Abstract
The two-dimensional laser-based Ångstrom method measures the in-plane thermal properties for anisotropic film-like materials. It involves periodic laser heating at the center of a suspended film sample and records its transient thermal response by infrared imaging. These spatiotemporal temperature data must be analyzed to extract the unknown thermal conductivity values in the orthotropic directions, an inverse parameter fitting problem. Previous demonstration of the metrology technique used a least-squares fitting method that relies on numerical differentiation to evaluate the second-order partial derivatives in the differential equation describing transient conduction in the physical system. This fitting approach is susceptible to measurement noise, introducing high uncertainty in the extracted properties when working with noisy data. For example, when noise of a signal-to-noise ratio of 10 is added to simulated amplitude and phase data, the error in the extracted thermal conductivity can exceed 80%. In this work, we introduce a new alternative inverse parameter fitting approach using physics-informed neural networks (PINNs) to increase the robustness of the measurement technique for noisy temperature data. We demonstrate the effectiveness of this approach even for scenarios with extreme levels of noise in the data. Specifically, the PINN-approach accurately extracts the properties to within 5% of the true values even for high noise levels (a signal-to-noise ratio of 1). This offers a promising avenue for improving the robustness and accuracy of advanced thermal metrology tools that rely on inverse parameter fitting of temperature data to extract thermal properties.
Keywords
Thermal conductivity, Thermal diffusivity, Thermal transport, Metrology, Data analysis, Noisy data, Artificial neural networks, Infrared imaging
Date of this Version
2024
Published in:
S. Sripada, A. Gaitonde, J.A. Weibel, A.M. Marconnet, Robust inverse parameter fitting of thermal properties from the laser-based Angstrom method in the presence of measurement noise using physics-informed neural networks (PINNs), Journal of Applied Physics 135, 225106 (2024).
Link to original published article:
https://doi.org/10.1063/5.0206247
Comments
This is the publisher PDF of S. Sripada, A. Gaitonde, J.A. Weibel, A.M. Marconnet, Robust inverse parameter fitting of thermal properties from the laser-based Angstrom method in the presence of measurement noise using physics-informed neural networks (PINNs), Journal of Applied Physics 135, 225106 (2024). Published CC-BY-NC-ND, the version of record is also available at DOI: 10.1063/5.0206247.