Fault Diagnosis of a Batch Process by Use of Pattern-Recognition Technique

Abstract

In the fault diagnosis of a batch process which has a few measured variables, the time-varying information (trend) of the measured variables is often used. Patternrecognition technique is useful in finding the correspondence between the trend and origins of failures occurring in the process. In the present work, a method to diagnose the fault in a batch process is developed utilizing pattern-recognition. In this method, the origin of failure is detected by matching the pattern observed in the batch process with a pattern prepared beforehand (typical pattern) for each origin. Then, the typical pattern is improved to make diagnosis more accurate. Experiments of fault diagnosis in the batch process confirm the effectiveness of the proposed method.