Support Vector Machine Fault Recognition Method And Its Application

The knowledge-based intelligent fault recognition method is one of the researches hot in the current nonlinear fault diagnosis field. The serious shortage of typical fault samples in present practical application as well as the discovery problem of diagnostic knowledge, which restrict the practical promotion of this technology. Because in fact fault diagnosis problem is one reflect of the small sample situation in the practical problem, its nature is a pattern recognition problem, Support Vector Machine attracts the attention of many researchers in the fault diagnosis field for its excellent performance to small samples situation. In the case of limited sample feature information, Support Vector Machines can explore as much as possible the classification knowledge. In terms of generalization performance, it is very suitable for the fault pattern recognition that is one of the actual engineering problems.Support Vector Machines was originally proposed for the classification of the two types of samples, but multi-class classification problems often exist in the practical applications, so this paper mainly studies multi-class classification method. The paper focuses on the Decision Directed Acyclic Graph (DDAG) multi-class classification method after studying "1-a-1" and "1-a-r" method. It aims at the problem which the root node of the traditional DDAG classifier tree is determined randomly, and puts forward the node optimization DDAG multi-classification method which is based on the distance between the class, and it is applied to steel plate fault pattern recognition. This improving method determines the root node of the DDAG classifier tree by calculating distance between classes, by combining with the ideology of optimizing the node. The main point of the paper is the key problem which needs to be solved in the application of fault intelligent diagnosis by using Support Vector Machine. In the follow ones, paper conducts a relative systematic and thorough research using steel plate fault datasets of the UCI database as experiment subjects in terms of the fault data preprocessing, the fault feature extraction based on Kernel Principal Component Analysis, the optimization of kernel function and it’s parameter, the establishing of the fault classification model and the implementation of the improving DDAG multi-class Support Vector Machine.The results of experiment show that feature extraction by using the kernel principal component analysis can reduce the data dimension and improve the performance of the classifier. The results also show the classification performance of the DDAG Support Vector Machines is much better than the classification performance of the Support Vector Machine structured by 1-a-1 and 1-a-r. In addition, the classification effect of the improvement DDAG method in this paper is better than that of the traditional DDAG method.