Research On Gearbox Bearing Compound Fault Diagnosis Method Based On New Deconvolution Theory

Gearbox,as the core component of rotating machinery,is used in wind power,coal mining,manufacturing and other fields widely.But in actual working conditions,the working conditions of gear transmission system are usually harsh,which can easily lead to faults of gear and bearing parts,further cause economic losses of enterprises,and even cause casualties and other major accidents.Therefore,it is of great significance to diagnose gearbox faults.Moreover,when one of the parts fails,other faults are induced in a complex mechanical environment.These fault characteristic signals will be intertwined and coupled with each other under the influence of complex transmission paths,and coupled with strong background noise,these factors increase the difficulty of mechanical fault diagnosis in industrial production.Therefore,it is of great theoretical significance and engineering application value to study the efficient noise reduction method and the separation technology of composite faults.In this paper,gearbox is taken as the research object,and new deconvolution theory and time-frequency analysis method are studied,such as minimum entropy deconvolution,maximum correlation kurtosis deconvolution,Multipoint Optimal Minimum Entropy Deconvolution Adjusted,Local Characteristic-scale Decomposition,dual-tree complex wavelet decomposition,combined modal function,etc.The main contents are as follows:1)Aiming at the single impulse easily appearing in MED results,the maximum kurtosis spectral entropy deconvolution(MKSED)is proposed with kurtosis spectral entropy as deconvolution target.The performance of MKSED is better than MED through simulation and experiment.First,the complex composite signal is decomposed into single-scale components by EEMD,and mode mixing is eliminated by modal recombination.Then,in order to overcome the influence of MKSED filter length on its result,the parameter adaptiveness of MKSED is realized by the improved particle swarm optimization algorithm.The modal recombination component is processed by adaptive MKSED to eliminate noise and extract impact pulses.Finally,fault features are extracted by envelope analysis and fault recognition is carried out.2)Aiming at the disadvantage that the effect of multi-point optimal minimum entropy deconvolution(MOMEDA)is susceptible to the influence of filter length,the filter length of MOMDEA is optimized by using kurtosis spectral entropy as fitness function through grid search optimization algorithm.Aiming at the disadvantage of MOMDEA that it is easy to fail in strong noise environment and can not process complex fault signals at the same time,a fault diagnosis method combining EEMD and MOMEDA is proposed.Firstly,signals containing complex faults are decomposed into intrinsic modal components of single fault information through EEMD,and meaningful components are selected through correlation analysis.Then,the adaptive MOMEDA optimized by grid search algorithm is used to process modal components and extract periodic fault pulses.Finally,the fault features are extracted through spectrum analysis,the location of the fault is located,and the feasibility and superiority of the proposed method are verified through simulation and experiment.3)Aiming at the defect that the filter length of minimum entropy deconvolution(MEDA)needs to be determined beforehand,the Grey Wolf algorithm is used to realize the parameter adaptation of MED A with the marginal power spectral kurtosis as the objective function.Aiming at the weakness of LCD which is susceptible to noise and weak shock extraction ability,a fault diagnosis method combining adaptive MEDA and LCD is proposed.Firstly,adaptive MEDA optimized by grey wolf is used to denoise complex signals and highlight fault pulses.Then,the complex mode signal is decomposed into single mode component by LCD,and the signal is further de-noised.Finally,the method of correlation analysis and mode reconstruction is used to overcome the disadvantage of LCD pseudo-component and mode aliasing.Finally,the effectiveness of the proposed method is verified by experiments.4)In order to overcome the disadvantage of uncertain decomposition layers of dual-tree complex wavelet(DTCWT),the number of decomposition layers of DTCWT is adaptively determined by residual energy ratio.Aiming at the disadvantage that DTCWT is easy to fail in strong noise environment,a frequency division DTCWT method is proposed,which can be well adapted to wider signal frequency band.The sub-bands obtained by DTCWT are divided into high frequency and low frequency by mutual information entropy,and noise is reduced by AR and SG filters respectively.The purpose of retaining the signal while noise is reduced is achieved to a great extent.