A Fault Diagnosis Approach For Rotor Systems Based On Support Vector Machine

With the rapid development of the relevant theoretical research,people payed more and more attention on the researches about condition monitoring and fault diagnosis.The superior performance of support vector machine to small samples attracted extensive attention.It searches the optimal solution between the complexity and learning ability of model,so it can achieve best outreach capacity and solve the overfitting problem effectively.Classifier based on SVM can be provided with good outreach and high accuracy rate even with small sample.Combined with the common faults simulated on the rotor experiment table,the paper used entropy band feature extraction of the fault vibration signal. In order to make the support vector machines have a higher classification accuracy,using particle swarm optimization to optimize the parameters of SVM. Aim at multi-category classification of the fault and accord the experimental analysis and theoretical algorithm, main contents and conclusions of the study were as follows:1) Four types of typical fault was simulated on the rotor experiment table,whose mechanism was analyzed.And carried on a filter for fault signal,spectralanalysis,axis path analysis. We can get the singular spectrum entropy value, power spectrum entropy value, wavelet energy spectrum entropy value and wavelet space feature entropy value after computed the entropy value band of fault signal in the time domain, frequency domain and time-frequency domain. And discuss a rotor system fault identification method is established based on entropy band .2) Because directly takes the entropy band as the SVM's training sample exist data redundancy question,therefore preprocessed the entropy band data as SVM's training sample, includes sample normalization and Principal component analysis. Subsequent experiments show that the entropy processed with data can not only reflect the vibration signal characteristics,but also suitable for SVM model training and fault classification.3) In order to construct a best clsssifier,systematically studied the PSO algorithm and GA algorithm influencing the classifier's accuracy after optimizing the parameters of SVM.Put the handled data into SVM and optimized the nuclear parameters and punish factor parameters of SVM by PSO and GA respectively.We found that SVM which optimized by PSO had the better classify ability than by GA.And time of training is shorter.4) As to the reseach is aiming at multi-category classification problem,and SVM is a two classifier,so designed a classifier that can separate four fault based on one against all method.Using PSO algorithm optimization to each classifier's parameters respectively.And developed a rotor fault diagnosis system based on the above procedures and used the MATLAB GUI to realized.One subsystem can analysis the vibration signal,including spectralanalysis,axis path analysis;another subsystem can optimize the parameter of SVM and distinct the unknown data sample.The experimental results verify the effectiveness of the algorithm and system.