Research On Multivariate Quality Diagnosis Model Based On Fuzzy Support Vector Machine For Manufacturing Process

In this condition that current product and manufacturing process are increasingly complicated, due to solving the problem of monitoring and diagnosing multivariate quality characteristics, a multivariate quality diagnosis method based on fuzzy support vector machine (FSVM) with complex manufacturing process oriented is proposed. This method mainly divides into two phases. In the first stage, multiple weighted moving average control chart with variable sampling interval (VSI-MEWMA) is used to identify process exception. If the control chart sends alarm signal, then the second stage begins. In this stage, FSVM method is used to identify the pattern classification of the process data stream that makes alarm signal, and eventually the process anomalous source is diagnosed out.First of all, the related theories about the VSI-MEWMA control chart were introduced. The one-dimensional and the two-dimensional Markov chain models of VSI-MEWMA were established, and two indicators of ATSo and SATS to measure control chart performance under the controlled and the uncontrolled states were given respectively. Furthermore, combined with the traditional MEWMA control chart performance in the test, the validity of VSI-MEWMA was verified.Then, the relevant theories of FSVM and its membership functions were summarized. The method of determining the membership function used in this paper was given. In order to make FSVM classifier have a good performance, K-fold cross validation method (K-CV) and particle swarm optimization (PSO) were used to optimize parameters of FSVM classifier. Then the simulation experiments were implemented by using the database in UCI data. Finally, classification accuracy of parameters optimization results were compared with that of FSVM classifier which is without parameters optimization.Finally, combined VSI-MEWMA control chart and FSVM classifier, multivariate quality diagnosis model based on FSVM for manufacturing process was built. In first stage, VSI-MEWMA is used for process monitoring to recognize whether a process is abnormal or not, and then FSVM classifier is used to identify the classification of the process data stream which causes abnormal process in the second stage. Then compared with the ANN method in the training test, the diagnosis effect of the model was indicated. Finally, with the shaft hole processing of the engine cylinder body taken as an example, the effectiveness and the practicability of this proposed method were verified.