Robust Monitoring of Lactic Acid Bacteria with Sequential Monte Carlo

Lactic Acid Bacteria (LAB) are commonly utilised in the dairy industry. Although the process is well established, the conditions in which the fermentation process takes place can strongly influence the production of LAB.

The objective of this manuscript is to apply the Sequential Monte Carlo (SMC) method for monitoring the fermentation process. The online monitoring system is based on a statistical method for the estimation of model parameters using importance sampling technique, making use of limited data and predicting a number of relevant but unmeasured process variables such as biomass and product yields. The SMC technique was evaluated using two case studies one a simulation study and the other is a pilot-scale data from LAB fermentation. To describe the process, a mechanistic fermentation model was developed and validated comprehensively before being used in SMC for online monitoring. The results show the application of SMC makes effective use of online data as batch progress and the quality of the predictions (for unmeasured state variables) improve as more information becomes available. This indicates the principle of Bayesian update of prior information on model parameters through importance sampling successfully works. In addition to process monitoring using limited online knowledge, the SMC technique allows convergence to the real value of the parameters of interest in the model. In fermentation studies with a limited online sensor, SMC offers a flexible alternative to traditional soft-sensors especially when the measurement errors are not necessarily normally distributed.