Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
Gemini 2.5 Pro
GPT-5
GPT-4o
DeepSeek R1 via Azure
2000 character limit reached

Physics-informed Multiple-Input Operators for efficient dynamic response prediction of structures (2505.07090v1)

Published 11 May 2025 in cs.LG

Abstract: Finite element (FE) modeling is essential for structural analysis but remains computationally intensive, especially under dynamic loading. While operator learning models have shown promise in replicating static structural responses at FEM level accuracy, modeling dynamic behavior remains more challenging. This work presents a Multiple Input Operator Network (MIONet) that incorporates a second trunk network to explicitly encode temporal dynamics, enabling accurate prediction of structural responses under moving loads. Traditional DeepONet architectures using recurrent neural networks (RNNs) are limited by fixed time discretization and struggle to capture continuous dynamics. In contrast, MIONet predicts responses continuously over both space and time, removing the need for step wise modeling. It maps scalar inputs including load type, velocity, spatial mesh, and time steps to full field structural responses. To improve efficiency and enforce physical consistency, we introduce a physics informed loss based on dynamic equilibrium using precomputed mass, damping, and stiffness matrices, without solving the governing PDEs directly. Further, a Schur complement formulation reduces the training domain, significantly cutting computational costs while preserving global accuracy. The model is validated on both a simple beam and the KW-51 bridge, achieving FEM level accuracy within seconds. Compared to GRU based DeepONet, our model offers comparable accuracy with improved temporal continuity and over 100 times faster inference, making it well suited for real-time structural monitoring and digital twin applications.

Summary

Physics-Informed Multiple-Input Operators for Efficient Dynamic Response Prediction of Structures

The paper, "Physics-informed Multiple-Input Operators for efficient dynamic response prediction of structures," presents a novel approach to predicting the dynamic responses of structures under moving loads using the Multiple-Input Operator Network (MIONet) architecture. This research addresses the intrinsic challenges associated with finite element modeling (FEM) in dynamic scenarios, where computationally intensive simulations often hinder real-time applications.

Overview

Finite Element Modeling (FEM) is vital for structural analysis, especially when analytical solutions are unavailable or too complex. Despite its accuracy, FEM proves costly in computational resources, especially under dynamic loading where repeated time-stepping processes are required. Traditional FEM simulations involve solving comprehensive Partial Differential Equations (PDEs) to achieve precise results, yet they remain inefficient for applications demanding real-time predictions, such as digital twins.

The emergence of operator learning approaches has shown promise in simulating static structural responses, yet adequately modeling dynamic structural behaviors poses additional complexities. Recent advancements have largely relied on Recurrent Neural Networks (RNNs) that suffer limitations due to fixed-time discretization, hampering their ability to capture continuous dynamic responses through sequential events.

Methodology

A significant contribution of this work is the introduction of the MIONet architecture, which utilizes a dual-trunk network design to encode spatial and temporal dynamics separately, allowing seamless predictions across varied spaces and times. The architecture enables input parameters—such as moving load characteristics, velocities, spatial configurations, and time steps—to be converted into continuous structural responses without requiring explicit PDE solutions.

To enhance efficiency and maintain physical fidelity, the authors incorporate a physics-informed learning framework leveraging precomputed matrices for mass, damping, and stiffness to enforce dynamic equilibrium. Besides, employing the Schur complement serves to reduce computational costs by training models in a reduced domain, hence accelerating simulations while preserving accuracy.

Results and Validation

The proposed MIONet is validated on a canonical beam structure and a real-world bridge (KW-51), demonstrating outstanding performance by reproducing FEM-level results swiftly. The comparative analysis with GRU-based DeepONet highlights MIONet’s proficiency in delivering temporal continuity and achieving computation speeds over a hundred times faster than conventional FEM practices.

Implications and Future Directions

The implications of applying MIONet for dynamic structural analysis are profound, offering a highly efficient alternative to traditional FEM. The architecture serves as a promising candidate for adapting real-time digital twin applications, enhancing structural health monitoring capacities in evolving conditions. Despite its efficacy in elastic scenarios, future developments may include integrating uncertainty and damage-informed inputs for broader applicability in real-world structural conditions, such as material degradation or unforeseen damages.

In sum, this research advances the use of novel neural operator architectures like MIONet to revolutionize dynamic response predictions, marrying precision, computational efficiency, and real-time operational potential for intricate structural applications.

Dice Question Streamline Icon: https://streamlinehq.com

Follow-up Questions

We haven't generated follow-up questions for this paper yet.