Research On Weak Fault Feature Extraction Method Based On Adaptive Sparse Signal

The mass data in the monitoring process of mechanical rotating equipment will bring a series of problems,such as data storage,data transmission,data mining and so on.Therefore,sparse decomposition of redundant data and extraction of high information and high value information by sparse representation are helpful to ensure the safe and stable operation of rotating machinery.In order to improve the adaptability of sparse complete dictionary,two problems exist in the research of sparse decomposition:(1)the problem of free radical atom acquisition of sparse complete dictionary based on the principle of data-driven;(2)the problem of fixed radical atom acquisition of sparse complete dictionary based on the premise of prior knowledge assumption.In this paper,TQWT(flexible Q-factor wavelet transform)is combined,TQWT)algorithm,from the perspective of self-adaptive,multi-level analysis filter bank,feature matching and so on,has been gradually studied in order to provide a new idea to solve the problem of poor robustness in sparse decomposition diagnosis of engineering practical signals relying on sparse complete dictionary.The main research contents and results are as follows:(1)In order to solve the problem that tqwt algorithm relies too much on prior knowledge to extract free radical atoms,a fault sparse feature extraction method based on adaptive tqwt is developed.In order to extract the free radical atom matching the sparse complete dictionary,the quality factor can be adjusted according to the characteristics of the signal in the tqwt decomposition,and the multi-resolution decomposition of the signal is realized based on the maximum kurtosis principle.Based on this,free radical atoms can be extracted from subbands without reverse reconstruction.(2)By iterating the filter banks of each order of tqwt algorithm,the adaptive decomposition of the target signal is realized.Based on this,the method of fault sparse feature extraction with single and double quality factor tqwt is studied.Based on the decomposition of single channel tqwt analysis filter,a dual quality factor tqwt decomposition method with high and low quality factors is introduced.In the optimized subband acquisition,the oscillating shock component is extracted as the free radical atom,and the self-adaptive complete dictionary consistent with the research target signal dimension is constructed through the extension of topspritz,which provides the research strategy basis for the sparse complete dictionary construction based on the fixed atom.(3)In order to ensure that tqwt sub-band can also get enough oscillation impact information without traversing the quality factor,a sparse decomposition method based on the combination of tqwt algorithm and VMD is developed.Through the idea of step-by-step optimization of signal decomposition,the sub-band obtained after tqwt decomposition is taken as the initial signal,which is further decomposed by VMD method,so as to achieve the effect of noise suppression to optimize the sub-band signal,and achieve the purpose of obtaining more ideal fault impact feature components.At the same time,it improves the research framework of TQWT.(4)In order to solve the problem that it is difficult to match the fixed base atom with the signal impact feature in sparse decomposition,a quality factor sparse fault feature extraction method based on correlation criterion is proposed.Based on the prior knowledge hypothesis,that is,the theoretical hypothesis that the quality factor can represent the impact characteristics in the fault signal,a rich sparse atom sample database is constructed under different parameters by using the oscillation characteristics of the quality factor,so as to ensure that the sparse atoms can adapt to the vibration impact characteristics of the signal.In the selection of the optimal atom,the correlation criterion is introduced,and the sparse complete dictionary based on the quality factor is constructed to obtain the sparse signal,which is verified in the diagnosis experiment of cross speed and cross fault mode.