Hybrid Abnormal Patterns Recognition of Control Chart Using Support Vector Machining

Abstract

A novel control chart pattern recognition system using support vector machine(SVM) is presented. Pattern recognition techniques have been wildly applied to identify abnormal patterns in control charts. Abnormal patterns exhibited by such charts can be associated with certain assignable causes affecting the process. Most of the existing recognition method are capable of recognizing a single abnormal pattern, however, a practical situation is concurrent patterns where two abnormal patterns may exist together. The presented method can enhance recognition capability and accuracy, and avoid the disadvantages, such us over-fitting, weak normalization capability, etc., of artificial neural network(ANN) method. Furthermore, it can recognize these hybrid abnormal patterns existing in control chart by combining voting and binary tree methods. Simulation experimental results are given to demonstrate that, compared with ANN recognition methods, the method proposed is superior in classifying shift, trend and cyclic patterns, and realized the recognition for hybrid abnormal pattern in control charts.