Product identification in industrial batch fermentation using a variable forgetting factor

For reliable operation and the optimization of production, industrial fermentation processes require appropriate tools for monitoring the process in real time. This work presents the structure and operation of a soft sensor for the on-line monitoring of biomass and product concentration during salinomycin and bacitracin fermentation in an industrial, 80-m3 batch reactor; moreover it provides a tool for evaluation of batch production verified in industrial application. The process estimation algorithm consists of decoupled growth and product models, which ensures an unbiased convergence of the estimator and the robustness of the model. The production of secondary metabolites is described with a non-structured model upgraded with a variable forgetting factor that demonstrated a successful estimation of the non-measured parameters and states of this highly interactive and interlinked system with complex dynamics. The possibility of using various input signals in product identification yields independent soft sensors. This serves to improve the reliability of the predictions, mutual sensor control and enables the detection of irregularities in the fermentation process before the broth becomes useless.