Multivariate Control Chart Pattern Recognition using Artificial Neural Network and Support Vector Machine

• Main Authors: Ping-Chen Chang, 張秉宸
• Other Authors: Chuen-Sheng Cheng
• Format: Others
• Language: zh-TW
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/33152038321375758999

碩士 === 元智大學 === 工業工程與管理學系 === 95 === In the past, univariate Shewhart control charts have been widely used to determine whether assignable causes of process variation are presented. Control chart pattern recognition is an important aspect in the application of control charts. The presence of non-random patterns indicates that a process is affected by assignable causes, and corrective actions should be taken. A particular non-random pattern is often associated with a set of assignable causes. Identification of non-random patterns can greatly narrow the set of possible causes that must be investigated, and thus the diagnostic search could be reduced in length. A recent research indicated that non-random patterns may also occur in multivariate control charts. Therefore, the pattern recognition of multivariate control chart is also an important research issue in multivariate process control. The purpose of this research was to investigate the feasibility of applying statistical learning algorithms for multivariate control chart pattern recognition. In this research, we considered two recognizers based on artificial neural network (ANN) and support vector machine (SVM). Furthermore, we used discriminant analysis as a baseline for comparison. The results showed that the performances of ANN and SVM were similar in classifying patterns of multivariate control charts. Both ANN and SVM can perform significantly better than discriminant analysis. In addition, two procedures were developed and compared in this research. The first procedure was used to recognize and classify both random data and non-random patterns. The second procedure was a two-stage approach. At the first stage, the ANN-based and SVM-based classifiers can be used to detect whether non-random patterns of process are presented. The second stage was to classify the types of non-random patterns. Results from our experiment showed that the second procedure performed better than the first procedure. Finally, this research investigated the effects of changing the parameters of ANN and SVM. The results exhibited that ANN-based and SVM-based classifiers are quite robust against changing of parameters.