固体火箭发动机药柱CT图像缺陷分析技术研究进展  

Review on CT image defect analysis technology for solid rocket motor

作　　者：戴俊杰 宣兆龙 李天鹏 胡利青 郭爱强 DAI Junjie;XUAN Zhaolong;LI Tianpeng;HU Liqing;GUO Aiqiang(Army Engineering University,Shi Jiazhuang 050003,China)

机构地区：[1]陆军工程大学,石家庄050003

出　　处：《固体火箭技术》2023年第1期138-146,共9页Journal of Solid Rocket Technology

摘　　要：药柱缺陷是影响固体火箭发动机安全工作的重要因素。为确保固体火箭发动机的安全性,需要对药柱进行无损检测,通过对三种主要的无损检测方法进行比较分析,认为工业CT探伤是目前最有效的检测手段。分别从图像缺陷分割和缺陷特征识别两方面综述传统的缺陷分析方法。深度学习的快速发展为药柱CT图像缺陷分析提供了新的研究思路,而且基于深度学习的方法将提高缺陷分析效率和算法的鲁棒性,为实现图像缺陷分割及缺陷识别一体化提供技术途径。分别从基于目标检测的缺陷特征识别、基于全卷积神经网络的缺陷分割和基于Mask R-CNN的缺陷分析三方面对深度学习在药柱CT图像缺陷分析的应用进行了总结和分析。The grain defects are an important factor affecting the safety of solid rocket motor.To ensure the safety of solid rocket motors,non-destructive testing of the grain is required.Through a comparative analysis of the three main non-destructive testing methods,it is concluded that industrial CT detection is currently the most effective testing methods.Traditional defect analysis methods were overviewed from two aspects:image defect segmentation and defect feature recognition.The rapid development of deep learning provides a new research direction for the analysis of defects in CT images of grain.The method based on deep learning will improve the efficiency of defect analysis and the robustness of algorithm.It provides a technical approach to achieve the integration of image defect segmentation and defect recognition.The application of deep learning in defect analysis of grain CT image was summarized and analyzed from three aspects:defect feature recognition based on target detection,defect segmentation based on full convolutional neural network and defect analysis based on Mask R-CNN.

关 键 词：固体火箭发动机 CT图像 图像缺陷分割 药柱缺陷识别 深度学习 

分 类 号：V435[航空宇航科学与技术—航空宇航推进理论与工程]