Parameter estimation and sensitivity analysis for dynamic modelling and simulation of beer fermentation

Beer fermentation efficiency improvements have the strongest potential to boost profitability, as its long batch time renders this particular unit operation the throughput bottleneck of this complex, multistage biochemical process which mankind has employed for several millennia. Accurate fermentation models are critical for reliable dynamic simulation and process optimization: empirical trial-and-error approaches are not viable, and incrementally altering proven recipes implies prohibitively expensive campaigns, in terms of equipment use, off-spec production and personnel time for sampling and analysis. This paper considers parameter estimation for a published beer fermentation model, demonstrating that estimating the complete unknown parameter set is an ill-posed problem, which can lead to inconsistent solutions. Systematic sensitivity analysis is pursued, elucidating the relative significance of parametric discrepancy on the validity of key species concentration trajectories. Parameters have been identified and ranked by decreasing importance, and a high-fidelity estimation is performed for a published dataset.