Abstract
The translation of kinetic parameters from laboratory-scale experiments to full-scale anaerobic digestion (AD) plants remains a critical barrier to the wide-scale adoption of mathematical modelling in the biogas industry. Full-scale facilities are characterised by dynamic loading, feedstock heterogeneity, and data scarcity where conditions are rarely replicated in controlled laboratory environments. This study addresses this scaling challenge by applying a mass-based Anaerobic Digestion Model No. 1 (ADM1) to two full-scale UK plants: Site 1 (stable maize and cow manure) and Site 2 (an 8-feedstock co-digestion plant). A laboratory ‘Physical Twin’ approach was initially evaluated, where kinetic parameters derived from a laboratory-scale digester fed a similar diet were applied directly to the full-scale model. While successfully predicting biogas outputs with mean absolute percentage error (MAPE) < 10%), it failed to capture internal process stability, significantly underestimating volatile fatty acid (VFA) accumulation (MAPE = 54%) and overestimating total ammonium nitrogen. Subsequent site-specific calibration, targeting sensitive parameters identified via Morris Global Sensitivity Analysis (GSA), effectively reduced VFA errors to 13.1% and nitrogen errors to 18.4% by mathematically compensating for the effects of spatial heterogeneity and non-ideal mixing at industrial scale. Furthermore, the framework demonstrated robustness when applied to the complex multi-feedstock regime of Site 2, accurately simulating the physicochemical buffering provided by pig slurry against the acidogenic load of high-yield crops. Due to increased biochemical complexity and seasonal feedstock shifts, Site 2 presented greater challenges, yielding moderate accuracy for biogas (13.4% MAPE) and pH (3.97%), but higher errors for VFAs (30.9%) and the IA/PA ratio (39.2%). Ultimately, these results establish a promising methodology for developing high-fidelity, offline calibrated mechanistic simulators. By directly addressing the physical constraints of industrial operations, this framework provides a foundational step toward mature Digital Twins capable of predicting gas production and process stability at scale.
[Display omitted]
•Lab kinetics predict biogas accurately but fail to capture VFA stability trends.•GSA-guided calibration reduced VFA prediction errors by 75% for site monitoring.•Digital twins must account for industrial mass transfer and non-ideal mixing.•Site-specific ADM1 calibration is essential for robust industrial process control.