Block-Structured Operator Inference for Coupled Multiphysics Model Reduction

What is it about?

This paper presents a block-structured formulation of the Operator Inference model reduction method as a way to learn structured reduced-order models for multiphysics systems. Coupled dynamical systems often exhibit a block structure, but this information is neglected by many data-driven model reduction methods. The block-structured Operator Inference method exploits the block structure of the governing equations to specify distinct structure and tailored regularization for each regime of the dynamical system, thus embedding knowledge of the coupling between physics regimes into the learning problem. Incorporating the block structure reduces complexity during both inference and prediction, which leads to lower computational costs compared to standard Operator Inference. The block-structured Operator Inference method demonstrates improved performance on the AGARD 445.6 wing aeroelastic benchmark case.

Featured Image

Photo by Volodymyr Hryshchenko on Unsplash

Photo by Volodymyr Hryshchenko on Unsplash

Why is it important?

Data-driven modeling is a popular and growing area of research, but models that are constructed only from data tend to neglect many useful features and structure that we already know should be present in a given model. This work embeds block structure into data-driven models to help combine the advantages of both data-driven and physics-informed modeling approaches within a single method.