Wind Turbine Fault Prognostics Based On Deep Learning

With the proposal of “Industry 4.0” and “Made in China 2025”,China has accelerated its industrial strategy with intelligent manufacturing as its core.In the field of machinery,wind turbine equipment is developing towards high-precision and high-efficiency.With the continuous development of data acquisition and storage technology,massive data has promoted fault prognostics into the era of “Big Data + Intelligence”.Traditional methods of fault prognostics are difficult to process and analyze large amounts of data with low density,diversity,and timeliness.Due to multi-layer nonlinear mapping ability,deep learning(DL)can adaptively extract and mine deeper internal features of data compared to traditional techniques.Thereby,the abnormal symptom information and the health state of the wind turbine can be intelligently recognized by DL.In this way,not only the number of wind turbine equipment failures can be effectively reduced,but also major failures can be avoided,and the power generation performance and operational reliability of the wind turbine can be improved.This paper takes the wind turbine as the research object,and carries out the following specific research.Fault prognostics of wind turbine blade based on DL.The deep network is carried out to extract features from the original unbalanced data,and the ensemble classifier with strong classification ability is selected to identify the fault based on those extracted features.One innovatively applies the proposed model to fault prognostics of the blade icing.Results show the ensemble classifier has better performance for fault recognition with the learned feature of the deep Boltzmann machine.However,the performance of ensemble classifier with learned feature of convolutional neural network and recurrent neural network has declined.Fault prognostics of wind turbine toothed belt evolution based on DL.From normal operating state to the complete failure state,it is a slow process of fault evolution for tooted belt.The prognostics of fault severity evolution can be realized by extracting and identifying the state parameters of the evolution process adaptively with convolution neural network.The proposed model is applied to tooth belt fault evolution prognostics.Results show that the model can effectively identify the fault states of tooth belt at different stages.Fault prognostics of gearbox based on DL.Firstly,based on the correlation between the operating parameters and bearing temperature of wind turbines gearbox output bearing,the main relevant operating parameters are selected.Secondly,the recurrent neural network which is most suitable for dealing with time series problem is selected to predict the trendy of output bearing.Then,the abnormal warning of gearbox fault is realized according to the threshold calculated by residual distribution.The experimental results of a wind farm show that the proposed model can accurately predict the output bearing temperature and realize early fault warning according to the threshold value.