Research On Intelligent Fault Diagnosis Method Of Rolling Bearing Based On Machine Learning

Rolling bearings are widely used in mechanical equipment due to their low frictional resistance,good interchangeability and high flexibility.It is very important for the safety of mechanical equipment.In order to avoid or reduce the influence of rolling bearing on the operating state of mechanical equipment as much as possible,In view of the shortcomings of traditional fault diagnosis methods,In this paper,the relevant machine learning theory were deeply studied in the intelligent fault diagnosis of rolling bearings.The details are as follows:(1)Typical structures and common failure modes of rolling bearings were introduced,The vibration mechanism was studied,The characteristic frequency was derived through dynamic analysis,And the time and frequency domain related features of vibration signal was analyzed.(2)In terms of over-fitting and local optimization caused by traditional neural networks in the diagnosis of rolling bearing faults,a novel method of optimizing BP neural network by mind evolutionary algorithm was proposed.The time domain and frequency domain characteristic parameters of the commonly used vibration signals were used as input of the BP neural network model optimized by the thought evolution algorithm.Then apply it to determine the type and the degree of failure for the bearing.Finally,the results of algorithm the optimized and the unoptimized were compared and analyzed.(3)With the single algorithm structure,the accuracy of multi-classification results of rolling bearing fault diagnosis is relatively low,a novel method by feature-reducing random forests was proposed.The time domain and frequency domain characteristic parameters of common vibration signals were input into the feature reduction-dimensional random forest method.Then apply it to determine the type and the degree of failure for the bearing.Finally,Compared with methods such as SVM and PCA-SVM,A comparative study was analyzed in terms of accuracy and timeliness.(4)In terms of the validity selection of features in traditional feature extraction,a convolutional neural network approach has been proposed.A deep convolutional neural network intelligent diagnosis model was designed and trained by original vibration data of rolling bearings.Then apply it to determine the type and the degree of failure for the bearing.The influence of parameters such as learning rate,number of iterations,proportion of training samples,and Batchsize on the model was explored.Finally,the intelligent diagnostic experiment by selecting the optimal parameters in the model of rolling bearings was carried out.The validity of the proposed method were to be verified,On the rolling bearing fault diagnosis test bench,The signal acquisition system for rolling bearing intelligent diagnostic vibration based on LabVIEW was designed.And the types and extents of various faults for rail vehicles under various operating conditions were simulated.At last,By conducting experiments on the test bench,The effects of different models on the diagnosis results were studied.And the validity of the proposed method was verified.more importantly,It can provide a reference for the prediction and fault diagnosis of rolling bearing condition.