Physics-Informed Gaussian Processes model for heat source identification

Thermal source identification is crucial in various engineering fields, particularly in material science, where understanding heat generation during mechanical loading is essential for comprehending material behavior. This study introduces a novel method for identifying self-heating sources in steel under mechanical loading by leveraging infrared thermography and Physics-Informed Gaussian Processes (PIGP). The method addresses the limitations of traditional filtering techniques, which struggle with noise sensitivity and lack integration with physical laws. PIGP uses Gaussian process regression to model noisy temperature data while embedding the heat equation directly into its framework, enabling robust and physically consistent estimation of heat sources. We present the theoretical formulation, training, and inference process of PIGP in detail. Synthetic data experiments demonstrate that PIGP delivers accurate heat source estimations even under significant infrared measurement noise ( σ = 0 . 084 K), with estimation errors remaining below 5%. These results make PIGP a highly competitive alternative to traditional filtering-based approaches, primarily due to its reduced number of tuning parameters. These results highlight PIGP’s potential as a robust and accurate tool for thermal source identification in complex, noisy environments.