Combining simulations and measurements in physics-informed engineering surrogates

Abstract

Engineering surrogates often combine inexpensive simulations, a smaller set of refined calculations, and limited physical measurements. These sources differ in numerical resolution, physical assumptions, observation operators, noise, and systematic discrepancy. Pooling them without representing those differences can produce a precise approximation to the wrong target. This review develops a multifidelity formulation for physics-informed flow and thermomechanical surrogates. Analytical examples show when adding a biased source helps or harms mean-square accuracy, why spatially averaged measurements cannot be treated as point values, and how a flexible discrepancy can confound physical parameter calibration. A proposed validation workflow separates fidelity transfer from case transfer and assigns computational and measurement budgets according to the engineering quantity of interest. Published PINN, thermal, and multifidelity studies connect the formulation to engineering applications.