Membrane-less microbial fuel cell: Monte Carlo simulation and sensitivity analysis for COD removal in dewatered sludge

The lack of access to clean electricity in developing nations has given importance to the development of low-cost, widely applicable energy technologies. As reported by Jollands, about 1.4 × 10⁹ people around the world had no access to electricity, of which 2.7 × 10⁶ of them depended on the traditional method of using biomass.¹ Most of these people originated from rural areas. It was predicted that this problem would continue to increase steadily until 2030. Of the 1.4 × 10⁹ people, the majority are living in Sub-Saharan Africa, Africa, India, and other developing Asian countries.² Sub-Saharan Africa continues to face the biggest challenge in terms of electricity access where only 31% of the population have an electricity source, and this is the lowest level recorded in the world.³ Tables I and II present the data of population without access to electricity and data of population without access to electricity in terms of region, respectively.

In order to avoid relying on non-renewable energy resources (crude oil, natural gas, etc.), energy-saving technologies need to be established.^3,4 Hence, renewable energy is believed to be the preferred choice as it contributes toward the sustainability of the environment as well. The world has clearly envisioned a brighter future in the energy industry by implementing more renewable energy sources (wind, hydro, and biomass) and upgrading their role in energy generation because they are green, clean, and environmentally friendly.^3,5 At the same time, energy production from microbial fuel cells (MFCs) is emerging as a viable source for electricity production as this system has the ability to regulate natural bacterial metabolism in order to produce electricity.⁶ In this study, the substrate used was dewatered sludge obtained from a municipal wastewater treatment plant (MWTP).

There are many unexplored energy producing resources available in the dewatered sludge that are made up of various forms of biodegradable organic matter.^8,9 Research in Canada reported that the sludge in a wastewater treatment plant located in Toronto consisted 9.3 times more energy than the amount that was used to treat the wastewater.¹⁰ Based on the finding of Logan, sludge originating from the wastewater of livestock and food industries is estimated to contain about 17 GW of energy, and this is equivalent to the energy that is needed by the U.S to power their entire water infrastructure. Due to this, producing and recovering energy from sludge are very promising as it can be used to power its own treatment plant.¹¹ Dewatered sludge is generated daily from the Indah Water Konsortium (IWK) treatment plant, and this sludge was analyzed for its capabilities to support the growth of electrogenic bacteria (EB) in order to initiate electricity generation.^12,13 If it is successful, the sludge can be used as a value added substrate instead of polluting the environment.

Almost 3 × 10⁶ cubic meters of sludge was produced annually, and it was also projected that IWK will be generating 7 × 10⁶ cubic meters of sludge by the year 2020.¹⁴ This makes it the most favorable raw material for bioconversion as it is renewable and lavishly available.⁸ The EB’s effectiveness and economic viability of converting the sludge into bioenergy rely on the sludge characteristics and compositions.^15,16 Previously, a large number of substrates were discovered to be suitable to be used as feed for MFCs, which range from artificial wastewater and real wastewater to lignocellulosic biomass. Zuo and his co-workers used the end product of stover in the steam-explosion process (e.g., hydrolysates or solids), and they crumpled the stover into powder before being used as an electricity producing substrate.¹⁷ These processes made lignocellulosic biomass easier to biodegrade but also resulted in a process that was less sustainable and hard to be scaled-up.¹⁸ Regarding this matter, it is a must to explore new and abundantly available substrates, such as dewatered sludge.^9,12 In this study, dewatered sludge was used as the main substrate that played the dual role of being the nutrient-rich anodic pole and the pseudomembrane. The usage of the membrane-less MFC (ML-MFC) configuration reduces the operation cost. The proximate composition of the tested dewatered sludge was also analyzed. The dewatered sludge is diverse in composition and varies in characteristics;¹⁴ in addition, the MFC performance has a tendency to fluctuate.¹⁸ Thus, it is possible to estimate a range of values in terms of the effectiveness of the MFC in producing electricity by applying uncertainty utilization substrate studies using Monte Carlo simulation.^19,20

A computerized mathematical technique, namely, Monte Carlo simulation, allows people to project the range of possible outcomes and the chances for any choice of action in terms of the effectiveness.²¹ The Monte Carlo algorithm was originally introduced based on the game of chance and got its name from a city in Monaco, “Monte Carlo,” which is famous for casino gambling. Owing to its flexibility and simplicity, the algorithm attracted many researchers to exploit its ability in various research fields. The core of the Monte Carlo algorithm is that it uses a random number to exemplify the variability or fluctuation of a system by a particular probability distribution.²² Researchers need to have a specific aim of research output when dealing with probability as it may vary the approach depending on the purpose of the study; thus, it is easier to project as the system is already clearly defined.²³ During World War II, scientists had clearly demonstrated this technique on the explosion of an atom bomb, and since its introduction, the Monte Carlo simulation has been used to model a variety of physical and conceptual systems. It is also being implemented in various areas, such as medical and health research,²⁴ material engineering,²⁵ chemical engineering, and product design,²⁶ due to its reliable projection.

Previous researchers found it hard to predict what the actual value of chemical oxygen demand (COD) removal in the ML-MFC would be, but the estimation can be done based on historical data. Two kinetic models have been tested to explain the biomass growth and the organic consumption (COD removal); namely, logistic and modified Luedeking–Piret model. These kinetic models gave the ability to researchers to monitor the significant parameter in ML-MFC technology, which covers the specific growth rate of EB, process yield, reaction of processes, process control criteria, and approaches for production of ML-MFC’s products and scaling up consideration.²⁷ In the Millennial era, scientists in bioprocessing fields had shifted to new approaches using more advanced and updated technology with smarter and better-controlled processes.²⁸ Therefore, this study is carried out using simple and easily comprehensible mathematical descriptions of the kinetic model of bacterial growth in a ML-MFC in order to examine the behavior of EB in a ML-MFC system. Meanwhile, the modified Luedeking–Piret equation model was implemented to describe the substrate utilization inside the system.²⁹ The modified Luedeking–Piret model was also used in the uncertainty analysis to create a more realistic picture of the COD removal, which has not been studied so far.

Dewatered sludge was collected from the Kerian Indah Water Treatment Plant, Parit Buntar, Perak, Malaysia. All preservation steps (kept at 4 °C in a cold room) and testing for sludge characterization follow the method used by the group research led by Muaz.³⁵ 

The ML-MFC was constructed using 10 × 10 cm² cylindrical PVC reactors. The electrodes were made of graphite felt (radius of 4.6 cm, thickness of 0.65 cm, and surface area of 0.0064 m²). The pH (6.0), moisture content [30% (v/w)], and electrode distance (3 cm) conditions were setup as used by Muaz et al. (Fig. 1).^38,39 Throughout the experiment, the electricity generated was measured using a digital multimeter (UNI-T UT33D).

The kinetic models proposed in this study for describing the growth and substrate utilization are the logistic model and modified Luedeking–Piret model, respectively. The models were fitted to the experimental data by non-linear regression. From the records, it is clear that the research led by Dhanasekar,²⁹ Elibol and Mavituna,⁴⁰ and Pazouki’s research group⁴¹ had also used the models for describing those processes (growth and substrate utilization). The Luedeking–Piret-like model also facilitates a bioengineer to predict and/or control microbial processes by helping to reduce the failure risk of a bioprocess from an empirical approach to a larger scale-up production. The goal was to make use of this mathematical modeling to reduce a complex (biological) system into a simpler (mathematical) system that can be analyzed in far more detail and from which key properties can be identified.

The linearization equation of each model (summarized in Table III) was derived to obtain the predicted initial condition value from the experimental result. For data analysis, the Polymath 5.1 software (CACHE Corporation., USA) was used.

The Monte Carlo simulation is particularly useful for problems that involve a large number of degrees of freedom. Thus, it is used to determine how an error or variability in the system could affect the final result.¹⁷ Most common problems in any research area stem out when the need to investigate the behavior of the system over a wider range of parameter values arises. In this situation, simulations were performed for more than one value of the parameters of interest.¹⁸ In this study, the Monte Carlo simulation is used to project the COD removal by the ML-MFC with the help of EB to obtain persistent outcomes. The values of the initial biomass (Xo), initial substrate concentration (So), maximum specific growth rate (μm), growth-associated substrate consumption coefficient (γ), and non-growth-associated substrate consumption coefficient (δ) were set similar to the values obtained from the ML-MFC kinetic models with variation around 5% standard errors; these values are aligned with that in the research carried out by Samsudin and Don²² in predicting the initial value kinetic parameter of S. cerevisiae in bioethanol production. To monitor the inconsistency and variability in the kinetic model parameters in COD removal, standard variation was used. Minitab 17 software version 17.1.0 was used to generate 100 000 replicates randomly to exemplify the variability of the COD removal in the ML-MFC with a normal distribution. Then, the simulated COD removal value was calculated accordingly.

All the kinetic parameters were tested for their sensitivity toward the COD removal performance via a sensitivity analysis test. During the analysis, each kinetic model parameter was varied to ±10% and ±50% of its original value except for the X_m value for which the changes were ±10% and ±50% of the increment from the X_o. This method was proposed by Samsudin and Don²² in order to see how each kinetic parameter contributed to the final result of the COD removal. Zhang and co-workers stated that the impact of each kinetic parameter on the COD removal was evaluated by calculating the value of response (S) as follows:

where S₂ is the COD removal value with single kinetic model parameters being varied to ±10% and ±50 and S₁ is the COD removal value with all the other model parameters remaining the same. S₁ and S₂ were calculated as shown in section Determination of COD removal.

The raw material, dewatered sludge, was the same substrate as used in the research of Muaz et al. (2019) and was taken from the MWTP Kerian Indah Water Treatment Plant, Parit Buntar, Malaysia.³⁵ The proximate analysis of the dewatered sludge was done using an elemental analyzer (PerkinElmer 2400 Series II), atomic absorption spectroscopy (AAS) (Shimadzu AA-6650), and a COD digester (Checkit Direct, Lovibond).

The growth of EB inside the ML-MFC was represented by the biomass and was evaluated using a volatile solid (VS) test (APHA 2017).⁴² The EB attach to the anode surface and form a biofilm. Samples of the biofilm were collected from the surface of the anode and subject to the VS test.

The isolation method carried out by Muaz and Murugan (2019) was used to extract the EB from the dewatered sludge.³⁸ The dewatered sludge was dissolved into a stock deionized water solution and diluted in serial dilution before being spread onto agar for isolation growth. The EB were streaked in an agar slant. To determine the growth, the phylogenic analysis was carried out by outsourcing it to a microbiology company Macrogen in South Korea with the primers used for the polymerase chain reaction analysis being 27F 5’ (AGA GTT TGA TCM TGG CTC AG) 3′ and 1492R 5′ (TAC GGY TAC CTT GTT ACG ACT T) 3’. The sequences obtained were analyzed using the National Center for Biotechnology Information (NCBI) online nucleotide BLAST tool and ribosomal database-II to identify the taxonomic hierarchy of the sequences. Taxonomically related 16S rRNA gene sequences were obtained from the NCBI nucleotide database. The sequences collected were aligned using the MUSCLE multiple sequence alignment algorithm. The phylogenetic tree was constructed and inferred using the neighbor joining method and validated using the bootstrap method (1000 replications). The evolutionary distances were computed using the maximum composite likelihood method and are presented as the number of base substitutions per site. All positions containing gaps and missing data were eliminated. All analyses were performed using MEGA6.

A digital multimeter was used to monitor the electricity generation. The voltage, power, and current were calculated using Ohm’s law,¹⁷ 

where V is the potential difference between two electrodes (in volts), I is the current (in amps), and R is the resistance applied (in ohms).

For Monte Carlo simulation and sensitivity analysis, the COD removal value was determined based on Eq. (3), which calculated how much COD had been consumed by EB in the ML-MFC using the modified Luedeking–Piret model.²² 

COD removal, S $mgl$⁠,

The dewatered sludge analysis using the elemental test showed that it consisted of 30% carbon, 6.7% nitrogen, 3.7% hydrogen, 43.4 mg/L phosphorus, and 2.5 mg/L potassium. Meanwhile, the AAS analysis discovered elements such as Fe, Zn, Cd, Mn, and Ni at 1.4, 4.3, 0.1, 1.8, and 15.6 ppm, respectively. These elements were considered to be the micronutrient source for the enhancement of bacterial growth, which contributed as the building compound and also co-factored for proton energy generation inside the cell.³⁰ The COD value of the dewatered sludge was found to be at 535 ± 9.01 mg/L.

Throughout the experiments, it was confirmed that an increase in incubation time resulted in a steady escalation of COD removal and generation of voltage. This is because the growth of the biomass and the generation of voltage were associated as both the growth and electricity generation increased in parallel during the incubation period. As shown in Fig. 2, consistent with the works done by Muaz and Vadivelu, EB have a long lag phase that spans from 0 h to 35 h due to their adaptation phase to the new environment in the ML-MFC.³⁸ The exponential phase then lasted up to 168 h prior to moving to the stationary phase. During the exponential phase, cell division was very fast, which increased the cell biomass productivity.³¹ Based on Fig. 2, it is clear that the EB reached their stationary phase after 168 h because after that the growth rate was equivalent to the death rate. Hence, no net increase in viable bacterial cell numbers was observed and cellular metabolic activity started to decline. The stationary phase then started as the essential nutrients required for bacterial growth were exhausted and waste by-products accumulated.^32,38 The maximum working voltage and EB biomass production recorded during the stationary growth phases were at 28.4 ± 0.35 mg/g and 914.1 ± 8.48 mV, respectively, after an incubation period of 168 h (Fig. 2).

The EBs colonized at the anode electrode were isolated and sent to the microbiology company Macrogen, South Korea. The analysis revealed the presence of genus Bacillus subtilis (BS). The presence of BS impacts the MFC as it has good biocatalyst ability in the ML-MFC and it can generate energy stably. It also secretes several enzymes that play a significant role in hydrolyzing the DS’s component. The enzymes secreted are protease, glycolipid, α-amylase, endo-β-glucanase, penicillin acylase, 50-inosine monophosphate, and riboflavin and will be used to strengthen the microbial biochemical pathway. This finding shows why a stable energy production was able to be obtained via this study as the presence of BS enhances the oxidation processes around anodes.

Figures 3 and 4 represent the simulated and experimental profiles of BS growth and substrate consumption (COD removal) when the dewatered sludge is used as the substrate in the ML-MFC. The BS growth trend obtained followed the typical cell growth trend explained by Mitchell et al. (2004)³³ and Zwietering et al. (1990),³⁴ which consists of three phases: (i) lag phase (adaptation phase) (0–35 h), (ii) log phase (86–168 h), and (iii) stationary phases. Obviously, the trend of BS growth and COD removal in this study exhibits a growth-consumption associated formation, where both processes (BS growth and substrate consumption trends) occurred concurrently.

Table IV presents the simulated profile of BS growth (logistic model) and COD removal (modified Luedeking–Piret model) and their kinetic model parameters with the calculated statistical error parameters of each tested model. From the observation of the high R² value range obtained (0.970–0.994) and low RMSE value, a competent trend of fitting with the experimental data is clearly defined by the observation of the high R² value range achieved (0.970–0.994) and low RMSE value; hence, it was adapted to be utilized to characterize the growth and COD elimination behavior. The growth-substrate associated consumption of BS could be further proven based on the profiling of ϒ values and δ values (ϒ values were higher than δ values) throughout the experiments.²⁹ When compared to the other research that focuses on the growth of a suspended microbial cell in a mixed substrate, Karin and Thomas (1998) stated that the modified Luedeking–Piret model could be used as a tool for the optimization process. The model managed to describe the substrate consumption in a conclusive manner throughout the experiment.

From Figs. 3 and 4, it is obvious that the COD removal by the bacteria impacted the voltage generation. Once a carbon source was consumed (oxidized) by BS, a higher potential was created between the anode and the cathode, which was recorded.^35,36 From the microbiological perspective, degraded carbon sources went through metabolic processes (glycolysis, citric acid cycle, and finally, oxidative phosphorylation) where the carbon sources were degraded to generate highly reduced bio-molecules (NADH), which act as electron transporters.³⁷ These processes occur at the surface of the anode through which the electrons from NADH (or any complex protein) are passed. The electrons transported from the electrode reacted with the oxygen and protons at the cathode, and thus, electricity was generated.³⁸ There was a need to monitor how much COD could each ML-MFC remove, and due to this, the uncertainty analysis had to be performed.

The highest COD removal value attained by ML-MFC was 178.7 mg/l. The COD removal data obtained fitted well and were used to describe the process behavior. From this analysis, the initial biomass value obtained (X_o) was 0.021 mg/g, the initial COD (S_o) present was 535 ± 9.01 mg/L, the growth-associated product formation coefficient (α) obtained was 2.719 mg/g, and the non-growth-associated product formation coefficient (β) was 0.011 mg/g h (Table V). It is found that the maximum specific growth rate (μ_m) and the maximum attainable biomass (X_m) values estimated were at 0.041 h⁻¹ and 31.1 mg/g, respectively.

The novelty of the study will be highlighted when the repeatability of the experiment could be predicted. If the ML-MFC is run by other researchers or the authors several times, the COD removal might be different due to the heterogeneous properties of dewatered sludge and the BS performance, which are reflected by system conditions. Hence, a consistent trend of the ML-MFC performance needs to be projected. Thus, the predictability analysis of COD removal was determined with the variability of the kinetic parameters as a basis using the Monte Carlo simulation (Minitab 17 software). The Minitab software version 17.1.0 blasted 100 000 random data (range deviation had 5% standard error), and the normal distribution curve (Table IV) represented the predicted values of X_o, X_m, μm, ϒ, and δ. The variability of the process parameters that contributed to the COD removal efficiency was calculated accordingly. The distribution of the predicted COD removal is shown with a histogram of data in Fig. 5.

The normal distribution exhibits that the COD removal is normally distributed and slightly negative for skewness (−0.01) and kurtosis (−0.011), which means that it is a left skew (left tail extends farther than the right tail). This indicated that the distribution obtained was sharper than the normal peak. After 100 000 samples were assessed, the predicted COD removal value (S) was around 169.99–187.88 mg/L with the X_m values being in between 29.545 and 32.655 mg/g (Table V). All data were presented in terms of means ± standard deviation. Clearly, the simulation result fits well with the experimental results whereby the experimental COD removal value of 178.7 mg/L falls within the estimated range. Hence, by implementation of the Monte Carlo algorithm on the experimental data, the COD removal in every ML-MFC can be predicted so that a consistent value will be obtained even though the dewatered sludge composition varied around 5% from the current samples in this study.

To strengthen the uncertainty analysis of COD removal, the study needs to determine the most significant kinetic parameter that contributed to the huge impact on the COD removal value; thus, sensitivity analysis was carried out. The range of COD removal value of the present study was quite wide (169.99–187.88 mg/L); thus, sensitivity analysis was further carried out to evaluate the significance of each parameter on the COD removal by varying each kinetic parameter to ±10% and ±50%. The outputs obtained are presented in Table VI. There would be an increment and decrement in COD removal value with the sign being positive and negative, respectively. This evaluation test was useful to identify the impact of experimental errors on the model output and clarified the most impactful (sensitive) parameters that affected the performance of COD removal of dewatered sludge in the ML-MFC. All six kinetic parameters in Table VI recorded the result involved parameter in the COD removal model, and each of the parameters shows the amount of effectiveness in the performance of the COD removal in the ML-MFC. The greatest impact on the COD removal was the maximum specific growth rate (μ_m) followed by the maximum biomass (X_m) value. As stated previously, the trend of BS and COD removal is a growth-substrate associated formation. This explains why the maximum specific growth rate (μ_m) contributed to a very significant impact on the COD removal. The sensitivity analysis proved that in order to have high efficiency to treat the dewatered sludge (reduce the COD value) using ML-MFC technology, it was important to have an optimum condition for the BS to let it grew well in the system.

This study demonstrates that long-term operation of a ML-MFC could stabilize the dewatered sludge by removing the COD. The highest COD removal value obtained experimentally was 178.7 mg/L. Since the amount of dewatered sludge from the wastewater treatment plant fluctuated, there must be a projected benchmark range so that the COD removal could be predicted. Thus, the application of the Monte Carlo simulation showed the simulated and the prediction trend and repeatability of the kinetics model parameters, which fluctuated around 5%. From 100 000 simulated samples tested, the COD removal value obtained was 172.58–185.02 mg/L. Sensitivity analysis clearly demonstrated that the performance of COD removal in the ML-MFC was highly dependent on the increment of BS biomass.

The authors would like to thank Universiti Sains Malaysia for the financial support via Research University Short Term Grant No. 304/PTEKIND/6315353 and support from the Ministry of Higher Education, Malaysia, under the Fundamental Research Grant Scheme (FRGS) (203/PTEKIND/6711823).

The authors have declared no conflict of interest for the manuscript.

Muaz Mohd Zaini Makhtar and Husnul Azan Tajarudin worked on the experimental setup plus the arrangement of the ideas for the manuscript. Vel Murugan Vadivelu focused on the cultivation of the micro-organism, while Mohd Dinie Muhaimin Samsudin, Noor Fazliani Shoparwe, and Nor ‘Izzah Zainuddin dealt with the statistical analysis of the experiment data.

The data that support the findings of this study are available within the article.