d2 = 0.822, d3 = 1.000 and D = 0.896) and mutual interactions of organic loading rate and operating
temperature on individual desirabilities (d1, d2 and d3) and the overall desirability (D).

temperature range should be conducted to achieve best results for optimization approach provided a useful tool for the UASB reactor
the anaerobic digestion and treatment efciency in the design and to optimize several responses simultaneously.
operation of UASB reactors. However, it should be noted that the The simultaneous optimization of three responses showed that
total amount of heat necessary to compensate heat losses through the maximum overall desirability (D) of 0.896 was estimated at the
the construction materials (i.e. walls, oor, roof, etc.) of the diges- operating temperature of 31.6 C, the hydraulic retention time of
ter and to heat the sludge to the temperature of the digester will 13.1 days, the inuent substrate concentration of 19.1 g COD/L
increase, when the difference between ambient temperature (air and the methane yield of 0.386 L CH4/g CODremoved. Accordingly,
or feed) and operating temperature increases (Eqs. (14) and (15)). the COD removal efciency of 81.7%, the methane production rate
On the contrary, methane production rate increases towards the of 0.46 L CH4/L/day, the efuent substrate concentration of
upper mesophilic temperatures (Eq. (21)). Based on the economic 3.48 g COD/L and the net operating cost of 16.24 /m3 were esti-
considerations, it should be noted that the increase in methane mated at the global optimal point. The predicted net operating cost
yield or production rate from the digestion process has to be is lower than the overall net treatment costs of 2431 per m3 pig
balanced against the increased energy requirement for maintain- slurry (liquid fraction: organic matter content 710 kg/m3 and
ing the reactor at the higher temperature, as also suggested by NH3/NH4 content 36 kg/m3) estimated by Schoumans et al.
others (Videla et al., 1990; Ward et al., 2008). From this point of (2010), but higher than those obtained as 56 per m3 of treated
view, this study demonstrated that an integrated multi-objective swine manure (equalization tank: COD of 6020 mg/L, total
100 K. Yetilmezsoy / Bioresource Technology 118 (2012) 89101
Kjeldahl nitrogen (TKN) of 681 mg/L) by Bortone (2009). However,
it should be noted that cost differences are due to the characteris-
tics of studied wastewaters, operating conditions and also sev-
eral economical and technical considerations specied by the
designer.
In order to gain a better understanding of the mutual interac-
tions of organic loading rate (LO) and operating temperature (T)
on individual desirabilities (d1, d2 and d3) and the overall desirabil-
ity, three-dimensional (3D) response surface plots were con-
structed based on the results of the multi-objective optimization
(Fig. 5). The nonlinear nature of all 3D response surfaces and the
respective contour plots demonstrated that there were consider-
able interactions between design variables and desirability values.
Under the optimal conditions (T = 31.6 C, LO = 1.457 g COD/L/
day, Ym = 0.386 L CH4/g CODremoved), the process-related re-
sponses (YV = 0.46 L CH4/L/day, Se = 3.48 g COD/L, gCODrem = 81.7%)
obtained from the computational analysis were found to be very
close to those ((YV)exp = 0.456 L CH4/L/day, (Se)exp = 3.57 g COD/L,
(gCODrem)exp = 81.2%) indentied by the experimental study
(Yetilmezsoy, 2008). The small deviations (between 0.73% and
2.52%) in the validation of the optimal conditions indicated the
suitability of the proposed integrated approach and suggested
that this methodology could be successfully adapted to the de-
sign and operation of a mesophilic anaerobic digester on PMW.
Consequently, the multi-objective optimization criteria within
the framework of desirability function and nonlinear regression-
based kinetic modeling demonstrated a useful tool to optimize
several responses for the UASB reactor simultaneously. The pro-
posed holistic approach can easily and efciently achieve the
optimization of a complex process, such as anaerobic digestion
process with multiple responses. Considering the effectiveness
of the integrated multi-objective optimization methodology, this
approach may be utilized for the large-scale implementations
(from pilot-scale to full-scale) to optimize more complex deci-
sions involving additional objectives (i.e. nutrient management,
processing of animal manure into marketable products) and mod-
el components (i.e. different construction materials, insulation
thickness, fuel type, etc.).
Finally, it should be noted that the holistic approach described
herein has several limitations. The values of desirability functions
reported in this paper are restricted to conditions reached in the
experimental work. Moreover, energy losses are determined for a
lab-scale reactor, but geometry and construction materials will
be changed in real-scale conditions. In order to extrapolate the re-
sults to full-scale congurations, where different technical and
thermophysical parameters will be used to calculate energy losses,
there is a need to obtain engineering parameters independently for
real-scale biogas plants. Furthermore, the equations presented in
this paper are limited to the experimental data working with spe-
cic organic and hydraulic loading values. Additionally, some lim-
itations to model extrapolation may be due to the fact that
volumetric organic loading rates experimented are relatively low
compared to the usual practice. Likewise, HRT values reported
are also large compared to the usual practice. However, it is noted
that operational conditions must be handled with care ensuring a
removal of at least 7580% of COD to meet the compliance with
the desired efuent quality. Although the present analysis de-
scribed herein are restricted by some assumptions and some
boundary conditions, the proposed holistic methodology can be
readily utilized as a useful preliminary analysis when transferring
the initial concepts from lab-scale to pilot-scale and, as a conse-
quence, to real-scale conditions. With regard to these facts, it can
be concluded that some additional measurements are required to
be quantied by future works to increase the precision of the pro-
posed methodology with fewer limitations. These issues have been
left as the subject of future work.
4. Conclusions
Multi-objective optimization of complex processes requires
development of novel approaches in order to improve the reliability
of design decisions. In this paper, a holistic method based on the
integration of kinetic modeling and desirability function approach
has been proposed as a useful tool for multi-objective optimization
of a challenging research topic by controlling a set of weighting
parameters. Considering the adaptability of several other con-
straints and additional design objectives into the proposed optimi-
zation scheme, any major improvements in this methodology will
enhance the robustness of this method for real-scale conditions of
the present particular design problem.
References
Abouelenien, F., Nakashimada, Y., Nishio, N., 2009. Dry mesophilic fermentation of
chicken manure for production of methane by repeated batch culture. J. Biosci.
Bioeng. 107, 293295.
Ahn, J.H., Forster, C.F., 2000. Kinetic analyses of the operation of mesophilic and
thermophilic anaerobic lters treating a simulated starch wastewater. Process
Biochem. 36, 1923.
Akalin, O., Akay, K.U., Sennaroglu, B., Tez, M., 2010. Optimization of chemical
admixture for concrete on mortar performance tests using mixture
experiments. Chemom. Intell. Lab. Syst. 104, 233242.
Basrawi, M.F.B., Yamada, T., Nakanishi, K., Katsumata, H., 2012. Analysis of the
performances of biogas-fuelled micro gas turbine cogeneration systems (MGT-
CGSs) in middle- and small-scale sewage treatment plants: comparison of
performances and optimization of MGTs with various electrical power outputs.
Energy 38, 291304.
Bayramoglu, M., Kobya, M., Eyvaz, M., Senturk, E., 2006. Technical and economic
analysis of electrocoagulation for the treatment of poultry slaughterhouse
wastewater. Sep. Purif. Technol. 51, 404408.
Bortone, G., 2009. Integrated anaerobic/aerobic biological treatment for intensive
swine production. Bioresour. Technol. 100, 54245430.
Buyukkamaci, N., Filibeli, A., 2002. Determination of kinetic constants of an
anaerobic hybrid reactor. Process Biochem. 38, 7379.
Chen, Y.R., Hashimoto, A.G., 1978. Kinetics of methane fermentation. Biotechnol.
Bioeng. Symp. 8, 269282.
Coffey, T., Gennings, C., Moser, V.C., 2007. The simultaneous analysis of discrete and
continuous outcomes in a doseresponse study: using desirability functions.
Regul. Toxicol. Pharm. 48, 5158.
Das, P., Mukherjee, S., Ganguly, S., Bhattacharyay, B.K., Datta, S., 2009. Genetic
algorithm based optimization for multi-physical properties of HSLA steel
through hybridization of neural network and desirability function. Comp.
Mater. Sci. 45, 104110.
Debik, E., Coskun, T., 2009. Use of the static granular bed reactor (SGBR) with
anaerobic sludge to treat poultry slaughterhouse wastewater and kinetic
modeling. Bioresour. Technol. 100, 27772782.
Hashimoto, A.G., 1982. Methane production from cattle waste: effect of
temperature, hydraulic retention time and inuent substrate concentration
on kinetic parameter (K). Biotechnol. Bioeng. 24, 20392052.
Hashimoto, A.G., 1984. Methane from swine manure: effects of temperature and
inuent substrate concentration on kinetic parameter (K). Agric. Waste. 9, 299
308.
Isik, M., Sponza, D.T., 2005. Substrate removal kinetics in an upow anaerobic
sludge blanket reactor decolorising simulated textile wastewater. Process
Biochem. 40, 11891198.
Kapdan, I.K., Erten, B., 2007. Anaerobic treatment of saline wastewater by
Halanaerobium Lacusrosei. Process Biochem. 42, 449453.
Kaynakli, O., 2008. A study on residential heating energy requirement and optimum
insulation thickness. Renew. Energ. 33, 11641172.
Malenovic, A., Dotsikas, Y., Makovic, M., JancicStojanovic, B., Ivanovic, D.,
Medenica, M., 2011. Desirability-based optimization and its sensitivity
analysis for the perindopril and its impurities analysis in a microemulsion LC
system. Microchem. J. 99, 454460.
Martn Santos, M.A., Lpez, J.A.S., Prez, A.F.C., Martn, A.M., 2010. Modelling the
anaerobic digestion of wastewater derived from the pressing of orange peel
produced in orange juice manufacturing. Bioresour. Technol. 101, 39093916.
Nijaguna, B.T., 2002. Biogas Technology. Principal JSS Academy of Technical
Education Bangalore, New Age International (P) Limited Publishers, New
Delhi, Bangalore, ISBN 978-81-224-1380-9, 300p.
Sakar, S., Yetilmezsoy, K., Kocak-Enturk, E., 2009. Anaerobic digestion technology in
poultry and livestock waste treatment a literature review. Waste Manage. Res.
27, 318.
Schoumans, O.F., Rulkens, W.H., Oenema, O., Ehlert, P.A.I., 2010. Phosphorus
recovery from animal manure, Technical opportunities and agro-economical
perspectives, Alterra report 2158, ISSN: 15667197, Wageningen, 111p.
Sponza, D.T., Ulukoy, A., 2008. Kinetic of carbonaceous substrate in an upow
anaerobic sludge sludge blanket (UASB) reactor treating 2,4 dichlorophenol (2,4
DCP). J. Environ. Manag. 86, 121131.
 K. Yetilmezsoy / Bioresource Technology 118 (2012) 89101
Thiers, S., Aoun, B., Peuportier, B., 2010. Experimental characterization, modeling
and simulation of a wood pellet micro-combined heat and power unit used as a
heat source for a residential building. Energy Buildings 42, 896903.
Turkdogan-Aydinol, F.I., Yetilmezsoy, K., Comez, S., Bayhan, H., 2010. Performance
evaluation and kinetic modeling of the start-up of an UASB reactor treating
municipal wastewater at low temperature. Bioproc. Biosyst Eng. 34 (2), 153162.
TWPCR (Turkish Water Pollution Control Regulation), Ofcial Gazette No. 25687
dated December 31, 2004.
Videla, S., Chamy, R., Navarrete, E., Lema, J.M., 1990. Design and operation of
anaerobic digesters using multi-objective optimization criteria. J. Chem. Tech.
Biotechnol. 49, 223231.
Wang, J., Wan, W., 2009. Application of desirability function based on neural
network for optimizing biohydrogen production process. Int. J. Hydrogen
Energy 34, 12531259.
Ward, A.J., Hobbs, P.J., Holliman, P.J., Jones, D.L., 2008. Optimisation of the anaerobic
digestion of agricultural resources. Bioresour. Technol. 99, 79287940.
Yetilmezsoy, K., 2008. Treatability of poultry manure wastewater using anaerobic
sludge bed reactor. PhD Thesis, Institute of Science, Department of
101
Environmental Engineering, Yildiz Technical University,
Istanbul.
Yetilmezsoy, K., Sakar, S., 2008a. Improvement of COD and color removal from UASB
treated poultry manure wastewater using Fentons oxidation. J. Hazard. Mater.
151, 547558.
Yetilmezsoy, K., Sakar, S., 2008b. Development of empirical models for performance
evaluation of UASB reactors treating poultry manure wastewater under
different operational conditions. J. Hazard. Mater. 153, 532543.
Yetilmezsoy, K., Sapci-Zengin, Z., 2009. Stochastic modeling applications for the
prediction of COD removal efciency of UASB reactors treating diluted real
cotton textile wastewater. Stochas. Environ. Res. Risk Assess. 23, 1326.
Yetilmezsoy, K., Ilhan, F., Sapci-Zengin, Z., Sakar, S., Gonullu, M.T., 2009.
Decolorization and COD reduction of UASB pretreated poultry manure
wastewater by electrocoagulation process: a posttreatment study. J. Hazard.
Mater. 162, 120132.
Yetilmezsoy, K., Turkdogan-Aydinol, F.I., Gunay, A., Ozis, I., 2011. Post treatment of
poultry slaughterhouse wastewater and appraisal of the economic outcome.
Environ. Eng. Manag. J. 10, 16351645.