Research On Fault Diagnosis Method Of Rolling Bearings Under Variable Conditions Based On Improved CNN

As a basic component of rotating machinery,rolling bearings are widely used in important fields such as construction machinery,aerospace and high-speed train.Setting up reliable health monitoring system is the key to ensure the safety of rotating machine in industrial processes.The load and speed of a rolling bearing are often changed in the practical engineering,and due to the complex and variable operating conditions,the distribution of vibration signal features is quite different.Traditional intelligent fault diagnosis methods based on signal processing and classifier cannot meet the requirements of feature extraction and fault identification under variable conditions,because those methods need expert experience and prior knowledge.Therefore,an end-to-end rolling bearing fault diagnosis method is proposed based on deep learning,for realizing the direct mapping from the original signal to the classification result.Firstly,the vibration data is mapped to a nonlinear spatial domain using the convolutional neural network(CNN),and the information related to the bearing running state is extracted from massive original signals.For the small displacement,scaling and other distortion forms of the input signal,CNN has invariant property,so,the fault features of the rolling bearing under variable conditions can be extracted adaptively.Secondly,according to the differences of bearing vibration signal under variable conditions,and the local abrupt characteristics of the signal when the fault occurs,it is proposed to integrate the attention mechanism into the CNN structure and establish the dependence between the feature channels.Reasoning of the correlation among different representational data,further improving the sensitivity of bearing vibration characteristics under variable conditions.Finally,the data augmentation method is used to obtain more abundant training samples,so that the proposed deep neural network can be more fully studied and effectively avoiding over-fitting.Through further analyzing the characteristics of the vibration signal.In order to learn the characteristics of the invariance of signal displacement,the selection method of hyperparameters in the attention mechanism CNN model is designed to reduce the design difficulty of the fault diagnosis algorithm.Experiments show that the proposed fault diagnosis model based on attention mechanism CNN can realize multi-state identification and classification of rolling bearings under variable conditions,and higher accuracy can be obtained compared with other methods.