Memory control chart based on machine learning technique for efficient process monitoring

Classical memory control charts (CCs), such as the Exponentially Weighted Moving Average (EWMA) CC, are effective in detecting small-to-moderate shifts in process parameters (location and/or dispersion). Nevertheless, the parameter (lambda, λ) in the classic EWMA CC optimizes its efficiency for particular shift sizes. The effectiveness of the standard EWMA CC with a fixed λ is typically compromised by shifts of different magnitudes that occur in real-world systems. To address this limitation, this study proposes a novel adaptive EWMA (AEWMA) CC to monitor shift in the process location parameter. To adjust to various shift of small sizes, the proposed CC makes use of dynamically varying parameter values based on current characteristics that have been learnt using machine learning (ML) techniques like support vector regression (SVR) and others. The SVR model is trained for each given λ over all possible values of characteristic, and then uses adaptive parameter to monitor process location more effectively. The SVR model is trained in MATLAB, and also algorithm is designed to calculate metrices which include average run length and standard deviation of run length to evaluate the performance of the proposed AEWMA CC against some other CCs. The proposed AEWMA CCs show superior performance against some other CCs. Additionally, a real-world process application is included to illustrate the practical applicability and broad utility of the proposed AEWMA CC.