Intelligent Automated Defect Detection

This book presents cutting-edge research on the integration of artificial intelligence, deep learning, and computer vision with traditional nondestructive testing (NDT) methods. It compiles recent advances in automated defect detection across critical infrastructure and industrial applications. Intelligent Automated Defect Detection: AI Solutions for Roads, Welding, and Industrial Inspection demonstrates how AI-driven solutions are transforming quality inspection from subjective, time-intensive manual processes into real-time, scalable, and objective automated systems. The book provides comprehensive coverage of machine learning applications in civil infrastructure inspection and industrial welding quality assurance, utilizing radiography, ultrasonic B-scans, and time-series sensing. The authors present state-of-the-art deep learning architectures including convolutional neural networks, transformer models, and foundation models like the Segment Anything Model (SAM), alongside practical guidance on developing explainable AI systems and overcoming data scarcity challenges. The book offers proven methodologies for transitioning from proof-of-concept studies to robust real-world applications that enhance inspection efficiency while reducing costs and failure risks. Real-world case studies are included throughout, with additional applications available on the authors' GitHub page. This book is intended for researchers, engineers, and practitioners working at the intersection of AI, NDT, and industrial quality assurance. It also serves as a resource for graduate students and early-career professionals interested in applying modern data-driven approaches to complex engineering problems.

Hassan Zargarzadeh is an Associate Professor in the Electrical and Computer Engineering Department at Lamar University, Beaumont, TX. He holds a PhD from Missouri University of Science and Technology (MS&T). His research focuses on applied machine learning in the fields of defect detection and robotics, where he has published several papers. He also has published numerous papers in the fields of advanced control systems and power electronics. Amir-M. Naddaf-Sh is currently a Ph.D. student in Computer Science at Kennesaw State University, Marietta, GA. He received his B.Sc. degree in Electrical Engineering from Ferdowsi University of Mashhad, Mashhad, Iran, in 2020, and his Doctor of Engineering degree in Electrical Engineering from Lamar University, Beaumont, TX, USA, in 2025. His research interests span deep learning, robotics, and computer vision. Sadra Naddaf-Sh is a Senior Machine Learning Engineer and Researcher at Truveta, Bellevue, WA, where he works on advancing efficient and robust deep learning and generative AI methods. His research interests include generative AI, language modeling, and data-centric AI. He received his Doctorate degree in Electrical Engineering from Lamar University, Beaumont, TX, USA, in 2022, and his B.Sc. degree in Computer Engineering from Ferdowsi University of Mashhad, Iran, in 2018. M-Mahdi Naddaf-Sh is a Research Scientist at the Laboratory for AI, Robotics, and Automation (LARA) at the University of California, Davis, CA. His research spans the intersection of artificial intelligence, robotics, and control systems, with a focus on autonomous systems and machine learning for robotic perception and decision-making. Before joining UC Davis, he worked at Ford Greenfield Labs in Palo Alto, CA, where he led the Advanced Robotics and Automation Lab within the Core AI/ML group. He received his bachelor's degree in electrical engineering from Ferdowsi University of Mashhad (Iran), his Master's in Artificial Intelligence from the University of Southampton (UK), and his Doctor of Engineering in Electrical Engineering from Lamar University (USA). He has authored several publications in peer-reviewed IEEE journals and conferences, with contributions recognized in the areas of autonomous robotics and intelligent manufacturing.

1 Introduction SECTION I: Roads 2 Optimized CNN for Real-Time Road Crack Mapping 3 Deep Learning for Efficient Road Damage Detection SECTION II Digital X-Radiography 4 AI-Powered Weld Assessment with Digital Radiography 5 Weld Defect Analysis via Digital Radiography and Deep Learning 6 Explainable Real-Time Object Detection in 2D Radiography SECTION III Arc Stud Welding 7 Explainable Defect Classification in Arc Stud Welding 8 Machine Learning in Automotive Stud Weld Defect Classification SECTION IV Ultrasonic Testing 9 AI-Driven Weld Inspection with Ultrasonic B-Scans 10 SAM for Weld Defect Detection in Ultrasonic B-Scans 11 Synthetic Data-Driven Weld Defect Classification