Page 116 - 《应用声学》2022年第1期

第 41 卷 第 1 期 Vol. 41, No. 1
2022 年 1 月 Journal of Applied Acoustics January, 2022

⋄ 研究报告 ⋄

焊缝超声相控阵检测数据深度学习降噪方法 ∗

朱甜甜 1,2 刘 建 3 宋 波 1† 桂 生 1 廉国选 1

(1 中国科学院声学研究所 声场声信息国家重点实验室 北京 100190)
(2 中国科学院大学 北京 100049)
(3 中国石油集团工程技术研究院 北京 102206)
摘要：超声相控阵检测技术在焊缝检测中具有广泛的应用。超声相控阵检测技术检测信号中常混入噪声导致
检测成像时难以分辨真实的缺陷特征。这些噪声主要为无关的反射信号和局部相关的结构噪声，传统的超声
图像降噪方法难以有效滤除这些噪声，且存在计算效率低、参数优化复杂等问题。该文提出了一种基于深度学
习的焊缝超声相控阵检测技术检测 S 扫图像的降噪方法，通过搭建深度神经网络降噪模型去除 S 扫图像中的
噪声。经过实验验证，该方法较传统的降噪方法能更有效去除焊缝超声相控阵检测技术检测 S 扫图中的噪声，
保留缺陷的图像细节，并且提高了计算效率，同时避免了人工对不同噪声水平的 S 扫图像进行参数优化。
关键词：超声相控阵检测；焊缝检测；深度学习；降噪
中图法分类号: TP399 文献标识码: A 文章编号: 1000-310X(2022)01-0112-07
DOI: 10.11684/j.issn.1000-310X.2022.01.013

Noise reduction method for weld PAUT detection data based on deep learning

ZHU Tiantian 1,2 LIU Jian 3 SONG Bo 1 GUI Sheng 1 LIAN Guoxuan 1

(1 State Key Laboratory of Acoustics, Institute of Acoustics, Chinese Academy of Sciences, Beijing 100190, China)
(2 University of Chinese Academy of Sciences, Beijing 100049, China)
(3 CNPC Engineering Technology R&D Company Limited, Beijing 102206, China)

Abstract: Ultrasonic phased-array inspection technology (PAUT) has a wide range of applications in weld
inspection, and the PAUT inspection signal is often mixed with noise, making it difficult to distinguish the
true defect characteristics during inspection imaging. These noises are mainly irrelevant reflection signals and
locally relevant structural noise, which are difficult to be effectively filtered out by traditional ultrasonic image
noise reduction methods and have problems such as low computational efficiency and complex parameter
optimization. In this paper, we propose a deep learning-based noise reduction method for S-sweep images
of weld PAUT inspection, and remove the noise in S-sweep images by building a deep neural network noise
reduction model. After experimental verification, the method can remove the noise in the S-sweep image of
weld PAUT detection more effectively than the traditional noise reduction method, retain the image details
of defects, and improve the computational efficiency, while avoiding the manual parameter optimization of
S-sweep images with different noise levels.
Keywords: Ultrasonic phased array testing; Weld testing; Deep learning; Noise reduction

2021-01-12 收稿; 2021-03-14 定稿
∗ 船舶建造焊缝质量数字化检测技术研究
作者简介: 朱甜甜 (1996– ), 男, 浙江金华人, 硕士研究生, 研究方向: 超声检测与信号处理。
† 通信作者 E-mail: songbo@mail.ioa.ac.cn