Fault Diagnosis Of Rolling Bearing Based On Wavelet Time-frequency Transform And Deformable Convolutional Neural Network

Rotating machinery plays an important role in aerospace,navigation,transportation,manufacturing and other fields.Rolling bearings are one of the important components and are widely used in rotating machinery.The running state of the bearing often determines the stability of the rotating machinery.Foreign matter entry,unsuitable assembly,insufficient lubrication,overload and other problems will cause the bearing to fail in advance,which may seriously affect the rotating machinery.Monitoring and fault diagnosis of rolling bearings can ensure that the rotating machinery is in good working condition and avoid serious accidents.Therefore,the fault diagnosis of rolling bearings has become a hot topic for discussion at home and abroad,which is of great significance to the development of modern industry.The traditional fault diagnosis methods are mainly signal analysis methods,but with the development of rotating machinery,the signal components are becoming more and more complex,and the methods that rely on manual feature extraction are no longer applicable.The fault diagnosis technology based on the intelligent algorithm method benefits from the development of big data technology.It has surpassed the signal analysis method in many aspects,saved a lot of resources,and is superior to the signal analysis method in terms of recognition rate and stability.Therefore,this paper proposes an intelligent algorithm method based on wavelet time-frequency analysis and deformable convolutional neural network,which is applied to rolling bearing fault diagnosis with an average accuracy rate of 99.9%,which has achieved good results.First of all,the article expounds the causes and categories of failures of rolling bearings,explains the harmonic components caused during the operation of rolling bearings,and analyzes the impact of harmonic components on the impact components.According to the difference in frequency between the harmonic component and the impact component,the Fourier dictionary-based OMP algorithm is used to remove the harmonic component,reducing the interference to the impact component.Secondly,for the vibration signals of rolling bearing faults with harmonic components removed,the idea of time-frequency analysis is used for signal processing,which enriches the fault characteristic information contained in the signals.The short-time Fourier transform was compared with the wavelet transform,and it was decided to use the wavelet transform.The Meyer wavelet and Morlet wavelet were selected for comparison experiments,and Morlet wavelet was selected as the mother wavelet function.Six types of fault signals were selected from the experimental data of rolling bearings published by the Western Reserve University.Time-frequency analysis was performed on the data of these six types of signals,and the comparison without the influence from harmonic component was made by the time-frequency analysis.Finally,this paper establishes an 11-layer convolutional neural network model and introduces a deformable convolution mechanism to enhance the model's ability to extract deep features.Taking the data set based on the intelligent algorithm method as input,the performance between the deformable convolutional neural network and the traditional convolutional neural network under the same data set are compared.In order to illustrate the superiority of the method in this paper,16 features in the time domain are extracted based on the previously selected 7 types of signals,a data set based on the signal analysis method is set up,and three sets of model pairs: SVM,BP neural network,and random forest are established for training the data set.Through comparative experiments,the superiority of the method in recognition rate and stability is proved.