State estimation for a penicillin fed-batch process combining particle filtering methods with online and time delayed offline measurements

Real time monitoring of physiological characteristics during a cultivation process is of great importance in the pharmaceutical industry. Measuring biomass, product, substrate and precursor concentrations continuously however is limited due to time-consuming laboratory analysis or expensive and hard-to-handle devices. In this work, a particle filter algorithm for estimating these difficult-to-measure process states in a Penicillium chrysogenum fed-batch cultivation is presented. The implemented particle filter represents a new algorithmic framework, combining several already existing methods and techniques for state estimation. It is based on nonlinear process and measurement models and takes into account both online measurements for state estimation and time delayed offline measurements, ensuring the observability of the considered system and being essential for the adaptation of dynamic model parameters. The application on real experimental data showed the convincing performance of the algorithm, estimating biomass, precursor and product concentration, as well as the specific growth rate, requiring standard measurements only. Furthermore, the positive effect of parameter estimation with respect to estimation quality was analyzed and the effect of the time delay was highlighted exemplarily. Despite of being computationally expensive due to time delayed data, the algorithm can be considered as an alternative monitoring strategy for industrial applications. Thus, it can be used further for process understanding and control.