CHAIR OF

THERMAL PROCESS ENGINEERING

Model predictive control of industrial processes is a field of active current research which implies that so as to optimize the operation of a process one needs to predict its state and future behavior by the analysis of measurement data. The prediction can be based on known physical and chemical relations. In addition to such knowledge-based modelling, machine learning algorithms that use artificial neural network analysis are increasingly used for the prediction of complicated processes, especially when sufficiently accurate physical models become too complex or are not available. A combination of both strategies, physical models and machine learning, can also be a promising way to enhance prediction.
The aim is to apply modern concepts and techniques of data analysis based on artificial neural networks, on the fluidized bed gasification plant at the Fraunhofer IFF, to analyze the correlation between various sensor data. The machine learning techniques require data from many sensors and over a longer period of operation. In the given case the database is not sufficient, CFD simulation models shall be used to provide artificial data to teach the machine learning algorithms.