Research On Fault Transient Feature Extraction Of Wind Turbine Transmission System Based On Sparse Representation

In recent years,the wind power industry has developed rapidly with the strong support of national policies.However,with the continuous increasing installed capacity of wind turbines,the failures of major components occur frequently due to insufficient operation and maintenance for wind turbines,which has seriously affected the economic and social benefits of wind power.The structure of wind turbine transmission system is complex and it is working in the complex conditions such as alternating load and gust impact in a long time,the key components are prone to failure,so effective monitoring of the operating state of wind turbines is of great significance to ensure the safe operation of the entire unit.The emergence of periodic transient features in vibration signal is an important failure symptom of wind turbines transmission system.Therefore,aiming for the fault transient feature extraction of wind turbines transmission system,the paper studies the sparse decomposition algorithm and dictionary learning method of periodic transient features based on the sparse representation theory,so as to realize the effective faults identification and diagnosis of wind turbines transmission system.The main work of this paper is as follows:(1)Aiming at the problem that the orthogonal matching pursuit(OMP)algorithm will select the wrong atoms,the correlation coefficient between atom and signal is used as the criterion for selecting atoms,and the correlated OMP algorithm is proposed,which avoids selecting the wrong atom and realizes the periodic transient features extraction of rolling bearing fault.(2)Aiming at the problem that the sparsity of randomized OMP algorithm is difficult to be accurately estimated,an adaptive randomized OMP algorithm is proposed.The algorithm achieves the adaptive determination of sparsity by pre-estimation of sparsity,improvement calculation method of residual and atomic index set,and solution of coefficients.By combining with the sliding window,the phenomenon of missed detection of weak transient impulse features is improved,and the periodic transient features extraction of gear fault and rolling bearing fault are realized.(3)Aiming at the problem that the atoms in dictionary obtained by vibration signal and K-SVD algorithm directly are similar to noise and interference under the influence of strong noise,and the periodic transient features are difficult to extract effectively,a transient feature extraction method based on EEMD and K-SVD is proposed.The harmonic interference in the signal and the unrelated atoms in the dictionary are removed by using EEMD and atom clustering method respectively,thus weakening the influence of harmonic interference and noise on feature extraction and realizing the effective extraction of periodic transient features of rolling bearing.(4)A fault diagnosis system of wind turbines transmission system is developed based on LabVIEW,and the algorithms proposed in this paper are used to realize periodic transient features extraction and fault diagnosis of generator rolling bearing of wind turbines.