Optimization and online-monitoring in industrial batch processes using data-mining methods

In general industrial batch processes are complex systems that have to be optimized due to several performance criteria. As control optimization based on the development of physical models is in many cases very time consuming and cost intensive or not even feasible an alternative approach consists in analyzing historical process data by means of computational intelligent methods. The aim is the identification of characteristic control patterns and classification of these patterns according to their contribution to process optimization. This paper presents a concept how the generated knowledge can be used for online monitoring production runs. The key idea is to derive a model from the historical process data which describes the impact of the most important features of the process variables (both manipulated and controlled variables) to a user-defined performance index (e.g. quality or losses of a production run). The set of relevant features is automatically found using an iterative algorithm based on Support Vector Machines. Further the model is used to calculate the optimal values of the relevant features for several segments. Based on the optimal values of the features an online monitoring of the process can be implemented. The proposed concept is applied to an industrial glass forming process.