Bioprocess Monitoring: A Moving Horizon Estimation Experimental Application

Abstract

Microbial fermentation processes are most often described by nonlinear time-varying dynamics, which require the implementation of nonlinear state estimators to infer unmeasured metabolites in the cultivation broth. Among the various nonlinear available estimator strategies, the Moving Horizon Estimator (MHE) is an on-line optimization approach that easily allows to enforce hard constraints, an important feature that helps to avoid unfeasible concentrations. In this work we implemented an MHE by using experimental data from a fed-batch cultivation process of Corynebacterium glutamicum. Available real-time measurements of biomass and CO2 formation were used to infer sugar concentrations by combining the available measurements with a simple Monod model. We found that the MHE was able to estimate all the three variables of interest, including the unmeasured sugar concentrations, during the entire fed-batch cultivation process. Moreover, we show that the estimates are accurate in comparison to the reference offline samples. This work demonstrates the benefits of MHE as a soft sensor that can monitor bioprocesses in real-time.