Fault Detection And Seperation Approach Based On Compressed Sensing

A large amount of vibration signals will be measured based on the traditional Shannon sampling principle during condition monitoring and fault diagnosis of mechanical equipment.It increases the burden of data storage,transmission and fault feature extraction.Moreover,the interaction between multi source coupling signals under underdetermined condition intensifies the difficulty of fault feature extraction.To solve these issues,the approaches with regard to the data storage and transmission,fault feature extraction and extraction of multi source coupling fault were carried out.The details are presented as below:(1)To solve the problem concerning the data storage and transmission of massive vibration signals,the reconstruction technique using sparse samples was performed.The requirements to achieve the accurate reconstruction using compressed sensing were discussed.Then,the compressed-sensing-based signal reconstruction approaches using different sparse dictionaries were carried out.To discuss the difference of vibration signal reconstruction results under different sparse dictionaries,the Fourier dictionary and wavelet dictionary were applied to investigate the sparsity of vibration signals,respectively.The conclusion can be drawn that the vibration signals are sparser under the Wavelet dictionary and its reconstruction effect are more accurate.However,the reconstruction results are random due to the undetermined sparsity of vibration signals.To deal with this problem,the vibration signal reconstruction method based on the block sparse Bayesian learning was studied.It can achieve the accurate reconstruction of vibration signals through knowledge of probability statistics.The results indicate that this method can achieve the vibration signal reconstruction without consider ation of sparsity.(2)To extract fault features from the large amount of vibration signals,the compressed-sensing-based fault diagnosis teniques were studied.The fault feature extraction approach based on the harmonic signal detection was investigated by analyzing the sparsity of harmonic signals under the Fourier dictionary.It is difficult to complete the construction of sparse matrix due to the large length of vibration signals,the symptom parameter wave can be constructed based on time-domain characteristic parameters,through which the vibration signals can be reduced with the faulty information remained.Then,the fault features can be extracted by detecting the harmonic signals.It is well-known that the existence of noise intensifies the difficulty of fault feature extraction and the inadequate sparsity of vibration signals.Thus,the fault diagnosis from compressed samples using tunable Q-factor wavelet transform was developed.The impact signal and noise can be separated effectively through tunable Q-factor wavelet with the help of spectral kurtosis,which can determine the Q-factor.Then,the fault features can be also extracted by detecting the harmonic signals.(3)To separate the multi source coupling faults under underdetermined condition,the compressed-sensing-based fault separation strategy was proposed.A unified model of underdetermined blind source separation based on compressed sensing was constructed via analyzing the equivalence between compressed sensing theory and blind source separation method.The number of source signals can be determined through potential function and the mixed matrix can be estimated via K-means.Then,the measurement matrix can be acquired by matrix transform and the fault separation and diagnosis can be achieved optimization strategy of compressed sensing.Then,the separation and synchronization detection method of multi source coupling signal was proposed on the basis of the above-mentioned approaches.Without complete reconstruction of source signals,it can extract the fault features by harmonic signal detection using the optimization method of compressed sensing.The simulated signal and vibration signals were used to validate the effectiveness of the proposed methods.