Fault Diagnosis Of Wind Turbine Generator Bearing Based On Deep Belief Networks

Wind power generation is currently one of the most well received renewable energy power technologies.The generator is the essential equipment for energy conversion.The bearings of the generator are prone to malfunction due to lubrication,installation,operation and other reasons,which affects the operation of the wind turbine.Because the nacelle that the generator is located far from the ground,it is not easy to maintain;On account of the cost of operation and maintenance become more and more expensive,the demand of reliability have become rising as well.Therefore,it is far-reaching practical significance to study fault diagnosis of wind turbine generator bearings which can extend service lifetime,save time and the cost of O&M,and raise efficiency.In response to this problem,the historical operating data of the SINOVEL SL1500 wind turbine generator in a wind farm of Longyuan Company was used as the research object to carry out the fault diagnosis method research.The main work and conclusions are as follows:(1)Analyze the generation principle and fault law of the generator bearing fault of the wind turbine,pick out the feature parameters of time domain,frequency domain and envelope spectrum can reflect the fault.and calculate the fault characteristic frequency.Build a data set based on the history of vibration signals.Prepare for follow-up experiments and research.(2)Use the particle swarm algorithm that the fuzzy entropy standard deviation as fitness functionto optimize the key parameters in the variational mode decomposition algorithm.This method can accurately obtain the optimal value of the parameter through iterative search.(3)Because the vibration signal of the wind turbine generator bearing is noisy and is affected by multiple vibration sources,a signal noise reduction method based on the compound entropy of the variational mode decomposition algorithm is introduced.Through the verification of three kinds of bearing fault data,this method can effectively eliminate the useless components in the signal and has a good noise reduction effect.(4)The deep belief network is used as the fault diagnosis model.Extract the fault feature parameters from the vibration signal after noise reduction and input them into the deep belief network and BP neural network,which proves that the deep confidence network has a good classification ability for bearing fault data.