This grant provides funding to develop, test, and evaluate a framework for process monitoring, diagnosis, and control. The unique feature of this framework is the integration of diagnosis, with monitoring and control. In existing process monitoring and control systems process compensatory adjustments are made without regard to the cause of process disturbances, or diagnosis is attempted without any guidance. This is not efficient. The multiresolution analysis capability of wavelet theory will be used as a "mathematical microscope" to glean process fingerprints, which are then mapped to a diagnosis. The mapping is done by extracting statistical features from wavelet transforms and utilizing neural networks. Diagnostic information obtained from the fingerprints will be used by a control decision module to make the "optimal" control decision. This framework will be tested and evaluated in the laboratory and in actual industrial settings. With manufacturing becoming more automated, many online sensors for process monitoring and control have been developed, however, the methods used for monitoring and control have not adapted to this, nor has the information available been used effectively. The results of this research, which uses both coarse and detailed views of processes, can be used for effective and efficient monitoring, diagnosis and control. This would reduce process variability, and improve process capability, resulting in higher quality and efficient production. This framework is applicable to other areas such as, equipment maintenance, medicine and finance, where the integration of monitoring and diagnosis offers considerable advantages.
