Research On Fault Diagnosis Method Of Wind Turbine Transmission System Based On Deep Learning

In recent years,with the increasingly serious problem of energy crisis and environmental pollution,the demand for the development and utilization of renewable and non-polluting new energy is urgent.As a green and renewable wind energy,it gradually appears in the world vision.At present,the wind power generation is developing rapidly,and the total installed capacity of the wind turbine is also increasing year by year.However,due to the poor operating environment and complex working conditions of the wind turbine,the wind turbine equipment is prone to failure.Among all the mechanical faults of wind turbine power generation system,the transmission system,as the key component of wind turbine energy conversion and transmission,has high probability of failure,long downtime and difficult maintenance,which seriously reduces the safety,reliability and power generation economy of the wind farm.Therefore,it is urgent to carry out targeted research on mechanical fault diagnosis of wind turbine transmission system.The vibration signal generated in the operation of the wind turbine transmission system contains important state information.If the vibration signal can be collected in real time during the operation of the wind turbine transmission system,and the vibration data can be analyzed in real time and the fault diagnosis can be made,the corresponding measures can be taken in time to reduce the downtime and the operation risk and loss of the wind turbine.This paper focuses on the research and improvement of the diagnosis accuracy and robustness of the mechanical fault diagnosis model of wind turbine transmission system.The main research content is to construct the deep neural network fault diagnosis model combined with the depth learning theory,which is applied to the transmission system of wind turbines,namely the gearbox and main bearing mechanical fault diagnosis.The main research work is as follows :(1)In aspect of gearbox fault diagnosis,aiming at the problem that the super-parameter tuning of most network models is based on subjective experience and experimental enumeration,and special test elections are carried out for specific diagnostic problems,resulting in weak generalization ability of the model,an improved fault diagnosis method of stacked sparse de-noising auto-encoders network(PSO-SSDAE)is proposed.The structure of the network is determined by using the particle swarm optimization algorithm to adaptively adjust the super parameters of the stacked sparse de-noising auto-encoders network,and the fault identification is carried out by using the Soft-max classifier.Finally,the effectiveness and accuracy of the method are verified by the gearbox dataset of Qianpeng QPZZ-II.(2)In aspect of main bearing fault diagnosis,aiming at the problem of data imbalance caused by small fault sample size,cross-domain adaptive problem under variable load condition and the influence of high background noise pollution,a novel fault diagnosis method(ACGAN-SDAE)is proposed,which combines auxiliary classifier generative adversarial network and stacked de-noising auto-encoders network.The one-dimensional convolutional neural network is used as a generator,and the category label is used as auxiliary information to help the generator capture the actual distribution of the original sample and generate high quality new samples with labels to expand the number of training fault samples.The stacked de-noising auto-encoders is used as a discriminator to identify the authenticity of the input sample and the fault category.During the training of the ACGAN-SDAE,the generator and discriminator are alternately optimized by adversarial learning mechanism,so that the model has significant diagnostic accuracy and generalization ability.Finally,the model is verified by CWRU bearing dataset.The results show that compared with the traditional fault method,this method has higher diagnostic accuracy and robustness.