Research

Surrogate Modeling for Complex Structures

Surrogate modeling aims to reduce the order of physics-based models in order to obtain a fast, sometimes near real-time, evaluation. For instance, limited by the accessibility of deformation, the 3D shape of an on-orbit space structure needs to be reconstructed from strain data. Compared to traditional FEM-based inverse problem, surrogate modeling provides a much faster response, so that the deflection information of the antenna truss is achievable in real-time for online communication compensation.

For civil infrastructures, quick awareness of the structural characteristics is often needed for operations and risk-informed decision-making. Metamodeling provides a fast approximation of the interested structural characteristics with acceptable accuracy. Some of the techniques that have been adopted include artificial neural network and Gaussian process emulator.