Research On Image Classification Diagnosis Method For Motor Bearing Faults

Nowadays,motors are not only an indispensable core driving device in the industrial field,but also a core device of various equipments in various fields of society.Due to the complexity of the mechanical and electrical components of the motor,failures and even damages often occur during its lifetime.Rolling bearings are one of the most critical components of motors.Their health actually has a great impact on the performance,efficiency,and stability of motors.Bearing failures are the most common cause of motor failures.Therefore,fault diagnosis of bearings is an important topic.This paper introduces a cutting-edge deep learning method and proposes an image-based classification and diagnosis method for motor bearing faults.The fault data used for research in this paper comes from the Rolling Bearing Data Center of Case Western Reserve University.Each piece of raw vibration data in the public data set of Case Western Reserve University contains only one independent bearing fault.But the actual situation is that the possibility of multiple faults occurring at the same time is very high.Therefore,this paper linearly superimposes the vibration signals of multiple different fault states under the same conditions in the public data of Case Western Reserve University to simulate the simultaneous occurrence of multiple faults.Condition.First of all,in order to solve the problem that the fault data with low vibration amplitude is easily submerged in the fault data with high intensity during the linear superposition process of the original data,the original vibration signal is smoothed and denoised.Then,for the problem that the one-dimensional fault signal fluctuation is not obvious and is not easy to find,the original one-dimensional vibration data of the motor bearing fault is converted into a two-dimensional time-frequency image by continuous wavelet transform,and the two-dimensional motor fault image can be based on its own characteristics.It can reflect the energy intensity of the fault vibration signal at different times and frequencies,and can display the detailed changes of the signal from multiple angles such as the frequency and amplitude of the fluctuation,so as to distinguish different faults from the details.This paper uses classification neural network and YOLO(You Only Look Once)target detection neural network method to realize bearing fault diagnosis.At the same time,the method of transfer learning is introduced to pre-train the model to improve the efficiency of training.The traditional classification neural network has high accuracy in motor bearing single fault diagnosis,but it does not perform well in concurrent fault diagnosis.Because in reality a variety of motor bearing faults may occur at the same time,classification neural network needs to define all fault categories in the full link layer when performing concurrent fault diagnosis.So all possible bearing fault combinations need to be fully considered,which will directly greatly increase the workload of the dataset and the time cost in the process of network training.Compared with classification neural networks,object detection networks require much fewer labels.So this wau can greatly save time and cost.In this paper,the YOLO series object detection network is used instead of the classification neural network for bearing fault diagnosis,and the concurrent fault detection and classification of bearings are completed on the basis of using fewer labels.In addition,this paper improves the traditional YOLOv4 by using GhostNet instead of the backbone feature network of YOLOv4.This way can effectively remove the redundant features generated during feature extraction and greatly improve the model accuracy.The improved GhostNet-YOLOv4 takes into account both training speed and detection accuracy.In addition,the method proposed in this paper can detect continuous bearing failures,which are difficult to detect with classification methods.