iToFD: An Automated Unsupervised Framework for Weld Defect Detection and Measurement Using Time-of-Flight Diffraction Data

Abstract Time-of-flight diffraction (ToFD) is a widely used nondestructive testing method in manufacturing industries due to its versatility in detecting various weld defects, applicability across a wide range of materials and thicknesses, and portability. However, ToFD data analysis poses significant challenges, including difficulty in distinguishing defects from noise and need of evaluating large volumes of data. Current manual inspection techniques are labor-intensive, prone to errors, and time-consuming, often requiring 8–10 h of analysis for a 15-m weld seam. Existing automated approaches are based on supervised learning and lack generalizability across diverse datasets. Furthermore, image-based interpretations exhibit poor precision, as accurate defect measurements require fine identification of the signal data peaks. Therefore, this article presents a robust unsupervised methodology that utilizes advanced signal processing and adaptive dynamic thresholding to detect both small, isolated flaws and continuous weld defects. It also provides precise measurement of defect dimensions and their location relative to the workpiece surface. Additionally, a dedicated software application, “iToFD,” has been developed implementing this framework, which offers a complete end-to-end solution for industrial implementation.