Pattern Recognition, Neural Network-based Control Charts

Day by day, the intense market competition and the complex production process, promoted the enterprise to bring advanced technologies as information technology, computer technology, intelligent control and so on into manufacture quality control system, which laid foundation of realizing modem quality control and quality guarantee for them. The competitive competence of an enterprise depends on the products quality, which is produced in products process. Thus it is necessary to manage products quality during the whole production process. Statistical Process Control (SPC) can manages products quality effectively, and increase the quality management competence of the enterprises. With the development of Computer integrated manufacturing environment and the advanced in data collection system, the automation of SPC implementation is considered unavoidable. Control charts are the key factor of SPC.This paper studies control charts recognition, which is part of the SPC automation system, including three tasks, monitoring the process, diagnosing the deviated process and taking corrective action if is necessary. The paper mainly includes four parts as following: Back prorogation (BP) neural network and its defects are studied, and three kinds of improved BP algorithm are discussed, and find a way to build a multiple neural network model to solve the problem, the results of the simulation indicate this method has high recognition accuracy and fast speed. The proposed models can recognize the six basic patterns of control charts, namely normal pattern, upward shift pattern, downward shift pattern, upward trend pattern, downward trend pattern and cycle pattern. The result of the simulation indicate the proposed model needs less time in training, has powerful recognition capability, and has good fault tolerance capability.An effective method is proposed to estimation the pattern parameters, which are useful in diagnosing the deviated process, and a spectrum method is also proposed. Compared with other methods it has great advantage. The proposed models in this paper can identify shift magnitude, trend slope, and cycle amplitude and cycle length. In order to avoid disturbance, three respective networks are used to estimate the pattern parameters.Control charts pattern recognition system is designed with MATLAB, VC, and VB. By the design of system interface, the user can adjust the structure and the parameters of the networks, and can choose different algorithm. The sample data also can be recomposed. For different sample data, the network can be retrained so as to achieve the best result. The system is easy to use and has a good adaptability, and has a good foundation to be used in on-line monitoring. This paper proposes an artificial neural network to recognize the abnormal control chart patterns and estimate the parameters of abnormal patterns, which make a good foundation for the realization of SPC automation system.