Scientific Machine Learning Estimation of Material Thermal Diffusivity from Non-Destructive Evaluation

Aging metallic structures in nuclear reactors degrade under high-temperature and radiation environments, leading to creep and embrittlement. Nondestructive evaluation (NDE) enables assessment of material state without altering its properties. We investigate thermal diffusivity estimation using one-sided pulsed infrared thermography (PIT), a non-contact NDE method that records transient surface temperatures after a rapid thermal pulse. As heat diffuses into the material bulk, surface temperature decays and is captured by a fast-frame infrared camera. Local variations in thermal diffusivity may indicate material degradation. An analytic solution for surface temperature transients in a plate, dependent on diffusivity and thickness, is fitted to PIT data for samples with known thickness. We apply a developed differential evolution (DE) optimization algorithm to solve the inverse problem and estimate thermal diffusivity for each pixel in a PIT image. The single-objective algorithm performs continuous optimization, where we minimize error over each iteration. Analyzing the spatial distribution of thermal diffusivity values reveals information about material state. Preliminary results validate the algorithm and show distinct thermal diffusivity patterns in simulated and experimental PIT data from both metallic and ceramic specimens. This supports DE as a promising computational method for structural health monitoring in nuclear reactors.