Fault Detection Of Wind Turbine Transmission System Based On Signal Decomposition And Data Drive

The load borne by the drive system of wind turbines is time-varying and impact,which leads to the high incidence of failure of the drive system.It is of great practical significance to diagnose the fault of wind turbine transmission system.In actual working conditions,the vibration signals of the key components of the transmission system generally contain strong noise,the frequency components are complex,and the structure and working mode are special,so the fault diagnosis of the transmission system of wind turbines is difficult.By analyzing the fault types and mechanism of the key components of the wind turbine transmission system,it is known that the extraction of fault features and construction of an effective fault identification model are helpful for fault diagnosis.Therefore,based on the two aspects of signal analysis and data driving,and focusing on signal decomposition,feature processing and sample data generation,an adaptive extraction method of fault feature frequency of wind turbine transmission system was proposed,and a multi-source and multi-perspective feature set was constructed that could characterize the state of key components of the transmission system.According to the fault mechanism,the fault diagnosis method for the key components of the transmission system under the condition of sample imbalance is established.The main contents are as follows:When the signal fault information is extracted by resonance sparse decomposition,there will be parameter uncertainty and post-decomposition component aliasing.Aiming at this problem,a resonance sparse decomposition method for optimization and reconstruction is proposed.In this method,an adaptive parameter selection strategy is constructed to reconstruct the signal subbands according to the time-frequency characteristics.Experimental results show that this method can effectively denoise signals,separate different signal components,and accurately extract fault characteristic frequencies of planetary gearbox.The variational mode decomposition and blind source separation algorithm were extended to the fault diagnosis of the planetary gearbox under the conditions of complex frequency component and unknown mixing method.The preprocessing process of the proposed extended variational mode decomposition algorithm makes the signals satisfy the narrow-band characteristic hypothesis of the algorithm,realizes the adaptive decomposition number,and effectively separates the signal components with similar center frequencies but different resonance characteristics.By constructing virtual channels and reducing the dimension of the original signal,the BSS extended algorithm makes the original signal meet the application conditions of the algorithm.Simulation and experimental results show that the two extended algorithms can effectively improve the accuracy and convergence speed of signal decomposition.Aiming at the problem that single source information can not fully reflect the running state of the equipment,the fault diagnosis method based on multi-source and multi-perspective information fusion is studied.Multi-source feature sets are constructed to explore the potential relationship between feature variables.The feature level and decision level fusion were carried out by using random forest and evidence theory,and the method was applied to the state recognition of rolling bearing faults.The results show that the constructed multi-source and multi-perspective feature set improves the recognition performance of the model,and the fusion of feature level and decision level further improves the accuracy of fault diagnosis.In order to solve the problem of low accuracy of classifier caused by unbalanced samples in deep learning,generative adversariant neural network,information fusion and transfer learning methods are adopted to solve the problem of insufficient samples in the two states of fault-free samples respectively.At the same time,lightweight network and migration model are adopted to solve the problems of high complexity,long time consuming and large memory of traditional deep learning model.