On the Hybrid Approaches for Innovative Process Analysis Technology

Abstract

This work contributes to the development of Hybrid Modelling for the purpose of Process Analysis Techniques within the Life Sciences. The hybrid modelling strategy is tailored to combine mechanistic and statistical process sensing into a hybrid abstract model. The model is developed in an abstract form such that it operates within an adequate process timeline. Within the life sciences, uncertainties are inevitable due to the complexity of the problems at hand, which cover multidisciplinary aspects of biology, chemistry and physics, thus a hybrid model is thought to provide better model diversity than a pure mechanistic or measurement- based model. From a mechanistic perspective, a model based upon different tempro-spatial levels is developed acquiring the required level of precision to model the main process characteristics. From a statistical perspective, different sensing data is analysed, correlated and calibrated with multivariate statistical algorithms to sense the main process characteristics. In a hybrid model, both types of models are combined into a high dimensional model representation (HDMR), which combines both complex paths (statistical and modelling) into an abstracted Sobol expansion. This model is trained by both measurements and mechanistic data as the input variables. The developed HDMR is thought to provide a robust alternative for process monitoring within the life sciences.