Research On Fault Diagnosis And Life Prediction Methods For Rolling Bearings Based On CNN

Rolling bearing is the core component of major equipment.Fault diagnosis and remaining useful life(RUL)prediction are the key technologies of prognostics and health management(PHM),which can effectively locate the fault location of rolling bearing and predict the remaining life.Condition-based maintenance of equipment is of great significance.This paper takes rolling bearings as the research object,and studies the prediction and health management technology of rotating machinery from two aspects: rolling bearing fault diagnosis methods and remaining useful life prediction methods.Firstly,starting from the original vibration signal of the rolling bearing,the fault diagnosis method of the rolling bearing is studied,without the feature extraction process,it can directly act on the original vibration signal.Secondly,from the two perspectives of one-dimensional signals of the same vibration direction and two-dimensional signals of different vibration directions,the method of predicting the remaining useful life of rolling bearings is studied.The main research contents are as follows:(1)Research on the fault diagnosis method of rolling bearing based on SNDCNN:Aiming at the problem of low fault recognition rate,poor performance under different loads and different noise environments in the process of rolling bearing fault diagnosis,a switchable normalization(SN)and deep convolutional neural network(DCNN)combined method(SNDCNN).The model can be directly applied to the original vibration signal.By increasing the width of the first layer of convolution kernels and stacking multiple layers of convolution kernels,it can effectively extract fault features and suppress high-frequency noise.The K-Max pooling operation is used in the pooling layer to avoid losing the strength information of the features.After each layer of convolutional layer,self-adapting normalization is used to solve the over-fitting problem and improve the generalization ability of the model.Experiments show that this method has a higher fault recognition rate than traditional diagnosis methods,reaches a diagnosis rate of more than 90% under different loads,and has better anti-noise performance.(2)Research on the prediction method of the remaining useful life of rolling bearings based on DCNN: In view of the problems that most of the remaining useful life prediction of bearings require complex feature extraction process,low prediction accuracy and poor noise resistance,this paper proposes a bearing based on deep convolutional neural network of remaining useful life prediction model,which can directly act on the original one-dimensional horizontal vibration signal,adopts the multi-layer convolution layer and the pooling layer superposition method to improve the feature extraction ability of the model,and adopts the batch standardization method to reduce the amount of model calculation,Improve the prediction speed and use smoothing to eliminate strong fluctuations in the prediction results.After the experimental verification of the bearing life cycle data set of the FEMTO-ST Research Institute,high prediction accuracy has been achieved.(3)Research on the prediction method of the remaining useful life of rolling bearings based on CBAM-SCNN: Although the prediction model of the remaining useful life of the bearing based on DCNN can predict the remaining useful life of the bearing better,it can also accurately describe the bearing performance under different load and noise environments.The degree of degradation,but there is a problem of large fluctuations in the prediction curve.In response to the above problems,this paper proposes a separable convolutional neural network(SCNN)and a convolutional block attention module(CBAM)Combined rolling bearing fault diagnosis method(CBAM-SCNN).For the horizontal acceleration vibration signal and the vertical acceleration vibration signal collected by the two sensors,a separable convolutional neural network is used to act on each input channel separately.In order to enhance the model’s ability to extract effective features and suppress noise,a convolutional attention mechanism module will be added to the model.Experiments show that the model proposed in this paper has high prediction accuracy in predicting the remaining useful life of the bearing,and it has been applied to different loads and noise environments to achieve better prediction results.(4)Rolling bearing fault diagnosis test bench design: The bearing fault diagnosis test bench was preliminarily designed,and the bearing data collection work under different fault types,different failure levels,different speeds,different loads and constant speed increase was carried out through the test bench.The complexity and diversity of bearing fault diagnosis data sets.According to the manufacturing information of the selected bearing,the characteristic frequency and multiplication frequency of the bearing are calculated,and compared with the amplitude frequency spectrum obtained by the fast Fourier transform,the validity of the bearing fault data collected by the bearing fault diagnosis laboratory is verified.Finally,the SNDCNN-based bearing fault diagnosis model proposed in Chapter 2 was verified with the collected single load and constant speed increase data,and both achieved a high fault diagnosis rate.