Research

Under prolonged high-temperature loading, deformation is governed by the coupled evolution of dislocation structures (multiplication, interactions, recovery), point defects (vacancies, interstitials), and grain boundaries. These mechanisms evolve continuously over time, depend strongly on prior loading history, and span multiple length and time scales. Existing computational methods either resolve these mechanisms using computationally intensive atomistic and crystal plasticity simulations or replace them with empirical constitutive equations that often fail when extrapolated to new alloy chemistries, processing routes, or service conditions.

The objective of this research is to develop a new computational paradigm for predicting history-dependent materials behavior via differentiable computational mechanics. Specifically, this research will develop a thermodynamically consistent constitutive evolution operator capable of learning the governing evolution physics of microstructural features during high-temperature creep. By learning the evolution of the material state (i.e., internal state variables), the resulting framework will remain physically interpretable, transferable across loading histories, and extensible to other operator-learning architectures. The long-term vision is to establish a differentiable digital twin capable of rapidly predicting high temperature deformation evolution, quantifying uncertainty, and accelerating the computational design of structural alloys for extreme environments.