Research On Rolling Bearing Fault Diagnosis Method Based On Multi-information Fusion

With the development of modern industry,rotating machinery and equipment have been found in every aspect of production and daily life.Rolling bearing is the main component of rotating machinery,and its reliability is the premise to ensure the normal operation of machinery.Once the bearing failure occures,it will affect the operation of the whole machinery and cause safety accidents.Therefore,fault detection of rolling bearing is particularly important.Multi-information fusion technology can fuse multisource information and improve the efficiency of data use.The deep learning algorithm can automatically mine the fault features in the data,and classify and identify the faults.This thesis takes the rolling bearing as the research object,takes the deep learning algorithm as the carrier,uses the multi-information fusion technology to diagnose the rolling bearing faults,and develops an online diagnosis system.The main contents of this thesis are as follows:Aiming at the problem of unbalanced data acquisition of rolling bearing fault information in actual operation,an improved SMOTE data enhancement method is proposed.This method uses the existing rolling bearing data to divide different kinds of data,and generates new data in the sparse space of a few classes and at the boundary of a few classes to strengthen the boundary between different classes.Through experimental comparison,the fault diagnosis recognition rate and the recognition rate of minority classes are improved after the introduction of this method.Aiming at the problem of fault diagnosis under the condition of constant speed,a method of WDCNN and FFT-LSTM feature fusion is proposed.This method makes full use of the time and frequency dual domain information in the vibration signal,extracts the features of different domains,fuses the features of the two domains,and further compresses and refines the fused information,strengthens the feature recognition in the vibration signal,improves the running speed of the network model,and enhances the learning efficiency.The experimental results show that this method not only has high bearing fault recognition rate under the same load,but also maintains high recognition rate under variable load.Aiming at the problem of fault diagnosis under the condition of variable speed,the MSKCNN-LSTM fault diagnosis model is proposed.The vibration signal under the condition of variable speed has the characteristics of non-periodic and nonlinear,and the common deep learning fault model is difficult to extract complete features.Therefore,the multi-scale kernel is introduced into the deep learning model to enhance the ability to extract the spatial features of the vibration signal,and the fused features are further compressed in the network,LSTM is introduced to enhance the feature extraction of time information in the signal and improve the accuracy of fault diagnosis.Experimental results show that this method can effectively improve the fault identification rate under variable load conditions.In order to ensure that the fault diagnosis algorithm can be used in actual production,the bearing fault online diagnosis system is designed and completed.The system is composed of a complete set of hardware equipment and software.It can realize the functions of real-time collection,display and storage of bearing vibration data,time and frequency domain analysis of vibration signals,online fault diagnosis,model training of new sample data and data enhancement.The stability and usability of the system are verified by testing on the experimental platform.