电磁热成像自适应稀疏微纹分离与量化研究

Accumulated stress has significant impact that resulted in micro-cracks on the near-surface of metal. The width of these cracks can be as small as micrometers and the number is unknown. Comparing with apparent cracks, they are more subtle and dangerous. Forefront research shows that the needs for automated with fast locate, identify and quantify large range of unknown surface and subsurface micro-cracks remain as challenges in nondestructive testing methods for metal components. This proposal is unique and combines the study of electromagnetic thermal analysis in eddy current pulsed thermography (ECPT), physical and mathematical modeling, and intelligent signal and image processing algorithms as well as other interdisciplinary technologies. The aims of the proposed research are to address and provide solutions to some of these urgent issues. In particular, the research will carry out : 1 ) in-depth study into the electromagnetic and thermal characteristics of different scaled metal surface and sub-surface micro-cracks through the ECPT; study the multidimensional features of spatial, time and frequency domain for electromagnetic and thermal characteristics of cracks, and establish the bridge between the physics observation model and the mathematical model of blind source mixing; 2 ) investigate the prior sparse characteristic of ECPT thermal pattern for different state of micro-cracks (such as different number, position, size, orientation, depth and distribution) in both time and frequency domains. Using the Bayesian theory, the proposed research work will derive an adaptive sparse control algorithm for effective identification of micro-cracks and separating these cracks from the metal components in the ECPT images; 3) develop an integration of pattern features and mathematical morphology algorithm for quantification of different micro-cracks. Finally, the natural micro-cracks metal samples served from aviation industry will be conducted to validate the established intelligent ECPT nondestructive testing system. The obtained results will be feedback to each stage of the proposed system which will be refined to render a more efficient overall system.

金属材料近表面积聚的应力会形成极为微小的裂纹，其宽度可小至微米量级并且数量未知。这种隐藏在材料近表面的微缺陷更具隐蔽性和危险性。研究表明，表面与亚表面未知数量微纹大范围快速自动定位与量化成为现有无损检测方法对金属构件健康监测的技术瓶颈。本课题结合脉冲涡流热成像（ECPT）无损检测技术中物理电磁热学分析、物理数学建模以及图像处理算法设计等多类跨学科技术，开展：1）研究金属材料表面与亚表面各尺度微纹在ECPT激励下电磁热分布物理特征；研究各物理特征空间、时序和频域分布特性，建立物理数学盲源混合模型；2）针对微纹状态（微纹数量、方位、大小、取向、深度和分布），研究各类微纹ECPT空间稀疏分布和时频域特征，研究基于贝叶斯理论的自适应稀疏控制时频域微纹分离算法； 3）融合数学形态学算法和盲源模式特征，研究微纹量化方法。针对航空工业核心金属零部件检测需求，建立自然微纹ECPT智能检测系统。

金属材料近表面积聚的应力会形成极为微小的裂纹，其宽度可小至微米量级并且数量未知。这种隐藏在材料近表面的微缺陷更具隐蔽性和危险性。研究表明，表面与亚表面未知数量微纹大范围快速自动定位与量化成为现有无损检测方法对金属构件健康监测的技术瓶颈。本课题结合脉冲涡流热成像（ECPT）无损检测技术中物理电磁热学分析、物理数学建模以及图像处理算法设计等多类跨学科技术，开展：1）研究金属材料表面与亚表面各尺度微纹在ECPT激励下电磁热分布物理特征；研究各物理特征空间、时序和频域分布特性，建立物理数学盲源混合模型；2）针对微纹状态，研究各类微纹ECPT空间稀疏分布和时频域特征，研究基于贝叶斯理论的自适应稀疏控制时频域微纹分离算法； 3）融合数学形态学算法和盲源模式特征，研究微纹量化方法。项目针对研究内容已完成电磁热成像表面亚表面裂纹检测仿真研究、人工缺陷试块方案设计，电磁热成像检测系统激励和传感器设计；结合工业金属人工和自然微纹试件，完成ECPT盲源微纹自动分离与量化系统。重要结果包括：研发了共生式磁轭架构的ECPT设备，可明显提高单次检测铁磁性材料缺陷的可检测面积、裂纹热对比度和辨识率，解决了高分辨率观察疲劳条件下导体材料的微观-宏观变化问题；提出并完成电磁热多物理场耦合效应在空间-时序-频域-状态多维信息中的物理数学模式成分混合模型，提出并完成基于变分贝叶斯的自适应稀疏和张量模式成分特征提取算法，解决疲劳状态评估和裂纹精确定位、特征提取，消除背景和噪声的干扰问题;提出并完成基于F-score的裂纹特征提取和量化性能判定策略，在此基础上提出并完成基于基因直方图的迭代阈值图像分割方法，实现对裂纹的有效自动量化检测；通过实验验证，人工裂纹自动检测可达0.2mm（长）x 0.1mm（宽）x 0.2mm（深），自然裂纹的检测包括微米级的核工业晶间腐蚀裂纹，风机叶片疲劳裂纹，刹车轴冲击疲劳裂纹和钢轨滚动接触疲劳裂纹，F-score平均检出率93%。科学与应用意义：研究成果针对导体构件裂纹电磁热多物理场表征对关键设备的服役性能、失效机理和演化规律、全寿命安全评定等方面进行研究具有重要的现实意义，对多物理场融合的深入挖掘和传感信号的非监督自适应解析有一定的理论价值。对进一步将电磁热成像系统推广，实现智能化、自动化、可评估预测化和图像化的定量无损评估有一定的应用价值。