Model Optimization For Support Vector Machine And Its Application To Intelligent Diagnosis Of Rotating Machinery

Support Vector Machine(SVM) gains wider attention and becomes one of the important technologies in the field of intelligent diagnosis of rotating machinery, because it has many unique advantages, such as its good performances in solving nonlinear, high dimensional pattern recognition and small sample problems. Feature selection and parameter selection are two important aspects of SVM optimization techniques. They guarantee the successful application of SVM in rotating machinery intelligent diagnosis. Feature selection is designed to preserve effective features and remove redundant features for improving efficiency and robustness of the model. Parameter selection is used to select the optimal parameter to improve the model’s generalization ability.This paper focuses on SVM model optimization technology and its application in intelligent diagnosis of rotating machinery. The main work includes the following three aspects:(1) For evaluating the characteristics of the nonlinear features efficiently and accurately, we propose effectiveness metric based on multi-dimensional statistics class separation, and a feature selection algorithm based on multi-metric fusion includes this metric. To avoid the limitation of single feature effectiveness metric, this algorithm evaluates both effectiveness and redundancy of the features for a comprehensive result. Moreover, it can identify non-linear characteristics. Experimental results show that the algorithm can accurately identify the sensitive fault features, reduce their redundancy and the model complexity for improving the classification accuracy of SVM.(2) For reducing the computational complexity and improving the convergence rate, we propose a parameter selection algorithm with good region recognition model. This algorithm is able to evaluate the irregular boundaries between good parameters and bad parameters for determining the performances of parameters without training SVM models. Experimental results show that the algorithm combined with grid search can accurately identify non-rectangular good region to reduce the parameter selection region and improve computational efficiency.(3) Select the rolling bearings as the research subjects, and design experiments for bearing inner race fault, outer race fault, rolling element fault and combination fault including all the three mentioned faults and collect data to extract features; train intelligent diagnosis model based on SVM; and apply the proposed feature selection and parameter selection algorithms to optimize the intelligent diagnostic model of rolling element bearings. Experimental results show that the proposed SVM optimization techniques can reduce the optimization time and improve the classification accuracy of the intelligent diagnostic model.