Damages Detection of Aeroengine Blades via Deep Learning Algorithms

Abstract

To solve the problem of detecting the damages of aero-engine blades in harsh environments, reduce the aviation safety hazards caused by visual reasons such as careless observation and delayed reporting of blade damages, the detection model of damages for aero-engine blades via deep learning algorithms is proposed in this paper. Firstly, the Gamma correction method is used to process the dataset captured by the borescope to enhance the characterization ability. Secondly, the improved Convolutional Block Attention Module (CBAM) is embed into the head and the end of backbone network of YOLOv7 model. Meanwhile, a branch is added to the channel attention module of CBAM to optimize its network structure. Finally, in order to improve the accuracy and convergence speed, CIOU is replaced by Alpha_GIOU as coordinate loss function in YOLOv7 model, and a new flow chart of detection for aero-engine blade damages is proposed. Detection experiment results demonstrate that the mean average precision (mAP) of improved YOLOv7 model in this paper is 96.1%, which is 1.0% higher than the original model. The improved YOLOv7 module has remarkable effects compared with YOLOv5s, YOLOv4, SSD and Faster R-CNN models. Meanwhile, the improved YOLOv7 model has better generalization performance, which provides a more reliable support for the real-time and visualization of damages detection of aero-engine blades.