Research On Fault Diagnosis Method Of Rotating Machinery Based On Signal Sparse Representation

Rotating machinery,as an important part of the mechanical structure,plays a vital role in the operation of equipment.However,the bad working environment easily leads to mechanical failure.At the same time,the weak faults of rotating machinery are not obvious and accompanied by strong background noise,which makes the weak faults not easy to be detected and brings many difficulties to the fault recognition.Aiming at the problems of weak fault identification and health condition monitoring of rotating machinery,the fault signal features are extracted based on sparse representation,based on sparse representation theory and extreme learning machine theory,fault pattern recognition and operation state monitoring are studied as follows:(1)In order to solve the problem that the parameters of wavelet dictionary are difficult to determine,a fault impulse matching method is proposed to identify wavelet parameters,after calculating the similarity between each subsignal and wavelets,the similarity sequence is obtained.Finally,the sequence is evaluated by kurtosis,and the wavelet parameters corresponding to the maximum kurtosis are selected to construct the dictionary.The method is applied to simulation signals and fault case signals,and the results show that the fault impulse matching method can identify the wavelet parameters quickly and accurately,and the wavelet dictionary can extract the fault impulse waveform obviously and diagnose the fault.(2)The fault impact matching algorithm is a ergodic algorithm,the step length is difficult to determine,and the calculation time is still long.An adaptive wavelet dictionary construction method is proposed for the fault diagnosis of rotating machinery by introducing the water cycle optimization algorithm into the diagnosis process.Firstly,the wavelet parameters are adaptively determined based on the fault shock matching method and the water cycle algorithm,then the orthogonal matching pursuit(OMP)is used to extract the fault shock waveform,and finally the fault type is diagnosed by envelope spectrum analysis.The diagnosis process reduces the humanintervention,avoids the use of prior knowledge,and increases the computational efficiency.The method is validated by bearing fault experiment and Gear Fault experiment.The experiment shows that the method can extract the fault feature and identify the fault type accurately.(3)A fault pattern recognition method based on sparse theory and extreme learning machine theory is proposed.Taking the training set as the learning sample of K-Singular Value Decomposition(K-SVD)dictionary,an adaptive dictionary is obtained,and the sparse features of each sample are extracted by using OMP method based on the dictionary.With sparse features as input,the classification models of multilayer extreme learning machine get better accuracy and can be used for fault pattern recognition,and the regression model of multilayer extreme learning machine can describe the bearing deterioration state well,it can be used for fault condition monitoring.