Problem Statement
Monitoring of biomass and product concentrations during a fermentation helps to ensure product quality as proposed by PAT guidance. As these concentrations are often difficult to be measured in real time directly soft sensors are used basing upon process models and secondary measurements. In this context Bayesian filtering algorithms play an important role such as the well-known Kalman or Extended Kalman filter. However, these algorithms may fail due to the nonlinear nature of bioprocesses.
Aim of the Project:
In this project the potential of Particle filter algorithms for state estimation in bioprocesses is revealed. These algorithms do not pose any requirements to the model or the noise distributions being however computaionally more expensive. Particle filter and EKF algorithms are compared theoretically and empirically with respect to estimation quality and time complexity. Simulations are carried out in in quasi real time mode for estimation of biomass concentration in a fed-batch P. chrysogenum process using real experimental data. The underlying mechanistic model takes into account the heterogenous structure of filamentuous fungi.
Contact
Univ.Prof. Dipl.-Ing. Dr.techn. Christoph Herwig
Dipl.-Ing. Dr.techn. Ines Stelzer