Research On The Construction Method And Application Of Sparse Matching Dictionary For Mechanical Non-stationary Signals

The non-stationary signal processing technology is an essential research direction for realizing fault diagnosis of mechanical equipment and its key components.In practical engineering applications,components associated with mechanical equipment failure often exhibit transient shock and sparsity throughout the signal structure.Sparse representation theory can effectively match the characteristics of shock signals when mechanical equipment fails,and it can capture the essence of information and achieve the most efficient expression.However,there are the following difficulties in improving the adaptability of sparse complete dictionaries in existing sparse representation studies:(1)the problem of obtaining fixed basis atoms of sparse complete dictionaries with the assumption of prior knowledge;(2)the problem of obtaining free basis atoms of sparse complete dictionaries with the assumption of data-driven.Therefore,this dissertation focuses on the main line of "sparse representation" and carries out the research of mechanical non-stationary signals fault feature extraction method based on matching dictionary sparse representation theory from the structural characteristics of fault signals.The main research contents of this dissertation are as follows:(1)Aiming at the problem that the traditional tunable quality factor wavelet transform overly relies on a priori knowledge for manually setting parameters to extract free quality factor wavelet basis atoms,which leads to poor matching with the signal,the sparse representation theory of adaptive double tunable quality factor wavelet transform is studied.In order to obtain the complete dictionary basis of double tunable quality factors matching the characteristics of non-stationary fault signals,the characteristics of the tunable wavelet basis of quality factors are sufficiently utilized.Take the maximum correlation kurtosis as the evaluation index,and combine the idea of multi-population innovation,propose the multi-population quantum genetic algorithm ergodic iteration optimize the quality factors.The optimal double tunable quality factor wavelet basis is selected as a complete dictionary for sparse decomposition to obtain optimal sub-bands,and the optimized sub-bands are reconstructed to ensure that the sub-bands have sufficiently impact feature information effect.Simulation and engineering experiments verify the feasibility of the proposed method.(2)To address the problem that the transient impact components associated with faults in mechanical non-stationary signals have both sparsity and certain structural characteristics in the time-frequency transform domain,and it is difficult for a single fixed dictionary to describe the integrity and physical ambiguity of the multi-physical structural features of non-stationary signals,a structure-enhanced Bayesian sparse representation method based on the adaptive composite dictionary is proposed.A structured enhanced Bayesian sparse representation model which can promote the structural characteristics of sparse reconstruction results is studied to achieve the sparse representation solution of signals.Meanwhile,according to the characteristics of mechanical fault vibration signals,a composite dictionary combining the Sin-Chirplet time-frequency dictionary and impact dictionary is designed,so that the dictionary can better match the analyzed signals and reduce the redundancy of the dictionary.The multi-population quantum genetic algorithm is introduced into the parameter search of the composite dictionary,providing an efficient atom selection strategy,and reducing the complexity of the sparse algorithm.Simulation and experimental analysis verify the effectiveness and universality of the proposed method.(3)For the mechanical non-stationary vibration signals,the frequency components are complex,the signal-to-noise ratio is low,the fault characteristics are not obvious,and the sparse representation results are affected by the internal structure correlation between atoms and signals in the predefined basis function dictionary,resulting in the difficulty in extracting the instantaneous periodic pulse components caused by local faults.A Bayesian biorthogonal sparse representation of an adaptive redundant lifting wavelet dictionary is proposed.The redundant lifting wavelet framework theory of multi-population quantum genetic optimization is investigated to construct an adaptive lifting wavelet dictionary that can be flexibly designed to match diverse fault features according to the physical prior of the intrinsic structure of the fault signal.The dictionary is integrated into the Bayesian biorthogonal matching tracking algorithm for Gram-Schmidt orthogonalization coordination to ensure that the selected atomic regressions are orthogonal to the solution of the selected component and the residual.Simulation and engineering experimental analysis show that the proposed method can achieve adaptive matching of the dictionary with the periodic pulse component waveforms generated by the target signal fault,which supports the capability and robustness of the proposed algorithm.(4)The dynamic signals of mechanical compound faults are manifested as the mutual coupling of multiple fault feature information,and the predefined basis dictionary does not have the ability to adaptively adjust the dictionary parameter to increase the sparse representation according to the complex signals to be processed,a variable window sparse representation of the adaptive multi-wavelet dictionary under the framework of symmetric lifting is proposed.Based on the symmetric lifting theory,a multi-wavelet symmetric lifting matrix is designed and an adaptive base function library with multiorder vanishing moments is constructed.According to the kurtosis optimal principle of local correlation with faults,the multi-wavelet lifting basis function is optimized by the multi-population quantum genetic algorithm,and a multi-wavelet adaptive lifting supercomplete dictionary matching the coupled feature components is constructed.At the same time,the variable window truncation technique is studied to detect the fault impact information hidden in the noise.Simulation signals and experimental analysis show that the proposed method can achieve one-time separation and enhanced extraction of mechanical compound fault coupling features under strong noise.