Research On Fault Diagnosis Method Of Motor Bearing Based On Vibration Signal

Motor bearings play an important role in industrial production,their operational reliability is directly related to the safe production and economic benefits of enterprises,fault diagnosis can provide a reliable guarantee for the normal operation of the motor.In this paper,the vibration signal of motor bearing was taken as the research object,the fault feature extraction and fault feature pattern recognition of motor bearing were studied respectively,and two fault diagnosis methods of motor bearing were proposed.Finally,an embedded platform was built as the hardware implementation of the proposed method.The main research contents are as follows:This paper first introduces the structure of motor bearing,fault evolution process and their common failure forms.Then the general methods of motor bearing fault diagnosis and the current research status at home and abroad were introduced.Furthermore,the fault feature extraction method and feature pattern recognition were explained in detail.In the process of extracting fault features of motor bearing: aiming to solve the mode aliasing problem of EMD method,EEMD and correlation coefficient method were used to extract fault features and construct feature vectors.To solve the problem that there are advantages,disadvantages and high randomness of the time domain parameter,the multi-parameter combination method was used to construct the time domain eigenvector.In the fault feature pattern recognition: aiming to solve the problem that when the number of continuous characteristic attribute values is too large,the decision tree algorithm in random forest has high complexity and becomes easy to overfit,an improved C4.5,CART decision tree algorithm was presented.A new integrated voting method was proposed to tackle the barrier that the simple voting method ignores differences between strong and weak classifiers,and it did not take the miss report rate into account.Then,an improved random forest algorithm was proposed.The general data set was used to verify the algorithm function of EEMD-improved random forest algorithm and time-domain feature-improved random forest algorithm,moreover,the traditional random forest algorithm and MLP multi-layer perceptron algorithm were used as pattern recognition comparison algorithms.The comparison results show that the improved random forest algorithm proposed in this paper can be used in a variety of feature extraction methods and has higher diagnostic accuracy and lower miss report rate than the comparison algorithm.An embedded motor bearing fault diagnosis system with Raspberry Pie 3B + as the main control node was established,and the system was used to diagnose motor bearing on-site.Therefore,the EEMD-improved random forest algorithm and the time-domain feature-improved random forest algorithm was verified for the second time.The verification results show that the improved random forest algorithm has a higher diagnostic accuracy and a lower false alarm rate than the other two algorithms when the number of samples is small,and as the number of collected samples increases,the improved random forest algorithm achieves higher accuracy and lower miss report rate earlier than the comparison algorithm.