针对钢结构焊缝存在夹杂、气孔、裂纹等缺陷与损伤、聚焦超声衍射时差焊缝检测中的噪声、回波重叠等难题，考虑超声衍射时差信号中仅含有少量回波而具有稀疏性，本项目建立基于信号稀疏表示和超声衍射时差法的钢结构焊缝缺陷与损伤的量化检测方法。首先，从超声衍射理论出发，考虑钢材晶粒散射、超声的频率依赖性衰减及频散效应，研究缺陷衍射回波的特征演化与缺陷几何信息之间的映射关系，建立超声回波的参数化模型；其次，提出参数化字典设计方法和基于稀疏贝叶斯学习的超声衍射时差信号稀疏表示方法，实现基于回波特征参数的超声衍射时差信号去噪、重叠缺陷回波的分离与辨识，解决近表、近底面及小尺寸缺陷与损伤的量化表征难题，提高超声衍射时差方法焊缝检测分辨率；最后，提出缺陷方位、大小的量化表征方法和缺陷类别的自动辨识方法。本项目研究将为多种钢材中多类焊缝缺陷与损伤的量化检测、表征提供完备、有效的解决方案，具有重要科学意义和实用价值。

This project aims at developing an ultrasonic time-of-flight diffraction (TOFD) technique in the framework of sparse signal representation for detection and quantitative characterization of defect and damage (porosity, slag inclusion, crack) in welded steel joints with focus on signal de-noising and separation of overlapped echoes, motivated by the fact that TOFD signals usually contain only a few echoes thus possess a sparseness property. First, parameters of defect-diffracted echoes (amplitude, time-frequency features) and their dependence on defect geometry are investigated, based on which parametric models are developed for characterization of the defect-diffracted echoes. Next, method for model-based over-complete dictionary design is proposed, the sparse representation of TOFD signals using sparse Bayesian learning is presented, and strategy for signal de-noising and separation of overlapped echoes is established using parametric information of echoes, which enables the quantitative characterization of near-surface, near-bottom and small-sized defects as well as resolution enhancement of TOFD method. Lastly, based on signal de-noising and echo identification, a technique to quantify the location, orientation, size and type of defects in weld is presented and a method for automatic defect classification is proposed. It is expected that the research in this project provide a complete and effective solution to the quantitative detection and characterization of multiple types of weld defect for various types of steel. The success of this proposal will have high scientific and practical values to our society.