# BioVolt Comprehensive Research ## A smart microbial fuel cell with a digital twin and artificial intelligence for turning wastewater into electricity > Prepared as the reference research for rebuilding the BioVolt website — September 2026 > Every number in this document is either (a) a fixed physical law, (b) a published value with its source, or (c) a calculation derived from those and shown step by step. Unverified values are explicitly marked. --- ## Contents 0. Executive summary 1. What the current site contains and what must change 2. The BioVolt idea in precise scientific terms 3. How a microbial fuel cell works 4. The bacteria: who they are, how they live, reproduce, die, and generate electricity 5. The complete calculations (the core of this research) 6. Real-world numbers: what these cells actually produce 7. The digital twin 8. The artificial intelligence 9. Fault catalogue: physical signatures, detection, and response 10. Hardware and sensors 11. Website rebuild plan 12. From solid organic waste to electricity 13. Disturbances and recovery: literature evidence and simulation 14. Reducing human intervention: what gets automated and what does not 15. Cell configurations and applications 16. Materials, cost, scale-up, and economics 17. Is it really clean energy? Environmental impact and sustainability goals 18. Digital twin standards and maturity levels 19. Data pipeline and experimental protocol 20. Safety 21. Coverage matrix: every sentence of the BioVolt idea and where it is proven 22. Glossary 23. Model limits and scientific honesty 24. References --- ## 0. Executive summary BioVolt combines four things: a real microbial fuel cell (MFC), sensors that measure its condition in real time, a digital twin that models it mathematically, and an artificial intelligence that reads the data and makes operating decisions. The main findings of this research: 1. **The calculations can be fully correct physically**, because they rest on fixed laws: Faraday's law (every gram of organic matter measured as COD carries exactly 12,061 coulombs), the Nernst equation (theoretical maximum voltage 1.10 V), and conservation of mass and electrons. These are not estimates. 2. **The biological part has measured constants, but as ranges** — for example bacterial growth rate (1.97 to 3.2 per day). The scientifically correct approach is to use published values as a starting point and then **have the digital twin calibrate them from the real cell's data**. This is precisely what justifies the digital twin's presence in the project. 3. **I built a complete mathematical model and tested it.** With no manual tuning it reproduced published experiments: open-circuit voltage 0.804 V (measured 0.798), maximum power 32 mW/m² for a cell with 1286 Ω internal resistance (measured 38–40), and a bacterial doubling time on the electrode of ≈ 7.5 hours (measured 6–8 hours). 4. **A pivotal finding**: in lab-scale cells the bacteria are not the bottleneck. Their capacity allows 3 to 7 A/m², while the cell only draws 0.1 to 0.4 A/m² from them. The real limiter is **the cathode and the internal resistance**. Therefore the most valuable thing the AI can do is **match the electrical load (MPPT)** and protect the bacteria from lethal conditions, not "feed them more". 5. **The honest numbers**: a cup-sized lab cell produces **tens to hundreds of microwatts**, not enough to light an LED continuously except through a supercapacitor discharged in bursts. The largest pilot projects in the world (1000 L) recovered about 0.033 kWh per cubic metre, i.e. 1.5–2% of the energy latent in the wastewater. The technology's real value today is **treating water with far less energy** while producing some electricity. 6. **Solid organic waste works, but slowly**: bacteria cannot eat solids directly; the solids must first be broken down (hydrolysis), and that step is the bottleneck. With a hydrolysis rate measured inside MFCs (0.0576 per day), a continuously fed cell with a one-day retention time breaks down only **5.4%** of the solids. Solid waste therefore needs **pretreatment** (fermentation or enzymes) or long retention times, and its best published power is 1–371 mW/m² (up to 1540 with fermentation pretreatment). 7. **Recovering from disturbances is where the AI proves its value**: the literature shows that faults caught within **15 minutes** recover fully, while acid shocks left alone took on average **60.7 hours** to recover and 3 cells out of 17 never recovered. In my simulation, a severe acid shock cut power by 100% and lost 11.9% of ten days' energy with a fixed resistor, while the full BioVolt controller (early detection + buffer dosing) kept the loss at **0%**. 8. **A warning found by simulation**: maximum power point tracking (MPPT) on its own, without a safety layer, made recovery **worse** (134 hours and 40.2% of energy lost), because the algorithm loses its signal when power falls to zero and drifts to the wrong resistance. Freezing MPPT during a collapse fixes this (46 hours). The safety layer is therefore mandatory, not optional. In normal operation MPPT raised energy by **36.5%** over a fixed 1000 Ω resistor. 9. **Environmental honesty**: life-cycle assessments show that a stand-alone MFC is not automatically cleaner than anaerobic digestion, because materials such as platinum, Nafion, and carbon cloth dominate its footprint. The environmental case depends on cheap, low-impact materials (an activated-carbon cathode costs ≈ $15/m² versus $140–700/m² of platinum alone). --- ## 1. What the current site contains and what must change ### 1.1 Current site structure | Page | Content | |---|---| | Home `/` | Animated hero, why, before/after, dashboard preview, AI advisor, how it works, applications, FAQ | | Simulation `/simulation` | Interactive reactor with 8 control sliders and 8 injectable faults | | Dashboard `/dashboard` | Live metrics, system health gauge, AI decision centre, fault panel | | Analytics `/analytics` | Time-series charts | | Hardware `/hardware` | Interactive Arduino Uno board and sensor cards | | Learn `/learn` | MFC explainer, electron transfer, biofilm, membrane, wastewater, glossary, quiz | | Research `/research` | Methodology, architecture, impact, physical build, future applications | Stack: Next.js 16 + React 19 + Zustand (simulation state) + Motion + Recharts. The simulation engine lives in `src/lib/simulation.ts` and the AI controller in `src/lib/ai-controller.ts`. ### 1.2 Scientific problems in the current simulation | Problem | Current state | Why it is wrong | Correct replacement | |---|---|---|---| | Temperature and pH effects | Gaussian bell curves with hand-picked coefficients | Not a published biological model; the tails are unrealistic | Rosso's cardinal temperature model (CTMI) and cardinal pH model (CPM) | | Bacteria | A "density %" growing logistically | No physical unit, no link to consumed food or to current | Biomass in g/m² growing in proportion to the electrons it actually delivers | | Open-circuit voltage | `0.8 × coefficients` | Not thermodynamic | Nernst equation for each electrode | | Internal resistance | `35 + (100−membrane)×2.6 + …` | A hand-assembled formula | Sum of ohmic resistance + cathode activation losses + Nernst–Monod losses at the anode | | Coulombic efficiency | `88 × activity × …` | Does not conserve electrons | Computed from its definition: electrons in the circuit ÷ electrons in the COD removed | | COD removal | Exponential function of "residence time" | Not tied to actual consumption | Mass balance in the chamber (CSTR) | | "Oxygen %" and "membrane efficiency %" | Abstract inputs | Correspond to no real measurement | Dissolved oxygen in mg/L, and an oxygen permeability coefficient for the membrane | | Current sensor | ACS712 | Sensitivity 185 mV/A and noise ≈ 114 mA; cell current is below 2 mA | Measure the voltage across the external resistor with an ADS1115, then I = V/R | Bottom line: the interface and design are excellent and most of them can be reused, but **the simulation engine must be replaced entirely** with the model described in section 5. --- ## 2. The BioVolt idea in precise scientific terms **BioVolt** is a smart microbial fuel cell system that turns organic waste or wastewater into clean electricity. The cell uses bacteria that break down organic material and release electrons, generating electrical current. The project improves this process with three layers: 1. **Measurement layer**: sensors that measure pH, temperature, voltage, and current in real time (ideally adding anode potential against a reference electrode, conductivity, and dissolved oxygen). 2. **Digital twin layer**: a virtual model of the real cell, continuously updated from the sensors, which allows new operating conditions to be tested and problems predicted **before** applying changes to the real cell. 3. **Artificial intelligence layer**: analyses the live data, detects changes that may reduce performance, and recommends or automatically applies adjustments to maintain stable energy production. Target benefits: more electricity generated, support for wastewater treatment, less human intervention, and faster recovery from disturbances. **Note on the name**: an open-source project called `biovolt-ai` exists on GitHub (an educational MFC research platform with an illustrative digital twin). It is not a commercial product, but it is worth knowing when choosing the final identity. --- ## 3. How a microbial fuel cell works ### 3.1 Components | Component | Function | |---|---| | **Anode chamber** | Oxygen-free, holds the wastewater and the bacteria. The bacteria oxidise the organic matter | | **Biofilm** | A bacterial layer 20–70 µm thick attached to the carbon electrode, transferring electrons to it directly | | **Anode electrode** | Carbon (cloth/brush/paper/graphite) with a large surface area that collects the electrons | | **Separator membrane (PEM/CEM)** | Passes protons from anode to cathode and limits oxygen leaking into the anode | | **Cathode chamber** | Contains oxygen (aerated or an air cathode). Oxygen accepts the electrons | | **External circuit** | Electrons flow from anode to cathode through a load (resistor, LED, capacitor charger) | ### 3.2 The reactions (example: acetate, the simplest food for electroactive bacteria) ``` Anode: CH3COO⁻ + 4 H2O → 2 HCO3⁻ + 9 H⁺ + 8 e⁻ Cathode: 2 O2 + 8 H⁺ + 8 e⁻ → 4 H2O Overall: CH3COO⁻ + 2 O2 → 2 HCO3⁻ + H⁺ ``` Each acetate molecule gives **8 electrons**. Glucose gives 24: ``` C6H12O6 + 6 H2O → 6 CO2 + 24 H⁺ + 24 e⁻ ``` ### 3.3 Why we measure the fuel as COD Wastewater is a complex mixture of organic matter. Instead of analysing every compound, we measure the **chemical oxygen demand (COD)**: the amount of oxygen needed to oxidise all the organic matter. Since one O₂ molecule (32 grams) accepts 4 electrons: ``` 1 mole of electrons = 8 grams of COD ``` This makes COD a precise **electron currency**: any quantity of COD can be converted into electrical charge exactly (section 5.3). --- ## 4. The bacteria: who they are, how they live, reproduce, die, and generate electricity ### 4.1 Who they are They are called **anode-respiring bacteria (ARB)** or **exoelectrogens**. The best known: | Species | Nature | How it transfers electrons | Maximum current density | |---|---|---|---| | *Geobacter sulfurreducens* | Anaerobic (tolerates air for 24 h, grows at ≤10% oxygen) | Directly via outer-membrane cytochromes and conductive nanowires | 10–15 A/m² | | *Shewanella oneidensis* | Facultative (lives with or without oxygen) | Via secreted shuttle molecules (flavins) | ≈ 0.16–0.34 A/m² | | Mixed communities from wastewater | Mixture | Both | 1–10 A/m² depending on maturity | In a real project the cell is inoculated with sludge from a treatment plant or with wastewater, and the electrode itself selects the electroactive bacteria over days to weeks. ### 4.2 How they breathe the electrode (the core idea) Ordinary bacteria breathe oxygen: they take electrons from food and hand them to oxygen, extracting energy from that voltage difference. Electroactive bacteria do the same, but hand the electrons to **a solid electrode** instead of oxygen. The electrode is their "lung". This is why **the anode potential** determines their speed, just as oxygen availability determines ours. That is what the Nernst–Monod equation describes (section 5.4.2). ### 4.3 How they live (survival conditions) | Factor | Safe range | Optimum | What happens outside it | Source | |---|---|---|---|---| | Temperature | ≈ 15–45 °C | 30–40 °C | Below 15 °C start-up fails (or takes >40 days); above 48 °C decline; 53 °C inactivation | Patil 2010; Min 2008; Liu 2011 | | Bulk pH | 6–9 | 7–8 | At pH 6 activity is only ≈ 18%; no activity at pH 3 or 11 | Patil 2011; Torres 2008 | | Oxygen in the anode | ≈ zero | zero | Steals electrons and drops coulombic efficiency from 40–55% to 9–12% | Liu & Logan 2004 | | Food (COD) | > 50 mg/L | 200–1000 mg/L | Starvation stops the current, and in stacks causes voltage reversal that kills the biofilm | Oh & Logan 2007 | | Anode potential | −0.3 to 0 V (vs SHE) | ≈ −0.1 V | Too low a potential means the bacteria are "suffocating" | Torres 2008 | | Conductivity | > 1 mS/cm | 5–20 mS/cm | High ohmic resistance | Rossi 2021 | **Important warning about pH**: protons accumulate inside the biofilm itself, so the pH near the electrode is about a full unit below the bulk (6.1 was measured near the electrode while the bulk was ≈ 7). We therefore hold the bulk pH at 7 or slightly above and use a buffer (phosphate or bicarbonate). Raising the buffer concentration from 12.5 to 100 mM raised the current from 2.2 to 9.3 A/m² (Torres, Kato Marcus, Rittmann 2008). ### 4.4 How they reproduce 1. They consume food (COD). 2. Most electrons go to the electrode (this is the current) and the cell extracts energy from them. 3. A small fraction of the electrons is used to build new cells (division). The split is set by the **growth yield Y**: how many grams of new bacteria per gram of COD consumed. For electroactive bacteria Y is low (0.1–0.3 theoretically, 0.02 measured net), meaning **more than 85% of the food's electrons go to electricity**, not to growth. This is the reason for their efficiency. - **Doubling time**: 6–8 hours for *Geobacter* under the best conditions (Bond & Lovley 2003), and 6 hours on the electrode in the early phase (Marsili 2010). - **Biofilm formation**: 5 µm after 65 hours, 17 after 79 hours, 50 after 129 hours (Jana 2014). Thickness then plateaus because of electron-conduction and proton-transport limits. - **Start-up time for a new cell**: 3.5 days at 35 °C, more than 40 days at 15 °C (Patil 2010). ### 4.5 How they die - **Natural death / endogenous respiration**: a constant rate b between 0.013 and 0.04 per day (Wilson & Kim 2016; Pinto 2010). That is, they lose 1.3–4% of their mass daily even in good conditions, offset by growth. - **Stress death**: acidification, high temperature, oxygen, toxic substances, and prolonged starvation. - **Detachment**: parts of the biofilm shear off under flow. - **Competition**: methanogens eat the same food and turn it into gas instead of electricity. ### 4.6 How they generate electricity, quantitatively Every gram of COD consumed by electroactive bacteria gives `12,061 × (1 − 1.42·Y)` coulombs to the electrode. With Y = 0.1 the fraction is 85.8%, i.e. **10,348 coulombs per gram of COD**. This links directly: ``` Current (amperes) = food consumption rate (grams COD/second) × 10,348 ``` --- ## 5. The complete calculations (the core of this research) ### 5.1 Physical constants (exact) | Constant | Symbol | Value | |---|---|---| | Faraday constant | F | 96,485 C/mol e⁻ | | Gas constant | R | 8.314 J/(mol·K) | | Molar mass of oxygen | M_O₂ | 32 g/mol | | Electrons per O₂ molecule | b | 4 | | COD per mole of electrons | — | 8 g | | Charge per gram of COD | F/8 | **12,061 C/g** = 3.35 Ah/g | | Electrons from acetate | n | 8 per molecule | | COD per gram of biomass | — | 1.42 g COD/g VSS | ### 5.2 Thermodynamics: the maximum possible voltage **Nernst equation**: ``` E = E° − (R·T / n·F) · ln(Q) ``` **Anode (acetate, 5 mM acetate, 5 mM bicarbonate, pH 7, 25 °C)**: ``` E_an = 0.187 − (8.314×298.15 / 8×96485) · ln( [CH3COO⁻] / ([HCO3⁻]² · [H⁺]⁹) ) = 0.187 − 0.003210 · ln( 0.005 / (0.005² × 10⁻⁶³) ) = −0.296 V (vs the standard hydrogen electrode, SHE) ``` **Cathode (oxygen, pO₂ = 0.2, pH 7)**: ``` E_cat = 1.229 − (R·T / 4F) · ln( 1 / (pO2 · [H⁺]⁴) ) = +0.805 V ``` **Theoretical maximum cell voltage**: ``` EMF = 0.805 − (−0.296) = 1.101 V ``` I verified these numbers computationally, and they match the table in Logan et al. 2006 exactly. **Effect of conditions** (my calculation): | Case | E_an | E_cat | |---|---|---| | 25 °C, pH 7 | −0.296 V | 0.805 V | | 30 °C, pH 7 | −0.304 V | 0.798 V | | 25 °C, pH 6 | −0.229 V | — | A drop in anode pH raises its potential (narrowing the gap to the cathode) = lower cell voltage. **Reality**: the highest open-circuit voltage measured is ≈ 0.80 V, and under load < 0.62 V. The cathode alone loses ≈ 0.6 V (its working potential under current is ≈ 0.2 V instead of 0.805). The reason is the slow oxygen reduction on carbon. ### 5.3 Faraday's law: from food to electricity (exact, 100%) **Theoretical charge from a quantity of COD**: ``` Q_theoretical (C) = ΔCOD (g) × 96485 / 8 = ΔCOD × 12,061 ``` **Coulombic efficiency (CE)**: the fraction of electrons that reached the circuit out of all the electrons in the COD removed. Batch system: ``` CE = (32 × ∫ I dt) / (96485 × 4 × V_an × ΔCOD) ``` Continuous system: ``` CE = (32 × I) / (96485 × 4 × Q × ΔCOD) ``` where I is in amperes, V_an the anode volume in litres, Q the flow in litres/second, ΔCOD in grams/litre. **Chemical energy in COD**: ≈ 14 kJ/g COD (pure compounds 13.9–14.7; real wastewater measured at 13–16 kJ/g depending on the drying method). **Overall energy efficiency**: ``` η_energy = ∫ V·I dt / (ΔCOD × V_an × 14,000 J/g) ``` ### 5.4 Bacterial kinetics #### 5.4.1 The Monod equation (eating rate as a function of food availability) ``` q = q_max · S / (K_S + S) μ = Y · q ``` - `S`: COD concentration in the chamber (mg/L) - `q_max`: maximum consumption rate = **22.3 g COD/(g VS·day)** (Lee, Torres, Rittmann 2009) - `K_S`: the concentration at which they eat at half speed = **119 mg COD/L** (intrinsic), 184 apparent in the biofilm (Lee 2009) - `Y`: growth yield = **0.10 g VSS/g COD** (theoretical range 0.1–0.3) **Self-consistency check**: ``` μ_max = Y × q_max = 0.10 × 22.3 = 2.23 day⁻¹ doubling time = ln(2) / μ_max = 0.693 / 2.23 = 0.31 day = 7.5 hours ``` This falls between Pinto's value (1.97) and Lee's (3.2), and matches the measured doubling time for Geobacter (6–8 hours). The constants are mutually consistent. #### 5.4.2 The Nernst–Monod equation (respiration rate as a function of electrode potential) This is the globally accepted model for anode-respiring bacteria (Kato Marcus, Torres, Rittmann 2007; Torres 2008): ``` j = j_max · [S/(K_S+S)] · 1 / (1 + exp( −(F/RT)·(E_an − E_KA) )) ``` - `j`: current density (A/m²) - `E_an`: anode potential - `E_KA`: the potential at which the bacteria work at half capacity = **−0.155 V vs SHE** at 30 °C (Torres 2008) **Inverting the equation** (to find the potential the anode needs for a given current): ``` E_an = E_KA + (R·T/F) · ln( j / (j_max − j) ) ``` Worked example at 30 °C: | j / j_max | E_an (V vs SHE) | |---|---| | 10% | −0.212 | | 50% | −0.155 | | 90% | −0.098 | | 99% | −0.035 | Meaning: the closer we push the bacteria to their maximum capacity, the higher the anode potential rises and the more of the cell voltage is lost. This is the "biological activation loss". #### 5.4.3 The maximum current the bacteria can sustain ``` j_max = (F/8) · (1 − 1.42·Y) · q_max · B ``` where `B` is the biomass on the electrode (g VS/m²). With a mature biofilm ≈ 54 µm thick at a density of 50,000 g/m³: ``` B_max = 50,000 × 54×10⁻⁶ = 2.7 g/m² q_max = 22.3 / 86,400 = 2.58×10⁻⁴ g COD/(g·s) j_max = 12,061 × 0.858 × 2.58×10⁻⁴ × 2.7 = 7.2 A/m² ``` This matches measured values (8.3 to 11.5 A/m²). Consistency confirmed. #### 5.4.4 Growth and death dynamics (the core equation for the digital twin) ``` dB/dt = Y · r_S · f_space − b · B ``` - `r_S = j / ((F/8)·(1 − 1.42Y))`: food consumption rate (g COD/m²/s), **directly tied to the actual current** - `f_space = 1 − B/B_max`: the electrode area is finite - `b`: death rate = **0.04 day⁻¹** (Pinto 2010; the lowest published value is 0.013) The important consequence: bacteria grow **in proportion to the electrons they actually hand to the electrode**. If the circuit is opened (no current) there is no growth, only slow death. This is experimentally true. ### 5.5 Temperature effects **Cardinal temperature model (CTMI, Rosso 1993)**: ``` f_T = (T−T_max)·(T−T_min)² / ((T_opt−T_min)·[(T_opt−T_min)·(T−T_opt) − (T_opt−T_max)·(T_opt+T_min−2T)]) ``` and zero outside [T_min, T_max]. Suggested values (inferred from published data, to be calibrated from experiment): **T_min = 5 °C, T_opt = 40 °C, T_max = 50 °C**. | Comparison | Relative f_T | Reference | |---|---|---| | 20 °C vs 30 °C | 0.42 | Arrhenius with activation energy 44.85 kJ/mol gives 0.54 | | 35 vs 30 | 1.27 | +80% from 30 to 40 °C (Liu 2010) | | 45 vs 30 | 1.16 | Optimum 45 °C for an acclimated biofilm (Liu 2011) | **Why does the cell's power drop less than the bacteria's rate?** Bacterial rate at 20 °C is ≈ half that at 30 °C, yet measured power dropped only 9% (Liu 2005) to 39% (Min 2008). The reason is that the bacteria have spare capacity and the limiter is the cathode. Our model reproduces this automatically. ### 5.6 pH effects **Cardinal pH model (CPM, Rosso 1995)**: ``` f_pH = (pH−pH_min)·(pH−pH_max) / ((pH−pH_min)·(pH−pH_max) − (pH−pH_opt)²) ``` A mathematical fit I performed on the published data (relative activity 0.18 at pH 6, 0.5 at 6.5, 0.89 at 9, normalised to pH 7): ``` pH_min = 5.85 pH_opt = 8.05 pH_max = 9.5 (fit error = 0.015) ``` These are values for **bulk pH**. Because protons accumulate inside the biofilm, safe practical operation is at a bulk pH of 7.0–7.5. ### 5.7 Oxygen and methanogens (the electron thieves) **Oxygen**: every gram of O₂ leaking into the anode consumes a gram of COD without a single electron passing through the circuit: ``` COD lost to oxygen (g/s) = k_O2 · A_membrane · C_O2,cathode ``` `k_O2` is the membrane's permeability coefficient (on the order of 10⁻⁴ cm/s for Nafion membranes). An exact value is calibrated per membrane. **Methanogens** (Pinto 2010): ``` q_m = q_max,m · S/(K_S,m + S) q_max,m = 8.2 g/g/day , K_S,m = 80 mg/L μ_m = Y_m · q_m μ_max,m = 0.1 day⁻¹ → Y_m = 0.012 ``` They grow slowly (0.1 per day versus 2.23 for the electroactive bacteria), but accumulate over time and steal part of the food. The VITO model calibrated on real wastewater (Applied Energy 2022) found that COD removal in an air-cathode cell split as: **electricity generation 21–25%, methane 9–14%, aerobic oxidation 21–22%, denitrification 44–45%**. That is, electricity takes only a quarter of the food, and that is the reality of the technology. ### 5.8 Mass balance in the anode chamber The chamber as a continuously stirred tank reactor (CSTR): ``` dS/dt = D·(S_in − S) − r_S·(A/V) − q_m·X_m − r_O2/V D = Q / V = 1 / HRT ``` - `S_in`: influent COD - `A/V`: electrode area ÷ chamber volume - `HRT`: hydraulic retention time ### 5.9 The electrical circuit: from bacteria to volts **Cell equation** (Logan 2006): ``` E_cell = E_cat − E_an − I·R_ohm = I · R_ext ``` **Cathode** (activation loss, Butler–Volmer in asinh form): ``` E_cat = E_cat,oc − (2RT/F) · asinh( I / (2·I₀,c) ) ``` - `E_cat,oc ≈ 0.50 V` (the typical measured value for an oxygen cathode, Logan 2006) - `I₀,c`: the cathode exchange current (depends on catalyst and area; **calibrated**) **Anode**: from Nernst–Monod (5.4.2), with a floor at the thermodynamic acetate potential. **Solution**: the current I is the unique root of: ``` g(I) = E_cat(I) − E_an(I) − I·(R_ext + R_ohm) = 0 ``` Solved by bisection at each time step. g is monotonically decreasing, so the root is unique and guaranteed. **Power**: ``` P = I² · R_ext = V² / R_ext power density = P / A_anode (mW/m²) ``` **Maximum power** (linear simplification): ``` P_max = OCV² / (4 · R_int) when R_ext = R_int ``` ### 5.10 A fully worked example **The cell**: 250 mL anode chamber, 25 cm² electrode (0.0025 m²), wastewater at COD = 500 mg/L. **1) Latent charge** (single batch, 80% removal): ``` ΔCOD = 0.5 g/L × 0.80 × 0.25 L = 0.1 g Q_theoretical = 0.1 × 12,061 = 1,206 C ``` **2) Actual charge** (coulombic efficiency 30%): ``` Q = 1,206 × 0.30 = 362 C run time at 1 mA = 362 / 0.001 = 362,000 s ≈ 100 hours ≈ 4.2 days ``` **3) Energy**: ``` E_electric = 362 C × 0.4 V = 145 J = 0.040 Wh E_chemical = 0.1 g × 14,700 J/g = 1,470 J energy efficiency = 145 / 1,470 = 9.8% ``` **4) Maximum power** (OCV = 0.75 V, R_int = 300 Ω): ``` P_max = 0.75² / (4 × 300) = 0.469 mW = 469 µW power density = 0.469 / 0.0025 = 188 mW/m² current there = 0.75 / 600 = 1.25 mA , voltage = 0.375 V ``` **5) Are the bacteria sufficient?** ``` required current = 1.25 mA / 0.0025 m² = 0.5 A/m² maximum bacterial capacity (mature biofilm) ≈ 7.2 A/m² utilisation = 7% ``` The bacteria are working at only 7% of their capacity. **The limiter is the internal resistance and the cathode.** **6) Wastewater energy per cubic metre**: ``` 500 mg/L × 14 kJ/g = 7 kJ/L = 1.94 kWh/m³ (latent energy) ``` ### 5.11 Simulation results (full model, 30 days, continuous operation) Setup: 25 cm², 250 mL, influent COD 500 mg/L, HRT = 24 h, 30 °C, pH 7, R_ohm = 200 Ω. **Start-up phase**: | Day | COD in chamber (mg/L) | Biofilm (g/m²) | Current (mA) | |---|---|---|---| | 0 | 500 | 0.01 | 0.05 | | 2 | 446 | 0.22 | 0.62 | | 6 | ≈ 400 | 0.85 | 0.68 | | 14 | ≈ 380 | 1.6 | 0.70 | | 28 | 366 | 2.23 | 0.72 | The biofilm matures in about two weeks, consistent with published start-up times (days to weeks). **Effect of external resistance (after stabilisation)**: | R_ext (Ω) | I (mA) | V (V) | P (µW) | Power density (mW/m²) | COD removal | CE | |---|---|---|---|---|---|---| | 100 | 1.49 | 0.149 | 222 | 89 | 33% | 26% | | **300** | **0.96** | **0.289** | **278** | **111** | 30% | 18% | | 500 | 0.72 | 0.359 | 257 | 103 | 29% | 14% | | 1000 | 0.45 | 0.445 | 198 | 79 | 27% | 9% | **Maximum power at ≈ 300 Ω.** This is what the AI should find automatically. **Validation against published experiments** (with no tuning): | Metric | Model | Measured | Source | |---|---|---|---| | Open-circuit voltage | 0.804 V | 0.798 V | Liu, Cheng, Logan 2005 | | Maximum power for a cell with R_int ≈ 1286 Ω | 32 mW/m² | 38–40 mW/m² | Min, Cheng, Logan 2005 | | Bacterial doubling time | 7.5 hours | 6–8 hours | Bond & Lovley 2003 | | Maximum current for a mature biofilm | 7.2 A/m² | 8.3–11.5 A/m² | Lee 2009; Torres 2008 | | Coulombic efficiency | 9–26% | 10–30% on wastewater | Liu & Logan 2004 | --- ## 6. Real-world numbers: what these cells actually produce ### 6.1 Lab-scale cells | System | Power density | Notes | Source | |---|---|---|---| | Two-chamber with Nafion membrane, acetate | 38–40 mW/m² | R_int = 1286 Ω, CE 19% | Min et al. 2005 | | Two-chamber with salt bridge | 2.2 mW/m² | R_int ≈ 19,920 Ω | Min et al. 2005 | | Air cathode, domestic wastewater, no membrane | 146 mW/m² | CE 20% | Liu & Logan 2004 | | Same with membrane | 28 mW/m² | CE 28% | Liu & Logan 2004 | | Air cathode, acetate 800 mg/L | 506–661 mW/m² | CE 10–31% | Liu, Cheng, Logan 2005 | | Brewery wastewater | 205 mW/m² at 30 °C, 170 at 20 °C | | Feng et al. 2008 | | Record (small cell, 100 mM acetate) | 4.30 W/m² | CE 83.5% | Fan, Han, Liu 2012 | **Scale rule**: cells smaller than 50 mL often exceed 500 W/m³, while those larger than 2 L are usually below 30 W/m³. ### 6.2 Large pilot projects | Project | Size | Energy recovered | Treatment | |---|---|---|---| | Liang et al. 2018 | 1000 L, 50 modules, over a year | **0.033 kWh/m³** | 70–90% COD removal | | Ge & He 2016 | 200 L, 96 tubular modules | net +0.003 to +0.006 kWh/m³ | COD > 75% | | He et al. 2019 | 1.5 m³ | 0.002 kWh/m³ | COD 91%, aeration used only 12% of activated sludge | | PEE POWER (Ieropoulos 2016) | 432 cells, 300 L of urine, Glastonbury festival | 300 mW average for lighting | Actually practical for lighting toilets | **The honest calculation**: domestic wastewater holds ≈ 1.7–2.1 kWh/m³. The best project recovered 0.033, i.e. **1.5–2%**. Conventional activated-sludge plants **consume** ≈ 0.3 kWh/m³. Hence the core value of MFCs: **treatment at near-zero energy + a small amount of electricity to run the sensors and control autonomously**. ### 6.3 Lighting an LED realistically - One cell: ≤ 0.8 V open and 0.3–0.5 V under load. Not enough to drive an LED (which needs ≈ 1.8–3 V). - The solution: two or more cells in series (watching for voltage reversal) or a boost converter (LTC3108 starts at 20 mV, BQ25504/BQ25570 need 600 mV to cold-start) plus a supercapacitor, then discharge in bursts. - Capacitor energy: `E = ½·C·(V₁² − V₂²)`. A 1 F capacitor from 2.5 to 1.8 V gives 1.5 J, enough for a 20 mW LED for 75 seconds. Charging it from a cell producing 250 µW takes ≈ 100 minutes (before conversion losses). ### 6.4 Voltage reversal in stacks (a real danger) When cells are wired in series, if one cell starves its voltage reverses (−0.58 V was measured while the other was at +0.6 V, leaving the stack at just 0.02 V) and its biofilm is damaged (Oh & Logan 2007). This is one of the most important faults the AI must detect early. --- ## 7. The digital twin ### 7.1 The precise definition A digital twin is not merely a simulation. Its four requirements: 1. **A physical entity**: the real cell. 2. **A virtual model**: the equations in section 5. 3. **Two-way data connection**: sensors update the model, and the model's decisions return to the cell. 4. **Continuous synchronisation**: the model constantly corrects itself to match reality. The current simulation on the site satisfies requirement 2 only. ### 7.2 Types of twin | Type | Description | Advantage | Drawback | |---|---|---|---| | Mechanistic (physical) | The equations of section 5 | Explains and extrapolates beyond the data | Constants need calibration | | Data-driven | Neural networks trained on measurements | Accurate within the data range | Black box, needs lots of data, fails outside its range | | **Hybrid (recommended)** | Physical equations + machine learning to correct the residual | Best of both | Slightly more complex | ### 7.3 Proposed architecture for the BioVolt twin ``` The real cell │ sensors every 1–10 s: V, I, pH, T, E_anode, EC, DO ▼ [1] Cleaning layer: noise filtering, sensor-fault detection ▼ [2] State estimation (extended Kalman filter, EKF) estimates what is not directly measured: biomass B, COD in the chamber S, internal resistance, cathode I₀ ▼ [3] The mechanistic model (section 5) + a machine-learning residual corrector ▼ [4] A "what-if" engine: running the model faster than real time to test 20–50 scenarios in seconds ▼ [5] The AI: picks the best safe action ▼ The real cell (execution: resistance, pump, heater, buffer dose) ``` ### 7.4 Calibration: how the model becomes "correct for this particular cell" 1. **Polarisation curve** (once a day to once a week): sweep the external resistance gradually from 10 kΩ to 10 Ω, waiting for steady state, recording V and I. The linear slope gives R_int, and the curve shape gives the cathode I₀ and the bacterial j_max. 2. **Current interrupt**: open the circuit momentarily; the instantaneous voltage jump = I × R_ohm. 3. **Manual COD measurements** (once daily): correct the estimate of S and compute the true coulombic efficiency. 4. **Bayesian estimation / Kalman filter** gradually updates the biological constants (q_max, K_S, b). --- ## 8. The artificial intelligence ### 8.1 The four layers #### Layer 1: State estimation (the virtual sensor) - **What**: estimates biomass and COD concentration from voltage, current, and pH. - **How**: an extended Kalman filter on top of the mechanistic model. - **Why**: COD is measured slowly in the lab (two hours) and biomass cannot be measured during operation. MFCs themselves are used as BOD sensors, so the relationship between current and food is known. #### Layer 2: Anomaly detection - **What**: detects the cell deviating from normal behaviour before performance collapses. - **How**: compare the measurement with the digital twin's prediction. If the difference (the residual) exceeds a statistical threshold (e.g. 3σ or a CUSUM test) an alarm fires. An autoencoder or Isolation Forest can be added on time windows. - **The advantage**: each fault has a **distinct physical signature** (section 9), so the cause can be diagnosed, not merely the problem detected. #### Layer 3: Prediction - **What**: forecast voltage, power, and COD removal for the coming hours. - **How**: the digital twin itself (with forecast inputs), complemented by an LSTM, Random Forest, or XGBoost on the residuals once enough data is collected. - **In the literature**: artificial neural networks (ANN), support vector regression (SVR), random forest regression (RFR), LSTM, and ANFIS have been used to predict cell power and COD removal (Discover Sustainability 2025 review; Green Energy & Resources 2025 review). The shared challenge: scarce standardised data and the difficulty of modelling biological complexity, which is why the hybrid model is preferred. #### Layer 4: Control and decision-making **a) Maximum power point tracking (MPPT)** — the most important and best-proven: - **Perturb and observe (P/O) algorithm**: change the external resistance by a small step, watch the power; if it rises keep the same direction, if it falls reverse. - **Published results**: - Woodward et al. 2010 compared P/O, gradient, and multi-unit optimisation: P/O was the easiest to tune and the most robust. - Pinto et al. 2011 (Water Research): cells with real-time optimised resistance produced markedly higher power and less methane, with coulombic efficiencies of 57% and 29% in two runs, and on real wastewater too. - Premier et al. 2011 (J. Power Sources): automatic load control raised power and efficiency. - Molognoni et al. 2014 (J. Power Sources): MPPT shortened start-up time and reduced energy losses. - A Korean study on flat-plate cells: MPPT raised power by ≈ 2.7 times compared with the value before it was applied. - **Important caution**: the cell's dynamics are slow (minutes to tens of minutes), so the P/O step must be slow (every 5–15 minutes) or the system oscillates. - **A hidden biological benefit**: an optimised resistance selects for more electroactive bacteria and reduces methane, meaning electrical control **improves the biology itself**. **b) Model predictive control (MPC) via the digital twin**: - Each cycle the twin tries several strategies (changing feed flow, temperature, buffer dose, resistance) over a 1–6 hour horizon and picks the best according to an objective function: ``` J = w1·(power) + w2·(COD removal) − w3·(risk to the bacteria) − w4·(intervention cost) ``` - This is exactly what "testing operating conditions on the twin before applying them to the real cell" means. **c) Safety layer (hard rules the AI must not cross)**: | Rule | Limit | |---|---| | Do not lower the resistance if cell voltage < 0.1 V (reversal/starvation risk) | hard | | pH protection: if it falls below 6.5, dose buffer immediately | hard | | Temperature: cool above 40 °C, heat below 20 °C | hard | | Do not change more than one major variable per cycle | hard | | Every automatic decision is logged with its reason and prediction (interpretability) | hard | **d) Reinforcement learning (future)**: an RL agent could be trained inside the digital twin (thousands of simulated years) and then transferred to the cell with care. This is still research; it should be presented as a future vision, not a current capability. ### 8.2 What is realistic for a student prototype | Capability | Realism | Note | |---|---|---| | Live monitoring of V, I, pH, T | ✅ immediate | | | MPPT via P/O over a resistor bank | ✅ scientifically proven | Highest impact and simplest to implement | | Anomaly detection via residuals | ✅ | Needs a calibrated twin | | Cause diagnosis via signatures | ✅ | Section 9 | | Calibrated mechanistic digital twin | ✅ | Section 5 is ready | | Machine-learning prediction | ⚠️ | After weeks of data | | Reinforcement learning on the real cell | ❌ for now | A future vision | --- ## 9. Fault catalogue: physical signatures, detection, and response | Fault | Cause | Sensor signature | Effect in the model | AI action | |---|---|---|---|---| | **Acidification** | Proton accumulation, weak buffer | pH falls gradually, current drops, anode potential rises | f_pH collapses (pH 6 → 18% activity) | Dose buffer, temporarily reduce feed | | **Food starvation** | Low influent COD | Current falls slowly, anode potential rises, pH stable | Monod S/(K_S+S) falls | Increase feed; block any resistance reduction | | **Overload** | High flow | Effluent COD rises, coulombic efficiency falls | Large D washes the food out | Slow the pump to restore HRT | | **Oxygen leak into the anode** | Damaged membrane/seal | CE falls sharply, current drops, DO in the anode > 0 | r_O2 consumes COD | Reseal, reduce aeration near the membrane | | **Overheating** | Cooling failure/sun | T > 40 °C then current collapses | f_T collapses above T_opt | Immediate cooling | | **Cold** | Winter | T < 20 °C, lower current | f_T and the cathode I₀ both fall | Heat if the energy budget allows | | **Membrane/cathode fouling** | Salt and biological deposits | R_ohm rises gradually (from current interrupt) | Higher R_ohm | Maintenance alert, adjust external resistance | | **Bacterial die-off (toxin)** | Chemical shock | Sharp current collapse with normal pH/T | B drops suddenly | Isolate the feed, run at high resistance to protect what remains | | **Methanogen competition** | Long operation at high resistance | CE declines gradually over weeks | X_m accumulates | MPPT reduces methane (Pinto 2011) | | **Voltage reversal (stack)** | One cell starving | Negative cell voltage | — | Disconnect the cell immediately, refeed it | | **Sensor fault** | Drift/disconnection | Physically impossible, frozen, or jumping reading | Large residual in one variable only | Fall back on the virtual estimate and lower confidence | --- ## 10. Hardware and sensors ### 10.1 The recommended list | # | Part | Why | |---|---|---| | 1 | **ESP32** (instead of Arduino Uno) | Wi-Fi to send data to the digital twin. Do not use its analogue converters for measurement (nonlinear and blind below 150 mV) | | 2 | **ADS1115** (16-bit converter) × 2 | Measures cell voltage to 31 µV on the ±1.024 V range. **Current = voltage ÷ external resistance** (the Logan 2006 method) | | 3 | **Precision resistor bank** (10 Ω to 100 kΩ) with signal relays or MOSFETs | For the MPPT algorithm and polarisation curves. Better than a digital potentiometer (limited to 1 mA) | | 4 | **Ag/AgCl reference electrode** (+0.210 V vs SHE for 3 M KCl) with a buffer amplifier | Measures the anode potential alone: separates bacterial problems from cathode problems. The most important "smart" sensor | | 5 | **DS18B20** waterproof × 2 | Temperature to ±0.5 °C, and for temperature compensation of pH and conductivity | | 6 | **DFRobot SEN0169-V2** for pH | Designed for continuous 24/7 immersion, ±0.1 pH, 0–3 V output safe for the ESP32 | | 7 | **DFRobot DFR0300** for conductivity | Conductivity determines the ohmic loss | | 8 | **DFRobot SEN0237-A** for dissolved oxygen | In the cathode chamber (or to detect a leak into the anode) | | 9 | **INA226** | To measure the boost-converter and LED side (milliamperes) | | 10 | **LTC3108 or BQ25570 + a 1–50 F supercapacitor** | To light an LED in bursts | | 11 | Peristaltic pump + small heater + buffer dosing pump | The actuators through which the AI executes its decisions | **What to drop**: ACS712 (its noise is ≈ 114 mA and it draws 10 mA itself), and the Arduino's internal converter for precision measurements (its step is 4.9 mV). ### 10.2 Required measurement precision (my calculation) A cell with OCV = 0.7 V and R_int = 1.3 kΩ gives a maximum power of ≈ 94 µW at ≈ 0.27 mA and 0.35 V. So the electronics must resolve **microamperes and millivolts**. --- ## 11. Website rebuild plan ### 11.1 What to reuse The technical stack (Next.js/Zustand/Motion/Recharts), the design system, the core UI components, and the tick loop. ### 11.2 What to replace entirely **The new simulation engine** `src/lib/biovolt-model.ts`: - **State variables**: B (electroactive bacteria, g/m²), X_m (methanogens, mg/L), S (COD, mg/L), pH, T, R_ohm (degrading with fouling). - **Controllable inputs**: feed flow (HRT), influent COD, target temperature, buffer dose, cathode oxygen, external resistance. - **Outputs**, all in real physical units: V, I, P, power density, anode and cathode potentials, true coulombic efficiency, COD removal, current doubling time, bacterial utilisation fraction. - **Solution**: bisection on the circuit equation at each step + Euler integration of the differential equations with a suitable step, plus time acceleration (one real second = minutes/hours of simulation) because the biology is slow. **The new AI** `src/lib/biovolt-ai.ts`: - Real MPPT via P/O. - Anomaly detection on the residuals between "the cell" (model + noise + hidden faults) and "the twin" (a clean model being calibrated). - Diagnosis by signature (the table in section 9). - A "what-if" engine showing the user the results of 3–5 scenarios before applying one. - A safety layer and an explained decision log. ### 11.3 Proposed pages | Page | What is new | |---|---| | Home | The BioVolt story: from wastewater to electricity, with honest numbers | | **Digital Twin** (replacing Simulation) | The real cell and the twin side by side, injecting hidden faults, watching the AI detect them | | **Life of the Bacteria** (new) | Interactive visualisation of growth, death, and reproduction: growth curve, doubling time, effect of temperature and pH, live equations | | **Calculation Lab** (new) | Interactive calculators: Faraday, Nernst, polarisation curve, coulombic efficiency, capacitor energy for lighting an LED | | Dashboard | Real physical metrics + AI decision centre + live power curve with the MPPT point | | Analytics | Polarisation and power curves, history, electron distribution (electricity/growth/methane/oxygen) | | Hardware | The corrected list (ESP32 + ADS1115 + reference electrode…) | | Learn | Content updated from this research + glossary + quiz | | Research | Methodology, model validation (table 5.11), references | | **Recovery Lab** (new) | Replay the disturbance scenarios of section 13 (acid, cold, starvation, overload) with and without the AI, and compare power drop, recovery time, and lost energy | --- ## 12. From solid organic waste to electricity The BioVolt idea mentions "organic waste or wastewater". Most MFC research uses wastewater because liquids are easy to handle and have low internal resistance. Solid waste (food waste, fruit peels, garden waste, dung) is possible, but it adds one extra biological step that changes the calculations. ### 12.1 The food chain inside the cell Electroactive bacteria only eat small dissolved molecules (acetate and similar). Solid waste consists of large polymers: cellulose, starch, proteins, and fats. Three groups of microbes therefore work in sequence: ``` Solids (cellulose, starch, protein, fat) │ 1) Hydrolysis: enzymes cut polymers into sugars, amino acids, fatty acids ← the slowest step ▼ Dissolved sugars and amino acids │ 2) Fermentation: fermenting bacteria turn them into acetate, H2, and other acids ▼ Acetate + H2 │ 3) Anode respiration: electroactive bacteria hand the electrons to the electrode ▼ Electric current ``` ### 12.2 Hydrolysis is the bottleneck (measured) In air-cathode MFCs inoculated with sewage (Velasquez-Orta et al. 2011): | Substrate | Power density | COD removal | Coulombic efficiency | |---|---|---|---| | Acetate (no hydrolysis needed) | 99 ± 2 mW/m² | 82% (acetate and glucose) | 26% (acetate and glucose) | | Starch (needs hydrolysis) | 4 ± 2 mW/m² | 60% | 19% | The combined hydrolysis–fermentation rate measured was **0.0024 per hour**, while fermentation alone was **0.018 per hour**: hydrolysis is 7.5 times slower and limits the current. ### 12.3 The hydrolysis equation and what it means Hydrolysis is modelled as a first-order reaction (the same form used in the international anaerobic digestion model ADM1): ``` dX_p/dt = D·(X_p,in − X_p) − k_h·X_p dissolved COD produced = k_h·X_p ``` - `X_p`: particulate (solid) COD in the chamber - `k_h`: hydrolysis constant. MFC measurement: 0.0024 h⁻¹ = **0.0576 d⁻¹**. ADM1 default values for high-rate mesophilic digesters are much higher (on the order of 10 d⁻¹). The true value depends strongly on the waste and the microbes, so it **must be calibrated**. **Consequences (my calculation with k_h = 0.0576 d⁻¹)**: In a continuously stirred cell, the fraction of solids hydrolysed at steady state is `k_h·HRT / (1 + k_h·HRT)`: | Retention time (HRT) | Fraction of solids broken down | |---|---| | 1 day | 5.4% | | 10 days | 36.5% | | 30 days | 63.3% | In a batch cell, the fraction is `1 − exp(−k_h·t)`, so reaching **50% takes ln 2 / 0.0576 = 12 days**. This is why food-waste MFCs either use long retention times or pretreat the waste. ### 12.4 Pretreatment | Method | Idea | Evidence | |---|---|---| | Biohydrogen (dark) fermentation first | A first reactor ferments the waste to acids and H₂; the liquid feeds the MFC | Highest reported power for food-waste leachate: 1540 mW/m² (Choi & Ahn 2015) | | Acidogenic fermentation | Similar, producing volatile fatty acids | 1205 (as reported), about 1.28 times lower than biohydrogen fermentation | | Enzymatic or fungal hydrolysis | Enzymes from fungi break polymers quickly | Ultra-fast fungal hydrolysis: estimated 192.5 million kWh/year plus biofertiliser from Singapore's food waste (Xin et al., an estimate, not a plant) | | Bioaugmentation | Adding cellulose-degrading strains (e.g. *Cellulomonas fimi*, *Bacillus subtilis*) | Improves carbohydrate removal in fruit-waste MFCs | | Thermal/chemical | Heat, acid, or alkali | Effective, but the energy and chemicals often cost more than the electricity gained | Additional practical rules: a **low solid-to-liquid ratio** improves efficiency, and waste composition varies widely, which is another reason the digital twin must calibrate. ### 12.5 Published numbers for different substrates | Substrate | Power density | Other results | Source | |---|---|---|---| | Food waste (various) | 1–371 mW/m² | COD removal 64–95%, CE up to 95% in 250 mL cells | Energies review 2022 | | Food-waste leachate after biohydrogen fermentation | 1540 mW/m² | Best pretreatment result | Choi & Ahn 2015 | | Glucose (anaerobic sludge inoculum) | 456.8 mW/m² | 508 mV, COD removal 94.3% | Pol. J. Environ. Stud. 2016 | | Soluble starch (same study) | 277.6 mW/m² | 396 mV, COD removal 79.4% | Pol. J. Environ. Stud. 2016 | | Brewery wastewater (acclimated to it) | 552 mW/m² | Glucose-acclimated cell on glucose: 1519 | Yu et al. 2014 | | Orange waste | 357 mV | Highest voltage among orange, banana, and mango | Energies review 2022 | **Net energy recovery by substrate** (averages from a meta-analysis by Ge 2015): acetate 0.40 kWh/kg COD, glucose 0.12, domestic wastewater 0.17, industrial wastewater 0.04. The rule: **the more complex the substrate, the lower the power**. ### 12.6 Worked example: one kilogram of food-waste slurry Assumption (for illustration only): the slurry contains 100 g of COD that the cell removes completely. ``` Latent chemical energy = 100 g × 14 kJ/g = 1,400 kJ = 0.39 kWh Theoretical charge = 100 × 12,061 = 1,206,100 C Actual charge (CE 30%) = 361,830 C Electrical energy at 0.4 V = 361,830 × 0.4 = 144.7 kJ = 0.040 kWh Overall efficiency = 144.7 / 1,400 = 10.3% ``` With k_h = 0.0576 d⁻¹, releasing even half of that COD takes about 12 days in batch mode without pretreatment. ### 12.7 What changes in the BioVolt model for solid waste Add one state variable `X_p` (particulate COD) and its hydrolysis equation (12.3). The dissolved COD `S` then receives `k_h·X_p` as an extra source. Everything else in section 5 stays the same. --- ## 13. Disturbances and recovery: literature evidence and simulation ### 13.1 What the literature measured | Disturbance | Effect | Recovery | Source | |---|---|---|---| | Low-pH shock (17 cells) | Power fell 52.7 ± 35.8% | 14 cells recovered in 60.7 ± 58.3 h; **3 never recovered** | Lesnik, Cai, Liu 2020 | | Acidification (bioelectrochemical cell) | Largest loss of all failure modes; permanent loss | Current stabilised after 115.1 ± 100.3 h on average | Satinover, Rodriguez, Borole 2020 | | Osmotic (salt) shock | Current 5.3 → 4.4 A/m² | Fully reversible within 48 h (back to 5.2 A/m²) | Satinover et al. 2020 | | Aeration and voltage reversal | Negligible if caught early | **Full recovery if caught in under 15 minutes** | Satinover et al. 2020 | | Starvation | Current falls to background | Acetate-fed cells recovered immediately, glucose-fed cells in about 1 day; 4.2% and 10.8% irreversible loss per starvation | Saheb-Alam et al. 2019 | | Long starvation (11.25 days) | — | 3.33 days to recover | Chang et al. 2004 (cited by Saheb-Alam) | | Voltage reversal in stacks | Cell collapse | Automatic reconfiguration of connections doubled power, halved charging time, and prevented reversal | Papaharalabos et al. 2017 | Two conclusions matter for BioVolt: 1. **Detection speed decides the outcome**: the same fault can be fully reversible or permanent depending on whether it is caught within minutes. 2. **Recovery can be predicted**: machine-learning models trained on microbial community data predicted recovery time within 5.8–8.7% of the observed values (Lesnik 2020). This supports the AI's prediction layer. **Honest note**: I did not find a published experiment that directly measured a feedback controller shortening MFC recovery time after a real disturbance. Published controller comparisons (for example, robust or adaptive voltage controllers) are simulations. BioVolt's recovery claim is therefore supported by (a) the "caught early recovers fully" evidence above and (b) the simulation below, and it should be verified experimentally. ### 13.2 Simulation setup The same validated cell from section 5.10–5.11 (25 cm² electrode, 250 mL, influent COD 500 mg/L, HRT 24 h, 30 °C), after 30 days of start-up (mature biofilm ≈ 2.0 g/m²). Ten days are simulated with a 60-second step, and the disturbance starts on day 2. Four operating modes are compared: | Mode | What it does | |---|---| | Fixed 1000 Ω | A fixed external resistor (a common lab default), no control | | MPPT only | Perturb-and-observe every 10 minutes with ±5% steps, nothing else | | MPPT with safety freeze | Same, but MPPT pauses when power falls below 20% of normal | | Full BioVolt AI | MPPT with safety freeze + detection with a 10-minute delay + buffer dosing (restores pH with a 30-minute time constant) + heater (1-hour time constant) | Recovery time is measured from the disturbance onset until the hourly average power returns to at least 90% of its pre-disturbance level. ### 13.3 Simulation results | Disturbance | Metric | Fixed 1000 Ω | MPPT only | MPPT + safety freeze | Full BioVolt AI | |---|---|---|---|---|---| | Mild acid (influent pH 5.0 for 12 h) | Max power drop | 9% | 11% | 11% | 2% | | | Recovery time | no loss > 10% | 14 h | 14 h | no loss > 10% | | | Energy lost (10 days) | 0.6% | 0.7% | 0.7% | 0% | | Severe acid (influent pH 4.0 for 24 h) | Lowest chamber pH | 5.10 | 5.10 | 5.10 | 6.78 | | | Max power drop | 100% | 100% | 100% | 1% | | | Recovery time | 45 h | **134 h** | 46 h | no loss > 10% | | | Energy lost | 11.9% | **40.2%** | 12.6% | **0%** | | Cold (ambient 15 °C for 24 h) | Max power drop | 12% | 15% | 15% | 3% (with heating) | | | Recovery time | 25 h | 25 h | 25 h | no loss > 10% | | | Energy lost | 0.9% | 1.2% | 1.2% | 0.1% | | Starvation (influent COD 0 for 24 h) | Max power drop | 21% | 75% | 75% | 75% | | | Recovery time | 27 h | 28 h | 28 h | 28 h | | | Energy lost | 1.0% | 1.9% | 1.9% | 1.9% | | Overload (influent COD 1500 for 24 h) | Max power drop | 2% | 2% | 2% | 2% | | | Energy lost | none (+0.5% gain) | none (+0.6% gain) | none | none | **Energy over 10 undisturbed days**: fixed 1000 Ω = 169.4 J; MPPT = 231.3 J, i.e. **+36.5%**. ### 13.4 What the simulation teaches (the lessons for the AI) 1. **Prevention beats recovery.** In the severe acid shock the full AI never let the chamber pH fall below 6.78, so the bacteria kept working and no energy was lost. Without it, pH fell to 5.10, below the bacteria's minimum (5.85), and power collapsed completely. 2. **MPPT without a safety layer is dangerous.** When power is zero, perturb-and-observe has no signal, so it drifts down to its lowest limit (20 Ω) and stays trapped there after the bacteria recover. Recovery took 134 hours instead of 45. Freezing MPPT during a collapse brings it back to 46 hours. This is why the safety rules in section 8.1 are mandatory. 3. **Running at maximum power removes the safety margin.** In the mild acid and starvation cases, the MPPT modes lost more power than the fixed resistor, because drawing maximum current uses more of the bacteria's spare capacity and more of the food. The AI's objective should therefore weigh power against resilience. 4. **The AI cannot create food.** In starvation all modes recover about 3–4 hours after feeding resumes. The AI's job there is protection (in stacks: disconnecting starving cells to prevent reversal), not recovery. 5. **Every action has an energy cost.** Holding a 250 mL cell at 30 °C against 15 °C ambient (with an assumed 2-hour thermal time constant) needs ≈ 0.25 × 4186 × 15 / 7200 ≈ **2.2 W**, i.e. ≈ **188 kJ per day**. That is about **815 times** the cell's entire 10-day output (231 J). Without heating, the cold snap cost only 1.2% of energy. The AI must therefore **not** heat a small cell unless external or waste heat is available; its objective function (section 8.1) must include the cost of each action. 6. **Mature biofilms are robust.** Because the bacteria run at a small fraction of their capacity, mild shocks and overloads barely affect power. Young biofilms and wastewater biofilms with lower capacity are more fragile, which is why the literature reports larger drops. ### 13.5 Recovery metrics BioVolt should report | Metric | Definition | |---|---| | Resistance | Maximum relative power drop during the disturbance | | Resilience (recovery time) | Time from onset until power returns to ≥ 90% of its previous level | | Lost energy | Energy produced compared with the same period without the disturbance | | Detection delay | Time from onset until the AI raises an alarm | | Irreversible loss | Permanent power reduction after full recovery | --- ## 14. Reducing human intervention: what gets automated and what does not ### 14.1 The tasks The table compares common manual lab practice with BioVolt. The frequencies are design targets based on typical lab routines, not measurements. | Task | Typical manual practice | BioVolt | |---|---|---| | Reading voltage and current | Multimeter readings a few times per day | Continuous logging every 1–10 seconds | | pH check and correction | Daily manual measurement and buffer addition | Continuous measurement + automatic dosing | | Setting the external resistance | Periodic polarisation curve, then manual setting | MPPT every 5–15 minutes, with a safety freeze | | Polarisation curve | Manual procedure lasting hours | Automatic sweep through a resistor bank on a schedule | | COD analysis | Lab test (about 2 hours) as often as daily | Continuous virtual estimate; lab test only weekly for calibration | | Fault diagnosis | An expert notices a drop, then investigates | Automatic residual detection + signature diagnosis within minutes (section 9) | | Temperature | Manual checks | Continuous monitoring; heating only if the energy budget allows (13.4) | | Maintenance planning | After failure | Planned from the trend of ohmic resistance and cathode performance | ### 14.2 What cannot be automated Inoculation and start-up, cleaning the cathode (a weak-acid wash restores activated-carbon cathodes to more than 85% of their initial power, Zhang et al. 2014), replacing membranes and electrodes, periodic sampling for calibration, and repairing leaks. ### 14.3 Evidence that low-intervention operation is possible - An MFC-type BOD sensor operated stably for **over 5 years without servicing** (Kim et al. 2003). - Automatic switching of stack connections, powered by the stack itself, ran with above 90% harvesting efficiency and no human action (Papaharalabos et al. 2017). - Real-time resistance optimisation raised power and reduced methane on real wastewater (Pinto et al. 2011), and MPPT shortened start-up (Molognoni et al. 2014). --- ## 15. Cell configurations and applications ### 15.1 Configurations | Configuration | Description | Typical performance | Strength / weakness | |---|---|---|---| | Two-chamber (H-type) with a membrane | Anode and cathode in separate vessels | 38–40 mW/m² (Min 2005) | Easy to study each electrode; high internal resistance | | Single-chamber air cathode | The cathode faces air directly; no aeration | 146 mW/m² on wastewater; 1220 mW/m² with an activated-carbon cathode | Cheaper and simpler; more oxygen leaks into the anode | | Brush anode (cube cell) | Graphite-fibre brush with a huge surface | 2400 mW/m² (Logan et al. 2007) | High surface area; still lab scale | | Tubular / modular | Many modules in series or parallel in a tank | 200 L and 1000 L pilots (section 6.2) | The route to scale-up | | Ceramic (urine) | Low-cost ceramic cylinders as separators | PEE POWER: 432 cells lighting toilets | Very cheap, practical | | Sediment / benthic | Anode buried in sediment, cathode in the water above | 26.5–47.3 mW/m²; powered a weather buoy (Tender 2008) | Runs for months unattended; low power | | Plant MFC | Plant roots feed the bacteria | 13–18 mW/m² of plant area, 1.1 mW per 1 m tube; drops below 6 °C | Permanent green power source for sensors | | Constructed wetland + MFC | Electrodes inside a treatment wetland | COD 82.8%, nitrate 87.13%, total nitrogen 78.13%, phosphorus 90.3% removal (lab scale) | Better treatment than either alone | ### 15.2 Related technologies - **Microbial electrolysis cell (MEC)**: the same bacteria plus a small added voltage produce hydrogen or other chemicals instead of electricity. A life-cycle study found an MEC producing hydrogen peroxide offers larger environmental benefits than an MFC (Foley et al. 2010). - **Anaerobic digestion**: produces biogas (methane) from the same waste. It is mature and economical at large scale; MFCs compete where low concentrations, low temperatures, or no-aeration treatment matter. ### 15.3 The MFC as a sensor (supports the digital twin) Since current rises with the amount of food, the MFC itself is a biochemical oxygen demand (BOD) sensor: | Result | Source | |---|---| | Linear response up to 100 mg/L BOD at HRT 1.05 h; about 60 minutes to a new steady state; repeatability within 10% | Chang et al. 2004 | | Response time 36 ± 2 minutes (25 mL cell) and 5 ± 1 minutes (5 mL cell) | Moon et al. 2004 | | Stable operation for over 5 years without servicing | Kim et al. 2003 | This is the physical basis of BioVolt's virtual COD sensor (section 8.1, layer 1). ### 15.4 Realistic applications 1. **Wastewater treatment with little or no aeration energy** (the main value). 2. **Self-powered sensors** in remote places: a sediment MFC producing 3.4 mW continuously ran a 2.5 W wireless sensor in bursts through capacitor storage (Donovan et al. 2011). 3. **Water-quality early warning** (BOD and toxicity: a sudden current drop signals a toxic input). 4. **Sanitation in off-grid areas** (urine MFCs for lighting). 5. **Education and research** (BioVolt's own case). --- ## 16. Materials, cost, scale-up, and economics ### 16.1 Materials and published costs | Part | Option | Cost | Performance / note | Source | |---|---|---|---|---| | Cathode base | Carbon cloth | ≈ $1000/m² | Classic lab material | Zhang et al. 2009 | | | Carbon mesh | $10–25/m² | 1355 mW/m² vs 1390 for carbon cloth | Luo et al. 2011 | | | Stainless-steel mesh current collector | ≈ $12/m² | Allows large electrodes | Yang et al. 2014 | | Cathode catalyst | Platinum (0.1–0.5 mg/cm²) | $140–700/m² | Loses activity over time: 250 mW/m² after 16 months | Zhang et al. 2009; Zhang et al. 2014 | | | CoTMPP | ≈ $180/m² | Platinum alternative | Zhang et al. 2009 | | | Activated carbon | ≈ $1.4/kg; complete cathode ≈ $15/m² | 1220 mW/m² vs 1060 for platinum; 960–970 mW/m² after 16 months (modified carbon) | Zhang 2009; Yang 2014; Zhang 2014 | | Separator | Nafion 117 | $1400–2200/m² | 40–60% of total cell cost | Membrane reviews | | | Ceramic (slip-cast clay) | 0.43 €/m²; 0.51 € per cell | 6 months with 29.4% power loss | Rodríguez et al. 2021 | | | Clay + coconut-shell activated carbon | ≈ $45/m² | Low-cost alternative | Ceramic membrane review 2022 | | | No membrane | 0 | Power 146 vs 28 mW/m² with membrane, but CE 20% vs 28% | Liu & Logan 2004 | | Anode | Graphite-fibre brush | — | 2400 mW/m² | Logan et al. 2007 | **Recommendation for BioVolt**: a carbon-brush or carbon-felt anode, an activated-carbon cathode on stainless-steel mesh, and either no membrane or a ceramic separator. Avoid platinum and Nafion. ### 16.2 Why scale-up is hard - **Power density falls with size**: cells below 50 mL often exceed 500 W/m³, while cells above 2 L usually stay below 30 W/m³ (section 6.1). - **Resistance grows** with electrode spacing and area. - **The practical solution is modules**: many small, identical units rather than one large cell (the 1000 L pilot used 50 modules). ### 16.3 The economics in numbers (my calculation) - **Electricity value**: 0.033 kWh/m³ (best pilot) at a typical electricity price of $0.10–0.20/kWh is worth **$0.003–0.007 per cubic metre**, under one cent. - **Aeration saved**: activated-sludge plants consume ≈ 0.3 kWh/m³, i.e. about **9 times** the electricity an MFC produces. Avoided energy consumption is the real economic value. - **Less sludge**: the growth yield of electroactive bacteria is 0.1 g VSS/g COD, while aerobic bacteria in activated sludge have a yield of 0.67 g COD/g COD (the standard ASM1 value) = 0.67 / 1.42 ≈ 0.47 g VSS/g COD. That is roughly **4.7 times less biomass** per gram of COD removed, meaning less sludge to dispose of. (This comparison is based on biomass yield only, not on measured plant sludge.) --- ## 17. Is it really clean energy? Environmental impact and sustainability goals ### 17.1 What life-cycle assessments found | Study | Finding | |---|---| | Foley et al. 2010 (industrial wastewater) | An MFC gave **no significant environmental benefit** over conventional anaerobic treatment with biogas. An MEC producing hydrogen peroxide did give significant benefits, provided it reaches 1000 A/m³. The result depends heavily on materials and performance assumptions | | Miwornunyuie et al. 2025 (domestic wastewater, lab scale) | A constructed wetland had the lowest global warming potential (142.26 kg CO₂-eq). The stand-alone MFC had higher burdens, mainly from **energy-intensive materials and fabrication**. The combined wetland + MFC gave the best treatment and 2.68 kWh of electricity. Costs: wetland $627/m³, wetland + MFC $718/m³ | ### 17.2 The honest environmental claim - The CO₂ released comes from organic waste (biogenic, short-cycle carbon), not fossil fuel. - A well-run MFC avoids methane (a strong greenhouse gas) and aeration energy. - But the materials matter: platinum, Nafion, and carbon cloth can cancel the benefit. **Clean materials are a condition for the "clean" claim.** - Correct wording for the site: "electricity recovered from waste with lower treatment energy", not "zero-emission power". ### 17.3 Sustainable Development Goals | Goal | BioVolt's contribution | |---|---| | SDG 6: Clean water and sanitation | Wastewater treatment with little aeration energy | | SDG 7: Affordable and clean energy | Electricity recovered from waste | | SDG 9: Industry, innovation, and infrastructure | Digital twin and AI for bioprocesses | | SDG 11: Sustainable cities | Decentralised treatment and self-powered sensors | | SDG 12: Responsible consumption and production | Turning food waste into a resource | | SDG 13: Climate action | Less aeration energy and less methane when operated well | --- ## 18. Digital twin standards and maturity levels ### 18.1 Three maturity levels (Kritzinger et al. 2018) | Level | Data flow | BioVolt stage | |---|---|---| | **Digital model** | No automatic data exchange; data is entered manually | The current website simulation | | **Digital shadow** | Automatic one-way flow: the real cell updates the model | Sensors + logging + automatic calibration | | **Digital twin** | Automatic two-way flow: the model's decisions also return automatically to the cell | Sensors + calibration + automatic actions (resistance, dosing, pump) | BioVolt only earns the name "digital twin" at the third level. A prototype can honestly be described as a "digital shadow" until the actions are automatic. ### 18.2 International standards - **ISO 23247 (2021, parts 1–4)**: a digital twin framework for manufacturing. It divides the system into an observable element (here, the cell), a device-communication domain (sensors and actuators), a digital-twin domain (models and synchronisation), and a user domain (dashboard). It describes a reference architecture but does not specify data rates or control behaviour. - **ISO/IEC 30173**: digital twin concepts and terminology. **BioVolt's layers mapped to ISO 23247**: | ISO 23247 domain | BioVolt layer | |---|---| | Observable element | The microbial fuel cell | | Device communication | ESP32, sensors, resistor bank, pumps, heater | | Digital twin | The mechanistic model, Kalman filter, what-if engine, AI | | User | The website dashboard and decision log | --- ## 19. Data pipeline and experimental protocol ### 19.1 Three time scales (why the system is multi-rate) | Process | Time scale | Consequence | |---|---|---| | Electrical (voltage, current) | Seconds to minutes | Sample every 1–10 s; MPPT step every 5–15 minutes | | Chemical (retention time, pH, COD) | Hours to days | pH every 10–60 s; COD estimated continuously and checked weekly | | Biological (growth, biofilm) | Days to weeks | Doubling ≈ 7.5 h; maturation ≈ 2 weeks; recalibrate biological constants daily to weekly | ### 19.2 Data record Every sample stores: time stamp, cell voltage, anode potential, current, external resistance, pH, temperature, conductivity, dissolved oxygen, the model's prediction, the residual, and quality flags. Every AI action stores: the reason, the predicted effect, and the actual effect (for the decision log). ### 19.3 Communication and security The ESP32 sends data over Wi-Fi (for example with the MQTT protocol) to a time-series database. Because the system can operate pumps and dosing remotely, connections must be authenticated and encrypted, and the safety rules must run **on the device itself** so they still apply if the network fails. ### 19.4 Start-up and test protocol 1. **Build and leak-test** the cell with clean water. 2. **Inoculate** with treatment-plant sludge or wastewater, or for schools with soil or sediment and a synthetic acetate medium. 3. **Start-up** with a fixed external resistance and regular feeding. Expect days to weeks (3.5 days at 35 °C, more than 40 days at 15 °C, Patil 2010). MPPT can shorten start-up (Molognoni 2014). 4. **Stable operation**: when the voltage repeats cycle after cycle, the biofilm is mature. 5. **Polarisation curve**: step through resistances, waiting for a steady voltage at each (typically tens of minutes), or use a slow linear sweep (about 1 mV/s). Calculate internal resistance and maximum power. 6. **Coulombic efficiency**: measure COD at the start and end of a cycle and integrate the current (section 5.3). 7. **Calibrate the digital twin** with these results (section 7.4), then enable the AI step by step: monitoring, then recommendations, then automatic actions. ### 19.5 Sensor maintenance pH electrodes drift and need regular two-point calibration with standard buffers (weekly is common practice). The reference electrode is checked against a spare one. The AI's sensor-fault rule (section 9) flags readings that are frozen, impossible, or inconsistent with the model. --- ## 20. Safety | Hazard | Source | Precaution | |---|---|---| | Pathogens | Real wastewater and sludge contain enteric microbes | Biosafety level 2 practices for lab work with wastewater (gloves, eye protection, disinfection, safe waste disposal); schools can use soil and a synthetic medium at biosafety level 1 | | Hydrogen sulphide (H₂S) | Sulphate-reducing bacteria in anaerobic chambers | Toxic gas: sealed cells, ventilation, never smell the chamber, gas detection at larger scales | | Methane and hydrogen | Small amounts from competing microbes | Flammable: vent the headspace, no flames nearby | | Electrical | Supercapacitors and boost converters | Cell voltage is safe, but charged capacitors can deliver large short-circuit currents | | Chemicals | Buffers, weak acid for cathode cleaning | Standard chemical handling and labelling | | Automated actuators | Pumps and dosing controlled by software | Hard limits on the device, a manual stop, and logging of every action | --- ## 21. Coverage matrix: every sentence of the BioVolt idea and where it is proven | Sentence in the BioVolt idea | Where it is covered | Verdict | |---|---|---| | A smart MFC system that turns organic waste or wastewater into clean electricity | 3, 5, 12, 17 | True, with honest numbers (6) and the materials condition (17) | | Bacteria break down organic material and release electrons that generate power | 3.2, 4, 5.3, 5.4 | Exact physics and measured biology | | Improved with artificial intelligence and a digital twin | 7, 8, 18 | Architecture, methods, and standards | | Sensors measure pH, temperature, voltage, and current in real time | 10, 19 | Plus anode potential, conductivity, and dissolved oxygen | | The digital twin allows testing operating conditions and predicting problems safely before applying changes | 7.3, 8.1 (MPC), 13 | Demonstrated by the what-if and disturbance simulations | | The AI analyses live data and detects changes that may reduce performance | 8.1 (layers 1–2), 9 | Residual-based detection and fault signatures | | Recommends or automatically applies adjustments | 8.1 (layer 4), 13, 14 | MPPT, dosing, safety layer | | Improves electricity generation | 5.11, 8.1, 13.3 | MPPT: +36.5% energy in simulation; published gains up to 2.7× | | Supports wastewater treatment | 5.11, 6.2, 15.1 | 70–90% COD removal in the 1000 L pilot | | Reduces human intervention | 14 | Task-by-task automation, with what remains manual | | Recovers more quickly from disturbances | 13 | Literature (caught within 15 min recovers fully) + simulation (0% vs 11.9% energy lost) | | Combines biotechnology, clean energy, AI, and digital twins into a smarter, more reliable system | All sections | Reliability depends on the safety layer (13.4) | --- ## 22. Glossary | Term | Meaning | |---|---| | MFC | Microbial fuel cell: bacteria convert organic matter directly into electricity | | Anode / cathode | The electrode receiving electrons from bacteria / the electrode where oxygen accepts them | | Biofilm | The bacterial layer attached to the anode | | Exoelectrogen / ARB | Bacteria that transfer electrons to a solid electrode | | COD | Chemical oxygen demand: organic matter measured as the oxygen needed to oxidise it; 8 g COD = 1 mol of electrons | | BOD | Biochemical oxygen demand: the part of the organic matter bacteria can consume | | HRT | Hydraulic retention time: chamber volume ÷ flow | | Coulombic efficiency (CE) | Fraction of the electrons in the removed COD that reached the circuit | | OCV | Open-circuit voltage: cell voltage without a load | | Internal resistance | Total of the cell's losses expressed as a resistance | | Polarisation curve | Voltage and power plotted against current across many loads | | MPPT | Maximum power point tracking: automatically matching the load to the cell | | Monod / Nernst–Monod | Equations for bacterial rate as a function of food / food and anode potential | | Doubling time | Time for the bacterial population to double (≈ 7.5 h here) | | Hydrolysis | Breaking solid polymers into dissolved molecules | | Methanogens | Microbes that turn food into methane and compete with electricity production | | Voltage reversal | A starving cell in a stack whose voltage turns negative | | Digital model / shadow / twin | No automatic data flow / one-way flow / two-way flow | | EKF | Extended Kalman filter: estimates unmeasured quantities from measured ones | | MPC | Model predictive control: choosing actions by simulating their future effect | | LCA | Life-cycle assessment: environmental impact from materials to disposal | | Resistance / resilience | Size of the drop caused by a disturbance / speed of recovery from it | --- ## 23. Model limits and scientific honesty - **The physical laws** (Faraday, Nernst, conservation of mass and electrons, Ohm) are **exact**. - **The biological constants** have published ranges, not single values. The model uses mutually consistent, verified values, but every real cell needs calibration (and this is the digital twin's role). - **Unverified values** are marked: T_min/T_opt/T_max and the pH coefficients (inferred by fitting the published data), the membrane's oxygen permeability coefficient, and the cathode exchange current I₀ (catalyst-dependent, to be calibrated). - **The model is one-dimensional** (it does not compute concentration gradients inside the biofilm). This is an accepted, published simplification (Pinto 2010 uses a comparable model). - **No exaggeration**: the site should not claim the cell "lights a house" or "solves the energy crisis". The honest claim: water treatment with less energy + a small amount of electricity + smart autonomous operation. - **pH is modelled as mixing** between the influent and the chamber, without buffer chemistry. Real buffers slow pH changes, so real shocks may be milder, while protons inside the biofilm can make them harsher (section 4.3). - **The recovery simulation does not include stress-induced death**: bacteria only lose activity while conditions are bad. Real severe shocks can kill part of the biofilm, so real recovery can be longer (the literature reports 60.7 h on average and some cells never recovering). - **The hydrolysis constant** differs by orders of magnitude between systems (0.0576 d⁻¹ measured in MFCs versus about 10 d⁻¹ in ADM1 digesters) and must be calibrated for each waste. - **The heating energy estimate** assumes a 2-hour thermal time constant for the 250 mL cell. - **The human-intervention frequencies** in section 14 are design targets, not measurements. --- ## 24. References ### Thermodynamics and methodology 1. Logan, B.E., Hamelers, B., Rozendal, R., et al. (2006). Microbial fuel cells: methodology and technology. *Environ. Sci. Technol.* 40(17):5181–5192. doi:10.1021/es0605016 2. Logan, B.E., Regan, J.M. (2006). Microbial fuel cells — challenges and applications. *Environ. Sci. Technol.* 40(17):5172–5180. 3. He, Z. (2017). Development of microbial fuel cells needs to go beyond "power density". *ACS Energy Lett.* 2:700. doi:10.1021/acsenergylett.7b00041 ### Bacterial kinetics 4. Kato Marcus, A., Torres, C.I., Rittmann, B.E. (2007). Conduction-based modeling of the biofilm anode of a microbial fuel cell. *Biotechnol. Bioeng.* 98(6):1171–1182. doi:10.1002/bit.21533 5. Torres, C.I., Kato Marcus, A., Parameswaran, P., Rittmann, B.E. (2008). Kinetic experiments for evaluating the Nernst–Monod model for anode-respiring bacteria in a biofilm anode. *Environ. Sci. Technol.* 42:6593. doi:10.1021/es800970w 6. Torres, C.I., Kato Marcus, A., Rittmann, B.E. (2008). Proton transport inside the biofilm limits electrical current generation by anode-respiring bacteria. *Biotechnol. Bioeng.* 100:872. doi:10.1002/bit.21821 7. Torres, C.I., et al. (2010). A kinetic perspective on extracellular electron transfer by anode-respiring bacteria. *FEMS Microbiol. Rev.* 34:3–17. doi:10.1111/j.1574-6976.2009.00191.x 8. Lee, H.S., Torres, C.I., Rittmann, B.E. (2009). Effects of substrate diffusion and anode potential on kinetic parameters for anode-respiring bacteria. *Environ. Sci. Technol.* 43:7571. doi:10.1021/es9015519 9. 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(14):5256–5265. doi:10.1016/j.biortech.2010.01.122 10. Wilson, E.L., Kim, Y. (2016). The yield and decay coefficients of exoelectrogenic bacteria in bioelectrochemical systems. *Water Res.* 94:233. doi:10.1016/j.watres.2016.02.054 11. Bond, D.R., Lovley, D.R. (2003). Electricity production by *Geobacter sulfurreducens* attached to electrodes. *Appl. Environ. Microbiol.* 69:1548. doi:10.1128/AEM.69.3.1548-1555.2003 12. Marsili, E., Sun, J., Bond, D.R. (2010). Voltammetry and growth physiology of *Geobacter sulfurreducens* biofilms as a function of growth stage and imposed electrode potential. *Electroanalysis* 22:865. doi:10.1002/elan.200800007 13. Esteve-Núñez, A., et al. (2005). Growth of *Geobacter sulfurreducens* under nutrient-limiting conditions in continuous culture. *Environ. Microbiol.* 7:641. doi:10.1111/j.1462-2920.2005.00731.x 14. Lin, W.C., Coppi, M.V., Lovley, D.R. (2004). *Geobacter sulfurreducens* can grow with oxygen as a terminal electron acceptor. *Appl. Environ. Microbiol.* 70:2525. doi:10.1128/AEM.70.4.2525-2528.2004 15. Jana, P.S., Katuri, K., Kavanagh, P., Kumar, A., Leech, D. (2014). *Phys. Chem. Chem. Phys.* 16:9039. doi:10.1039/c4cp01023j 16. Malvankar, N.S., et al. (2011). Tunable metallic-like conductivity in microbial nanowire networks. *Nat. Nanotechnol.* 6:573. doi:10.1038/nnano.2011.119 ### Temperature and pH 17. Rosso, L., Lobry, J.R., Flandrois, J.P. (1993). An unexpected correlation between cardinal temperatures of microbial growth highlighted by a new model. *J. Theor. Biol.* 162:447–463. 18. Rosso, L., Lobry, J.R., Bajard, S., Flandrois, J.P. (1995). Convenient model to describe the combined effects of temperature and pH on microbial growth. *Appl. Environ. Microbiol.* 61:610–616. 19. Liu, H., Cheng, S., Logan, B.E. (2005). Power generation in fed-batch microbial fuel cells as a function of ionic strength, temperature, and reactor configuration. *Environ. Sci. Technol.* 39:5488. doi:10.1021/es050316c 20. Min, B., Román, Ó.B., Angelidaki, I. (2008). Importance of temperature and anodic medium composition on microbial fuel cell performance. *Biotechnol. Lett.* 30:1213. doi:10.1007/s10529-008-9687-4 21. Patil, S.A., Harnisch, F., Kapadnis, B., Schröder, U. (2010). Electroactive mixed culture biofilms in microbial bioelectrochemical systems: the role of temperature. *Biosens. Bioelectron.* 26:803. doi:10.1016/j.bios.2010.06.019 22. Patil, S.A., et al. (2011). Electroactive mixed culture derived biofilms in microbial bioelectrochemical systems: the role of pH. *Bioresour. Technol.* 102:9683. doi:10.1016/j.biortech.2011.07.087 23. Franks, A.E., et al. (2009). Real-time spatial gene expression analysis within current-producing biofilms. *Energy Environ. Sci.* 2:113. doi:10.1039/B816445B ### Performance and pilot projects 24. Min, B., Cheng, S., Logan, B.E. (2005). Electricity generation using membrane and salt bridge microbial fuel cells. *Water Res.* 39:1675. doi:10.1016/j.watres.2005.02.002 25. Liu, H., Logan, B.E. (2004). Electricity generation using an air-cathode single chamber microbial fuel cell in the presence and absence of a proton exchange membrane. *Environ. Sci. Technol.* 38:4040. doi:10.1021/es0499344 26. Liu, H., Cheng, S., Logan, B.E. (2005). Production of electricity from acetate or butyrate using a single-chamber microbial fuel cell. *Environ. Sci. Technol.* 39:658. doi:10.1021/es048927c 27. Feng, Y., Wang, X., Logan, B.E., Lee, H. (2008). Brewery wastewater treatment using air-cathode microbial fuel cells. *Appl. Microbiol. Biotechnol.* 78:873. 28. Fan, Y., Han, S.K., Liu, H. (2012). Improved performance of CEA microbial fuel cells with increased reactor size. *Energy Environ. Sci.* 5:8273. doi:10.1039/C2EE21964F 29. Rossi, R., et al. (2021). *ACS Sustain. Chem. Eng.* doi:10.1021/acssuschemeng.0c09144 30. Liang, P., et al. (2018). One-year operation of 1000-L modularized microbial fuel cell for municipal wastewater treatment. *Water Res.* 141:1–8. doi:10.1016/j.watres.2018.04.066 31. Ge, Z., He, Z. (2016). Long-term performance of a 200 liter modularized microbial fuel cell system treating municipal wastewater. *Environ. Sci.: Water Res. Technol.* 2:274. doi:10.1039/C6EW00020G 32. He, W., et al. (2019). *Water Res.* 155:372–380. doi:10.1016/j.watres.2019.01.062 33. Ieropoulos, I., et al. (2016). Pee power urinal — microbial fuel cell technology field trials in the context of sanitation. *Environ. Sci.: Water Res. Technol.* 2:336. doi:10.1039/C5EW00270B 34. Oh, S.E., Logan, B.E. (2007). Voltage reversal during microbial fuel cell stack operation. *J. Power Sources* 167:11. doi:10.1016/j.jpowsour.2007.02.016 35. Heidrich, E.S., Curtis, T.P., Dolfing, J. (2011). Determination of the internal chemical energy of wastewater. *Environ. Sci. Technol.* 45(2):827. doi:10.1021/es103058w 36. Whole-cell model for wastewater-fed single-chamber MFCs (2022). *Applied Energy* (VITO open access). ### Control and artificial intelligence 37. Woodward, L., Tartakovsky, B., Perrier, M., Srinivasan, B. (2009). Maximizing power production in a stack of microbial fuel cells using multiunit optimization method. *Biotechnol. Prog.* 25:676–682. doi:10.1002/btpr.115 38. Woodward, L., Perrier, M., Srinivasan, B., Pinto, R.P., Tartakovsky, B. (2010). Comparison of real-time methods for maximizing power output in microbial fuel cells. *AIChE J.* 39. Pinto, R.P., Srinivasan, B., Guiot, S.R., Tartakovsky, B. (2011). The effect of real-time external resistance optimization on microbial fuel cell performance. *Water Res.* 40. Premier, G.C., Kim, J.R., Michie, I., Dinsdale, R.M., Guwy, A.J. (2011). Automatic control of load increases power and efficiency in a microbial fuel cell. *J. Power Sources* 196:2013–2019. doi:10.1016/j.jpowsour.2010.09.071 41. Molognoni, D., et al. (2014). Reducing start-up time and minimizing energy losses of microbial fuel cells using maximum power point tracking strategy. *J. Power Sources* 269. 42. A brief review on factors affecting the performance of microbial fuel cell and integration of artificial intelligence (2025). *Discover Sustainability*. doi:10.1007/s43621-025-01619-6 43. Exploring the application of artificial intelligence for bioelectrochemical systems: a review of recent research (2025). *Green Energy and Resources*. doi:10.1016/j.gerr.2025.100141 ### Hardware (manufacturer datasheets) 44. TI ADS1115, INA219, INA226, BQ25504, BQ25570; Analog Devices LTC3108, DS18B20; Allegro ACS712; DFRobot SEN0169-V2, DFR0300, SEN0237-A, SEN0165; Microchip MCP41100. ### Solid organic waste and substrates 45. Velasquez-Orta, S.B., Yu, E.H., Katuri, K.P., Head, I.M., Curtis, T.P., Scott, K. (2011). Evaluation of hydrolysis and fermentation rates in microbial fuel cells. *Appl. Microbiol. Biotechnol.* 90(2):789–798. doi:10.1007/s00253-011-3126-5 46. Organic waste substrates for bioenergy production via microbial fuel cells: a key point review (2022). *Energies* 15(15):5616. doi:10.3390/en15155616 47. Wang, S., Adekunle, A., et al. (2021). Synthesizing developments in the usage of solid organic matter in microbial fuel cells: a review. *Chem. Eng. J. Adv.* 48. Choi, J., Ahn, Y. (2015). Enhanced bioelectricity harvesting in microbial fuel cells treating food waste leachate produced from biohydrogen fermentation. *Bioresour. Technol.* 183. 49. Performance of single-chamber microbial fuel cells using different carbohydrate-rich wastewaters and different inocula (2016). *Pol. J. Environ. Stud.* 25(2):503–510. doi:10.15244/pjoes/61115 50. Yu, J., Park, Y., Kim, B., Lee, T. (2014). Power densities and microbial communities of brewery wastewater-fed microbial fuel cells according to the initial substrates. *Bioprocess Biosyst. Eng.* doi:10.1007/s00449-014-1246-x 51. Ge, Z. (2015). Energy-efficient wastewater treatment by microbial fuel cells: scaling up and optimization. PhD dissertation, Virginia Tech. 52. Batstone, D.J., et al. (2002). Anaerobic Digestion Model No. 1 (ADM1). IWA Scientific and Technical Report No. 13. IWA Publishing. 53. Pant, D., Van Bogaert, G., Diels, L., Vanbroekhoven, K. (2010). A review of the substrates used in microbial fuel cells (MFCs) for sustainable energy production. *Bioresour. Technol.* 101(6):1533–1543. ### Disturbance, recovery, and automation 54. Lesnik, K.L., Cai, W., Liu, H. (2020). Microbial community predicts functional stability of microbial fuel cells. *Environ. Sci. Technol.* 54(1):427–436. doi:10.1021/acs.est.9b03667 55. Satinover, S.J., Rodriguez, M., Borole, A.P. (2020). Microbial electrolysis cell recovery after inducing operational failure conditions. *Biochem. Eng. J.* doi:10.1016/j.bej.2020.107800 56. Saheb-Alam, S., Persson, F., Wilén, B.M., et al. (2019). Response to starvation and microbial community composition in microbial fuel cells enriched on different electron donors. *Microb. Biotechnol.* 12(5):962–975. doi:10.1111/1751-7915.13449 57. Papaharalabos, G., et al. (2017). Autonomous energy harvesting and prevention of cell reversal in MFC stacks. *J. Electrochem. Soc.* 164(3). doi:10.1149/2.0081703jes 58. Chang, I.S., Jang, J.K., Gil, G.C., Kim, M., Kim, H.J., Cho, B.W., Kim, B.H. (2004). Continuous determination of biochemical oxygen demand using microbial fuel cell type biosensor. *Biosens. Bioelectron.* 19:607–613. 59. Kim, B.H., Chang, I.S., Gil, G.C., Park, H.S., Kim, H.J. (2003). Novel BOD (biological oxygen demand) sensor using mediator-less microbial fuel cell. *Biotechnol. Lett.* 25:541–545. 60. Moon, H., Chang, I.S., Kang, K.H., Jang, J.K., Kim, B.H. (2004). Improving the dynamic response of a mediator-less microbial fuel cell as a biochemical oxygen demand (BOD) sensor. *Biotechnol. Lett.* 26:1717–1721. doi:10.1007/s10529-004-3743-5 ### Materials, cost, and configurations 61. Zhang, F., Cheng, S., Pant, D., Van Bogaert, G., Logan, B.E. (2009). Power generation using an activated carbon and metal mesh cathode in a microbial fuel cell. *Electrochem. Commun.* 11:2177–2179. doi:10.1016/j.elecom.2009.09.024 62. Yang, W., He, W., Zhang, F., Hickner, M.A., Logan, B.E. (2014). Single-step fabrication using a phase inversion method of poly(vinylidene fluoride) (PVDF) activated carbon air cathodes for microbial fuel cells. *Environ. Sci. Technol. Lett.* 1:416–420. 63. Zhang, X., et al. (2014). Long-term performance of chemically and physically modified activated carbons in air cathodes of microbial fuel cells. *ChemElectroChem*. doi:10.1002/celc.201402123 64. Luo, Y., et al. (2011). Power generation using carbon mesh cathodes with different diffusion layers in microbial fuel cells. *J. Power Sources*. 65. Logan, B.E., Cheng, S., Watson, V., Estadt, G. (2007). Graphite fiber brush anodes for increased power production in air-cathode microbial fuel cells. *Environ. Sci. Technol.* 41:3341–3346. 66. Rodríguez, J., et al. (2021). Comprehensive characterization of a cost-effective microbial fuel cell with Pt-free catalyst cathode and slip-casted ceramic membrane. University of Cagliari (IRIS 11584/309209). 67. The implications of membranes used as separators in microbial fuel cells (2021). *Membranes* (MDPI). PMC8539572. 68. Next-generation proton-exchange membranes in microbial fuel cells: overcoming Nafion's limitations (review). 69. Current outlook towards feasibility and sustainability of ceramic membranes for scaling-up and practical applications of microbial fuel cells (2022). *Renew. Sustain. Energy Rev.* (University of Liverpool repository). 70. Tender, L.M., et al. (2008). The first demonstration of a microbial fuel cell as a viable power supply: powering a meteorological buoy. *J. Power Sources* 179(2):571–575. doi:10.1016/j.jpowsour.2007.12.123 71. Donovan, C., Dewan, A., Peng, H., Heo, D., Beyenal, H. (2011). Power management system for a 2.5 W remote sensor powered by a sediment microbial fuel cell. *J. Power Sources* 196(3):1171–1177. doi:10.1016/j.jpowsour.2010.08.099 72. Long-term performance of pilot-scale tubular plant-microbial fuel cells in a brownfield-constructed wetland (field study). 73. Enhancing the power performance of sediment microbial fuel cells by novel strategies: overlying water flow and hydraulic-driven cathode rotating (2019). *Sci. Total Environ.* ### Environment, economics, and standards 74. Foley, J.M., Rozendal, R.A., Hertle, C.K., Lant, P.A., Rabaey, K. (2010). Life cycle assessment of high-rate anaerobic treatment, microbial fuel cells, and microbial electrolysis cells. *Environ. Sci. Technol.* 44(9):3629–3637. doi:10.1021/es100125h 75. Miwornunyuie, N., Alamu, S.O., Mao, G., Benani, N., Hunter, J., Oguntimein, G. (2025). Comparative life cycle and techno-economic assessment of constructed wetland, microbial fuel cell, and their integration for wastewater treatment. *Clean Technol.* 7(3):57. doi:10.3390/cleantechnol7030057 76. Henze, M., Gujer, W., Mino, T., van Loosdrecht, M.C.M. (2000). Activated Sludge Models ASM1, ASM2, ASM2d and ASM3. IWA Scientific and Technical Report No. 9. 77. Kritzinger, W., Karner, M., Traar, G., Henjes, J., Sihn, W. (2018). Digital twin in manufacturing: a categorical literature review and classification. *IFAC-PapersOnLine* 51(11):1016–1022. doi:10.1016/j.ifacol.2018.08.474 78. ISO 23247-1 to -4:2021. Automation systems and integration — Digital twin framework for manufacturing. 79. ISO/IEC 30173:2023. Digital twin — Concepts and terminology. ### Safety 80. American Society for Microbiology. Guidelines for Biosafety in Teaching Laboratories; and "Powerful Soil: utilizing microbial fuel cell construction and design in an introductory biology course" (2015). *J. Microbiol. Biol. Educ.* 16(2). doi:10.1128/jmbe.v16i2.934