The Method Research Of Main Bearing Health Condition Evaluation Based On SCADA Data Of A Wind Turbine

Because of bad operation environment,imperfect operation control strategy and defects in design and installation,the cost of operation and maintenance is high in wind farms.In recent years,wind power technology has been greatly developed,with the application in wind power industry of mechanical engineering,artificial intelligence and other disciplines.People have began the study of the health condition evaluation of the wind turbines.In order to achieve the early fault warning,adjust the operation and arrange the maintenance reasonably,and reduce the cost of operation and maintenance.According to the statistics,the downtime and maintenance costs caused by the main bearing failure are high.At present,most methods assess the main bearing health by measuring the temperature,and then compare it with the preset threshold.However,wind turbine has the characteristics of complexity and variability due to the influence of wind.environmental temperature,electrical and other factors.The preset threshold method can not meet the requirements of the health condition evaluation of the wind turbine main bearing.In this thesis,a health condition evaluation method of main bearing based on Supervisory Control And Data Acquisition(SCADA)data of a wind turbine is studied.The method is divided into off-line part and online part.Each part includes three processes:SCADA data preprocessing,main bearing temperature prediction and health condition evaluation.The off-line part is used to establish the health condition evaluation model of the main bearing,and the online part can achieve the health condition evaluation of the main bearing.The main contents and conclusions are as follows:(1)Preprocessing of SCADA data is divided into five steps:data integration,data reduction,data cleaning,data transformation and data partitioning.The data used in the thesis are derived from a single data source(SCADA system),which simplifies the process of data integration.The data reduction uses the Pearson correlation coefficient method to accurately screen the health state variables of the main bearing.The data cleaning uses the modified Optimal Interclass Variance(OIV)algorithm to get the normal data sets of the related variables.The data transformation uses the minimum maximum normalization method to convert data format into a format suitable for the modeling.The data partitioning uses wind speed division method to identify the operating conditions of the main bearing.After data preprocessing,the normal data sets of main bearing health state variables can be obtained.(2)The temperature of main bearing is predicted.Using Particle Swarm Optimization(PSO)BP neural network algorithm,established the temperature prediction model of main bearing.By comparing with other algorithms,it is concluded that PSO-BP neural network algorithm has stronger global search ability,faster training speed and higher prediction accuracy.(3)The health condition evaluation of the main bearing is evaluated.The fuzzy comprehensive evaluation method is applied to establish the health condition evaluation model of main bearing,and an example is analyzed.The results show that the health condition evaluation method proposed in this thesis,can detect the potential failure of the early main bearing,and track the trend of deterioration of main bearing more effectively.