Research On Wind Turbine Gearbox Operation Condition Assessment Based On Machine Learning

Wind turbine gearbox is one of the most critical assemblies working in a harsh environment,and suffering variable loads.Therefore,wind turbine gearbox always suffers high failure rates in practice.On the other hand,the actions of operation and maintenance of wind turbines are time-consuming,the cost of which is extremely high.Due to this reason,not only manufacturers but also service providers want to improve the reliability of wind turbine gearbox and reduce the cost of operation and maintenance.In reality,the condition monitoring method is commonly used in wind turbines,alike CMS and SCADA.However,they are just used to collect the data from wind turbines and cannot assess the health status and locate the faults of wind turbines.These systems are also quite expensive for manufacturers.For this reason,we propose to develop a novel method that can assess the health status of wind turbines and identify potential faults.The monitoring data is used to assess the health status using machine learning and fuzzy comprehensive evaluation method.The results of my research show that the proposed method is of considerable significance to master the real-time operational status of the wind turbine gearbox and to identify potential faults.The results are realistic and in line with reality.Hence,the proposed method in this dissertation is promising in engineering application.The main contents of this dissertation are organised as follows:(1)The evaluation indices are obtained by analyzing the monitoring data of wind turbines in the north of China.Meanwhile,the prediction dimensions of each evaluation index are determined using correlation analysis.Cluster analysis on KF and RMS is carried out to determine the number of operating condition levels.Besides,the training data samples of each evaluation index of the prediction model are subjected to dimensionality reduction preprocessing.(2)Prediction models of evaluation indices are established based on support vector machine and BP neural network.In this research,the grid method and particle swarm optimization algorithm are utilized to optimize the parameters of the support vector machine.The data samples are trained and predicted using optimized parameters,and compared with the prediction results of BP neural network.Following this,the best optimization and prediction schemes of each evaluation index are determined via the error comparison analysis.The results show that the support vector machine can obtain higher classification accuracy than BP neural network.(3)The membership function is determined by half echelon and half hill combined subjection functions based on the state level of the gearbox,which could simulate the internal more correctly.Moreover,the rules for solving the combined weight are elaborated using the analytic hierarchy process,the variable weight theory and the equalization function.Then using the forecast data,measured data,threshold value and fuzzy comprehensive theory of the evaluation indicators,a real-time evaluation model for the operational health status of wind turbine gearbox is established.(4)According to the historical data of a wind turbine operating status,the combined evaluation method and the traditional evaluation method are used to evaluate the operating status of the wind turbine gearbox to verify the scientificity and accuracy of the combined method.The results show that the degradation degree based on the prediction data can overcome the defects caused by the fixed alarm threshold in the SCADA system;the combination evaluation method gives prominence to the fault information hidden by the small constant weight value.It can accurately detect the faults of wind turbine earlier.