Research On Anti-interference Diagnosis Method Of Bearing Faults

In industrial production,the safe operation of mechanical transmission equipment is the core requirement.Equipment fault may cause serious economic losses or even catastrophic accidents.Bearing is an indispensable part of mechanical transmission,so it is necessary to monitor and diagnose it.However,due to the harsh working environment of the bearing,such as strong noise,unbalanced load,and the mutual coupling between multiple faults of the bearing itself,the bearing reliability diagnosis is challenged.With the rise of artificial intelligence,it has made major breakthroughs in robotics,control systems,pattern recognition and other fields.Therefore,it is of extraordinary significance to propose a kind of intelligent bearing fault diagnosis method which combines artificial intelligence algorithms with bearing fault diagnosis.In this paper,the anti-interference diagnosis of bearing faults is realized under different interference sources.The main research work includes the following three points.First,the bearing fault diagnosis under the condition of multiple faults interference is studied,and a new diagnosis method based on the characteristics of industrial vibration signals is proposed,which is called hybrid time series convolutional neural networks(HTSCNN).This method uses the method of estimating the overall proportion to adaptively determine the number of model training samples to ensure that the proportion of the input fault sample information is not distorted;then the time series fragments are randomly combined and input to the convolutional neural networks(CNN).This method solves the problem of insufficient extraction of non-adjacent signal information for bearing faults,thereby improving the model's ability to recognize multiple faults.Then,the bearing fault diagnosis under the background of noise is studied,and a bearing fault diagnosis method based on feature fusion and feature selection is proposed.This method combines the time-domain and frequency-domain global features of the original bearing vibration data with the local features extracted by CNN,makes full use of the bearing vibration signal information,and overcomes the shortcomings of traditional CNN that the overall signal is easy to be ignored.Then the robust discriminant features are selected from the fusion features through the max-relevance and min-redundancy feature selection algorithm(m RMR),and used for support vector machine(SVM)classification.Finally,through experimental analysis of bearing data under different signal-to-noise ratios,the experimental results prove that the method has better robustness under noisy environments.Finally,the bearing fault diagnosis of small sample data under unbalanced load is studied,and a bearing fault diagnosis method based on gradient penalty wasserstein generative adversarial network(WGAN-GP)and convolutional neural network with selfattention mechanism(Se CNN)is proposed.This method uses WGAN-GP to generate new samples similar to the training sample distribution,enriches the original training set,and improves the performance of bearing fault diagnosis based on the deep learning model;then,on the basis of CNN,the anti-interference ability of the model against unbalanced load is further reduced by adding self-attention mechanism.Finally,experiments verify that the method can achieve accurate diagnosis of bearing faults with small sample data under unbalanced load.