Research On Bearing Fault Diagnosis Method Based On Deep Neural Network

The development of mechanical and electrical equipment is an important factor in measuring the comprehensive national strength of the country.The rolling bearing is a key component of mechanical rotating parts,which are widely used in aviation,agriculture,railways,etc.Their health is related to the safety and stability of the entire mechanical equipment.Because the bearing has complex structure,strong coupling,and harsh operating environment,once a failure occurs,it will affect the damage and production of the bearing and the entire component,resulting in huge property losses,and may even cause casualties.Therefore,the bearing fault diagnosis becomes a key task.In recent years,with the continuous development of computer technology,deep learning technology has played a huge potential in bearing fault diagnosis based on vibration signals.Fault diagnosis technology has become more intelligent,systematic and automated.Traditional methods usually include three steps: feature extraction,feature dimensionality reduction,and classification.They rely too much on professional knowledge and cannot ensure versatility,which can no longer meet actual needs.The fault diagnosis method based on deep learning still has some shortcomings.The traditional single network topology lacks the distinguishability of feature extraction and noise robustness,which reduces the accuracy of fault diagnosis.In practical applications,the working conditions are diverse,and most of the current research methods are faced with a single working condition,resulting in the lack of self-adaptability in fault diagnosis.In view of the above problems,this paper proposes a bearing fault diagnosis method based on deep neural network.The main work is as follows:1.A fault diagnosis method based on multi-scale neural network is proposed,which improves the learning ability of features.Firstly,this method uses the original time-domain signal as input.Next,we use the Piecewise Aggregate Approximation algorithm to down-sample the input signal to obtain signal representations of different scales.Then,we use full convolutional neural networks and long short-term memory networks to extract features from multi-scale signal.Local multi-scale and global multi-scale methods are proposed,and the long and short-term memory network is used to extract the time sequence information in the signal.Finally,the multi-classification function is used to realize the fault diagnosis.This method was tested on public datasets.Experiments show that this method has high accuracy and noise immunity.2.A fault diagnosis method based on the attention mechanism is proposed.Two types of attention mechanisms are used to solve the dependence between signals and between multiple scales,which effectively improves the adaptability of bearing fault diagnosis.Firstly,in terms of feature extraction,this method differs from the fault diagnosis method based on multi-scale neural network in that the attention mechanism is added to the long and short-term memory network to solve the correlation between the signals.Then,after the feature extraction and before the fusion of the features,it focuses on multiple the performance of scale features is different.Combining the idea of attention mechanism and multi-scale structure,promote the connection between multiple scales,better establish the dependence between multi-scale features.Finally,we use the multi-classification function for classification.The proposed method further improves the accuracy and noise resistance of bearing fault diagnosis,and has high adaptability under variable working conditions.In summary,this paper proposes a bearing fault diagnosis method based on deep neural network that can directly implement end-to-end bearing fault diagnosis from the original time-domain vibration signal,which is different from traditional methods that require feature extraction,feature dimensionality reduction,and classification.By testing on multiple data sets and comparing with existing methods,the method in this paper has high accuracy,high robustness under noisy environments and high adaptability under different working conditions.