Research On Fault Diagnosis Of Bearing Cage Based On Deep Learning Algorithm

The level of industrialization has a profound impact on the development of the economy,and has became an important indicator of the degree of development of a modern country.Bearings,as basic and key components,are indispensable in important fields such as numerical control equipment,mechanical equipment,aviation equipment research,ship manufacturing and rail transit design.Therefore,in the era of electromechanical big data,the research on the fault diagnosis algorithm of bearings and their components is of great significance to the successful transformation of our country from a large manufacturing country to a powerful country with intelligence.With the development of manufacturing equipment in the direction of large-scale,highspeed,integration and intelligence,the working environment of rolling bearings is becoming more and more severe,and the probability of failure is getting higher and higher.As an important part of the cage,once failure occurs,other parts will also fail,making it difficult to analyze the cause of the failure.In the process of bearing operation,as the early failure of the cage gradually expands,the cage will eventually break and fail.Once the cage is broken,it will often cause catastrophic accidents and great harm.Therefore,in order to maintain the safe and stable operation of the bearing equipment,it is of great significance to carry out the research of the cage fault diagnosis algorithm taking the cylindrical roller bearing as an example.Aiming at the three characteristics of current mechanical and electrical products bearing fault diagnosis,big data,diversified data types and close parts association,combined with advanced theoretical methods of feature extraction,in order to achieve efficient and accurate detection of cage faults.The research focuses on the diagnosis of cylindrical roller bearings with cage failures,combined with deep learning algorithms,to realize the intelligent diagnosis of cage failures.The specific research content of this paper is as follows:(1)Build a rolling bearing cage failure vibration signal acquisition system,and perform bearing failure processing under different cage failure states according to the type of bearing selection according to the cage failure forms in the actual production process,and then collect through the built experimental platform Complete the establishment of the fault database according to the vibration signal of the fault,and perform the corresponding time-frequency domain characteristic analysis on the fault data.(2)In view of the instability,non-impact characteristics and failure characteristics of the rolling bearing cage failure vibration signal,it is difficult to obtain the problem.The symmetrical point mode information fusion method is used to perform characteristic information fusion on the EMD inherent modal components of the cage fault vibration signal to show the time-frequency characteristics of different cage fault vibration signals,and then the SDP characteristic images of the rolling bearing cage under different fault states are studied.Differences,and then combined with a two-dimensional CNN model with strong adaptive feature learning capabilities for SDP image recognition,thereby designing a CNN bearing cage fault diagnosis method model based on the fusion of EMD and SDP features.(3)In order to reduce the difficulty of implementing the diagnosis model,combined with the characteristics of the vibration signal of the bearing cage fault,a one-dimensional convolutional neural network fault diagnosis model based on the "end-to-end" identification of the original vibration signal in the time domain is researched and proposed.The model first performs data enhancement through overlapping sample segmentation,and then uses the designed one-dimensional convolutional neural network to achieve adaptive feature extraction and dimensionality reduction of the vibration signal,and finally outputs the diagnosis result through the classifier.Experimental results show that the algorithm can achieve a fault recognition rate of more than 99%,and can effectively complete the task of fault classification.