Term 252

Drone-Based Structural Crack Detection and Classification Inspection System

Project Type: Self-Initiated

Project Description

Manual structural inspections are often slow, expensive, and risky because engineers must access hard-to-reach areas using scaffolding or lifts. This project develops a semi autonomous drone-based inspection system that captures images and converts them into actionable crack information for concrete structures. The workflow is designed as two missions: first, the drone scans a user-defined area and collects high-resolution images for crack detection. Then, after cracks are identified, the system revisits only the detected crack locations for close-range inspection to support depth-based severity screening, focusing on cracks with measured depth of at least 5 mm. On the software side, a lightweight YOLO based model is trained to detect cracks and classify them into eight structural crack types, targeting at least 70% model accuracy, with classification performance of at least 60% (target 70%) on a held-out test set. The results are mapped into a structured inspection report to support maintenance decisions. The overall design is constrained by local operational requirements and practical project limits, including a maximum takeoff weight of 2.5 kg, a total budget cap of 18,850 SAR, and limited computing resources, which guide model selection and system architecture