Dynamic Process Monitoring Using Machine Learning Control Charts

Abstract

Machine learning methods have been widely used in different applications, including process control and monitoring. For handling statistical process control (SPC) problems, the existing machine learning approaches have some limitations. For instance, most of them are designed for cases in which in-control (IC) process observations at different time points are assumed to be independent and identically distributed. In practice, however, serial correlation almost always exists in the observed sequential data, and the longitudinal pattern of the process to monitor could be dynamic in the sense that its IC distribution would change over time (e.g., seasonality). It has been well demonstrated in the literature that control charts could be unreliable to use when their model assumptions are invalid. In this chapter, we modified some representative existing machine learning control charts using nonparametric longitudinal modeling and sequential data decorrelation algorithms. The modified machine learning control charts can well accommodate time-varying IC process distribution and serial data correlation. Numerical studies show that their performance are improved substantially for monitoring different dynamic processes.