Rao Jing, Tao Yangji, Sun Yan, Miao Cunjian, Wang Wenlong
Key Laboratory of Precision Opto-mechatronics Technology of Education Ministry, School of Instrumentation and Opto-Electronic Engineering, Beihang University, Beijing 100191, China; School of Engineering and Information Technology, The University of New South Wales, Canberra, ACT 2600, Australia.
Zhejiang Academy of Special Equipment Science, Hangzhou 310020, China.
Ultrasonics. 2023 May;131:106930. doi: 10.1016/j.ultras.2023.106930. Epub 2023 Feb 4.
Accurate detection and characterization of defects in high-density polyethylene (HDPE) pipe materials are very important in assessing the structural integrity of critical structures in the nuclear industry. One specific challenge here is the presence of the viscoelastic attenuation of this material, which can lead to resolution degradation and loss of detail in ultrasound imaging. In this work, an effective ultrasonic imaging technique using the least-squares reverse time migration with preconditioned stochastic gradient descent (LSRTM-PSGD) is developed to improve image quality. Compared with standard ultrasonic imaging methods which only consider the direct ray path of ultrasound, least-squares reverse time migration (LSRTM) is a powerful wave-equation-based approach and it has the ability to account for rapid spatial velocity variations and to utilize all wavefield information. The LSRTM is an inversion method, which iteratively updates the reflectivity model by minimizing the modeled data and measured data. The proposed LSRTM-PSGD combines the advantages of stochastic gradient descent (SGD) and adaptive learning rate. The SGD updates the parameter on each transmitter and the fluctuation of SGD can enable it to reach a better minimum, thus improving the imaging quality. Compared with the conventional LSRTM algorithm using a fixed step size, the proposed LSRTM-PSGD algorithm can use the adaptive moment estimation to calculate the adaptive learning rate for the parameter, thereby updating the parameter appropriately. The performance of the LSRTM-PSGD algorithm is tested with experimental data. The results show high-quality reconstructed images with good resolution for defect identification in HDPE pipe materials, especially for deep defects.
准确检测和表征高密度聚乙烯（HDPE）管材中的缺陷对于评估核工业关键结构的结构完整性非常重要。这里的一个具体挑战是这种材料存在粘弹性衰减，这可能导致超声成像中的分辨率下降和细节丢失。在这项工作中，开发了一种有效的超声成像技术，即使用带有预处理随机梯度下降的最小二乘逆时偏移（LSRTM-PSGD）来提高图像质量。与仅考虑超声直接射线路径的标准超声成像方法相比，最小二乘逆时偏移（LSRTM）是一种基于波动方程的强大方法，它能够考虑快速的空间速度变化并利用所有波场信息。LSRTM是一种反演方法，它通过最小化模拟数据和测量数据来迭代更新反射率模型。所提出的LSRTM-PSGD结合了随机梯度下降（SGD）和自适应学习率的优点。SGD在每个发射器上更新参数，并且SGD的波动可以使其达到更好的最小值，从而提高成像质量。与使用固定步长的传统LSRTM算法相比，所提出的LSRTM-PSGD算法可以使用自适应矩估计来计算参数的自适应学习率，从而适当地更新参数。使用实验数据测试了LSRTM-PSGD算法的性能。结果显示了高质量的重建图像，具有良好的分辨率，可用于识别HDPE管材中的缺陷，特别是对于深层缺陷。
