Data Recovery And Fault Classification Based On Sparse Sampling

To recover the damaged data in signal acquisition,reduce the effect of signal sparsity on data recovery and recognize the mechanical fault under under-sampling condition.Inspired by sparse representation theory,the data recovery approaches based on sparse sampling and sparse adaptive algorithm were carried out in this paper.Besides that,the fault classification method based on sparse sampling was developed.The main contents are as follows:(1)The data recovery methods based on sparse sampling was proposed.In actual vibration signal acquisition,some data may be lost due to signal acquisition failure.Aiming at the damaged data in signal acquisition,combined with the compressed sensing framework,a data recovery model based on sparse sampling was established.Firstly,the sparse dictionary was selected according to the waveform characteristics and prior knowledge.Then based orn the unit matrix,the observation matrix was constructed under the missing data model.Finally,the damaged data was recovered by the efficient sparse solution algorithm.Explored the recovery results under different sparse dictionaries,analyzed the sparse characteristics of vibration signals under different dictionaries.The efficiency of the proposed method was validated by simulation signal and practical signal.The experimental results showed that the recovery signal is more suitable for fault diagnosis.Compared with traditional data recovery methods,the proposed method has more advantages in data recovery accuracy.(2)The data recovery method based on sparsity adaptive algorithm was proposed.The data recovery method with compressed sensing(CS)theory was served for damaged signal.However,the sparsity of vibration signal is unknown,which impedes its application.Aiming at the problem mentioned above,a sparsity-adaptive data recovery method was developed for the data reconstruction.Discussed the performance of sparsity adaptive matching pursuit algorithm(SAMP)under different iteration step and termination coefficient,analyzed its application conditions.However,the SAMP algorithm are greatly influenced by termination condition,which will lead to unsatisfied results.In this case,the modified SAMP algorithm based on termination criterion was proposed.The efficiency of the proposed method was validated by simulation signal and practical signal.The experimental results showed the modified algorithm has better performance on reconstruction accuracy and computing efficiency.Besides that,the modified algorithm also outperforms orthogonal matching pursuit(OMP)and regularized orthogonal matching pursuit(ROMP).(3)The fault classification method based on sparse sampling is proposed.The Shannon sampling theorem lays the foundation for signal processing technology,but also brings much pressure for data transmissing and processing.Thus,a fault classification method based on sparse representation was studied to realize mechanical fault diagnosis under undersampling.Discussed the sparse representation method based on redundant dictionaries,and established the sparse representation classification(SRC)model based on redundant dictionary.In the case of the problem that the redundant dictionary of SRC is difficult to be constructed,a sparse representation method based on wavelet transform modulus maxima(WTMM)transform was performed.Therefore,the complex redundant dictionary was replaced by the unit matrix,which reduced the difficulty of the method.In addition,to overcome the influence of signal time shift on the classification results,the-maximum cross-correlation was used as the classification criterion.The efficiency of the proposed method was validated by bearing and gear signal.Compared with other classification algorithms,this method does not need to design classifiers or calculate feature parameters,which avoid the influence of improper parameters on classification results.