Enhancing Mechanical Behavior Assessment in Porous Thermal Barrier Coatings using a Machine Learning Fine-Tuned with Genetic Algorithm

Abstract

In this study, a Genetic Algorithm-Enhanced Machine Learning (GAML) model has been established to predict stress variations (σ_ave) and equivalent strain (ε_cr) in porous thermal barrier coatings (TBCs) subjected to diverse thermal loading conditions. The input parameters encompass loading parameters, geometrical characteristics, and porosity features. Remarkable predictive performance was observed, with determination coefficient values of 0.971 for ε_cr and 0.939 for σ_ave, emphasizing a robust correlation between predicted and actual values. The hierarchical nature of the GAML model allows latent patterns and relationships within the data to be effectively unveiled. Moreover, the study illustrated that the relevance of each input parameter undergoes substantial changes with variations in output target values, indicating unique sensitivities of each output to specific input parameters. Specifically, at high stress levels, the weight factors of porosity features became more significant in predicting σ_ave due to their direct influence on stress concentration effects, while thermal loading parameters are more effective in predicting ε_cr. Lastly, through an illustrative example, the model’s utility in facilitating coating design and parameter adjustment for achieving desired mechanical properties was demonstrated.