| Intelligent fault diagnosis of rolling bearings is an important technology for ensuring stable and safe operation of rotating machinery,which can not only improve factory productivity and reduce equipment maintenance costs,but more importantly,prevent safety accidents.In the actual production process,rolling bearings often operate under complex working conditions of heavy load,high speed,and strong impact,resulting in frequent bearing failures.Therefore,it is crucial to study the fault diagnosis techniques of rolling bearings.Traditional methods are susceptible to signal pre-processing techniques,manual feature analysis and labeled samples in the process of fault diagnosis.With the development of computer equipment and the emergence of various intelligent algorithms,fault diagnosis methods based on deep learning have attracted widespread attention.Compared with traditional fault diagnosis methods,deep learning fault diagnosis methods can achieve automatic feature extraction,efficient data representation,and accurate fault diagnosis.Therefore,this article studies the deep neural network-based rolling bearing fault diagnosis and life prediction method,and the main research contents are as follows.(1)Aiming at the problems that the rolling bearing fault diagnosis method cannot achieve adaptive noise reduction and has poor fault diagnosis effect under strong noise environment,a strong noise fault diagnosis method based on feature-enhanced noise reduction and residual neural network is proposed.Firstly,an adaptive signal noise reduction module based on feature enhancement is constructed,which transforms a one-dimensional signal into a two-dimensional matrix and performs a low-order enhancement of the matrix,thus this module strengthens the periodic impulse characteristics of the signal and achieves adaptive noise reduction of the signal.Then,a noise-reducing residual module is constructed by the signal noise reduction module and the residual module,which not only improves the noise immunity of the model,but also reduces the risk of model overfitting through the residual connection.Finally,a diagnostic model based on feature-enhanced noise reduction and residual neural network is constructed by using multiple noise-reducing residual modules,which has a powerful feature learning capability and noise resistance.The strong noise environment fault diagnosis experiments are conducted by using Case Western Reserve University and Southeast University bearing datasets,and the results show that the proposed method has a higher rolling bearing fault diagnosis accuracy.(2)Aiming at the problem that the rolling bearing working conditions are complex and the traditional fault diagnosis method is difficult to effectively distinguish the nuances of different load data,a rolling bearing variable load fault diagnosis method based on multi-cascade and multi-scale convolutional attention network is proposed.First,a multi-layer cascade is used to connect each convolutional layer in the multiscale convolution,and a multi-cascade and multi-scale convolutional module is built to extract the multi-scale features in the signal more efficiently.Then,the channelspace attention module is constructed by combining channel attention and spatial attention,which can not only process channel features and spatial features in parallel,but also improve the efficiency of multi-scale convolutional feature learning through the mask function of attention.Finally,the multi-cascade and multi-scale convolutional modules and the channel-space attention modules are used to construct a multi-cascade and multi-scale convolutional attention network,which can effectively extract the discrepancy information in different load data.Multiple sets of variable load experiments are conducted with four load bearing datasets from Case Western Reserve University,and the results show that the proposed method has strong generalization capability and can achieve accurate rolling bearing variable load fault diagnosis.(3)Aiming at the problem that the traditional Temporal Convolutional Network(TCN)is difficult to accurately predict the Remaining Useful Life(RUL)of rolling bearings in a noisy environment,a rolling bearing RUL prediction method based on the shrinking attention mechanism and temporal convolutional memory network is proposed.First,a shrinking attention mechanism is designed,which uses the attention mechanism to adaptively adjust the threshold parameters of the soft threshold function,thus effectively reducing the influence of environmental noise on the deep learning model.Then,a TCN threshold module with noise immunity is built by embedding a shrinking attention mechanism in the TCN module.Finally,a prediction method based on the shrinkage attention mechanism and temporal convolutional memory network is established by multiple TCN threshold modules and long short-term memory modules,which can learn not only the long-time features in vibration data but also the backward and forward dependent features in vibration data.Bearing RUL prediction experiments are conducted by using IEEE PHM 2012 and Xi’an Jiao tong University dataset,and the results show that the proposed method can effectively predict the degradation trend of bearings with lower prediction error compared with other prediction methods. |
