Machine learning based parameter-free adaptive EWMA control chart to monitor process dispersion

Area of Science:

• Industrial Engineering
• Statistical Process Control
• Machine Learning

Background:

• Traditional control charts use fixed parameters, limiting adaptability during online monitoring.
• Adaptive control charts dynamically adjust parameters for enhanced process control.
• Effective monitoring of process dispersion is crucial across diverse operational environments.

Purpose of the Study:

• To develop and evaluate an adaptive exponentially weighted moving average (AEWMA) control chart.
• To integrate support vector regression (SVR) for adaptive parameter adjustment in control charting.
• To improve the sensitivity and reliability of process dispersion monitoring.

Main Methods:

• Implementation of an adaptive exponentially weighted moving average (AEWMA) control chart.
• Utilizing support vector regression (SVR) with linear, polynomial, and radial basis function (RBF) kernels.
• Adapting the smoothing constant based on detected shifts in process dispersion.

Main Results:

• The proposed SVR-based AEWMA control chart demonstrates enhanced performance in detecting process dispersion shifts.
• The RBF kernel within the SVR framework proved particularly effective for adaptive monitoring.
• Validation using real-life data confirmed the method's adaptability and reliability.

Conclusions:

• The SVR-based AEWMA control chart offers a robust and adaptive solution for process dispersion monitoring.
• This approach improves upon conventional methods by dynamically adjusting control parameters.
• The study highlights the potential of machine learning integration in statistical process control.