柔性涡流阵列中的远近场信息融合成像检测方法研究

With the development of manufacturing technology, various shapes of functional materials are applied to equipment. Especially in aerospace and railway transportation, minor damage to these materials can cause serious accidents and cause significant loss of life and property. Flaw detection of curve surface is a difficult problem in current nondestructive testing. In this paper, a 3D reconstruction imaging method of defect detection based on flexible eddy current array is proposed for the curve surface structure. In order to locate and visualize the defect in the curve surface structure, it should build mathematic coupling model based on the relationship between the close sensors. The information of the near field and remote field will be taken into consideration to the model so as to ensure the detection result accurately. In order to enhance the resolution rate of the image, the neural net algorithm based method for processing the digital image is proposed. In this method, the eddy current array sensors are considered as neurons. The source pixels are divided into many units and a new image is set up in the method. By using the neural net algorithm, the pixels of the new image is reestimated so as to improve the resolution rate of the defect image. The frequency information of the eddy current array sensors are studied to measure the deep of the defect based on the machine learning algorithm. By combining the deep and the corresponding image of the defect, it designs a 3D image of the defect which could be used to analyze the harmful to the equipment. This project is of great significance for aerospace equipment maintenance thus assisting the development of eddy current non-destructive test on curve surface material.

随着制造技术的不断发展，各种不同形状的功能材料被应用在关键设备当中。特别在航空航天以及铁道交通运输领域，这些材料的微小损伤都会导致严重事故而造成重大生命财产损失。对于任意曲面结构的缺陷检测是当前无损检测技术的一个难点。本项目针对曲面结构的缺陷提出了基于柔性涡流阵列的远、近场信息融合的成像检测。通过对涡流阵列传感器耦合关系建模，融合远场涡流检测技术对缺陷水平面的定位优势及传统涡流（近场涡流检测）对缺陷边沿的检测精度，对缺陷进行准确定位及成像；为了提高成像检测的分辨率，通过构建网络节点模型，设计基于神经网络计算方法的数字图像处理方法，对实测的图像矩阵重新分割并估计；研究涡流阵列信号的频率信息，设计基于机器学习算法的深度测量算法，获取对应图像的深度信息，实现对缺陷的三维重构成像。项目的研究对于指导航空航天、轨道交通等设备维护，助力涡流无损检测技术在曲面异性结构的缺陷检测上的发展具有重要意义。

电磁涡流无损检测技术作为目前检测速度快、检测精度高的前沿科学技术，受到越来越多的学者关注。在针对导电材料进行缺陷检测过程中，电磁涡流无损检测技术能够激发出磁场和热场。由于涡流场与缺陷的紧密联系，磁场和热场作为不同反应这种联系的有效可测场信号，能够为缺陷检测带来更多有利条件。基于此，本项目利用涡流、磁场和热场之间的关系，设计柔性涡流传感检测技术，通过对不同曲面和不同类型缺陷进行了基于电磁涡流技术的成像检测，利用远、近场多场信息，开发了基于机器学习和基于电磁物理模型的缺陷增强检测方法。完成了基于电磁涡流阵列的缺陷成像检测技术、开发了基于涡流磁响应的磁记忆法成像检测技术、开发了基于深度学习的电磁涡流磁光成像检测技术、开发了基于涡流热响应物理模型的机器学习缺陷检测方法，增强了缺陷图像的对比度，提升了缺陷轮廓检测精度，对电磁热成像缺陷检测技术的量化提供了理论和实践依据和支撑。