Developments in anaerobic digestion modelling

Doctoral Thesis


Permanent link to this Item
Journal Title
Link to Journal
Journal ISSN
Volume Title
Anaerobic digestion (AD) is notorious for being susceptible to failure and is regarded as unstable and sensitive. Thus, to avoid failure, anaerobic digesters are frequently operated far below their optimal level. In order to run a digester closer to capacity, a better understanding of AD failure is required. Under conditions approaching failure, or during startup, intermediate products such as acetate, propionate and hydrogen accumulate. Successful AD modelling during failure requires the AD model to be suitably calibrated. Some AD models have been calibrated to the initial slow rate-limiting hydrolysis step only with the result that these models cannot be used to predict AD failure. Without adequate modelling of AD dynamics, the AD model cannot be used to model digester start-up, digester failure or even upflow anaerobic sludge bed (UASB) reactors (which have temporary failure conditions at the bottom of the bed). This study aims to develop an AD model capable of predicting failure and digester start-up conditions. Development of an improved model was accomplished by means of calibrating the AD model to a UASB reactor dataset wherein temporary failure conditions are present in the bottom of the reactor, evident by the presence of the abovementioned intermediate products. After comparing and contrasting available AD models to identify one for further development, the AD model subset (PWM_SA_AD) of plantwide model South Africa (PWM_SA) was selected because (1) it characterizes the organics’ composition using routine wastewater treatment measurements rather than carbohydrates, lipids and proteins, which is typical of other models, (2) external speciation reduces model stiffness, (3) includes aqueous, gas and solid phases for pH calculation, gas evolution and mineral precipitation and (4) contains the same components as PWM_SA enabling plant-wide modelling without needing component transformers between process units. Before calibration, PMW_SA_AD was rigorously tested for mass balance, stoichiometric and kinetic correctness. Because the UASB reactor undergoes temporary failure observed by the accumulation of AD intermediate substrates in the bottom of the bed, the glucose fed UASB reactor system of Sam-Soon et al. (1989) was modelled to calibrate the Monod kinetic constants of the acidogens, acetoclastic methanogens, acetogens and hydrogenotrophic methanogens. This required coding into WEST® (MikebyDHI, 2016), the platform on which PWM_SA runs, a six-in-series completely-mixed AD system with a solids retention factor for each digester that retained a fraction of the reactor’s solids. Determination of the parameters that required calibration was identified with sensitivity analysis. Due to the complexity of the physical, biological and aqueous interactions, many model simulations were needed to identify the important parameters with the lasso (least absolute shrinkage and selection operator) feature selection method. Not unexpectedly the most important parameters that required calibration were the retention factor for each digester in the series; and the maximum specific growth rates and the half-saturation coefficients for the four AD biomass groups, which were global parameters, i.e. for each biomass group the same values in each digester of the series applies. Stability of the complex six-in-series UASB reactor needed the initial masses in each digester to be reasonably close to the final steady-state masses. Steady-state Microsoft Excel AD spreadsheet models were set-up to calculate these initial masses. Following the calibration procedure wherein the modelled AD intermediate products matched the measurements from the UASB reactor dataset, it was expected that the pH also would be predicted well. However, this was not the case. So, the assumption of equilibrium between the headspace CO2 partial pressure and aqueous phase CO2 concentration was replaced by a rate-controlled CO2 evolution. With this correction, the predicted pH matched well with that observed along the height of the UASB reactor. The calibrated model was then tested to observe how the UASB reactor system fails irrecoverably by gradually decreasing the influent alkalinity from the dataset value of 6000 mg/L as CaCO3. Irrecoverable failure occurred at an influent alkalinity of 4200mg/L as CaCO3 because the specific growth rate of the acetoclastic methanogens, which progressively decreases the further the pH falls below 7, fell below the minimum required to utilise the high acetate concentration. The role of the sensitivity of the acetogens to hydrogen in digester failure was also tested. Counter-intuitively, it was found that this in and of itself did not cause failure but that to a degree, postponed failure because the acetogen inactivity at high hydrogen concentration delayed the acetate load on the acetoclastic methanogens. To verify the acetogen effect on failure, acetogen sensitivity to hydrogen was increased, and the alkalinity was gradually decreased. Irrecoverable failure now occurred at an influent alkalinity of 4000mg/L as CaCO3. The above modes of UASB reactor failure predicted by PWM_SA_AD were compared with ADM1 (Anaerobic Digestion Model No. 1). ADM1 was coded into PWM_SA_AD as an independent subset using the same external speciation routine. Although ADM1 has previously been documented to be incapable of predicting AD failure, this comparison showed that ADM1 predicted the same failure modes as PWM_SA_AD but at a higher influent alkalinity of 5000mg/L as CaCO3. One of the main reasons why ADM1 fails at sooner is that the specific growth rate of the acetoclastic methanogens in ADM1 is slower than in the PWM_SA_AD model calibrated to the UASB reactor data. The UASB reactor system calibrated PWM_SA_AD model was applied to model digester start-up with primary sewage sludge. This was done by adding to the single completelymixed anaerobic digester a percentage of seed and filling the rest of the volume with wastewater treatment plant effluent with 250 mg/L as CaCO3 alkalinity. The percentage of seed was the percentage of the anaerobic digester volume, which was filled with seed sludge containing the same biomass concentrations as those at steady state after start-up is complete and the set digester sludge age is reached. The hydrolysis rate of biodegradable particulate organics (BPO) was modelled with saturation kinetics (also known as Contois (1959) kinetics) with constants obtained from Sötemann et al. (2005b). Three different startup cases were investigated (1) setting the influent pump at the final steady-state flow rate but switching it off and on with either a pH controller or a Ripley ratio controller, (2) increasing the influent flow by a fixed proportion of the final steady-state flow daily (t1/t -1), where t is the start-up duration, (3) same as (2) but adding either a pH controller or a Ripley ratio controller. Two modes failure, resulting in an inability to start up, were found: (1) BPO overload which causes acetoclastic methanogen overload and surprisingly (2) acetoclastic methanogen starvation. BPO overload results in a high acetate concentration and low pH, which slow the acetoclastic methanogens below the tipping point to start up. It is exacerbated by low percentage of seed and during the slow hydrolysis and acidification of BPO, or a setpoint which is not sufficiently conservative. Acetoclastic methanogen starvation is as a result of a too conservative setpoint which prevents flow from entering the digester, thereby depriving the organisms of the substrate. Plotting the specific growth rate to the maximum specific growth rate ratio of the acetoclastic methanogens indicated that under starvation conditions, the ratio is extremely low. The reason for the low ratio is due to the low bulk liquid concentrations on which Monod kinetics depends. The limitations of Monod kinetics are made apparent here because the initial seed amount is significantly below the final steadystate mass. So, for these cases, further investigations are required to identify if saturation kinetics will allow better predictions. Through the development of the model, although the model was capable of predicting failure and start-up in line with the expected principles, it is not possible to find a unique set of kinetic constants, resulting in a degree of freedom with the choice of a maximum specific growth rate of acidogens. This degree of freedom may have been eliminated if sufficient measurements were available. Overall, the investigation provided useful insight into the mode of AD failure and difficulties regarding modelling digester start-up. There is, therefore, the scope for further additions to the study, with a specific focus on the residual COD, sludge bed measurements, gas flow and hydrogen concentration in the bed and modelling the acidogen, acetogen, acetoclastic methanogen and hydrogenotrophic methanogen specific growth rates with saturation kinetics. This will enable greater insight into the failure modes and the effect of hydrogen and growth rate kinetics on the AD system failure.