Fault Diagnosis Of Planetary Gearbox Based On Measurement Point Optimization And Deep Learning

Planetary gearbox has the characteristics of strong bearing capacity and large transmission ratio.It is widely used in industrial fields such as coal energy,advanced manufacturing and wind power generation.However,due to the extremely bad working environment in these work fields,the complex internal transmission structure of planetary gearbox and the influence of noise interference,the phenomenon of mutual coupling of vibration signals often occurs during work.These comprehensive factors make the condition detection and fault maintenance of planetary gearbox more difficult.In addition,in the fault diagnosis test of planetary gearbox,the number of sensors directly affects the diagnosis cost,so it is particularly important to use an effective number of sensors to accurately diagnose the fault.Aiming at the problems of high cost of planetary gearbox fault diagnosis,difficult fault signal acquisition,mutual coupling of vibration signals and inaccurate fault diagnosis,combined with the powerful data mining ability of deep learning,a new method of planetary gearbox fault diagnosis based on measuring point optimization and deep learning is proposed.Taking the planetary gearbox as the key research object,this paper studies the fault diagnosis technology combined with the advantages of deep learning algorithm.Firstly,it deeply studies the deep learning theory,compares and analyzes the two deep neural network model structures of Alex Net and Res Net,and applies the two network structure models to the fault diagnosis of planetary gearbox.Then the fault simulation experiment is carried out on the planetary gearbox fault diagnosis test-bed in the laboratory.11 signal acquisition and measurement points are arranged on the planetary gearbox and helical gearbox,and the single fault working condition plus composite working condition are artificially set.A total of five working conditions are used for vibration signal acquisition.Then,the vibration signals of five working conditions are processed by multidimensional integrated global empirical mode decomposition（MEEMD）and local mean decomposition（LMD）.Then,the characteristic correlation analysis method is used to analyze the correlation of the effective components and optimize the measuring points.The optimization results are verified by ALNAFSA-BP network.Secondly,the fusion features are composed of the effective components decomposed by MEEMD and LMD.The fusion features are classified by using the above two deep convolution neural network structures,and the better network is compared and analyzed.Set the original data of the preferred measuring points,the effective component features of the single decomposition method and the diagnostic accuracy of the fusion features for comparative analysis.The results show that the effective component fusion features of the optimized measuring points and the application of depth residual network（Res Net）in the fault diagnosis of planetary gearbox have achieved good results and high recognition accuracy.Finally,by comparing with the latest excellent diagnostic method on the same experimental platform,it is found that this method has the highest diagnostic accuracy under the condition of large samples,reaching 97.2%.The conclusion shows that combining the advantages of deep learning algorithm and using the fusion features of better measuring points selected by feature correlation analysis for recognition and diagnosis,a new set of better fault diagnosis method of planetary gearbox based on measuring point optimization and deep learning can be formed.