Beyond Corrosion: Autonomous Real-Time Detection of Structural Cracks and Degradation Using Robotic Sensing and Digital Twins

Structural crack detection and environmental hazard monitoring are essential in non-destructive testing (NDT) for
civil infrastructure. This paper presents a real-time, AI-powered multi-sensor dashboard that receives data from a
mobile climbing robot. The system integrates optical, thermal infrared (IR), and gas sensing, each supported by
dedicated deep learning models. AI inference is performed server-side to classify surface cracks, segment IR-based
crack regions, and detect gas anomalies. Models include MobileNetV2 for crack classification, U-Net for IR
segmentation, and a gas classifier using MQ-135 data. The dashboard fuses predictions with visual overlays and
quantitative measurements, enabling faster, safer, and more robust infrastructure inspection