Study On Feature Extraction Method Of Vibration Signal And Its Application In Fault Diagnosis

In order to more accurately extract fault features of rolling bearing and realize the intelligent fault diagnosis of rolling bearing,this thesis has carried on the new research of feature extraction method,combined with optimization of support vector machine(SVM)method to explore the effectiveness of intelligent fault diagnosis method based on the different characteristic parameter vector,and this thesis’ s main work includes the following four parts:1)Firstly,the background and significance of this research were given,and then the methods in the field of fault diagnosis of rotating machinery were summarized and discussed.Then,the research status and common feature extraction methods of rolling bearing were introduced,with emphasis on the application of scale theory in the field of fault diagnosis.2)Research on signal feature extraction.Analyzing the principle of extreme value increment sequence to highlight the original signal characteristics and theoretical derivation was given.And a method called super order analysis which introduces the extreme value increment sequence to DFA was put forward,and the method was proved by simulation and experimental data.3)The traditional SVM classification method was optimized in terms of data preprocessing and parameter selection,and the validation of the optimization was carried out by model accuracy which obtained by means of cross-validation.And two optimization methods were used to synthetically optimize the classification process of SVM,and it was found that the obtained parameter model from optimized SVM had better classification ability.4)Research on the fault diagnosis method based on optimized SVM and feature combination.The dimensionless parameters of vibration signal in time domain were selected as the normal characteristic parameters.Then by combining the scale-law index obtained by the super order analysis method with the conventional characteristic parameters,and a hybrid feature vector that can better represent the fault characteristics of signals was constructed.Also by combining with the optimized SVM method,a fault diagnosis method of support vector machine rolling bearing based on the hybrid feature optimization is proposed.The experimental data sets of the two methods were verified,and it was found that the classification effect of the hybrid feature vectors was greatly improved.