Control Chart Pattern Recognition Using Small Window Size for Identifying Bivariate Process Mean Shifts

Abstract

There are many traits in the manufacturing technology to assure the quality of products. One of the current practices aims for monitoring the in-process quality of small-lot production using Statistical Process Control (SPC), which requires small samples or small window sizes. In this study, the recognition performance of bivariate SPC pattern recognition scheme was investigated when dealing with small window sizes (less than 24). The framework of the scheme was constructed using an artificial neural network recognizer. The simulated SPC samples in different window sizes (8 ~ 24) and different change points (fixed and varies) were generated to study the recognition performance of the scheme based on mean square error (MSE) and classification accuracy (CA) measures. Two main findings have been suggested: (i) the scheme was superior when recognizing shift patterns with various change points compared to the shift patterns with fixed change point, with lower MSE and higher CA results, (ii) the scheme was more difficult to recognize smaller window size patterns with increasing MSE and decreasing CA trends, since these patterns provided insufficient information of unnatural variation. The outcome of this study would be helpful for industrial practitioners towards applying SPC for small-lot-production.