Machine Learning Approaches for Statistical Process Control

Abstract

Machine learning approaches are widely used in different applications, including the ones involving statistical process control (SPC). In a traditional SPC problem (e.g., online monitoring of a production line), a small set of in‐control (IC) process observations is routinely collected before online process monitoring for estimating certain parameters of the IC process distribution. This dataset, however, does not contain any out‐of‐control (OC) process observations. Thus, supervised machine learning methods would be inappropriate to use in such cases since they require a training dataset that contains both IC and OC process observations. To overcome this difficulty, some machine learning methodologies specific for SPC have been developed based on one‐class classifications, artificial contrast, real‐time contrast, transparent sequential learning, and more. This article provides an overview on these methods.