Session: 02-06-01: Machine Learning in Structural Dynamics and Aeroelasticity

Paper Number: 188543

188543 - Neural Network-Aided Vibration Suppression for Flexible Structures 

Flexible elastic structures such as beams, cables, deployable components, and lightweight spaceframe elements are highly sensitive to external disturbances, making vibration suppression a persistent challenge in many aerospace engineering systems. Real-time control is difficult because high-fidelity finite element models (FEM) are computationally expensive, while classical reduced-order approaches often lose accuracy under broadband, nonlinear, or transient excitations.

To overcome these limitations, we build upon recent advances in mesh-based neural surrogates, particularly MeshODENet—a general framework that synergizes the spatial reasoning of Graph Neural Networks (GNNs) with the continuous-time dynamics modeling capability of Neural Ordinary Differential Equations (Neural ODEs). Traditional GNN surrogates for structural dynamics often rely on autoregressive rollout, which makes long-horizon prediction vulnerable to error accumulation and numerical instability. In contrast, MeshODENet embeds the governing dynamics into a learned continuous-time vector field, enabling stable integration over long time horizons and greatly improving predictive robustness. Prior results demonstrate that MeshODENet achieves substantial computational speed-ups over FEM solvers while significantly outperforming baseline neural models in long-term accuracy for challenging one- and two-dimensional elastic bodies undergoing large, nonlinear deformations.

Building on this foundation, this work presents a learning-augmented framework that integrates a MeshODENet-style GNN surrogate for steady-state vibration response with a differentiable, gradient-based boundary control strategy. We extend the neural structural simulator to incorporate controllable boundary conditions, enabling the surrogate to function as a fast, differentiable digital twin that maps both external forcing and boundary actuation to the resulting vibration profiles. Leveraging full end-to-end differentiability, we formulate vibration mitigation as a continuous optimization problem and directly compute optimal boundary control forces using automatic differentiation—eliminating the need for hand-crafted control laws or linearized reduced models. Closed-loop evaluations using the learned surrogate show that the proposed approach can effectively suppress vibration amplitude with only a single controllable boundary input, even under complex loading conditions. These results highlight that combining mesh-based GNN/Neural-ODE surrogates with differentiable optimal control offers a practical, computationally efficient, and highly flexible pathway for real-time vibration mitigation in flexible aerospace structures.

Presenting Author: Leixin Ma Arizona State University

Presenting Author Biography: Leixin Ma joins SEMTE as an assistant professor in 2023. Before joining ASU, she was a postdoctoral fellow in the Department of Mechanical and Aerospace Engineering at UCLA. She received her BSc in Naval Architecture & Ocean Engineering from Shanghai Jiao Tong University in 2015, an S.M. degree, and a Ph.D. in mechanical engineering from MIT in 2017 and 2021, respectively.

Her research interest includes fluid-structure interactions of deformable structures, physics-informed machine learning, and the machine learning-aided design of programmable structures and robots.

Her research approach emphasizes a synergic integration of machine learning into traditionally expensive simulations and experiments for nonlinear mechanics problems. Her career goal is a data-driven and physically consistent approach to modeling and designing programmable smart structures and robots, especially under complex fluid flow conditions. Her work has been featured in the Physics of Fluids Journal.