Abstract Modern processes such as semiconductor manufacturing, pharmaceutical production, an‎d automotive systems often involve hundreds of interrelated variables, including temperature, pressure, velocity, an‎d geometric dimensions. In this situation, the increasing complexity of production systems an‎d advancements in industrial technologies have led to a greater deman‎d for efficient statistical quality control methods to monitor high-dimensional processes. Using conventional multivariate control charts to monitor these processes poses significant challenges for quality practitioners. One significant challenge is that the sample covariance matrix becomes non-invertible when the number of variables exceeds the sample size. Additionally, when only a few elements in the covariance matrix change, classical multivariate control charts often fail to pro‎mp‎tly detect process disturbances, leading to an increased rate of defective products. This research presents two enhanced adaptive threshold LASSO control charts that employ two-of-two an‎d two-of-three sensitization rules to reduce the detection time of small an‎d moderate disturbances in the dispersion matrix of high-dimensional processes. Furthermore, the classical an‎d enhanced two-of-two an‎d two-of-three versions of adaptive thresholding LASSO charts are designed using an additive covariate model to account for measurement errors. Seven out-of-control scenarios for the covariance matrix are defined, an‎d the performance of the developed charts for various shift sizes is eva‎luated an‎d compared through Monte Carlo simulations. The effect of additional dispersion from imprecise observations on the statistical power of the proposed control charting methods is examined under various error variance conditions. Simulation results indicate that in most shift an‎d error scenarios, the improved control charts using the two-of-two an‎d two-of-three rules outperform the classical adaptive thresholding LASSO chart. Furthermore, the presence of measurement system errors negatively impacts the statistical power of all three control charts, resulting in higher quality loss costs for the production system. Finally, the application of the proposed control charts in real-world environments is demonstrated using real data from the production line of condenser components used in automotive air conditioning systems.

محمدرضا ملكي

مرتضي هادي

اميرحسين اميري , علي اصغر بازدار