Aero engine blade damage detection based on improved YOLOv8

It is of great significance to find and identify internal engine damage accurately for engine maintenance. Borehole inspection technology is one of the main means of aero-engine fault detection and maintenance. The traditional borehole inspection relies on manual experience and proficiency， which has a large error and low accuracy. To solve this problem， an aero engine blade damage recognition model based on improved YOLOv8 was proposed. In this study， data enhancement techniques were used to construct the engine blade damage datasets containing damage images and annotation information. The Convolutional Block Attention Module（CBAM） attention mechanism module is introduced in the backbone network to solve the problem that the proportion of blade damage in the image is small and the information is easily lost. The WIoU loss function was used to replace the original function to optimize the model classification performance and improve the robustness of the model. In order to improve the adaptability and generalization ability of the model to different types of damage， DCNv3 deformable convolution module was introduced to construct C2f_DCNv3 module. The results show that the average detection accuracy of the proposed improved model is 92.30% （mAP@50%）， which is 2.4% higher than that of the original YOLOv8n model， 9.57% higher than that of YOLOv5 and 10.75% higher than that of YOLOv6. When the calculation amount of the model is about 72.47% less than that of YOLOv8s， the accuracy is improved by 1.37%， which proves that the improved model has higher comprehensive performance under the condition of taking into account the calculation amount and detection accuracy， and is conducive to improving the automation and intelligence level of aero-engine hole detection.