Research On Data-driven Diagnosis Method For Bearing Faults Of In-wheel Motor

With the rapid development of electric vehicles,the distributed driving technology of in-wheel motors is more and more widely used.Compared with the centralized driving technology of traditional cars,this technology integrates the functions of driving,braking and carrying into the motor and installs it in the wheel hub.Its flexible control and efficient transmission are the important development trends of electric vehicle driving methods in the future.However,due to the special installation method of the in-wheel motor,the key component of the wheel motor,the bearing,needs to bear both radial and axial loads,and the complex and variable operating environment of the electric vehicle will also increase the impact of the in-wheel motor bearing,which makes it to produce local wear and degradation of performance to cause failure.Once the in-wheel motor bearing fails,it will affect the running performance of the In-wheel motor and endanger the driving safety of the electric vehicle.Therefore,it is very important to take research on the diagnosis methods for the bearing faults of in-wheel motor.Aiming at the operation safety of electric vehicles driven by in-wheel motors,this paper takes the in-wheel motor bearing as the research object,focuses on the three aspects of fault feature extraction,fault feature dimensionality reduction and fault state recognition,and proposes a data-driven fault diagnosis method of the in-wheel motor bearing.The method can effectively realize the diagnosis and recognition of the fault state of the in-wheel motor bearing.First of all,for the problem of difficult to control variables in the actual vehicle test,based on the actual operating environment of the in-wheel motor of the electric vehicle,the in-wheel motor bearing fault test system was built based on the ZHIDOU D1 electric vehicle as a prototype,and the in-wheel motors with the common bearing faults were customized.The test plans were designed accordingly to provide data support for the subsequent fault diagnosis research work.Secondly,for the problem that it is difficult to extract the fault feature information of the in-wheel motor bearing under intermittent strong interference,a fault feature extraction method based on optimized resonance sparse decomposition is proposed.Taking the ratio of the smoothness index of the high resonance component and the kurtosis of the low resonance component in the resonance sparse decomposition as the optimization objective function,the decomposition factor optimized by the wolf pack algorithm is used to decompose the original signal to extract the fault feature information.Then,for the dimension reduction problem of multi-dimensional fault feature parameter set,a fault feature dimensionality reduction method based on improved t-distribution stochastic neighborhood embedding is proposed.Based on the traditional t-distribution stochastic neighborhood embedding,the gradient is accelerated by the Barnes-Hut algorithm,which can effectively reduce the time complexity from O(N~2)to O(NlogN)while ensuring the accuracy of dimensionality reduction features,and greatly improve the operation speed.Finally,for the problem of noise data interference in the fault diagnosis process of the in-wheel motor bearing,a fault state recognition method based on an artificial hydrocarbon networks is proposed.The characteristics of the hydrocarbon packaging information are used to compare the characteristic parameter set in the database with the state of the wheel motor bearing by supervised learning.Combined with the experimental data,through comparison with other classifier algorithms,it is verified that this method not only has a high fault recognition rate,but also has strong robustness in processing noisy data.