Thermal dominance in wave reflection in functionally graded micropolar composites: A machine learning-driven sensitivity analysis
Materials Today Communications, cilt.54, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 54
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.mtcomm.2026.115672
- Dergi Adı: Materials Today Communications
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Chemical Abstracts Core, Compendex, INSPEC
- Anahtar Kelimeler: Functionally Graded Materials, Micropolar Thermoelasticity, Parameter Ranking, Sobol Sensitivity Analysis, Surrogate Modeling, Three-Phase-Lag Theory
- Recep Tayyip Erdoğan Üniversitesi Adresli: Evet
Özet
Functionally graded micropolar composite materials are increasingly used in harsh environments, including nuclear reactors and turbine engines, where wave propagation under coupled thermo-mechanical-rotational conditions is a determining factor in material performance and durability. However, owing to the large number of parameters in these multiphysics systems, it is challenging to ascertain the relative importance of each physical parameter on system behavior. Here, a deep neural network surrogate model is employed in conjunction with global variance-based sensitivity analysis to quantitatively evaluate the relative importance of eight thermo-mechanical parameters that govern wave reflection in a rotating, incompressible, functionally graded micropolar half-space subjected to three-phase-lag heat conduction. A three-hidden-layer neural network surrogate model, trained on 20,000 high-fidelity simulations, was employed to compute Sobol sensitivity indices by Monte Carlo sampling. Over the physically realistic parameter ranges for carbon-epoxy composites, the thermal gradient lag (τv) and thermal displacement lag (τt) account for approximately 92% of the total variance in the reflection coefficients. However, a normalized range sensitivity study (±50% for all parameters) reduces this to approximately 78%, indicating that the degree of thermal dominance is sensitive to the chosen parameter bounds. The conclusion of thermal primacy is therefore conditional on the specific material system and parameter regime studied, not universally unconditional. Furthermore, variance-based Sobol indices are limited in that they can hide narrow but physically critical resonances in this case, a rotational resonance at Ω≈1.2 rad/s where rotation locally dominates. This limitation must be recognized in design applications. From a timescale perspective, these thermo-mechanical parameters maintain the system in a state of persistent thermodynamic non-equilibrium, where thermoelastic dissipation is a dominant energy-controlling mechanism. This study presents a quantitative ranking of parameters for this specific physics combination. It is evident that machine-learning-based sensitivity analysis is a powerful tool for extracting essential physics from high-dimensional material models, provided its limitations are acknowledged.