基于ICT 的大尺寸工业构件图像重建及缺陷检测研究

Defect detection for large industrial components has been one of the important problems in the field of aeronautics and space. Traditional nondestructive testing (NDT) methods, such as ultrasonic, vortex flow, magnetic particles, permeate test and so on, cannot effectively detect defect and its expansion on large industrial components. Compared with other NDT methods, Industrial CT (ICT) has an unparalleled advantage and is regarded as the best NDT method in the world. Aiming at the problems of defect detection for large industrial components, a novel method for defect detection based on ICT image sequence is proposed. Some key problems will be studied via theory analysis, simulation and real CT scanning experiments. .Aiming at the incomplete projection data of large industrial components, we will propose a fast and parallel image reconstruction method for cone beam volume integral model of algebraic reconstruction techniques (ART) algorithm. Also, we will solve the problem of choosing the optimal relaxation factor utilizing neural network, and realize fast image reconstruction for cone beam ART utilizing parallel computation technique of GPU and multi-core. Based on the above research, we will realize the defect detection and monitoring for large industrial components via neural network, image registration and digital subtraction. The research of this project will realize NDT for large industrial components of aeronautics and space, and will have great significance in ensuring their safety and reliability.

大尺寸工业构件的缺陷检测是航空航天领域迫切需要解决的重要课题之一。常规无损检测方法无法对工业构件的缺陷及其扩展进行有效检测。ICT技术具有其他无损检测方法无法比拟的优越性，被视为当今世界最佳无损检测手段。针对大尺寸工业构件缺陷检测问题，提出基于ICT图像序列的缺陷检测新方法，并通过理论分析、模拟仿真和CT扫描实验相结合的方法对若干关键问题进行研究。针对工业构件非完整扫描投影数据，提出锥形束体积分模型快速并行图像重建方法，实现基于神经网络的最佳松弛因子选择；在此基础上，通过图像配准及数字减影等方法实现缺陷的检测与监测。本项目的研究对于实现航空航天产品等大尺寸工业构件的无损检测，确保其安全性和可靠性等具有重要意义。

本项目围绕大尺寸工业构件的图像重建及缺陷检测展开相关研究。在图像重建方面，针对二维SART算法，提出了一种基于查找表的快速反投影方法，与传统投影及反投影方法相比，取得了约6倍的总体加速比；针对三维SART算法，提出了一种线积分模型快速三维前向投影算法，投影计算时间与Siddon算法相比提高了约18倍，总体图像重建时间提高了近8倍；在此基础上，通过多核并行技术在6核处理器上取得了约5.9倍的重建加速比。这些研究成果对于实现大尺寸工业构件的快速高质量图像重建具有重要意义。在缺陷检测方面，提出了一种基于CT图像序列的大型工业构件缺陷检测方法，首先对工业构件扫描重建得到一副原始基准图像，在后续不同时刻对构件进行扫描重建得到系列CT切片图像，并将这些图像与基准图像进行配准，然后与基准图像进行数字减影，得到一系列减影图像，通过减影图像可以对缺陷的产生及扩展进行检测。研究结果表明，本项目提出的缺陷检测方法是可行的，对于确保工业构件的安全性和可靠性等具有重要意义。在ICT仿真方面，首先针对二维Joseph算法提出了一种快速投影计算方法，然后将该方法扩展到三维，提出了三维Joseph算法快速投影计算方法，在保持投影精度的同时，与传统方法相比取得了约2.3倍的加速比，其研究对于离散体素模型快速投影仿真具有重要意义。