A Box-Behnken design-based model for predicting power performance in Microbial

Fuel Cells using wastewater

E.J. Martínez-Conesa1,, V.M. Ortiz-Martínez2,, M.J. Salar-García2,, A. P. de los Ríos3, F.

Corresponding authors E-mail: eusebio.martinez@upct.es; victor.ortiz@upct.es;

Abstract

Although modeling is regarded as a useful tool to understand the performance of

Microbial Fuel Cells (MFCs), the number of MFC models remains very low compared

with the number of experimental works available in the literature. Moreover, there are

very few MFC modeling attempts dealing with the use of wastewater as fuel in these

devices, which is essential for the practical implementation of MFCs since the potential

of this technology lies in the two-fold benefit of wastewater treatment and bioenergy

generation. In this work, a four-factor three-level Box-Behnken design was developed to

model the electrochemical power generation in two-chamber MFCs using wastewater as

fuel. The optimum values of temperature, external resistance, feed concentration and

anodic pH that maximized power output were investigated. Optimum conditions were

found at T = 35 °C and R = 1 kΩ, corresponding to a maximum power density of 0.88

W.m-3, while feed concentration and pH did not show statistical significance in the ranges

studied. Thus, a Box-Behnken design-based model as empirical approach could provide

an effective tool for the optimization study of MFC systems.

1
KEYWORDS: Microbial Fuel Cells; Modeling; Optimization; Response Surface

Methodology; Power density; Wastewater treatment.

1. INTRODUCTION

Microbial fuel cells (MFC) are a promising technology dealing with two of the most

pressing issues that modern society has to face such as demands for renewable non-fossil

fuels and the needs for usable water (Cheng et al., 2014; Logan et al., 2004; Salar-García

et al. 2016). In a MFC, bacteria degrade organic matter present in wastewater producing

electrons and protons (Liu et al., 2004; Menicucci et al., 2006). If bacteria are properly

attached to a conductive electrode material, electrons released as result of the oxidation

reaction at the anode can be transferred to the cathode through an external circuit

(Duteanu et al., 2010; Guo et al., 2014; Ortiz-Martínez et al., 2015a). In order to balance

the cell, protons go from the anode to the cathode through an internal membrane or

separator. Electrons and protons are consumed at the cathode usually in an oxygen

reduction reaction (ORR) to form water. Thus, there is an electromotive force in the cell

due the half-reactions taking place at the cathode and anode electrodes (Deng et al., 2010;

Hernández-Fernández et al., 2015; Yu et al., 2007).

The levels of power output generated in these devices are still relatively low and

therefore research efforts are being concentrated on improving MFC performance

(Degrenne et al., 2012; Kiely et al., 2011; Lanas et al., 2014; Wang et al., 2013).

Modeling is a powerful tool for studying and optimizing the operation of MFCs because

mathematical models can describe the processes that occur in these systems, covering

multiple scenarios with significant savings in terms of cost and time. Although there has

2
been a growing interest in MFC modeling in the last years with a resulting increase in the

number of MFC models released, this type of work remains scarce in comparison with

the number of experimental studies available (Oliveira et al., 2013). Among them, several

computational models focusing on the anode as limiting factor have been reported given

the importance of biofilm formation in MFCs (Ortiz-Martínez et al., 2015b). The model

developed by Marcus et al. (2007) can be highlighted as one of the most prominent

models belonging to this group. Their approach focuses on the electrical conduction

properties of the biofilm formed in the anode chamber. This one-dimensional model

employs the Monod and Nernst equations, anode-based mass balances and Ohm's law to

describe the electrochemical performance of the system. Picioreanu et al. (2007) also

presented a model based on the microbial activity of the anode, incorporating the Butler-

Volmer expression to calculate the current density generated in an electrochemical

mediator-based oxidation process. Picioreanu et al. (2008) integrated the before

mentioned model with the IWA Model (ADM-1) (Batstone et al., 2012) to cover the

competition between anodophilic and methanogenic microbial communities, since

methanogensis phenomenon severely limits MFC efficiency. Pinto et al. (2010) reported

a two-population model taking into account the competition between these two types of

microbial communities.

On the other hand, there are a few models studying both the anode and the cathode

compartments from an overall viewpoint, avoiding any a priori assumption that the anode

is the limiting factor of the system. In this context, Zeng et al. (2010) used the Monod

and Bulter-Volmer equations to describe the electrochemical performance of double-

3
chamber MFCs. With a similar approach, Oliveira et al. (2013) followed the model

proposed by Zeng et al., (2010) including the study of heat transport phenomena.

Sirinutsomboon (2014) presented a comprehensive model for single-chamber MFCs in

the absence of separator, comprising the modeling of biofilm formation by the Nernst-

Monod equations and distinguishing between endogenous and exogenous respiration

processes.

Finally, there are other works mainly focusing on the modeling of a key process or

component, such as ionic transport through ion exchange membranes used as separator in

MFCs (Harnisch et al., 2009) or the polarization response of these systems (Wen et al.,

2009). Predictive techniques and mathematical algorithms have been also used in this

field. Stratford et al. (2014) investigated the capability of several biological indexes in

predicting MFC power performance, and Yan and Fan (2013) used fuzzy control

combined with PID control to study double-chamber MFCs.

As can be seen above, several modeling approaches concerning MFCs have been

developed. However, the majority of them employ pure substrates and synthetic

wastewater prepared in the laboratory (e.g. acetate as carbon source and Shewanella

putrefaciens as bacteria population). There are very few attempts to model MFCs using

wastewater, despite the importance of this approach for the practical implementation of

MFCs since the great potential of this technology lies in the twin advantage of treating

wastewater and generating bioelectricity (Wen et al. 2009). These include the work by

Aljiveh et al. (2015a), who proposed a one-dimensional model based on the spatial

4
distributions of the different microorganisms including syntropic interactions. They also

combined several approaches from previous models in order to predict the performance

of MFCs using simple and complex substrates such as dairy wastewater (Aljiveh et al.

2015b). Besides, to the best of our knowledge, there is no attempt in the open literature

providing Box-Behnken-based empirical models to these devices. This empirical

approach offers a useful substitute for pilot scale or scale-up optimization studies.

Response surface design methodology (RSM) is based on a sequential set of designed

experiments to achieve the optimal response, allowing the relationships between

controlled and response variables to be studied for the optimization of a given process.

Box-Behnken designs are a well-known optimization tool that has been applied to many

chemical processes (Chaichi et al., 2013; Ferreira et al., 2007; Li et al., 2010). This work

presents a four-factor three-level Box-Behnken design to describe the electrochemical

performance of an experimental two-chamber MFC, studying factors and operational

parameters in MFC technology such as temperature, external resistance, feed

concentration and anodic pH that maximize power output.

2. MATHERIALS AND MEHTODS

2.1. Fuel And Reagents

Wastewater from the primary clarifier of a local wastewater treatment plant was used as

organic matter source to inoculate the MFCs set up in the present work (Murcia-Este

Plant, Spain). The wastewater used was characterized with a soluble chemical oxygen

demand (COD) of 430 mg.L-1, total organic carbon (TOC) of 48.5 mg.L-1, volatile

suspended solids (VSS) of 122 % and suspended solids (SS) of 126 mg.L-1. The value of

5
Soluble COD was fixed at the desired level before each experiment by mixing raw

wastewater with high COD wastewater from a local brewery industry (COD = 4100

mg.L-1). The final concentration was obtained by solving the set of equations in V1 and

V2:

V1C1 V2 C 2 VT CT
[1] and [2]
V1 V2 VT

where CT is the desired COD in composite water, VT is the desired volume in composite

water, C1, and C2 are the known COD values for wastewater and brewery water,

respectively, V1, and V2 are wastewater and brewery water volumes, respectively and VT

is the desired volume in composite water.

Cathode compartments were filled with phosphate buffer solution, pH 7.0 (monobasic

and dibasic potassium phosphates, Sigma-Aldrich, USA). The pH in the anode chamber

was adjusted by adding dropwise solutions of acetic acid (for acidic setpoints) and

sodium hydroxide (for basic setpoints) up to the desired value and monitoring the pH of

the mixture by using a pH meter. When COD and pH adjustments were both required,

COD adjustment is performed in the first place followed by pH adjustment. These

chemicals were purchased at the highest purity available (Sigma-Aldrich, Spain).

2.2. MFC Construction And Operation

The experimental MFC set-up consisted of double-chambered cells, each comprising two

250 mL glass bottles to form the anode and the cathode chambers, respectively.

Experiments were run in batch mode in cycles of 216 h in the absence of anode
6
recirculation and stirring. The bottles were water-jacketed to keep the operating

temperature constant (Schott, Germany). Both chambers were connected with a straight

glass tube acting as a bridge (inner diameter of 1.5 cm). A proton exchange membrane

Nafion-117 (DuPont Co., USA) acting as separator was placed between the anode and the

cathode chambers by using rubber gaskets and 30 mm rounded metal joint clips (J.P

Selecta, Spain). The temperature of the reactors was controlled with a circulation bath

Frigiterm-10 (J.P Selecta, Spain) (minimum cooling temperature of 10 ºC and a

maximum heating temperature of 100 ºC). An aquarium aerator and porous diffusers

were used to supply oxygen to the cathode chamber. Figure 1 shows a schematic

representation of the MFC system used for the present work.

The anode electrodes were made of carbon cloth (E-TEK, USA) while the cathodes

consisted of platinised titanium (William Gregor Ltd., UK), both prepared with the same

dimensions of 3x2 cm2. Nafion-based membranes were pre-treated in oxygen peroxide

(30% H2O2) and deionized water at 80 ºC for one hour, respectively, followed by 0.5 M

H2SO4 at 80 ºC for a further hour. Both electrodes and membranes were kept in deionized

water before each use. Electrodes were connected with titanium wires of 30 cm in length

(Sigma-Aldrich, USA). In experiments requiring resistance load, anodes and cathodes

were connected in closed circuit with an external resistance load of 1 k .

A central hole was drilled on the anode chamber cap to house the electrode. Oxygen was

removed from anode chamber by bubbling nitrogen during the start-up of the MFCs

investigated through the anode. Nut-and-septum inserts and polypropylene tubing were

used for anode sampling to avoid the exposure of the anode chamber to the open air.

7
Anode chambers were covered with aluminium foil to avoid typical light and temperature

day-night cycles.

2.3. Analytical Methods And Measurements

Organic soluble matter concentration was measured as COD (mg.L-1), which is defined as

the amount of oxygen necessary to completely oxidize the organic matter contained in a

sample. COD measurements were carried out by using the method described in APHA et

al. (2005). Anode sampling was conducted on a daily basis. 3 mL of sample were 0.45

μm filtered (Millipore, Spain) and then added to a test tube containing enough amounts of

COD reagents (Merck, Germany), with a final sample concentration in the 25-1500 mg.L-
1
range. Samples were digested during two hours at 150 ºC in an ECO16 thermoreactor

(Velp Scientifica, Italy) until reaction was complete. COD was measured in a

Spectroquant NOVA 30 (Merck, Germany). The percentage of elimination of soluble

organic matter (% COD) is expressed as a percentage with respect to the initial COD.

pH measurements were monitored with a pH electrode (Crison Cat. Nº 52-04) connected

to a pH/conductivity measurement device with temperature compensation (Crison

micropH 2000). The accuracy of the measurement was ± 0.01 pH units.

Voltage was measured by using a DVM-891 digital polymeter (HQ Power, Germany)

clipped to both sides of the external resistance load. Current (I) was calculated from

Ohms law:

Ecell
I [3]
Rext

8
where Ecell is the potential of the cell and Rext is the external resistance load.

Thus, power (P) can be calculated as:

2
Ecell
P [4]
Rext

Power is usually normalized to one of the characteristics of the MFC reactors. This way,

power outputs from different systems can be compared. In this work, power output was

normalized to the volume of the reactor. This allows engineering calculations for size and

costing of reactors to be performed:

2
Ecell
PV [5]
V ·Rext

where Pv is the power density (W.m-3) and V is the total reactor volume (empty bed

volume).

3. RESULTS AND DISCUSSION

3.1. Box–Behnken Design Based Model

The experimental design was selected according to a Box-Behnken matrix (see

Supporting Information) with high and low levels determined by the maximum and

minimum of the experimental range for each variable, respectively. Temperatures in the

experimental space ranged from 15-35 °C, wastewater feed concentration from 500-1500

mg.L.1, external resistance from 1-100 kΩ and pH from 5 to 9 units. Having these four

factors to be investigated, the number of possible scenario are 28, including 4 replicates

in order to assess experimental error (experimental numbers 11, 14, 26 and 28 in Table

1). Experiments were executed in blocks of 8 experiments and in a random order between
9
blocks so that any possible bias from previous results was avoided.

The experimental operation of fed-batch MFCs in closed circuit goes through several

stages. After an induction phase for the accommodation of microorganisms, there is an

exponential growth followed by a stationary phase and a final declining phase. The

duration of each phase depends on the initial content of organic matter and the type of

oxidant. Examples of the experimental results are depicted in Figures 2 and 3, in which

power density and COD removal evolution are plotted as a function of time for some

selected experiments. Response used for the model was maximum power density, which

typically develops after 24 – 72 hours depending on the experimental conditions.

In this model, the levels of four variables, temperature (T), external resistance (R),

wastewater initial concentration (C) and anodic pH was investigated for the maximization

of power density. A summary with all experimental results is shown in Table 1, which

includes the maximum power obtained for the conditions established in each scenario.

According to the RSM method, a quadratic model where ε is the experimental error will

suit for the purpose (Eq. 4.):

k k k j 1
2
y 0 i xi ii xi x xj
ij i [6]
i 1 i 1 j 2 i 1

This empirical model approach is very useful since it is easy to estimate and apply,

especially when little is known about the process. If Eq.6 is rewritten in matrix form for

our particular case:

[ PV ] [ X ][ ] [7]

10
Where matrix [X] represents the independent terms (namely, temperature, external

resistance, wastewater concentration and pH), matrix [PV] is the volumetric power

density (W.m-3) and matrix [ ] is the matrix of the coefficients of the model. As [X] is

not a square matrix (and consequently it lacks inverse), transposition of matrix X is

needed in order to solve the equation in [β]:

[ ] [[X T ][X]] 1[X][PV ] [8]

3.2. Model Resolution

In this paper, several forms of equation [6] were solved with the aid of commercial

software Sagata. These forms were linear, linear + interactions, quadratic (without

interactions) and full quadratic (including interactions). An ANOVA analysis was

performed in all cases and regression coefficient (r2), residual sum of squares (RSS) and

lack-of-fit (p-value) were registered and are summarized in Table 2 and in detail only for

the full quadratic model in Table 3.

As can be seen from Table 2, the values of r2 increases and the residual sum of squares

(S) reduces as number of parameters in the model rises. Only quadratic and full quadratic

models showed a significant correlation (p-value for lack-of-fit > 0.05). However, even

with those models regression coefficients, r2 values were relatively low (around 90 %).

The four replicates conducted (experiments numbers 11, 14, 26 and 28) ranged from a

power density from 0.029 W.m-3 to 0.179 W.m-3 with a mean of 0.102 W.m-3 and a

standard deviation of 0.057 W.m-3. This limited repeatability is recognized in the

11
literature (Larrosa et al., 2009) and is responsible for the relatively low r2 and RSS values

found with ANOVA.

A plot of the main effects for the linear model is represented in Figure 4. These plots can

be used to compare the magnitudes of the various main effects. A main effect occurs

when the mean response changes across the levels of a factor (the red dotted line is the

grand mean of the response data). Therefore, main effect plots can be used to compare

the relative strength of the factors. On the other hand, as can be observed in Figure 4,

both initial wastewater concentration and pH were not statistically relevant (within the

experimental space) for predicting power density as compared to temperature and

external resistance. Also, although the full quadratic model showed a polished

correlation, these parameters showed to be non-significant (p-values > 0.05, 95 %

confidence). Therefore, in a further refinement they were removed. Finally, the best

model for predicting power density was:

Pv (W/m3 ) 0,09117 0,02257·T - 0,35942·lo gR

In this expression, T represents the cell temperature, log (R) is the common logarithm of

the external resistance, C is the wastewater initial concentration and P, the pH of the

contaminated fluid. The ANOVA for this model (see Table 4) shows an acceptable r2

value balanced with a close to minimal residual sum of squares and all its parameters are

statistically significant (p-values < 0.05, 95% confidence).

12
A view of the response surface is illustrated as a contour plot in Figure 5a and in a three-

dimensional space in Figure 5b. As can be observed, power density increases with

temperature while it decreases as LogR increases, which would indicate that the response

of the system would be favored by high and low values of temperature and external

resistance load, respectively, within the ranges studied. Also, Figure 6 depicts the

predicted values vs. observed values. These are scattered randomly along the diagonal not

following any pattern, which is in agreement with homoscedasticity hypothesis, in which

these designs are based. Normality hypothesis of the data is also guaranteed as can be

inferred from the plot of the quartiles of the normal distribution vs. the residuals (Figure

7).

Once the prediction equation was shown to be sufficiently strong, it was used to obtain

the conditions that maximize power density for the experimental MFCs. In order to make

this optimization, partial derivatives of Eq. 7 were taken with respect to each one to the

factors and were equated to zero.

P
0.02257 0.01593·log R 0
T
[10]
P
0.35942 0.43654·log R 0.01593·T 0
log R

The resulting set of two equations was subjected to constrained optimization by using

Sagatac. The search was forced into the experimental space in order to avoid finding of

unrealistic optimum values. Accordingly, operating conditions that yield maximum

power density within the experimental design were found to be T = 35 °C and R = 1 kΩ.

Initial COD concentration and pH do not significantly affect the power within the

13
experimental space studied. Under these conditions the maximum power density was

found to be 0.83 W·m-3. The trends marked in this study point out towards optimal

performance for the MFC systems studied at high temperatures (i.e. hot climates) and

external resistances relatively low within the interval analyzed (microelectronics). On the

other hand, COD elimination was higher than 80 % in all cases, showing its potential use

in wastewater treatment. Finally, initial wastewater concentration within 500-1500 mg/L

did not reveal statistical significance on power response and could be considered

adequate for the MFC systems investigated. Other works that have studied this parameter

in a wider range have reported that feed concentration must be balanced, since high

organic load can favor the growth of methanogenic bacteria instead of anodophilic

microbial populations. Methanogenic bacteria promote the production of CH4 in the

anode instead of boosting electrical performance, thus affecting the power output

obtained (Pinto et al., 2010). In the same way, the effect of pH within the interval 5-9 on

power performance was not significant, even when subjected to slightly acid pH

conditions (pH=5). Compared with other modeling approaches commented in the

introduction part, Box-Behnken design based models can present some limitations such

as the lack of understanding of the dynamic behavior of the system. However, it can be

especially useful when little is known about the process (Ferreira et al. 2007). Moreover,

the optimum values could be used as starting point for searching optimum values with

more sophisticated non-linear models.

4. CONCLUSIONS.

14
In this work, a four-factor three-level Box-Behnken design was built to predict the

electrochemical power generation in two-chamber MFC systems using wastewater as

fuel. Optimal values for temperature (T), external resistance (R), wastewater initial

concentration (C) and anodic pH were investigated using a quadratic model including

linear, linear + interactions, quadratic (without interactions) and full quadratic (including

interactions) forms. Operating conditions that yield maximum power density within the

experimental design were found to be T = 35 °C and R = 1 kΩ, while Initial COD

concentration and pH did not significantly affect the power within the experimental range

tested. More MFC models are expected to be developed in the future due to their great

advantages for off-line process optimization, especially models dealing with wastewater

for the practical application of MFC technology for simultaneous production of energy

and water treatment.

ASSOCIATED CONTENT.

Supporting Information

Box-Benhken methodology. This material is available free of charge via the Internet at

http://pubs.acs.org.

ACKNOWLEDGEMENTS

This work has been supported by the Ministry of Science and Innovation of Spain

(MICINN), ref. CICYT ENE2011-25188, and the Séneca Foundation Ref. 18975/JLI/13.

V.M. Ortiz-Martínez and M.J. Salar-García are supported by grants FPU12/05444 and

BES-2012-055350, respectively.

15
 REFERENCES

1. Alavijeh, M.K., Yaghmaei, S., Mardanpour M.M. (2015a). A combined model for

large scale batch culture MFC-digester with various wastewaters through different

populations. Bioelectrochem., 106, 298-307.

2. Alavijeh, M.K., Mardanpour M.M, Yaghmaei, S. (2015b). One-dimensional

Conduction-based Modeling of Bioenergy Production in a Microbial Fuel Cell Engaged

with Multi-population Biocatalysts. Electrochem. Acta, 184, 151-163.

3. APHA, AWWA, WPFC. (2005). Standard Methods for examination of water and

wastewater, 21th edition, American Public Health Association, Washington DC.

4. Batstone, D.J., Keller, J., Angelidaki, I., Kalyuzhnyi, S.V., Pavlostathis, S.G.,

Rozzi, A., Sanders, W.T.M., Siegrist, H., Vavilin, V.A. (2002). The IWA Anaerobic

Digestion Model No 1 (ADM1). Water Sci. Technol., 45, 65-73.

5. Chaichi, M.J., Azizia, S.N., Alijanpoura, O., Heidarpoura, M., Qandaleeb, M.

(2013). Application of Box–Behnken design in the optimization of new peroxyoxalate

H2O2 chemiluminescence system using furan derivatives as blue activators. J. Lumin.,

138, 65–71.

6. Cheng, S., Ye, Y., Ding, W., Pan, B. (2014). Enhancing power generation of

scale-up microbial fuel cells by optimizing the leading-out terminal of anode. J. Power

Sources, 248, 931-938.

7. Degrenne, N., Buret, F., Allard, B., Bevilacqua, P. (2012). Electrical energy

generation from a large number of microbial fuel cells operating at maximum power

point electrical load. J. Power Sources, 205, 188-193.

16
8. Deng, Q., Li, X., Zuo, J., Ling, A., Logan, B.E. (2010). Power generation using an

activated carbon fiber felt cathode in an upflow microbial fuel cell. J. Power Sources,

195, 1130-1135.

9. Duteanu, N., Erable, B., Kumar, S.M.S., Ghangrekar, M.M., Scott, K. (2010).

Effect of chemically modified Vulcan XC-72R on the performance of air-breathing

cathode in a single-chamber microbial fuel cell. Bioresour. Technol., 101, 5250-5255.

10. Ferreira, L.C., Bruns, R.E., Ferreira, H.S., Matos, G.D., David, J.M., Brandão,

G.C., da Silva, E.G.P., Portugal, L.A., dos Reis, P.S., Souza, A.S., dos Santos, W.N.L.

(2007). Box-Behnken design: An alternative for the optimization of analytical methods.

Anal. Chim. Acta, 597, 179–186.

11. Guo, X., Zhan, Y., Chen, C., Zhao, L., Guo, S. (2014). The influence of microbial

synergistic and antagonistic effects on the performance of refinery wastewater microbial

fuel cells, J. Power Sources, 251, 229-236.

12. Harnisch, F., Warmbier, R., Schneider, R., Schröder, U. (2009). Modeling the ion

transfer and polarization of ion exchange membranes in bioelectrochemical systems.

Bioelectrochem., 75, 136-141.

13. Hernández-Fernández, F.J., Pérez de los Ríos, A., Salar-García, M.J., Ortiz-

Martínez, V.M., Lozano-Blanco, L.J., Godínez, C., Tomás-Alonso, F., Quesada-Medina,

J. (2015). Recent progress and perspectives in microbial fuel cells for bioenergy

generation and wastewater treatment. Fuel Process. Technol., 138, 284–297.

14. Kiely, P.D., Cusick, R., Call, D.F., Selembo, P.A., Regan, J.M., Logan, B.E.

(2011). Anode microbial communities produced by changing from microbial fuel cell to

17
microbial electrolysis cell operation using two different wastewaters. Bioresour. Technol.

102, 388-394.

15. Lanas, V., Ahn, Y., Logan, B.E. (2014). Effects of carbon brush anode size and

loading on microbial fuel cell performance in batch and continuous mode. J. Power

Sources, 247, 228-234.

16. Larrosa, A., Lozano, L.J., Katuri, K.P., Head, I., Scott, K., Godínez, C. (2009).

On the repeatability and reproducibility of experimental two-chambered microbial fuel

cells. Fuel, 88, 1852–1857.

17. Li, M., Feng, C., Zhang, Z., Chen, R., Xue, Q., Gao, C., Sugiura N. (2010).

Optimization of process parameters for electrochemical nitrate removal using Box–

Behnken design. Electrochim. Acta, 56, 265–270.

18. Liu, H., Ramanarayanan, R., Logan, B.E. (2004). Production of electricity during

wastewater treatment using a single chamber microbial fuel cell. Environ Sci. Technol.,

38, 2281-2285.

19. Logan, B.E., Hamelers, B., Rozendal, R., Schröder, U., Keller, J., Freguia, S.,

Aelterman, P., Verstraete, W., Rabaey K. (2006). Microbial Fuel Cells: Methodology and

Technology. Environ. Sci. Technol., 40, 5181-5192.

20. Marcus, A.K., Torres, C.I., Rittmann, B.E. (2007) Conduction-based modeling of

the biofilm anode of a microbial fuel cell. Biotechnol. Bioeng., 98, 1171-1182.

21. Menicucci, J., Beyenal, H., Marsili, E., Veluchamy, R.A., Demir, G.,

Lewandowski, Z. (2006). Procedure for determining maximum sustainable power

generated by microbial fuel cells. Environ. Sci. Technol., 40, 1062-1068.

18
22. Oliveira, V.B., Simões, M., Melo, L.F., Pinto, A.M.F.R. (2013). A 1D

mathematical model for a microbial fuel cell. Energy, 61, 463-471.

23. Ortiz-Martínez, V.M, Salar-García, M.J, de los Ríos, A.P., Hernández-Fernández,

F.J., Egea, J.A., Lozano, L.J. (2015a). Developments in microbial fuel cell modeling.

Chem. Eng. J., 271, 50-60.

24. Ortiz-Martínez, V.M, Salar-García, M.J, Hernández-Fernández, F.J., de los Ríos,

A.P. (2015b). Development and characterization of a new embedded ionic liquid based

membrane-cathode assembly for its application in single chamber microbial fuel cells.

Energy, 93, 1748-1757.

25. Picioreanu, C., Head, I.M., Katuri, K.P., Loosdrecht, M.C.M., Scott, K. (2007). A

computational model for biofilm-based microbial fuel cells. Water Res., 41, 2921-2940.

26. Picioreanu, C., Katuri, K.P., Head, I.M., Loosdrecht, M.C.M., Scott, K. (2008).

Mathematical model for microbial fuel cells with anode biofilms and anaerobic digestion.

Water Sci. Technol., 57, 965-971.

27. Pinto, R.P., Srinivasan, B., Manuel, M.-F., Tartakovsky, B. (2010). A two-

population bio-electrochemical model of a microbial fuel cell. Bioresour. Technol., 101

5256-5265.

28. Salar-García, M.J, Ortiz-Martínez, V.M, Hernández-Fernández, F.J., de los Ríos,

A.P. (2015). Scaled-up continuous up-flow microbial fuel cell based on novel embedded

ionic liquid-type membrane-cathode assembly. Energy, 101, 113-120.

29. Sirinutsomboon, B. (2014). Modeling of a membraneless single-chamber

microbial fuel cell with molasses as an energy source. Int. J. Energy Environ. Eng., 5, 1-

9.

19
30. Stratford, J.P., Beecroft, N.J., Slade, R.T.C., Grüning, A., Avignone-Rossa, C.

(2014). Anode microbial community diversity as a predictor of the power output of

microbial fuel cells. Bioresour. Technol., 156 84-91.

31. Wang, X., Gao, N., Zhou, Q., Dong, H., Yu, H., Feng, Y. (2013). Acidic and

alkaline pretreatments of activated carbon and their effects on the performance of air-

cathodes in microbial fuel cells. Bioresour. Technol., 144 (2013) 632-636.

32. Wen, Q., Wu, Y., Cao, D., Zhao, L., Sun, Q. (2009). Electricity generation and

modeling of microbial fuel cell from continuous beer brewery wastewater. Bioresour.

Technol., 100, 4171-4175.

33. Yan, M. and Fan, L. (2013). Constant voltage output in two-chamber microbial

fuel cell under fuzzy PID control. Int. J. Electrochem. Sci., 8, 3321-3332.

34. Yu, E.H., Cheng, S.A., Scott, K., Logan, B.E. (2007). Microbial fuel cell

performance with non-Pt cathode catalysts. J. Power Sources., 171, 275-281.

35. Zeng, Y., Choo, Y.F., Kim, B.-H., Wu, P. (2010). Modeling and simulation of

two-chamber microbial fuel cell. J. Power Sources. 195, 79-89.

20
Table 1. Summary table with all experimental results.

Test T Log R C pH PV, max Imax

(°C) (mg·L-1) (W·m-3) (*) (A·m-2)

1 25 1 500 9 0.111 3.002·10-5

2 25 1 1500 9 0.172 3.739·10-5

3 35 1 1500 7 0.056 2.141·10-5

4 35 1 1000 9 0.091 2.727·10-5

5 35 2 1000 7 0.015 3.538·10-6

6 25 2 1500 7 0.018 4.752·10-6

7 15 1 500 7 0.056 2.141·10-5

8 15 0 1000 7 0.241 1.406·10-4

9 25 1 500 5 0.327 5.179·10-5

10 35 1 500 7 0.178 3.795·10-5

11 25 1 1000 7 0.029 1.531·10-5

12 15 1 1000 9 0.019 1.239·10-5

13 15 1 1500 7 0.009 8.315·10-6

14 25 1 1000 7 0.070 2.352·10-5

15 15 1 1000 5 0.217 4.221·10-5

16 25 2 500 7 0.024 4.435·10-6

17 15 2 1000 7 0.012 3.094·10-6

18 25 0 1000 9 0.848 2.474·10-4

19 35 0 1000 7 0.881 1.628·10-4

21
20 25 2 1000 9 0.006 2.294·10-6

21 25 1 1500 5 0.082 2.598·10-5

22 25 0 500 7 0.641 2.293·10-4

23 25 0 1000 5 0.713 2.396·10-4

24 25 0 1500 7 0.508 2.012·10-4

25 25 2 1000 5 0.005 1.912·10-6

26 25 1 1000 7 0.131 3.301·10-5

27 35 1 1000 5 0.130 3.276·10-5

28 25 1 1000 7 0.179 3.817·10-5

(*) Conversion from W·m-3 to W·m-2 may be accounted by multiplying volumetric power

density by a 1.1·10-4 factor which includes anode specific surface, anodic chamber

surface-to-volume ratio and graphite density.

22
Table 2. Basic statistics of the model studied.

r2 (%) S Lack-of-fit (p-value)

Linear 61.5 0.1638 0.068

Linear+interactions 66.6 0.1790 0.040

Linear+quadratic 84.2 0.1165 0.165

Full quadratic 90.1 0.1127 0.163

23
Table 3. a) Regression and b) ANOVA data for the full quadratic model.

a. Regression

Term Coef SE Coef T P

Constant 0.1022 0.05637 1.814 0.095

T 0.0565 0.03565 1.584 0.139

Log R -0.3111 0.03565 -8.725 0.000

C -0.0410 0.03255 -1.260 0.232

pH -0.0106 0.03255 -0.325 0.751

T*T -0.0504 0.04827 -1.045 0.317

Log R*Log R 0.2129 0.04827 4.411 0.001

C*C -0.0009 0.04660 -0.019 0.985

pH*pH 0.0790 0.04660 1.695 0.116

T*Log R -0.1294 0.07130 -1.815 0.095

T*C -0.0187 0.05637 -0.333 0.745

T*pH 0.0397 0.05637 0.705 0.494

Log R*C 0.0317 0.05637 0.563 0.584

Log R*pH -0.0585 0.05637 -1.038 0.320

C*pH 0.0765 0.05637 1.357 0.200

S = 0.1127 r2 = 90.1 %

b. ANOVA

Source DF Seq SS Adj SS Adj MS F P

Regression 14 138.259 138.259 0.098756 7.77 0.001

24
Linear 4 0.94458 0.98931 0.247327 19.46 0.000

Square 4 0.34727 0.36040 0.090100 7.09 0.004

Interaction 6 0.09073 0.09073 0.015122 1.19 0.374

Lack-of-Fit 9 0.13940 0.13940 0.015489 3.54 0.163

Pure Error 3 0.01312 0.01312 0.004374

Total 26 153.511

25
Table 4. a) Regression and b) ANOVA data for the definitive model.

a. Regression

Term Coefficients SE Coef. T P

Constant 0.1161 0.02846 4.079 0.000

T 0.0664 0.03549 1.871 0.045

Log R -0.3210 0.03549 -9.045 0.000

Log R*Log R 0.2183 0.04549 4.798 0.000

T*Log R -0.1593 0.06970 -2.285 0.032

S = 0.1138 R2 = 85.4 %

b. ANOVA

Source Degrees of Seq SS Adj SS Adj MS F P

freedom

Regression 4 125.009 125.009 0.31252 24.12 0.000

Linear 2 0.92307 106.429 0.53214 41.08 0.000

Square 1 0.25940 0.29828 0.29828 23.02 0.000

Interaction 1 0.06763 0.06763 0.06763 5.22 0.032

Lack-of-Fit 3 0.08753 0.08753 0.02918 2.81 0.097

Pure Error 19 0.19749 0.19749 0.01039

Total 26 153.511

26
Figure 1. Schematic representation of the MFC system used.

27
Figure 2. Volumetric power density as a function of time for some selected experiments.

28
Figure 3. COD evolution as a function of time for some selected experiments.

29
Figure 4. Main effects plots for the linear model.

30
Figure 5. Graphical representations of the response surface as a function of T and R: a)

Countor plot; b) Wireframe plot.

31
Figure 6. Model prediction vs. observed values.

32
Figure 7. Normality test.

33