ROI-driven thermal hyperplane analysis for automated non-destructive evaluation via pulsed thermography

This paper proposes a novel methodology for structural fault detection utilising pulsed infrared thermography data. The approach systematically scans thermal image sequences using Regions of Interest (ROIs) with variable sizes, adjusted according to the expected fault dimensions. All temporal frames are considered during the analysis. For each ROI, a transformation is performed to linearise the thermal response, followed by a reconstruction of the data in a flattened space combining spatial coordinates, time, and temperature. These reconstructed hyperplanes are subsequently evaluated by a Convolutional Neural Network to classify the presence or absence of faults. Experimental validation demonstrates that the proposed method achieves a fault detection accuracy of 96%, with only one false positive identified. The results highlight the method’s potential for enhancing the reliability and automation of structural health monitoring systems using infrared thermography. • ROI-ThermNet: Novel fault detection via variable-sized ROI scanning. • Achieved 96% fault detection accuracy with only 1 false positive. • No manual preprocessing; uses raw thermal data directly. • Flexible CNN accepts variable-sized ROIs without retraining. • Full temporal data in ROIs boosts NDE automation and application generalization.