Integrated Use of ICA and ANN to Recognize the Mixture Control Chart Patterns in a Process

Abstract

The quality of a product is important to the success of an enterprise. In process designs, statistical process control (SPC) charts provide a comprehensive and systematic approach to ensure that products meet or exceed customer expectations. The primary function of SPC charts is to identify the assignable causes when the process is out-of-control. The unusual control chart patterns (CCPs) are typically associated with specific assignable causes which affect the operation of a process. Consequently, the effective recognition of CCPs has become a very promising research area. Many studies have assumed that the observed process outputs which need to be recognized are basic or single types of abnormal patterns. However, in most practical applications, the observed process outputs could exhibit mixed patterns which combine two basic types of abnormal patterns in the process. This seriouslyraises the degree of difficulty in recognizing the basic types of abnormal patterns from a mixture of CCPs. In contrast to typical approaches which applied individually artificial neural network (ANN) or support vector machine (SVM) for the recognition tasks, this study proposes a two-step integrated approach to solve the recognition problem. The proposed approach includes the integration of independent component analysis (ICA) and ANN. The proposed ICA-ANN scheme initially applies ICA to the mixture patterns for generating independent components (ICs). The ICs then serve as the input variables of the ANN model to recognize the CCPs. In this study, different operating modes of the combination of CCPs are investigated and the results prove that the proposed approach could achieve superior recognition capability.