摘要
固体火箭发动机(SRM)作为现代军事和航空航天领域的关键动力装置,在储存和运输过程中,药柱易受载荷和环境影响,导致裂纹、气泡和脱粘等缺陷。这些潜在缺陷可能会影响SRM安全服役。基于X射线的计算机层析成像(CT)技术因能够提供详细的内部结构图像,是评估药柱健康状态及安全服役能力的有效手段。鉴于SRM缺陷样本稀缺的挑战,该文提出一种适配于SRM药柱CT图像无监督范式的基于分数生成模型的缺陷检测与定位算法。该算法通过模拟正向和反向扩散过程,精确采样复杂分布以实现缺陷检测。实验结果表明:该算法在脱粘、裂纹和气孔缺陷检测任务中均获得良好的性能,缺陷检测精度达到95%以上,缺陷定位精度达到86%以上。该算法在样本效率上的优势,以及在复杂场景下的鲁棒性和稳健性,使其在SRM质量控制和故障诊断方面具有广泛的应用前景。
As a key power device in modern military and aerospace fields,the grain of solid rocket motor(SRM)is susceptible to load and environmental impact during storage and transportation,resulting in defects such as cracks,bubbles and debonding.These potential defects may affect the safe service of SRM.X-ray computed tomography(CT)technology,capable of providing detailed internal structural images,is an effective means of assessing the health status and safe service capability of grains.Given the challenge of the scarcity of SRM defect samples,this paper proposes an unsupervised score-based generative model algorithm tailored for SRM grain CT images for defect detection and localization.The algorithm achieves defect detection by accurately sampling complex distributions through the simulation of forward and reverse diffusion processes.Experimental results indicate that the algorithm performs well in detecting cracks,bubbles and debonding,with a defect detection accuracy of over 95%and a defect localization accuracy of over 86%.Its advantages in sample efficiency,along with its robustness and stability in complex scenarios,suggest that it holds significant applicable in SRM quality control and fault diagnosis.
作者
李毅红
孙雪琴
陈平
LI Yihong;SUN Xueqin;CHEN Ping(School of Mathematics,North University of China,Taiyuan 030051,China;State Key Lab of Extreme Environment Optoelectronic Dynamic Testing Technology and Instrument,North University of China,Taiyuan 030051,China;School of Information and Communication Engineering,North University of China,Taiyuan 030051,China)
出处
《中国测试》
北大核心
2025年第8期147-154,共8页
China Measurement & Test
基金
国家重点研发计划资助项目(2023YFE0205800)
国家自然科学基金资助项目(62201520,62301508,62301507,U23A20285,62471442)
山西省自然科学基金资助项目(202303021222096,202203021222052,202303021222094,202303021212207,202303021211149)
山西省重点研发计划资助项目(202302150401011)
中央引导地方科技发展资金项目(YDZJSX2024D037)。
关键词
固体火箭发动机
CT图像
缺陷检测
分数生成模型
无监督学习
solid rocket motor
CT image
defect detection
score-based generative model
unsupervised learning
作者简介
李毅红(1983-),女,山西临汾市人,副教授,博士,研究方向为无损检测、CT成像等;通信作者:陈平(1983-),男,安徽池州市人,教授,博士,研究方向为无损检测、X射线成像探测等。
