A Multi-Day Thermal Signature and Machine Learning Framework for Predicting Defect Detectability in Passive Infrared Thermography of Reinforced Concrete Slabs
Bildiriler > Tam Metin Bildiri
Yazarlar: Aseel Abdulazez Abdulridha, Ninel ALVER, Kazuma Shibano, Tetsuya Suzuki
International Conference on Road and Rail Infrastructure, , , Uluslararası, Tam Metin Bildiri, cilt 9, ss 533-540
DOI: 10.5592/co/cetra.2026.1957
Özet
Passive infrared thermography is widely used for non-destructive evaluation of concrete structures; however, its reliability under real outdoor conditions is limited by environmental variability and the transient nature of defect visibility. As a result, image-based interpretation often leads to inconsistent assessments of subsurface damage. This study presents a time-resolved, weather-aware framework to quantify the detectability of defects in passive IRT using multi-day surface temperature monitoring. A full-scale concrete slab including artificial delaminations at depths of 50, 100, and 250 mm was tested outdoors over seven daytime periods with one-minute temporal resolution. A physics-informed Random Forest regressor was used to model and remove the Weather-driven baseline surface temperature to isolate residual thermal responses. An objective, sound-referenced detectability score was defined, enabling threshold-based detection without manual labeling. Persistence-based metrics, including duty cycle and detectability windows, and supported by Probability-of-Detection (PoD) analysis, were used to evaluate the detectability. Results demonstrate that passive IRT detectability is fundamentally time-dependent; all delaminations at 250 mm depth, as well as smaller delaminations at shallower depths, remained non-detectable and provided a stable intact-reference group. These findings highlight the necessity of temporal reliability metrics for defensible interpretation of passive thermographic monitoring.