Condition Monitoring And Fault Diagnosis Of Wind Turbines Based On Twin Support Vector Machines

The gearbox is a key component in the drive system of the wind turbine,which has a complex structure and is placed inside the nacelle which is far away from the ground.When an abnormality occurs,it is inconvenient to identify and maintain the fault,resulting in a long downtime of the wind turbine and high equipment maintenance costs.Therefore,it is necessary to select an efficient wind turbine gearbox condition monitoring and fault diagnosis method.Based on the classification advantages of twin support vector machines,this paper conducts condition monitoring and fault diagnosis of wind turbine gearboxes.The main contents are as follows:1.A unilateral residual EWMA control chart is constructed to monitor the operating state of gearbox bearing temperature.Due to the autocorrelation of the gearbox bearing temperature collected by the Supervisory Control Data Acquisition and Control(SCADA)system,the use of the residual EWMA control chart can eliminate the influence of the data autocorrelation.According to EWMA control chart,which has the same characteristic of parameter offset direction,a unilateral residual EWMA control chart is proposed.The average running chain length when out of control is used as the evaluation index,and Monte Carlo is used to simulate the evaluation index.When the running process is out of control,compared with the traditional unilateral EWMA control chart,the average running chain length of the unilateral residual EWMA control chart constructed in this paper is smaller.2.A twin support vector machine(IPSO-TWSVM)prediction model with improved particle swarm optimization is constructed to predict the bearing temperature of wind turbine gearbox.To monitor the gearbox bearing temperature using the one-sided residual EWMA control chart,it is necessary to obtain the residual series of the gearbox bearing temperature.Based on this,an IPSO-TWSVM prediction model was constructed to predict the bearing temperature of the wind turbine gearbox.Compared with other prediction models,the IPSO-TWSVM prediction model has stronger fit,higher prediction accuracy and faster convergence speed.Comparing the predicted value of the bearing temperature of the wind turbine gearbox with the actual value,the residual sequence of the bearing temperature of the gearbox is obtained.After calculating the statistics of the control chart,the unilateral residual EWMA control chart is drawn.Compared with the traditional unilateral EWMA control chart,the unilateral residual EWMA control chart constructed in this paper can issue a warning in advance and improve the early warning performance of the control chart in the out-of-control state.3.A genetic algorithm-based twin support vector machine(GA-TWSVM)control chart pattern recognition model is constructed to diagnose gearbox faults.Based on the statistical features and shape features of the samples,three different feature combination schemes are designed.The simulation experiment is carried out through the data sample points of the basic control chart model obtained by Monte Carlo simulation.Compared with other common recognition models,the GA-TWSVM pattern recognition model has advantages in classification accuracy and recognition time.Therefore,the GA-TWSVM control chart pattern recognition model is applied to the gearbox fault diagnosis.By establishing the mapping relationship between the abnormal mode of the control chart and the fault,and according to the identified abnormal control chart mode,the relevant factors leading to the failure of the wind turbine gearbox are determined,and the Lock the cause of the failure.