On Approaches for Monitoring Categorical Event Series

Abstract

In many manufacturing applications, the monitoring of categorical event series is required, i. e., of processes, where the quality characteristics are measured on a qualitative scale. We survey three groups of approaches for this task. First, the categorical event series might be transformed into a count process (e. g., event counts, discrete waiting times). After having identified an appropriate model for this count process, diverse control charts are available for the monitoring of the generated counts. Second, control charts might be directly applied to the considered categorical event series, using different charts for nominal than for ordinal data. The latter distinction is also crucial for the respective possibilities of analyzing and modeling these data. Finally, also rule-based procedures from machine learning might be used for the monitoring of categorical event series, where the generated rules are used to predict the occurrence of critical events. Our comprehensive survey of methods and models for categorical event series is complemented by two real-data examples from manufacturing industry, about nominal types of defects and ordinal levels of quality.