A Defect Detection Method for a Boiler Inner Wall Based on an Improved YOLO-v5 Network and Data Augmentation Technologies

During the long-term operation of a coal-fired boiler, some defects of its inner wall are unavoidable. The traditional manual detecting method is time-consuming and not safe for maintenance engineers. In this paper, we propose an automatic detection method to deal with inner wall defects based on an improved YOLO-v5 network and data augmentation technologies. Specifically, some shallow features and original deep features are fused on the basis of the original YOLO-v5 network for the small objects. Meanwhile, a squeeze-excitation (SE) attention module is added behind the network’s backbone to improve the feature extraction efficiency of the network, and a varifocal loss function is adopted to make it easier for the network to detect those dense objects. Moreover, 176 images including four types of typical inner wall defects (castables falling off, anti-wear layer damage, perforation and bruise) are collected from a power plant boiler, and five data augmentation technologies are introduced to increase the number of samples. The experimental results demonstrate that the proposed method can effectively detect various defects of a boiler inner wall with a satisfactory accuracy, and bring a great facilitation to the maintenance of a power plant.