Real-time monitoring of multi-mode industrial processes using feature-extraction tools

A real-time monitoring solution is presented for detection of faulty status in industrial processes which uses the input-output data features to set monitoring boundaries. The proposed solution uses independent components analysis (ICA) in order to extract features of data samples measured from the actual plant. The features extracted from the simulated model of the plant at normal operating condition will be used to compute values of the monitoring criteria. The detection threshold and monitoring boundaries are then determined based on the monitoring criteria obtained for normal condition. Measured data of the plant are fed into the monitoring system in order for the plant's status to be detected for each and every sample time. The suggested approach will be implemented on a model of Continuous Stirred-Tank Reactor (CSTR) with two modes of normal operation for monitoring and fault detection. Simulation results are presented and performance of the proposed approach is discussed.