LOWERING THE BARRIER TO AI-DRIVEN INSPECTION: A NO-CODE WORKFLOW FOR AUTOMATED STRUCTURAL DEFECT DETECTION

Structural health monitoring (SHM) is critical for maintaining the integrity of modern engineering systems. While traditional visual inspection is the standard, it is labor-intensive and subjective. Recent advancements in computer vision, particularly deep learning architectures like YOLO (You Only Look Once), have enabled real-time, accurate defect detection. However, these modern tools typically require significant programming expertise for data labeling, model configuration, and deployment, creating a barrier for domain experts in civil and mechanical engineering.

The authors introduce YOLOEZ, an open-source, GUI-based software designed to bridge the gap between advanced AI capabilities and practical, code-free application. YOLOEZ integrates the entire machine learning pipeline—data annotation, model training, and inference—into a single, intuitive interface. By abstracting away the underlying Python scripts and command-line operations, the tool allows engineers to focus on structural analysis rather than software development, while also promoting reproducible research practices through a standardized workflow.

To validate the tool, the researchers compared YOLOEZ against a manually tuned classical morphological image processing pipeline. Using scanning electron microscope (SEM) images of additively manufactured tungsten, they tested the ability of both methods to detect micro-scale cracks. YOLOEZ demonstrated superior performance across key metrics, including recall (0.6065 vs. 0.4372), F1 score, and Intersection over Union (IoU), despite being trained on a small dataset of only 20 labeled images. Furthermore, YOLOEZ showed robust generalization when tested on out-of-distribution images from a different tungsten sample, proving its reliability for real-world SHM scenarios.