Fault Diagnosis Of Rolling Bearing Based On Feature Combination Optimization And Convolution Neural Network

With the development of automation level of mechanical equipment,the complexity of equipment is higher and higher,and the requirement of intelligent fault diagnosis technology of mechanical equipment is increasingly demanding.As one of the most widely used parts in mechanical equipment,rolling bearing is also one of the most prone parts to be faulted.Therefore,it is very important to carry out intelligent fault diagnosis for rolling bearing.The extraction and selection of fault features greatly affect the accuracy of the diagnosis results.If an optimal feature combination can be constructed from many feature extraction methods such as time domain,frequency domain and time-frequency domain,it can not only improve the accuracy of the diagnosis results,but also greatly improve the operation efficiency of the diagnosis model.Based on this,the fault diagnosis method of rolling bearing based on feature combination optimization and convolution neural network is proposed in this paper.The research results are as follows:(1)For the non-linear and non-stationary characteristics of the original vibration signal of rolling bearing,the vibration data collected by direct observation can not be used for fault analysis,and the single domain feature extraction information is limited,which makes it difficult to accurately characterize the running status of the bearing.The hybrid feature combination of time domain and frequency domain is established,which can fully characterize the running state of bearings,and the feature matrix is extracted from the vibration signals of bearings in different operating states.The feature selection problem of rolling bearing fault data is studied in this paper.The searching strategy of feature subset and the selection basis of feature evaluation criteria are mainly analyzed.On this basis,an optimal feature combination filtering model is established.(2)In view of the feature extraction of the original vibration signal by using multi domain feature combination,the feature selection method based on random forest and maximum correlation minimum redundancy is proposed based on the optimal feature combination selection model.Firstly,the importance of each kind of feature in the classification process is calculated and sorted according to the random forest algorithm.Some features are eliminated by combining the classification error rate criterion and the sequence forward search.Then,the maximum correlation minimum redundancy criterion is used to calculate the redundancy between features and the correlation between features and classification variables.The optimal feature combination is obtained by combining distance criterion and sequence forward search.The simulation results show that the feature set of rolling bearing obtained by this method not only reduces the data dimension of original signal,but also has better class aggregation and class dispersion compared with other common feature combinations.(3)In the traditional fault diagnosis method of rolling bearing,the method based on single domain feature can not fully represent fault information and easily produce misjudgment.The end-to-end method based on deep learning has the characteristics of many training times and complex network structure,and can not effectively utilize the correlation between data in the feature extraction process,and there is a low degree of feature division.In view of the above problems,a fault diagnosis method of rolling bearing based on feature combination optimization and convolution neural network is proposed.Firstly,the original feature combination is optimized by the feature selection method based on random forest and the maximum correlation minimum redundancy.The optimal feature combination suitable for the rolling bearing to be diagnosed is obtained,the corresponding fault label is established,and then one-dimensional convolutional neural network is constructed.The neural network model is trained and the classification precision is tested by the feature matrix extracted by the optimal feature combination Degree.The common methods include the method of combining the time domain statistical features with k-nearest-neighbor classifier,the method of using the original signal directly with convolutional neural network,and the fault diagnosis method based on time-frequency feature extraction and cyclic neural network.The simulation results show that the method and other common methods are effective in fault diagnosis of rolling bearing under composite condition.The feasibility and effectiveness of the method are verified.The function of the optimized multi domain hybrid feature combination in the process of fault diagnosis is summarized.