Application of Machine Learning in Statistical Process Control Charts: A Survey and Perspective

Abstract

Over the past decades, control charts, one of the essential tools in Statistical Process Control (SPC), have been widely implemented in manufacturing industries as an effective approach for Anomaly Detection (AD). Thanks to the development of technologies like the Internet of Things and Artificial Intelligence (AI), Smart Manufacturing (SM) has become an important concept for expressing the end goal of digitization in manufacturing. However, SM requires a more automatic procedure with capabilities to deal with huge data from the continuous and simultaneous process. Hence, traditional control charts of SPC now find difficulties in reality activities including designing, pattern recognition, and interpreting stages. Machine Learning (ML) algorithms have emerged as powerful analytic tools and great assistance that can be integrating to control charts of SPC to solve these issues. Therefore, the purpose of this chapter is first to presents a survey on the applications of ML techniques in the stages of designing, pattern recognition, and interpreting of control charts respectively in SPC especially in the context of SM for AD. Second, difficulties and challenges in these areas are discussed. Third, perspectives of ML techniquesbased control charts for AD in SM are proposed. Finally, a case study of an ML-based control chart for bearing failure AD is also provided in this chapter.

A Review on Statistical Process Control in Healthcare: Data-Driven Monitoring Schemes

“…Design and develop novel control charts, such as k-charts, derived from support vector data description (SVDD), and deviation charts, using kernel-based methods (e.g., SVM, NN, KDE, clustering algorithms, one-class classifiers, among others) for their formulation [67].…”

“…Recognition of abnormal patterns in the data that are hidden or unknown to the human eye expert using kernelbased methods, specially NN and SVM, among others [67].…”

“…Adequate modeling of highly autocorrelated datasets using kernel-based methods, NN, and SVM, among others, to decrease the false alarm rate of the control chart [67].…”

“…The ability to perform forecasts and build monitorable residuals of data with unknown distributions using NN or PLS [46], [54], [67]; or with special characteristics, such as autocorrelated data with NN and SVM [67], or survival and censored data with the Cox proportional hazards model [44], [68].…”

In-depth discussion on pitch doubling flow CpK improvement through process integration

“…Statistical Process Control (SPC) chart is a mature tool widely implemented in the industry over the decades to monitor the inline CD performance based on real-time data. Recently, SPC, integrated with machine learning (ML), facilitates the technology development like Internet of Things (IoT) and Artificial Intelligence (AI) [6]. Figure 2 shows a typical SPC chart with 5 main components: Upper Specification Limit (USL), Upper Control Limit (UCL), Central Line (CL or Target), Lower Control Limit (LCL) and Lower Specification Limit (LSL).…”