A Support Vector Regression Based Control Procedure for Detecting a Range of Unknown Mean Shifts

Abstract

Cumulative sum (CUSUM) control scheme is an effective alternative to the Shewhart control chart to detect small process shifts. It is well known that the performance of CUSUM charts mainly depends on the pre-specified size of the shifts in the monitored quality characteristics. Since the shift size is usually unknown in advance, a control method capable of detecting a range of shifts is therefore required. The combination of several CUSUM charts (called CUSUM multi-chart) has been proposed to deal with this problem. This paper presents a support vector regression (SVR) based method to monitor the stability of process mean. The input vector of SVR comprises a mixture of charting statistics and statistical features that allow for the quick detection of a range of mean shifts. The performance of the proposed SVR was evaluated in terms of average run length (ARL). An extensive simulation study shows that the proposed method is better than the CUSUM multi-chart in detecting a range of mean shifts, especially for moderate to large shifts.