Model-based reasoning in diagnostic expert systems for chemical process plants

The class of system safety techniques which fall under the heading of fault diagnostic methods may vary in their approach, though they all attempt to achieve the same goal, that of locating the causal source(s) of a set of system abnormalities. This work examines the use of expert system technology in developing new methodologies for the automation of fault diagnosis in chemical processes. The expert system which results possesses the following four desirable characteristics: (1) The system reasons from device models and as such it has the potential to diagnose a variety of process configurations. (2) The system can diagnose a wide diversity of fault combinations since it need not assume that all observed process abnormalities are attributed to a single causal origin. (3) The system's reasoning is transparent, i.e., the system can justify its diagnostic strategy to a human operator. (4) The system is robust; it need not fail abruptly when faced with the diagnosis of a novel pattern of faults. Rather, the system can offer reasonable partial solutions.; Several expert system architectures are tested and sample results are presented for prototypical chemical processes. The most successful of these shows that diagnostic speed and reliability can both be achieved by using a knowledge base comprised of experiential knowledge (process-specific knowledge) and first-principles knowledge (process-general knowledge). A failure-driven learning strategy has been devised to give the expert system the capability to learn so that the system can automatically refine its knowledge base in order to improve its diagnostic skills. A scheme for incorporating troubleshooting knowledge into a diagnostic system to broaden the scope of the system is also discussed.