Wind turbine blade fault diagnosis technology based on Efficient-YOLOv5s algorithm

Abstract

Abstract The daily operation and maintenance of wind turbine blades in wind power plants requires the application of target detection algorithms that are both accurate in detection and fast in training. For the current wind turbine blade fault detection algorithms in the dataset when the number of samples of the problem of slow training time through the bi-directional feature pyramid network of the fast weighted normalised fusion to improve the training speed, for the small size of the fault detection of the low accuracy of the problem through the bi-directional feature pyramid network of the better use of multiscale information to improve the detection accuracy and through the coordinate attention to generate bidirectional complementary feature maps to enhance the object of interest to further improve the accuracy of detection. In this way, an efficient wind turbine blade fault detection model (Efficent-YOLOv5s) is proposed on the basis of the YOLOv5s model. Through the experimental validation of the homemade dataset, the results show that compared with the YOLOv5s model, the detection map0.5 value and accuracy of the efficient wind turbine blade fault detection model are improved by 0.5% and 0.6%, respectively, and the average training time is shortened by 4s/epoch, which verifies that the model is effective.