Automatic Acquisition of Diagnostic Knowledge for Process Monitoring and Diagnosis

The task of programming software for the fault detection and diagnosis of industrial processes is considered to be an art; mainly because it is highly dependent on the specific process at hand. Therefore, many critical and high performance automated processes are still supervised by human; as the consequences of their unexpected failure can be dramatic to both man and machines. We propose the pattern theory as a systematic paradigm for the automatic programming and extraction of diagnostic logic from process data. The proposed approach combined elements of the model-based process analysis techniques with the learning aspect of the pattern theoretic paradigm in order to extract diagnostic knowledge from fault scenarios and their symptoms. The performance of the pattern-finding algorithms in solving three benchmark problems is evaluated in the Phase I. Our team's experience in integrating computational learning schemes to industrial applications and our background in, and contributions to, the state-of-the-art in process control and identification enhance our chance of resolving many of the multidisciplinary issues of the proposed approach.