摘要
Fault isolation is an essential procedure in multivariate statistical process monitoring, which is used to locate the detected fault. Fault isolation identifies the crucial variables responsible for the detected fault. Accurate isolation results are useful for process engineers in diagnosing the root cause of the fault. Recent studies have revealed the equivalence between the fault isolation task and the variable selection problem in discriminant analysis. Inspired by this idea, a nonnegative garrote-based fault isolation strategy is developed to identify the criticality of each process variable to the detected fault, which is further revised to a more robust version by adopting a robust nonnegative garrote. The critical variables can be identified even when the historical process data are contaminated by outliers using the method proposed in this study. The Tennessee Eastman process was used to illustrate the validity of the proposed method.