Method For Monitoring Icing Of Wind Turbine Blades Based On Image Processing

Since the "Thirteenth Five-Year Plan",high-altitude wind power resources in the southern China have been rapidly developed.Compared with the northern regions,these areas have humid winter air,and the wind farm ice problem is prominent,which seriously affects the safe and stable operation of wind turbines.At present,with regard to the monitoring of icing on wind turbine blades,domestic and foreign researches mostly focus on model calculation and sensor monitoring.There are problems such as limited accuracy,high cost,and difficulty in monitoring the wind turbines that have been put into production.The research on the icing monitoring of wind turbine blades from the perspective of image identification and experiment has important theoretical significance and help engineering applications.The main content and conclusions of the paper are:(1)According to the characteristics of the ice coating of the wind turbine blades to be processed and the test results,the choice of image processing noise reduction and enhancement methods is selected and improved,mainly including defogging,grayscale transformation,histogram correction,Gaussian filtering,Five aspects of Plath edge enhancement effectively restore image information and improve image quality.(2)A multi-scale wavelet edge detection algorithm was used to perform edge detection on the icing image of the blade,which strike a balance between accurate edge positioning and noise reduction.The effects of wavelet transform and five traditional edge monitoring methods such as Sobel are evaluated.The results show that multi-scale wavelet edge detection is significantly better in terms of overall visual perception,signal-to-noise ratio,positive detection rate,and quality factor.(2)An ice-covering type recognition method based on the combination of ULBP feature vector extraction and multi-class SVM classification models was proposed,and natural icing image recognition experiments were carried out several times.The results show that when the training sample is large enough,the recognition results agree with the icing types predicted by the environmental parameters,and the recognition accuracy of the three icing types of glaze,soft rime,and hard rime reaches 100%.(3)A method for calculating the blade ice thickness based on the pre-calibration and the true length / pixel coordinate difference was proposed,and several experiments were performed in the natural ice-covered environment.The average measurement error value at short distance(1m)/ long distance(20m)is 0.30mm/7.4mm,and the average measurement error rate is 2.56%/4.58%.(4)A three-dimensional reduction method of ice coating was proposed.Based on the reduced three-dimensional model,the ice coating volume was measured and the ice coating weight was calculated.Multiple measurement tests were performed in a natural ice-covered environment.The measurement error was 2.08 g / 8.78 g / 14.87 g when the ice was slightly / medium / heavy.The error rate was 4.23%/4.75%/7.57%.