Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Bioresource Technology
journal homepage: www.elsevier.com/locate/biortech
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 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.
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 for concrete design (Akalin et al., 2010), and optimization of a
process congurations (Sakar et al., 2009). microemulsion liquid chromatographic system (Malenovic et al.,
Anaerobic digestion has been reported as a benecial, feasible 2011).
and advantageous process in treatment and disposal of poultry To the best of the authors knowledge, however, there are al-
and livestock wastes (Sakar et al., 2009). In particular, this technol- most no papers in the scientic literature specically devoted to
ogy have been successfully applied to treat a variety of solid, a study of an integrated multi-objective optimization approach
semi-solid and liquid poultry waste with the use of different reac- based on nonlinear regression-based kinetic modeling and desir-
tor congurations, such as mesophilic digestion of poultry manure ability function for the optimization of UASB reactor treating poul-
slurry using up-ow anaerobic sludge blanket (UASB) reactor try manure wastewater (PMW). Thus, in this study, bio-kinetics of
(Yetilmezsoy, 2008) and dry fermentation of chicken manure using the present digestion process were rst explored by using two of
batch type digester (Abouelenien et al., 2009). Although anaerobic the most widely used kinetic models (ChenHashimoto and mod-
digestion has been proven as an established biotechnological ied StoverKincannon), and then the several responses were opti-
method for degrading a variety of polluting organic wastes, how- mized simultaneously by the method of desirability function, with
ever, it is a complex process requiring careful operation and mon- the purpose of obtaining the optimal conditions for the methane
itoring the several parameters, such as pH, alkalinity, temperature, production rate, the efuent substrate concentration, and the net
volatile fatty acids, organic loading rate, hydraulic retention times, operating cost.
biogas composition, etc. For this purpose, development of inte-
grated solutions is worthy of investigation for the optimization 2. Methods
and simulation of complex anaerobic digestion processes in a
straightforward manner. 2.1. Summary of previous ndings
Kinetic models can be used to control and predict the treatment
plant performance, to optimize the plant design and to determine The experimental methodology including acclimation step, feed
the importance of relationships between variables in fundamental preparation and characteristics, seed sludge, basal medium, re-
research of anaerobic processes (Isik and Sponza, 2005). These agents, equipments, sample collection, chemical analyses, reactor
models can also provide a good insight into reaction mechanisms conguration, gas collection system, and other technical and
by reducing laborious and complex experimental data to simple operational details was fully documented in the authors previous
and convenient mathematical expressions (Turkdogan-Aydinol studies on the anaerobic treatability of high strength PMW
et al., 2010). Therefore, a growing number of publications have re- (Yetilmezsoy, 2008; Yetilmezsoy and Sakar, 2008a, 2008b;
ported the application of several kinetic models in the anaerobic Yetilmezsoy et al., 2009). For this numerical study, technical and
digestion of various waste streams which have different character- operational data about the UASB system treating PMW are summa-
istics and higher contents of solids. So far, numerous kinetic mod- rized in Table 1.
els (i.e. Monod-type kinetic, Grau second-order multicomponent, On the basis of the steady-state data obtained from the previous
StoverKincannon, ChenHashimoto, Contois, Haldane, pseudo- work (Yetilmezsoy, 2008), in the present study, an integrated mul-
mechanistic, LineweaverBurk (Sundstrom model), Barthakur, ti-objective optimization using kinetic modeling, nonlinear regres-
Dewalle and Chian, Young and McCarty, Benjes, etc.) have been sion methodology, cost analysis and desirability function method
introduced in anaerobic digestion to explore biokinetics and to was proposed as a straightforward approach to optimize a specic
determine the importance of relationships between the design digestion process with multiple responses.
data and experimental results (Yetilmezsoy, 2008).
Several factors such as operating temperature, initial pH, organ-
2.2. Kinetic modeling
ic and hydraulic loading rates and substrate concentration can sig-
nicantly inuence the stability of anaerobic digestion process,
In the present study, ChenHashimoto and modied Stover
because they can directly affect the activity of microbial consor-
Kincannon models were implemented to the UASB reactor for
tium. For this reason, optimal ranges of process-related parameters
determination of bio-kinetic coefcients. The proposed models
should be chosen in design and operation of anaerobic digesters to
were formulated based on the following hypothesis and consider-
increase the ability of bacterial population and to produce methane
ations: (i) the reactor was completely mixed during the feeding
as a valuable source of renewable energy. Therefore, the multi-var-
and operated in a daily-continuous mode without solids recycle,
iate optimization of complicated biosystems has been received
(ii) no microbial solids were contained in the inuent substrate,
increasing attention in recent years (Wang and Wan, 2009).
(iii) the effective volume of the reactor must constant, (iv) hydrau-
The performance of a biosystem is associated with a set of sev-
lic retention time hh is dened as the volume of the reactor divided
eral process variables, which are dened operationally and exper-
by the ow rate of the inuent and (v) steady-state conditions
imentally, often interrelated with different signicance and
prevail throughout the system. Implementation of the above-men-
measured in different scale of units. At this point, the concept of
tioned kinetic models in the proposed multi-objective optimiza-
desirability scale arises when there is a need to combine the mag-
tion scheme is discussed in brief in the following sections.
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 2.3. ChenHashimoto model
as a useful tool to optimize several responses simultaneously when
there are many responses to be optimized for a bioprocess (Wang For a completely mixed reactor without solids recirculating,
and Wan, 2009). It must also be remembered that an appropriate Chens and Hashimotos equation is expressed as follows (Videla
modeling of the process data often becomes a prerequisite for et al., 1990; Yetilmezsoy, 2008):
the simultaneous optimization of multiple responses by the meth-
1 RY m Si K
od of desirability function (Wang and Wan, 2009; Das et al., 2009). YV 1 1
hh hh lm 1 K
Recently, this methodology has been successfully utilized to
explore the optimal process conditions for several challenging where YV is the volume of methane produced per unit volume of di-
problems, such as simultaneous analysis of neurotoxicological gester per unit time (L CH4/L/day), R is the fraction of refractory
endpoints (Coffey et al., 2007), optimization of chemical admixture (non-biodegradable) COD, Ym is the volume of methane produced
K. Yetilmezsoy / Bioresource Technology 118 (2012) 89101 91
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
i1
2.6. Desirability function approach
where D is the overall desirability, n is the number of responses, Ii is
In a specic system, every performance variable (xi), and there- the relative importance that varies from the least important a value
fore every desirability (di) function, may be related to particular of 1, to the most important a value of 5 (Akalin et al., 2010). The
94 K. Yetilmezsoy / Bioresource Technology 118 (2012) 89101
simultaneous optimization of several responses can be achieved by A schematic of the proposed multi-objective optimization proce-
using various numerical techniques (i.e. response surface design dure is depicted in Fig. 1.
(RSM), linear or nonlinear programming (LP or NLP), genetic
algorithm (GA), etc.) to nd a global point that maximizes the over- 2.7. Statistical analysis
all (composite) desirability (Wang and Wan, 2009; Akalin et al.,
2010). In derivation of objective functions (/i), the experimental data
In this study, the method of desirability function was used (i) to was evaluated by a multiple regression software package (DataFit
maximize the volume of methane produced per unit volume of di- V8.1.69, Copyright 19952005 Oakdale Engineering, PA, RC167),
gester per unit time (Y V ! /1 : L CH4/L/day), (ii) to minimize efu- containing 298 two-dimensional (2D) and 242 three-dimensional
ent substrate concentration (Se ! /2 : g COD/L) and (iii) to (3D) nonlinear regression models. The regression analysis was con-
minimize the net cost (C nc ! /3 : /m3). For the application of the ducted according to the LevenbergMarquardt method with dou-
method, converted responses /i ! di were then combined on a ble precision, as similarly done in the authors previous studies
single overall (global) desirability (D), which is maximized to nd (Yetilmezsoy and Sakar, 2008b; Yetilmezsoy and Sapci-Zengin,
the optimum set of input variables. 2009).
It is apparent from the literature that there are almost no papers The experimental data was imported directly from Microsoft
investigating the effect of weight factor (wi) on individual (di) and Excel used as an open database connectivity data source, and then
composite (D) desirability values in the optimization of a specic the regression analysis was performed. As regression models were
biosystem. Considering this gap in the literature, this study also in- solved, they were automatically sorted according to the goodness-
tends to explore the global optimal conditions by shifting the of-t criteria into a graphical interface on the DataFit numeric
shape of each individual desirability function by means of the computing environment. Moreover, t-ratios and the corresponding
weight factor. In the present study, a total of 19 different weight p-values were determined to appraise the signicance of the
factor scenarios were implemented computationally for a number regression coefcients. Descriptive statistics of the residual errors
of weight factors ranging from 0.1 to 8. For computational conve- were also provided for the appraisal of the multiple regression
nience and comparative purposes, the numerical solution of the model performance. An alpha (a) level of 0.05 (or 95% condence)
overall desirability function was conducted on an equally impor- was used to determine the statistical signicance of the predicted
tant response surface (I1 = I2 = I3) to avoid possible knowledge bias. 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.
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. ship between the observed methane production rate and efuent
Robert J. Vanderbei from Princeton University, NJ, USA), running COD concentration from the experiments carried out under
on a Pentium 4 CPU (Intel Atom Processor 3.00 GHz, 480 MB steady-state conditions and predicted methane production rate
of RAM) PC. SigmaPlot software (V10.0.0.54, Copyright 2006 and efuent COD concentration calculated by using Eqs. (21) and
Systat Software, Inc., GmbH, Germany) was used to create graphs (22) based on ChenHashimoto and modied StoverKincannon
for the computational results. kinetics.
11:83Si 11:83Si
3. Results and discussion /2 ! Se Si Si 22
13:02 Si =hh 13:02 LO
3.1. Kinetic modeling For the ChenHashimoto model, the value of determination
coefcient (R2 = 0.999) indicated that only 0.1% of the total varia-
Nonlinear regression-based results including several descriptive tions were not explained by the regression model. In addition,
statistics such as standard error of the estimate (SEE), sum of resid- the value of adjusted determination coefcient (R2a 0:999) was
uals (SR) residual average (RA), residual sum of squares (RSS), also very high, showing a high signicance of the model, as simi-
determination coefcient (R2), adjusted determination coefcient larly reported by Yetilmezsoy et al. (2009). In experimental terms
(R2a ) and number of nonlinear iterations (NNI) for ChenHashimoto (Yetilmezsoy, 2008), the effectiveness of the present process in
and modied StoverKincannon models are summarized in Table terms of COD removal was above 83% on average, meaning that
2. Moreover, regression variable results including standard error, most of the CODremoved was employed for a substantial quantity
t-ratios and the corresponding p-values for the implemented ki- of methane production in the volumetric organic loading rate
netic models are given in Table 3. range studied. It is noted that the maximum volume of methane
The dimensionless parameter (K) in the ChenHashimoto produced per gram of CODremoved ((Ym)max, L CH4/g CODremoved) is
model was proposed for the rst time for PMW as a new empirical determined according to the criterion that 0.35 L of methane is
formulation of volumetric organic loading rate in Eq. (21). The theoretically produced (at 1 atm, 0 C) per gram of CODremoved
efuent substrate concentration (Se) was predicted as a function (Martn Santos et al., 2010). Therefore, considering the biomass
of inuent substrate concentration, volumetric organic loading growth and cell maintenance stable, 0.392 L of methane is theoret-
rate, maximum utilization rate constant and saturation constant ically produced (at 1 atm, 32.5 C) per gram of CODremoved. The
[Se = f (Si, LO, Umax, KB)] in Eq. (22). Fig. 2a and b shows the relation- experimental results (Yetilmezsoy, 2008) indicated that about
Table 3
Regression variable results for ChenHashimoto and modied StoverKincannon models.
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
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
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.
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
Thiers, S., Aoun, B., Peuportier, B., 2010. Experimental characterization, modeling Environmental Engineering, Yildiz Technical University,
and simulation of a wood pellet micro-combined heat and power unit used as a Istanbul.
heat source for a residential building. Energy Buildings 42, 896903. Yetilmezsoy, K., Sakar, S., 2008a. Improvement of COD and color removal from UASB
Turkdogan-Aydinol, F.I., Yetilmezsoy, K., Comez, S., Bayhan, H., 2010. Performance treated poultry manure wastewater using Fentons oxidation. J. Hazard. Mater.
evaluation and kinetic modeling of the start-up of an UASB reactor treating 151, 547558.
municipal wastewater at low temperature. Bioproc. Biosyst Eng. 34 (2), 153162. Yetilmezsoy, K., Sakar, S., 2008b. Development of empirical models for performance
TWPCR (Turkish Water Pollution Control Regulation), Ofcial Gazette No. 25687 evaluation of UASB reactors treating poultry manure wastewater under
dated December 31, 2004. different operational conditions. J. Hazard. Mater. 153, 532543.
Videla, S., Chamy, R., Navarrete, E., Lema, J.M., 1990. Design and operation of Yetilmezsoy, K., Sapci-Zengin, Z., 2009. Stochastic modeling applications for the
anaerobic digesters using multi-objective optimization criteria. J. Chem. Tech. prediction of COD removal efciency of UASB reactors treating diluted real
Biotechnol. 49, 223231. cotton textile wastewater. Stochas. Environ. Res. Risk Assess. 23, 1326.
Wang, J., Wan, W., 2009. Application of desirability function based on neural Yetilmezsoy, K., Ilhan, F., Sapci-Zengin, Z., Sakar, S., Gonullu, M.T., 2009.
network for optimizing biohydrogen production process. Int. J. Hydrogen Decolorization and COD reduction of UASB pretreated poultry manure
Energy 34, 12531259. wastewater by electrocoagulation process: a posttreatment study. J. Hazard.
Ward, A.J., Hobbs, P.J., Holliman, P.J., Jones, D.L., 2008. Optimisation of the anaerobic Mater. 162, 120132.
digestion of agricultural resources. Bioresour. Technol. 99, 79287940. Yetilmezsoy, K., Turkdogan-Aydinol, F.I., Gunay, A., Ozis, I., 2011. Post treatment of
Yetilmezsoy, K., 2008. Treatability of poultry manure wastewater using anaerobic poultry slaughterhouse wastewater and appraisal of the economic outcome.
sludge bed reactor. PhD Thesis, Institute of Science, Department of Environ. Eng. Manag. J. 10, 16351645.