Research On Weak Fault Extraction And Identification Of Rolling Bearing

As mechanical equipment continues to be integrated and complicated,fault detection technology also needs to advance with the times in order to ensure both processing efficiency and product quality.Rolling bearings are widely used in machinery and equipment.Any minor damage may affect the entire equipment.Early detection and repair of bearing failures can effectively reduce safety risks and economic losses.Therefore,it is particularly important to study the early fault diagnosis technology of rolling bearing.The characteristics of early bearing fault signals are weak,and they are vulnerable to factors such as noise and human interference.It is impossible to directly identify the status of bearings.Therefore,in order to accurately grasp the working state of the rolling bearing,this article studies the two aspects of feature extraction and state recognition.Firstly,the APLCD-WPT method based on APLCD(Adaptive Partly-ensemble Local Characteristic-scale Decomposition)and WPT(Wavelet Package Transform)and the AGSR(Adaptive Genetic Stochastic Resonance)method are used to extract early weak fault characteristics of bearings.The APLCD-WPT algorithm uses APLCD to adaptively decompose the vibration signal into multiple ISC(Intrinsic mode components).Each ISC component represents the characteristics of different frequency bands of the original signal,and the WPT is used to correct the ISC components that are still modally aliased after decomposition.AGSR uses genetic algorithms to select and optimize multiple parameters of the stochastic resonance system in parallel.It can adaptively filter the stochastic resonance system parameters that best match the input signal,thereby accurately extracting fault features.The paper also introduced a deep learning method,using the SSAE(Stacked Sparse Autoencoder)to identify patterns of the extracted features and achieve intelligent diagnosis of early failure of rolling bearings.The SSAE network is composed of multiple Auto-encoders.By training the network layer by layer,it can adaptively learn the characteristics of various types of failures from massive data.Then the whole network is optimized through supervised back-propagation algorithm,and finally the feature is input into the Softmax classifier to determine the health status of the rolling bearing.The article compares APLCD-WPT algorithm with LCD and other feature extraction methods,it shows that APLCD-WPT has certain advantages in suppressing modal confusion and improving the accuracy of feature extraction.The AGSR algorithm uses genetic algorithm for parameter optimization,and the feature extraction effect is better than the traditional stochastic resonance method.In addition,the feature samples extracted by APLCD-WPT and AGSR were used for SSAE network training.In the experiment,the optimal network model that found by adjusting parameters such as the number of AE layers,the number of hidden layer nodes,and the learning rate,finally,achieves efficient intelligent diagnosis of early failures of rolling bearings.