Application and detection of generative adversarial network-based surface defect generation on blade surface

Wind turbine generator equipment is mainly distributed in more remote and harsh environment areas. The paddle of wind turbine operates all day for a long time and suffers from wind and sand attacks, resulting in paddle surface damage. It has become mainstream to further study detection algorithms on the surface defects of the object in place of manual inspection. After the fieldwork and the dataset making, it is found that there are less actual data samples collected, which is not enough to make the detection model training make the training of the detection model better. In this paper, in order to solve the problem of insufficient defect data required for the training of the detection model, a defect generation method combining the Transformer idea and the generative adversarial network is proposed to generate a large amount of defect data closer to the real distribution. And then the generated data is combined and divided with the original data, and trained and tested by the YOLOv7 detection algorithm. The experimental results show that after the training of the data samples generated in this paper, the mAP value of the detector reaches 87.8% in the training results, the recall rate increases by 1.6%, and the final test precision rate improves from 80.4% to 84.1%, and the leakage rate is 0%. Compared the results after training with the original data, it improves the accuracy and stability of defect detection.