A probabilistic model-based soft sensor to monitor lactic acid bacteria fermentations

A probabilistic soft sensor based on a mechanistic model was designed to monitor S. thermophilus fermentations, and validated with experimental lab-scale data. It considered uncertainties in the initial conditions, on-line measurements, and model parameters by performing Monte Carlo simulations within the monitoring system. It predicted, therefore, the probability distributions of the unmeasured states, such as biomass, lactose, and lactic acid concentrations. To this end, a mechanistic model was developed first, and a statistical parameter estimation was performed in order to assess parameter sensitivities and uncertainties. The model coupled a biokinetic and a mixed weak acid/base model to predict biological variables and chemical variables like the pH, respectively. In the soft sensor, the limited available on-line measurements, namely the quantity of added ammonia and pH, were used to update the model parameters that were then used as input to the mechanistic model. The soft sensor predicted both the current state variables, as well as the future course of the fermentation, e.g. with a relative mean error of the biomass concentration of 8 %. This successful implementation of a process analytical technology monitoring system opens up further opportunities, including for on-line risk-based monitoring and control applications.