AI-Driven 3D Defect Mapping and Identification in Fracture-Critical Structural Welds

Phased Array Ultrasonic Testing (PAUT), a leading nondestructive evaluation (NDE) method for weld inspection, offers higher accuracy and repeatability compared to conventional ultrasonic testing (UT). Yet, rapid and reliable identification of defect type remains a persistent challenge in assuring the safety of fracture critical structural welds. This research introduces an AI-assisted framework for 3D defect mapping and classification using raw PAUT C-scan data. MATLAB-based visualization and data processing are integrated with a lightweight k-nearest neighbor (kNN) classifier and physics informed feature extraction to recognize weld flaws such as porosity and cracks. While this study focuses on interpretable mathematical and kNN-based classification, the workflow established a validated dataset and processing pipeline to support future development of a convolutional neural network (CNN). By integration AI with advanced NDE, this work aims to reduce inspection time, improve reliability of defect characterization, and lower the risk of undetected weld failures. This framework demonstrates how physics informed AI can accelerate the transition toward autonomous inspection systems, ultimately strengthening resilience in structural steel infrastructure.