Speaker
Description
nfrared thermography captures surface temperature fields that can visualize subsurface defects, yet conventional analyses are hampered by sensor noise and imperfect heat-conduction models. The physics-guided (informed) neural reconstruction framework can be developed embedding heat-transfer priors within a deep network to recover a denoised, physically consistent background from raw thermal sequences. Differential maps between this reconstruction and the original measurements selectively amplify defect contrast while suppressing background clutter. Thanks to examples, effectiveness is demonstrated on laboratory-fabricated and handmade specimens, yielding improved background restoration, more reliable defect characterization, and the estimation of unknown physical parameters such as, e.g., the material’s thermal diffusivity. Comparison with standard (well-recognized) techniques are also discussed.
