A review and analysis of fault detection and attribution metrics in MSPC literature

In multivariate statistical process monitoring, many methods have been developed for detecting faults and then attributing the cause of a fault to particular monitored variables. Consequently, a variety of metrics are used to assess the efficacy of these methods. In this paper, we summarize the most commonly used metrics for assessing the performance of such methods in fault detection and attribution, and we standardize the notation and language. We illustrate the features that each metric is designed to measure and highlight. Metrics are demonstrated in simulated data, and we discuss the benefits of establishing new metrics that (1) evaluate a method's ability to detect and attribute a fault simultaneously and (2) quantify the cost associated with different types of errors.