Abstract

Mathematical models have been proven to be a key factor in optimizing production processes in recent years. However, in the case of biochemical processes the design is usually done using heuristics, since these systems show complex internal regulation mechanisms and strongly nonlinear behavior. This makes it difficult to find an appropriate model. In those cases, where a structured biochemical model has been successfully identified, the yield of the process can be increased significantly. Obtaining a suitable production model is usually a difficult and time consuming task, especially for biochemical systems. In this contribution the concept for an automation tool is presented which starts with the few noisy measurements of initial experiments to perform a model evolution from run to run. Thereby, the first unstructured model candidates are used for an optimal production-orientated process design whose realization will provide additional information about the dynamic behavior within the production area, thus, leading to new and improved model candidates. Due to the difficult measurement situation in biochemical processes many different model candidates may show a similar fit to the data why it is unwise to focus on one model candidate for process design, only. Furthermore, the use of more than one model candidate for the design procedure represents a kind of robustness for the planning. This cyclic procedure enables an optimal production design corresponding to the available measurement information at any time.