Abstract

The use of mathematical models for analysing complex processes is a powerful tool to gain a deep system understanding. However, this approach requires realistic, predictive mathematical models. During the model development phase, scientists have to cope with numerous challenges, e.g., limited knowledge about the underlying mechanisms, lack or exorbitance of dynamic or static experimental data, large experimental and process variability. Given a specific model class, a plethora of many different methodologies to optimally identify a specific model class structure have been developed since the mid of 20th century. This includes on the one hand methods for discrimination of competing structures but also methods for parameter estimation. We would like to discuss, whether further methodologies in the direction of model-based design are still needed, and if yes, to what extent. Further, given the trend of gathering massive data of a system of interest we highlight the analogy of DoE for systems identification and big data analysis. What does this mean for classic experimental design principles and where are we in need for further research?