Reference: D Lee et al, ‘DiffectNet: diffusion-enabled conditional target generation of internal defects in ultrasonic non-destructive testing (NDT)’, Mechanical Systems and Signal Processing, Vol 240, 113454, November 2025. DOI: 10.1016/j.ymssp.2025.113454
System reliability and safety are critical across industries such as semiconductors, energy, automotive, and steel, where even microscopic cracks or defects can severely compromise performance. Because these flaws are hidden from direct visual inspection, engineers have long relied on non-destructive testing (NDT) to assess the internal health of materials and structures without causing damage. Despite its importance, accurately identifying the precise location, size, and nature of internal defects remains a formidable challenge.
In practical applications, signals collected by physical sensors—such as ultrasonic or electromagnetic waves—are often distorted by complex factors including geometry, material heterogeneity, and harsh operating environments. These effects impose fundamental physical limits on conventional non-destructive testing (NDT), making it difficult to achieve detailed and reliable defect characterization. As a result, critical anomalies may go undetected or be inaccurately interpreted.
Inspired by the question of whether artificial intelligence can “see” what humans cannot, a research team from South Korea led by Sooyoung Lee, Assistant Professor at Chung-Ang University and Principal Investigator of the Industrial Artificial Intelligence Laboratory, has introduced a groundbreaking approach. The team developed DiffectNet, a diffusion-enabled conditional target generation network capable of producing high-fidelity, defect-aware ultrasonic images. Their findings were published online on 30 September 2025.
Professor Lee explained that by leveraging AI’s learning and reasoning capabilities, the inherent limitations of traditional inspection methods can be overcome. Rather than simply enhancing existing workflows, the proposed technology represents a fundamental shift: a generative AI system that can reconstruct hidden cracks inside structures in real time, surpassing the physical constraints of conventional non-destructive testing (NDT).
The implications are far-reaching. In power plants, where even a minor crack can trigger catastrophic failures, AI-driven real-time monitoring could enable early warnings and preventive action. In semiconductor and advanced manufacturing facilities, internal defects could be virtually reconstructed without interrupting operations, improving quality control while maintaining productivity. The technology also holds promise for continuous monitoring of buildings, bridges, and other infrastructure, contributing to safer and more resilient cities.
Together, these advances illustrate how AI is expanding the boundaries of engineering practice. By enabling machines to act as the “eyes” of structures, this research opens new possibilities for real-time defect reconstruction and prediction in reliability-critical fields such as aerospace, energy, manufacturing, and civil infrastructure. As Professor Lee concluded, artificial intelligence is no longer just a tool for analysis—it is becoming an active agent that reshapes the future of engineering itself.
