Training and Integrating a Machine-Learning-Based Shell Element in Reinforced Concrete Simulations


Loading...

Date

2025

Publication Type

Conference Paper

ETH Bibliography

yes

Citations

Altmetric

Data

Abstract

The finite element analysis (FEA) provides state-of-the-art solutions for any reinforced concrete (RC) shell structure. Emerging surrogate models leveraging physics-based machine learning (ML) may provide viable alternatives by improving speed and enabling differentiability while maintaining physical consistency. This study develops a linear elastic ML-FEA hybrid framework for RC layered shell elements to create a Surrogate 3D-Shell element. A Deep Operator Network based on data generated via local simulation yields the most accurate and generalisable hybrid model among all tested approaches. With a case study of a plate in combined membrane and bending action, the ML-FEA hybrid model is shown to be a promising alternative to traditional FEA.

Publication status

published

External links

Book title

Proceedings of the fib Symposium 2025

Volume

71

Pages / Article No.

2136 - 2145

Publisher

International Federation for Structural Concrete

Event

fib Symposium 2025: Concrete Structures - extend lifespan, limit impacts

Edition / version

Methods

Software

Geographic location

Date collected

Date created

Subject

Organisational unit

09469 - Kaufmann, Walter / Kaufmann, Walter check_circle
09620 - Coros, Stelian / Coros, Stelian check_circle
02219 - ETH AI Center / ETH AI Center

Notes

Conference lecture held on June 16, 2025

Funding

Related publications and datasets