航空科学技术2025，Vol.36Issue(2)：29-37,9.DOI:10.19452/j.issn1007-5453.2025.02.002
基于扩散模型的航空发动机叶片损伤智能检测
Diffusion Model-Based Intelligent Damage Detection of Aero-Engine Blade
摘要
Abstract
As an important component of turboshaft and turboprop engines,accurate,fast and reliable damage detection of high-speed rotating multi-layer array blades is of great significance to ensure the safety of aero-engine operation.Aero-engine blade damage is characterized by multiple types and multiple sizes.The traditional borescope inspection method relies on manual experience and visual inspection,which has high labor cost,low inspection efficiency.It is difficult to meet the real-time,high-efficiency and accuracy requirements of the in-situ inspection of aero-engine and maintenance.Existing intelligent detection algorithms usually rely on manually set predefined candidate bounding boxes,which require people to empirically set the number,size,scale and other hyperparameters.Therefore,this paper proposes a diffusion model-based detection(DMbD)method for intelligent damage detection of aero-engine blade,which decomposes the damage detection task into two parts:image feature extraction and damage detection and assessment,where the image feature extraction network uses the residual module to extract feature representations from the input image,and the damage detection and assessment network utilizes the diffusion process to take the noisy boxes as inputs and predict the damage categories and coordinates.Therefore,the network realizes the generation of object boxes directly from the noise candidate boxes and uses them for damage classification and localization,without the need to manually set the parameters of the predefined candidate boxes.DMbD method is applied to aero-engine blade image for damage detection and evaluation,and the experimental results show that this method can quickly and accurately recognize aero-engine blade damage,and the mean average precision(mAP)is 0.988.Compared with traditional methods and existing intelligent methods,the proposed model avoids the manual setting of bounding boxes,and can predict bounding boxes from noise distribution to achieve intelligent damage detection of aero-engine blade.关键词
航空发动机叶片/损伤检测/扩散模型/孔探检测Key words
aero-engine blade/damage detection/diffusion model/borescope inspection分类
航空航天引用本文复制引用
商红兵,杨淇琇,孙闯,王诗彬..基于扩散模型的航空发动机叶片损伤智能检测[J].航空科学技术,2025,36(2):29-37,9.基金项目
航空科学基金(20200046070002) Aeronautical Science Foundation of China(20200046070002) （20200046070002）
