复合材料结构件缺陷的红外成像检测机理研究

Due to unique advantages such as contactless imaging, real-time defective region detection, capable of large area monitoring, infrared thermography provides an effective technique for nondestructive testing (NDT) of composite components. However, it is a challenging task to perform quantitative evaluation of detected defects, which still remains the bottleneck for applications of thermal NDT. In this research, effective infrared signal processing techniques are proposed, as an effort to both improve the quality of capture thermal data and interpret input data to compute accurate defect diagnosis results. Firstly, we systematically investigate non-uniform heating effect in the process of active thermal pulse stimulation, and further develop an effective non-uniformity correction method for temperature shifts compensation, so as to improve quality of input thermal data for further analysis. The proposed method makes use of a novel guided filtering algorithm to extract gradient components caused by non-uniform heating effect and then computes the optimal correction model by fitting the derivatives of correction model to the refined image gradients. Then, we plan to develop an effective solution for accurate temperature measurement of complex structural components. We study pixel-wise data fusion of 2D infrared image and 3D depth information to produce 3D thermal model in real-time. Then temperature distributions of the target are re-built to compensate surface emissivity variations based on 3D viewpoint normalization. Finally, we study effective solutions to analyze infrared image sequence for quantitative evaluation of composite defects. A multi-layer Pulse-coupled neural networks (PCNN) model is proposed to process image sequence in both time-domain and space-domain for small infrared target detection. Irregular temperature variances caused by defects are accurately detected and temporally tracked to calculate important quantitative evaluation results of defects in composite components.

红外热成像检测由于非接触、快速诊断、检测面积大等优点应用于复合材料构件的无损检测，但存在缺陷量化诊断精度不高的瓶颈问题。为此，本项目旨在探索红外图像的有效处理分析方法，在提升采集红外温度信号质量的同时探索复合材料缺陷诊断定量化的有效途径。首先，针对红外检测过程中的外源加热空间不均匀的问题，分析其形成机理和导致信号偏移的形态规律，提出基于温度场梯度信息最小差异拟合和导向性滤波处理的温度场信号偏移矫正方法，从源头上改善采集红外信号的质量。其次，将观察目标的二维温度场信息和三维深度信息进行像素级准确匹配，研究红外温度场的三维视角补差理论和方法，对不同曲率表面的温度信息进行补偿和重建，实现对复杂结构件温度场的准确测量。最后，研究基于多层脉冲耦合网络的时间域和空间域混合信号分析算法，对动态温度场中构件缺陷产生的局部温度梯度进行精确定位和时域追踪，在此基础上研究复合材料构件的内部缺陷诊断和定量化方法。

红外热成像检测由于非接触、快速诊断、检测面积大等诸多优点被广泛应用于复合材料构件的无损检测。但存在红外图像信噪比低、检测信息源较单一、缺陷量化诊断精度不高等瓶颈问题。基于红外热成像的复合材料构件无损检测的关键在于如何对复杂构件表面温度场进行快速、准确、动态的测量，以及如何对采集的红外温度场信息有效处理获得复合材料缺陷的定性、定量检测结果。因此该项目积极开展基于红外、可见光、深度多源异构信号处理和融合互补的复合材料构件无损检测技术研究。在红外无损探测过程中的信息采集、信息优化、信息分析三个重要环节进行创新理论研究。首先，探索长波红外成像系统的噪声产生机理和特征分布规律，提出红外信号固定模式噪声的补偿算法和低像素特征重建模型，解决制约红外热成像仪器工作性能的瓶颈问题。其次，探索多源异构数据中互信息特征的构建机制，研究复杂场景中多源异构数据的时空精确匹配方法，支持红外温度测量任务中信息采集方式的维度扩展。最后，探索多源信息的融合互补方法，研究基于大数据深度学习的多源信号特征提取技术，提出在复杂背景噪声干扰下微弱/隐形缺陷检测深度网络模型的有效训练方法。研究成果为提升红外温度信号质量和解决缺陷诊断量化问题提供技术途径，可有效提升红外无损检测技术的应用范围和诊断精度。项目研究成果共发表/录用SCI论文15篇（JCR Q1论文3篇），英文著作章节1章，EI论文2篇。其中1篇论文被Applied Optics期刊选为2018年6月的编辑精选论文（Editor's Pick Paper），1篇论文被Optics Express期刊选为2018 Top Downloaded Articles In Imaging Systems and Displays，并应邀在国内测试领域期刊《振动、测试与诊断》上发表题为“红外热成像信号处理技术的研究进展”的专家论坛特邀论文。相关科研成果获得国内外相关领域专家关注，目前与中国航天科工集团二院二部、华为科技有限公司，法国ULIS（欧洲最大红外芯片制造厂商）积极开展科研合作。