High-fidelity simulations of complex engineering systems — finite-element analysis, computational fluid dynamics, and coupled multiphysics models — are accurate but computationally expensive, which makes large-scale design exploration, optimization, and real-time decision-making intractable. Surrogate models (emulators) learn the input–output behavior of these systems from data, approximating them at a fraction of the cost.
This stream of research develops surrogate models that capture spatio-temporal and evolving physics, fuse heterogeneous and multi-fidelity data sources, and quantify predictive uncertainty. These models enable efficient design optimization, Bayesian optimization, and trustworthy Digital Twins, spanning methods from Gaussian processes and latent-variable models to neural operators for high-dimensional, time-dependent systems.
Selected journal articles
