HIERARCHICAL DECISION-PROCESS FOR FAULT ADMINISTRATION

Development of an expert system, called FAX, to address the following problems is reported: (1) determination of the root causes of process malfunctions; and (2) performance or suggestions for corrective action(s) to cope with the abnormal process situation. A hierarchical system decomposition approach called the goal tree-success tree model is used to represent deep-knowledge for fault diagnosis and process operations. The Bayesian theorem is also applied to the inference algorithms for induction of uncertain evidences. The inference algorithms contain two separate search routines, a depth-first search for fault diagnosis including uncertainty consideration, and an adaptive search for determining the most appropriate control action(s) which recover the process upset condition. The performance of a suggested recovery control action will be monitored for signs of improvement. If necessary, FAX resorts to other control actions.