Afshin Ashofteh

Multilayer thermal barrier coatings provide greater flexibility for tailoring coating performance, but the large number of possible material combinations and layer configurations makes architectural optimization increasingly difficult through experimental trial and error alone. In this study, an experimentally generated dataset spanning single-layer, bilayer, multilayer, graded, variable-thickness, and physically mixed coating architectures containing conventional yttria stabilized zirconia (YSZ), nanostructured YSZ, and ceria yttria stabilized zirconia (CYSZ) was analyzed using machine learning to investigate the relationship between coating architecture and degradation-related performance. Surrogate models were developed for multiple degradation responses and combined with model interpretation, consistency assessment, and multiobjective optimization to examine the influence of composition, layer arrangement, and thickness distribution on coating behavior. The analysis showed that different degradation metrics provide distinct perspectives on coating performance and should not be treated equally during optimization. It also identified the architectural variables with the greatest influence on degradation resistance and durability, allowing practical architecture level design trends to be identified from the experimental data. Global performance–relative-cost screening of the investigated 450 μm architectures selected a YSZ/nanostructured YSZ bilayer as a practical candidate for balancing performance and cost while reducing final degradation and preserving high cyclic durability, making it technically viable for further development and engineering applications.