Research On Operational Reliability Assessment Of Rolling Bearings

At present,with the gradual progress of China towards becoming a manufacturing powerhouse and the rapid development of big data,the manufacturing industry is in a new stage of becoming intelligent.Intelligent manufacturing technology has become a major trend and core content of the future development of manufacturing.As a key component in modern machinery and equipment,rolling bearings are widely used in the manufacturing industry.The reliability of rolling bearings affects and restricts the development of the industry.The demand for intelligent diagnosis and assessment of rolling bearings is growing.How to assess the reliability and intelligent maintenance of rolling bearings quickly has attracted the attention of many scholars.The traditional reliability assessment method is not suitable for the real-time operation reliability assessment of rolling bearings because of its strict application conditions.Most data-driven methods have high requirements for model parameters,and the methods are not universal.In this paper,the method research on the operational reliability of rolling bearings is carried out.The main contents are as follows:(1)Aiming at the fault pattern recognition of rolling bearings,this paper proposes a fault pattern recognition method based on ensemble learning and Stacked Auto-Encoder(SAE).Based on the vibration data generated during the running of the rolling bearings,a series of basic learners based on SAE are constructed through the disturbance of the activation function and training data.Then,the classification performance of each class of the base learners is evaluated by the precision.The output probability of each base learner is corrected and filtered by the precision,and finally the failure pattern recognition of the rolling bearings is realized.Through the case of bearing fault diagnosis,this ensemble strategy can make full use of the classification ability of each basic learner for each class,reduce the design of hyper-parameters,and improve the final classification accuracy by complementing the advantages of each basic learner.(2)Aiming at the operation reliability assessment of rolling bearings,this paper proposes a feature parallel learning and selection method based on SAE for operation reliability assessment of rolling bearings.In this method,the time-domain signals of rolling bearings are converted into frequency-domain signals by fast Fourier transform as the input of the feature parallel learning model.After that,several parallel learning models are constructed through the disturbance of the activation function.Then,the learned features are classified by correlation analysis,and the typical features in each category are selected.Finally,the operation reliability of rolling bearings is evaluated through the definition of reliability based on Mahalanobis distance.Through the case of operation reliability assessment of rolling bearings,it can be concluded that this method can effectively select the characteristics related to rolling bearing degradation,and the evaluation results can well show the trend of bearing degradation.(3)In view of the fact that a single signal source cannot comprehensively evaluate the performance degradation of rolling bearings,a method for evaluating the operational reliability of rolling bearings based on a deep-coupled autoencoder is proposed.In this method,multi-source signals of rolling bearings are used as input to train multiple models respectively.Each model stacks three autoencoders to extract features from multi signal respectively.The last layer plus the coupled autoencoders jointly extract common features from multi-source signals,and then the similarity measure is used to fine-tune the entire deep model.Then,combined with the definition of Mahalanobis distance,the operation reliability of rolling bearings is evaluated respectively.Finally,multiple results are fused to get the final assessment result.the case of operation reliability assessment of rolling bearings,it can be proved that the greedy layer-by-layer training method and the fine-tuning of the overall model based on the feature similarity measure can effectively extract the direct relevant features of multi-source information,and can represent the operational reliability of rolling bearings from multiple aspects.