Incorporation of process-specific structure in statistical process monitoring: A review

Abstract The incorporation of process-specific structure in monitoring activities has the potential to improve fault detection and fault diagnosis in modern industrial scenarios. By including causal information in the normal operation conditions (NOC) models, more effective use of the large amount of data can be made, leading to the detection of finer deviations from normal behavior. Furthermore, embedded in these methods is the fundamental cause effect information necessary to efficiently guide fault diagnosis and troubleshooting activities. This capability is absent from classical process monitoring methods based on acausal (or noncausal) NOC models, limiting to a large extent their fault diagnosis performance. In this context, a variety of approaches incorporating process-specific structure have emerged in the applied statistics, engineering, and machine learning communities and, in this article, we propose a classification and provide a systematic review of this developing trend in process monitoring. We also discuss the main features and limitations of each category of methods, propose a mapping for defining their application contexts and illustrate their use with an example.