Research On Bearing Condition Evaluation Method Of Rotating Machinery Based On Deep Learning

Rolling bearings are important parts of rotating machinery,and among the failures of rotating machinery,bearing failures account for a large proportion.Once the bearing fails,if it is allowed to develop freely,it will cause unimaginable and immeasurable waste of resources and property losses.Therefore,timely understanding of its operating status and fault diagnosis of the bearing can provide a reliable basis for equipment repair and maintenance.,Thereby extending the service life.The vibration signal contains a lot of time series information,and the features in the one-dimensional vibration signal can be directly extracted.Based on this,a onedimensional convolutional neural network(1DCNN)fault diagnosis model is constructed,and its adaptive learning features and classification capabilities are used to achieve “end-to-end” diagnosis from data to recognition results,avoiding The interference of human factors.Two tasks were carried out based on the failure data of rolling bearing of different pit sizes(simulating different failure degrees)of Case Western Reserve University:(1)Research on fault diagnosis of bearing with the same load.Through data enhancement,the experimental data set is enriched.Through the time-frequency analysis of different fault states,the change law of the characteristic frequency amplitude after the development of the fault is grasped,and then the data set is sorted in proportion.The rolling bearing vibration signal is directly used as the input of the one-dimensional convolutional neural network(1D CNN).The results show that the fault type and fault state with a clear degree of damage can be learned and identified by extracting features.When the load remains unchanged,the accuracy of the test set under each load has reached 100%.(2)Research on fault diagnosis of bearing with the variable load.The working conditions of the bearing in actual work are changing.The fault test data at different loads is used to simulate 12 different working conditions,and perform fault diagnosis through the built fault diagnosis model.The results show that the fault identification is accurate when the load changes.The lowest rate is 94.5%,and the overall average recognition rate reaches 97.97%,indicating that the 1D CNN rolling bearing fault diagnosis model constructed in this paper has good generalization ability.When a rolling bearing is in service,its degradation process is a continuous evolution process.Diagnose the entire degradation process through the life cycle data of the University of Cincinnati rolling bearings.First,perform time-frequency analysis on it,and confirm its fault characteristic frequency.A degradation state judgment method based on time domain indicators and envelope analysis is proposed.Take the performance degradation process of the outer circle as an example.According to the trend of the standard deviation,the fault status of the outer circle is initially classified,namely normal,slight degradation stage,moderate degradation stage,severe degradation stage and complete failure.Fault state recognition,the results show that the accuracy of the non-intersection test set is over 96%.Introducing the confusion matrix,further analysis revealed that the misdiagnosis mainly occurred between the slightly degraded stage of the outer ring and the normal state.Using the same method,the failure state of rolling bearings is divided into three degradation stages: normal,degradation and failure,so that it is consistent with the on-site alarm classification,and each type of failure has achieved better accuracy;Finally,the outer ring faults of the two sets of experiments are used as a test set for each other to prove that the trained model can be used to identify the operating status of other bearings of the same model,with an accuracy rate of 88%.Besides,the Alex Net network,a deep learning method,was used to conduct comparative research on external faults,and its accuracy rate was 90.98%,which is lower than the one-dimensional convolutional neural network model,which proves the effectiveness of the proposed method.The paper has 35 pictures,21 tables,and 89 references.