Fault Diagnosis Of Wind Turbine Gearbox Based On Sparse Filtering With Current Signals

The gearbox is an important mechanical component of the transmission system in the wind turbine.Its operating state directly affects the operating state and efficiency of the whole wind turbine.Gearbox works in serious condition for a long time,so it is easy to break down.Gearbox failure will inevitably lead to unnecessary shutdown and severe economic loss.Therefore,timely and precise fault diagnosis of the gearbox is of great importance to ensure the safe and reliable operation of the wind turbine.At present,the fault diagnosis of wind turbine gearbox is mainly based on vibration signal analysis.However,the vibration signal is easily contaminated by the environment noise,and the installation of the sensor increases the system cost.Compared with the vibration signal,the current signal has the advantages of non-invasion,low monitoring cost,and high stability.Nevertheless,the current signal has a large fundamental frequency component and a low signal-to-noise ratio.The fault information contained in the current signal is easily submerged by the modulation of the fundamental frequency and harmonic components,which makes it difficult to extract the fault features.Based on sparse filtering,the latest achievement of unsupervised learning,a series of methods are researched to learn the features of current signals and realize the high-precision classification diagnosis of the gearbox.The main contents of this paper are as follows:(1)This paper systematically cards the common fault types and causes of the wind turbine gearbox,analyzes the fault detection principle of current signal in detail,and summarizes the performance characteristics of the current signal under different health conditions of the gearbox.(2)A fault diagnosis method based on sparse filtering feature fusion is proposed to mine the features which carry abundant fault information from the current signal.Firstly,a local feature learning network based on sparse filtering is designed to learn different fault features from the original current signal and envelope signal respectively.Then,fusing the features of original signal and envelope signal to increase the diversity of fault features.Finally,the fused features are input to support vector machine to realize the intelligentrecognition and diagnosis of different fault types of gearbox.The effectiveness of the proposed method is verified by the gearbox fault simulation experiments on the wind turbine gearbox test-rig.(3)In order to further enhance the feature learning ability of sparse filtering,multi-view learning and sparse filtering are combined organically,and a multi-view sparse filtering(MVSF)unsupervised feature learning method is proposed based on the local feature learning network.The MVSF method first automatically learns the useful and complementary features from different views of the original current signal,and then increases the complementary of the features by fusing the multi-view features learned in parallel,so as to improve the fault diagnosis performance.The effectiveness of the proposed MVSF method is verified by experiments on the wind power gearbox experimental platform,and the proposed method is compared with the traditional feature extraction method and deep learning method.