Integrating machine learning based EWMA control charts for multivariate process monitoring

Memory-type control charts, such as the multivariate EWMA (MEWMA), are recognized for their effectiveness in detecting small to moderate changes in the process mean vector. In comparison, Adaptive Multivariate EWMA (AMEWMA) control charts offer superior capabilities over traditional methods. This study proposes an adaptive multivariate EWMA (SAMEWMA) control chart based on Machine Learning (ML) techniques such as Random Forest (RF), K-NN, and Support Vector Regression (SVR) to monitor specifically small shifts in the process mean vector. The results show that the proposed chart demonstrates exceptional performance in detecting small shifts in the mean vector compared to various existing charts, such as AMEWMA-I and AMEWMA-II. Investigations also prove the superiority of the SVR method among other ML approaches. The Average Run Length (ARL) matric is employed to determine the performance of the control charts using the Monte Carlo (MC) simulation technique. Two real-world examples are presented to demonstrate the effectiveness and superiority of the proposed control chart in detecting variations in the process mean vector.