Research For Quality Abnormal Pattern Recognition In Dynamic Process Based On PCA

The most important thing in manufacture processing is process quality control, which is the effective measurement to ensure the quality of productions. Quality abnormal pattern recognition of dynamic process plays a very important role in monitoring both the manufacturer process running in its intended mode and the presence of abnormal patterns and realizing online quality diagnose of automatic production process. Since modern industries, such as petroleum, metallurgy, machinery and other industries, have become more large-scale, complex and continuous, the monitoring and diagnosis of the dynamic process has attracted many scholars’ attention. The research at home and abroad show that artificial intelligence technology has broken the limitation of traditional statistical control methods, and gradually becomes the new development direction of quality intelligent control and diagnosis in dynamic process. Because its effective for denoising and reducing dimension, principal component analysis combined with support vector machine method becomes the research hotspot in the field of quality recognition and diagnosis in dynamic process. There are two key problems of quality abnormal pattern recognition in dynamic process. One is how to extract the feature from dynamic data flow’s variation tendency with PCA effectively. The other is how to establish the suitable MSVM recognition model to monitor and diagnose the manufacturing process.Based on the analysis of the relevant literature, the principal component analysis methodology and support vector machine as the theoretical background, studying systematically the theory of quality abnormal pattern recognition in dynamic process based on PCA. Firstly, based on the research of pattern recognition in dynamic process, the basic principle of PCA method and SVM method are introduced and the six quality patterns of dynamic process are defined. Secondly, through the analysis of the characters of quality abnormal pattern in dynamic process, the framework and implementation process of quality abnormal pattern recognition in dynamic process based on PCA are presented. Thirdly, the principal feature vector of quality abnormal pattern of dynamic process is extracted with PCA method. After optimizing the parameters in the support vector machine classifiers with particle swarm optimization, the MSVM model of quality abnormal pattern recognition in dynamic process based on PCA is established. At last, the empirical analysis of quality abnormal pattern recognition model in dynamic process based on PCA is applied based on MATLAB software platform. In the simulation experiment, the dynamic dataflow is generated by Monte-Carlo simulation method. Simulation results show that the recognition model proposed in this paper has very high recognition accuracy for all patterns. Moreover, the overall average recognition accuracy is 97.36% and the best principal extraction level is under the variance contribution rate level of 45%.The characteristics and innovation of this paper has shown from three aspects: One is that the framework of quality abnormal pattern recognition in dynamic process based on PCA is presented; Two is that after extracting the principal feature vector from dynamic dataflow with PCA and optimizing the parameters in the recognition model with PSO, the MSVM model of quality abnormal pattern recognition in dynamic process is established. Three is that the quality abnormal pattern recognition model in dynamic process based on PCA is proved with empirical analysis, andthe simulation results show that the recognition model proposed in this paper has very high recognition accuracy for all patterns under the variance contribution rate level of 45%.Research of this paper not only puts forward a set of operational MSVM model and diagnosis methods of quality abnormal pattern recognition in dynamic process based on principal component analysis, but also provides the theoretical model and empirical analysis results for quality monitoring and diagnosis in dynamic process based on pattern recognition method.