Physics-Infused Reduced Order Modeling of Hypersonic Aerothermal Loads for Aerothermoelastic Analysis

This paper presents the Physics-Infused Reduced Order Modeling (PIROM) methodology. It integrates 1) a physics-based component that guarantees its robustness and generalizability for the entire operating envelope and 2) a data-driven component that achieves an accuracy comparable to high-fidelity methods. This work generalizes the PIROM methodology that encompasses the indirect approach developed in the authors’ previous work, and a new, more efficient direct approach. The PIROM incorporates a data-driven differential-algebraic component, i.e. neural ordinary differential equations (NODEs), for the learning of nonlinear functional relations, that are used to enhance the accuracy of the basic physics model. An efficient gradient-based optimizer aided by adjoint methods is employed to train the NODE together with the physics model. The PIROM methodology is applied to develop a new hypersonic aerothermal model that augments the classical boundary layer theory models. The superior accuracy and robustness of the new PIROM-based aerothermal models are demonstrated on fully-coupled aerothermoelastic responses of structures with realistic boundary conditions, with comparison to the conventional kriging-based aerothermal surrogates. Finally, potential applications to multi-disciplinary optimization of PIROM and its current limitations are discussed.