Remaining Useful Life Estimation For Offshore Wind Turbine Bearings

Offshore wind turbines have been widely installed for wind energy utilization.However,they often break down due to the harsh working conditions.Their operation reliability receives widespread attention.Bearings play the critical role in wind turbines.The failure of bearings may lead to the entire wind turbine cannot run properly.Thus,there is a growing interest in estimation of bearing remaining useful life(RUL).Bearings' current health status can be estimated and their future working status can be predicted via these works.Therefore,wind turbine can be controlled to run safely and catastrophic failures can also be avoided.In this thesis,the current development status of wind turbine industry was briefly reviewed.The role of bearings in wind turbines was introduced and the consequence of bearing failure was analyzed yet.Vibration signals of bearings were analyzed to construct a health index reflecting bearing health status.With the aid of anomaly threshold and moving windows,two different stages(normal working stage and degradation stage)of bearing life were distinguished.These helped to estimate bearing's RUL in the degradation stage.Then two different approaches were proposed to predict the RUL.They are Bayesian framework-based and neural networks-based approaches.In Bayesian framework-based approach,a state-space model was constructed by curve fitting of the degradation data.Parameters of the model were initialized by using Dempster-Shafer method,and then updated by Kalman filter and particle filter algorithms for RUL estimation.In neural networks-based approach,the mind evolutionary algorithm(MEA)was used to optimize the initial weights and parameters of the back-propagation(BP)neural network.The time and health indexes were recognized as the input and output data of neural network in order to train the BP neuralnetwork.The trained neural network was then used to track and predict the health conditions of bearings.The proposed approaches were validated by two experimental bearing life data.The works in this thesis are studied by analyzing of vibration signals,constructing a health index,utilizing Bayesian filtering technique and neural networks.Results show that the proposed approaches can divide different stages in bearing life,track the degradation trend of bearings,and predict their future health status.