Application of model-based redesign of experiments for reactor characterization

Abstract

In Biotechnology, correlations between critical quality attributes (CQAs) and critical process parameters (CPPs) are essential in view of the quality by design (QbD) approach. Models provide the opportunity to save and display such correlations in a clear and unambiguous way. Thereby, the development of models and model-based methods is in focus of research. Experiments are an essential part of the model building workflow. Thus, design of experiments is critical for an effective model development and the time to market of model-based methods. Due to their nonlinearity and complexity, statistical DoE is not optimal for most correlations of CQAs and CPPs. An alternative is model-based redesign of experiments (MBreDoE). MBreDoE uses the model equations to predict and optimize the information content of an experiment. Further, it utilizes the measured data immediately for redesigning the further experiment instead of executing an experiment in the usual schedule (Design  Execution  Evaluation). The hypothesis tested in this thesis is, that MBreDoE can be used to reduce the effort of reactor characterization. Therefore, an algorithm performing MBreDoE was developed. Particular attention was devoted to a generic structure of the algorithm to enable a universal applicability. The algorithm uses the Fisher information matrix and optimization methods to maximise the information content of an experiment. Due to its importance in bioprocesses, the oxygen transfer rate (OTR) is the reactor characteristic tested in this thesis. To develop and test the algorithm, a simulation study was successfully executed. Minimalized calculation effort and robust data transfer were detected during the simulation study to be the key attributes of an algorithm performing MBreDoE. The simulations showed that MBreDoE enables a faster reactor characterization compared to statistical DoE. The improved algorithm was tested on a microbial and a cell culture reactor setup. The algorithm proved to be robust and fast enough for real-time application. Further, the results of these experiments confirm the hypothesis that MBreDoE can be used to reduce the effort of reactor characterization. MBreDoE reduces the duration and effort of reactor characterization, by designing more informative experiments and better utilization of prior knowledge. Further, it enables a high degree of automation and reduces experimental costs.