F. Blanchard (1), M.J. Kadi, E. Bousser (1), B. Baloukas (1), M. Azzi (1), J.E. Klemberg-Sapieha (1), L. Martinu (1)
Surface and Coatings Technology. Volume 452 (15 January 2023). DOI: https://doi.org/10.1016/j.surfcoat.2022.129080

Keywords

Thermal barrier coatings, Optical properties, Sintering, Deep-learning segmentation, FDTD modeling

Abstract

Thermal radiation represents a significant portion of the thermal heat transfer from the hot gases of an engine to its metallic components. Thermal barrier coatings (TBCs) are used to protect those components from heat and must, therefore, be effective blockers of radiative heat. The porous microstructure of TBCs causes them to be highly reflective in the visible and near-infrared wavelength ranges through the scattering of light. This microstructure, however, is susceptible to degradation. In this study, we establish a clear link between the porosity of a TBC and its ability to reflect heat through radiationscattering by extracting the scattering coefficient of two coatings with different microstructures and quantifying their pore space. Using deep learning trained image segmentation models on high resolution SEM cross-sections, we identify the proportion of the pore space comprised of pores 2 μm in diameter or smaller and show that these are responsible for most of the scattering. The pore size distribution is also confirmed with mercury infiltration porosimetry. The coatings are then subjected to cyclic heat treatments at 1450 K for a total of 1111 h to induce sintering of the microstructure. This results in a reduction of the scattering coefficient by around 20 % for both samples, which is attributed to a reduction in the space occupied by pores 2 μm and less in diameter. Finally, a model is built for finite-difference time-domain (FDTD) simulations using high resolution SEM cross-sections to calculate the reflectivity and transmission of the TBCs. The results exhibit good agreement with our experimental data, showing that such models could be used in the future to predict the effect of degradation on a TBC's optical properties.

How Our Software Was Used

A deep learning model based on a U-net neural network was trained on small sections of labeled images to distinguish between ceramic, large pores, and small pores. The fully trained model was then applied to the panoramic cross-sections of the Cold and Hot samples.

Author Affiliation

(1) Department of Engineering Physics, Polytechnique Montréal, Montreal, QC H3T 1J4, Canada