IMPROVED BAYESIAN METHOD FOR DIAGNOSING EQUIPMENT PARTIAL FAILURES IN-PROCESS PLANTS

A new methodology, called the Improved Bayesian (IB) method, for diagn osing equipment partial failures in process plants is described. The p artial failure of an equipment unit generally implies a partial loss o f its function(s). A partial failure can show different symptoms when it occurs, according to the level (i.e. failure strength) at which the function is lost, and this fact, in turn, makes the diagnosis even mo re difficult. Among diagnostic inference techniques, it has been shown that the Bayesian method is a theoretically superior method. The Baye sian method's construction is based on a rigorous probabilistic interp retation of data and expert judgment. The Bayesian method is adopted i n many diagnostic applications, and produces reasonable results in mos t cases. However, the assumption of independence of symptoms, normally employed in applying the Bayesian method to process malfunction diagn osis, can lead to erroneous conclusions. This problem can become worse when large numbers of partial failures are involved in the diagnosis; A subsidiary model called the F-curve model is developed in an effort to apply the Bayesian method more accurately in diagnosing partial fa ilures. The F-curve model utilizes the knowledge of the symptom variat ion with respect to failure strength, hence visualizing the degree of adverse influence (i.e. severity) of a failure on the process. When th e Bayesian method is modified by the F-curve model, it is referred to as the Improved Bayesian (IB) method. An application example is presen ted to verify that the proposed IB method can yield more accurate resu lts than the Bayesian method in diagnosing partial failures. (C) 1998 Elsevier Science Ltd. All rights reserved.