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

With the rapid development of industrial society,mechanical equipments play an important role in human life scenes.As a necessary part of mechanical equipment,rolling bearings are prone to various damages in long-term and high-load operating environments.Once a failure occurs,it will cause incalculable losses.Therefore,we need to perform accurate fault diagnosis on this kind of precision rotating machinery.Traditional fault diagnosis methods perform time-frequency domain feature analysis on the collected time-domain signals,relying heavily on manual extraction of features,which is time-consuming and labor-intensive.Thus,we cannot adapt traditional diagnosis methods to the industrial environment with high noise and multiple working conditions.Intelligent fault diagnosis technology fused with deep learning algorithms can adaptively extract features,identify and filter features.This paper focuses on the problems of rolling bearings in complex operating environments and operating conditions that are prone to failure,and the accuracy of fault diagnosis is not high,and the following issues are studied:(1)Explore and explain the algorithm principles of machine learning and deep learning.Compared with machine learning,the shallow non-linear expression network layer has limited effect when expressing complex features.The deep and complex network structure of deep learning can better extract the effective features in complex rolling bearing signals,and at the same time can solve the complex operation in the industry.The task of bearing fault diagnosis under the state;in addition,the basic principles and hyperparameter selection of the optimization algorithm,loss function,learning rate and optimization strategy in the debugging process of the deep learning algorithm are introduced.In the network model reasoning stage,the model Some evaluation indicators are explained in detail;finally,the data set used in this article is introduced,and the characteristics and usage scenarios of the data set are introduced.(2)Propose a fault diagnosis algorithm model based on domain adaptive multi-scale convolutional neural network.In order to extract the fault information of different time scales in the original vibration signal and improve the model’s recognition rate of the fault samples under different loads,this paper improves the classic one-dimensional convolutional neural network and uses a variety of small convolution kernels of different sizes.Extract features to improve the fault recognition rate.In the training process,in order to enhance the network model’s recognition rate of fault samples from different loads,adaptive batch standardization is introduced to reduce the difference in sample distribution between the source domain and the target domain,and enhance the generalization ability of the model.Experiments show that the model has a high fault recognition rate and strong domain adaptation ability.(3)Study the influence of the fault diagnosis and recognition rate of the convolutional neural network model fused with different attention mechanisms.The attention mechanism is similar to the "eyes" of neural networks,which can adaptively assign higher weights to channels with rich fault information.The core idea is to obtain a feature map of attention weight distribution through some network layer transformations.In order to improve the model’s ability to discriminate important features,this paper compares the principles and experimental results of channel attention mechanism,spatial attention mechanism,non-dimensionality reduction attention mechanism and cross attention mechanism.Different attention mechanisms have different network focuses.Different attention mechanisms have different network focuses.Experiments show that the non-dimensionality reduction attention mechanism has the highest diagnostic accuracy,while ensuring the lightness of the model The channel attention mechanism has a rather lower diagnosis accuracy rate.