Research On Fault Diagnosis Of Fan Gearbox Based On Deep Neural Network

With China’s wind power engineering technology matures,wind power development ushered in the spring,at the same time also have a few problems,the gear box is the key transmission parts of the whole unit,failure rate is higher,and it is not easy,is a major cause of affecting the social production and wind power benefits,so the research on fault diagnosis of gearbox and the prediction is of great significance.The paper takes the vibration signal of the gearbox as the research object,studies the fault diagnosis method of the gearbox and does the simulation and experimental verification.In this paper,the structure of gear box in wind turbine is studied deeply,the root cause of the fault is analyzed,the fault types and manifestations of the main parts of the gear box are expounded in detail,and the frequency characteristics of vibration signals when the gear and rolling bearing fail are studied.Because the actual vibration signal collected by the sensor will have noise information,it will affect the accuracy of feature parameter extraction.Therefore this paper proposes an improved wavelet threshold denoising method,by constructing uniform sine signal with gaussian white noise and white noise to simulate the failure of gear vibration signals,signals of gearbox gear fault simulation experiment,the signal-to-noise ratio(SNR)and root mean square error(RMSE)to evaluate denoising effect,to verify the effectiveness of the proposed method.In order to reflect the fault information comprehensively,the feature vector set is composed of time domain feature index and frequency domain feature index when the fault signal is extracted.The feature vectors obtained after feature extraction have redundancy and conflict problems,so it is necessary to reduce the dimension of the feature vector set,and Principal component Analysis(PCA)is introduced to reduce the dimension.The problem of one-sided information of a single eigenvector can be solved when the reduced eigenvector is input into the diagnosis and prediction model of RBF neural network.In order to solve the problem that RBF neural network is easy to fall into local optimum,a genetic particle swarm optimization algorithm(GA-PSO)with selection,crossover and mutation operation of genetic algorithm is used for parameter optimization process of RBF neural network to identify and judge the fault type.Taking the vibration data obtained from the gear box experimental equipment of a domestic wind farm as the original data,the experimental simulation of 6 different types of gear faults is carried out.The results show that the fault diagnosis model based on GA-PSO-RBF neural network is more effective.