Analysis of variations in a multi-variate process using neural networks

In the area of quality control, several correlative quality characteristics may be considered simultaneously in a manufacturing process. In such a case, the Schewhart control charts neglect the influence of the correlation among these quality characteristics, leading to an incorrect judgment. Numerous methods have been proposed to control a process with multiple quality characteristics, however, most research emphasises only the variation of a process mean and devotes much less attention to the variation of a variance, relatively speaking. In this study, a neural network procedure is proposed for detecting variations in the variances, with the assumption that the mean value of the multiple quality characteristics of a process is under control. Two criteria, the average run length (ARL) and the average number of abnormal cases found, are used to evaluate the performance of the proposed procedure. The developed neural model is compared with the traditional |Σ| control method with regard to the aspect of detecting the variations in the variances in a process. The simulation results demonstrate the superiority of the proposed procedure in process control while multiple quality characteristics are simultaneously considered.