Failure Identification Procedure using Redundant Rules

Abstract

At the first stage of failure diagnosis, it is very important to confirm the correctness of input data, because the failure diagnosis using such a rule-based expert system identifies the cause of system failure based on them. The operator must check whether the input data are correct or not. This paper develops a simple method to identify an abnormal event using redundant rules, which help the operator understand the current system state. The proposed method requires only a relation table, which shows the values of input data under each abnormal event to be identified. Redundant rules, which are composed of minimal combinations of input data that can identify an abnormal event, can be obtained from the relation table using the concept of “failure distinguishability” criterion. Using redundant rules, the operator can estimate easily not only what state the system is under, but also which input datum may be wrong. The construction of redundant rules can also show which input datum is important and must be improved for its validation. Illustrative examples show the details and merit of the method in the operator’s aide.