An Experimental Research On Fault Diagnosis Of Planetary Gearbox Based On Feature Dimension Reduction

Planetary gearboxes are important components in machine driven systems.Faults in planetary gearboxes may have important influence on the normal operation of equipments,and even cause more serious consequences.With the increasing requirements for safety,fault diagnosis of planetary gearboxes becomes extremely important.However,due to the complicated physical structures and severe working conditions,fault diagnosis for planetary gearboxes becomes very difficult.Taking the sun gear of the planetary gearboxes for consideration,a fault diagnosis experiment relying on spectra quests wind turbine drivetrain diagnostic simulator is designed in this thesis.Under the application of machine learning fault diagnosis method,for the problem that the increasing dimension of feature set leads to the decrease of classification accuracy,the fault diagnosis research of planetary gearboxes based on the feature dimension reduction technology is carried out.The main contents are as follows:(1)Under the premise of applying support vector machine(SVM)for fault classification,a hybrid dimension reduction algorithm based on feature selection and kernel principal component analysis(KPCA)is proposed to solve the problem that the high dimensional sample set leads to the decrease of classification accuracy rate.In order to reduce the complexity of calculation and the negative effects of some unnecessary features in the sample,a multi-criterion feature selection method is used to eliminate the irrelevant features.The low dimensional feature subset with high information density is built through KPCA.Then,fault is recognized by putting the feature subset into the SVM classification.The proposed method is applied to the planetary gearbox fault diagnosis experiment,and the experimental results show that the fault diagnosis method based on the hybrid dimension reduction algorithm improves the fault classification accuracy of planetary gearbox and the proposed dimension reduction algorithm outperforms the ones which employ feature selection or KPCA separately.(2)When using sparse representation-based classification(SRC)for fault classification,a fault diagnosis method based on improved kernel principal component analysis(IKPCA)and dictionary learning is presented to improve the real-time performance and accuracy rate of the fault diagnosis in the planetary gearboxes.Considering the condition that the number of dictionary atoms is much larger than the dimension of dictionary,IKPCA is used to reduce the dimensionality of data set.Some time domain and frequency domain features are combined into a feature vector to represent a sample,which can reduce the computational burden and enhance the real-time performance of fault classification.The feature set is transformed into a low dimensional feature set with high information density through IKPCA,which can improve the precision of fault classification.Then,the training samples are used to implement dictionary learning,and the testing samples are taken as the input of the SRC for classification.The planetary gearbox fault diagnosis experiments are presented to verify the effectiveness of the proposed methods.