Discovery of Hyperelastic Constitutive Laws from Experimental Data with EUCLID Journal: Experimental Mechanics

Arefeh Abbasi, Maurizio Ricci, Pietro Carrara, Moritz Flaschel, Siddhant Kumar, Sonia Marfia, Laura De Lorenzis · 2025

Read Discovery of Hyperelastic Constitutive Laws from Experimental Data with EUCLID Journal: Experimental Mechanics by Arefeh Abbasi, Maurizio Ricci, Pietro Carrara, Moritz Flaschel, Siddhant Kumar, Sonia Marfia, Laura De Lorenzis (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-24747.txt.

Introduction

Arefeh Abbasi, Maurizio Ricci, Pietro Carrara, Moritz Flaschel, Siddhant Kumar, Sonia Marfia, and Laura De Lorenzis test a tool that finds rubber laws straight from test data. EUCLID, as the tool is known, picks both the law form and its fit values in one pass. The team checks it on real lab data, not just fake code runs.

They bent and stretched raw rubber parts in shapes from plain to hard, and logged both whole force and full field shift maps. Laws from EUCLID were matched with laws fit the old way, where a form is picked first. They checked which laws guess best on new shapes, and how noise and state range play in.

Catalog record

CREATORArefeh Abbasi, Maurizio Ricci, Pietro Carrara, Moritz Flaschel, Siddhant Kumar, Sonia Marfia, Laura De Lorenzis
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE94 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2510-24747

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Discovery of Hyperelastic Constitutive Laws from Experimental Data with EUCLID Journal: Experimental Mechanics Arefeh Abbasi, Maurizio Ricci, Pietro Carrara, Moritz Flaschel, Siddhant Kumar, Sonia Marfia, Laura De Lorenzis Abstract We assess the performance of EUCLID (Efficient Unsupervised Constitutive Law Identification and Discovery), a recently proposed framework for automated discovery of constitutive laws, on experimental data. Mechanical tests are performed on natural rubber specimens spanning simple to complex geometries, from which we collect both global (force–elongation) and local (full-field displacement) measurements. Using these data, we obtain constitutive laws via two routes: (i) the conventional identification of unknown parameters in a priori selected material models, and (ii) EUCLID, which automates model selection and parameter identification within a unified model-discovery pipeline. We compare the two methodologies using global versus local data, analyze predictive accuracy, and examine generalization to unseen geometries. Moreover, we quantify the experimental noise, investigate the coverage of the material state space achieved by each approach and discuss the relative performance of different datasets and different a priori chosen models versus EUCLID. 1 Introduction Material models are an essential ingredient of any engineering simulation aiming at the prediction of mechanical phenomena. Thus, their accurate determination is of great relevance across a wide range of engineering applications. For hyperelasticity, where the material response is specified via a strain energy density function, a variety of models have been proposed, ranging from phenomenological to physically- and statistically-based approaches ( Ogden et al. 2004 ; Marckmann and Verron 2006 ; Ricker and Wriggers 2023 ) . Among the most widely used are the generalized Mooney-Rivlin ( Mooney 1940 ; Rivlin 1948 ) , the Gent-Thomas ( Gent and Thomas 1958 ) and the generalized Ogden ( Ogden 1972 ) models. The conventional approach to finding a suitable material model requires pre-selecting its functional form (a task often referred to as model selection ) and determining its unknown parameters (i.e., parameter identification ) by minimization of a suitable objective function, which quantifies the discrepancy between experimentally measured quantities and their counterparts as predicted by the model. Different parameter identification approaches differ primarily in the choice of this objective function. Some of them rely on global measured quantities (such as forces and elongations); in order to convert these quantities into stress and strain measurements, they require simple experiments, e.g., uniaxial tensile, pure shear, or equibiaxial tensile tests, interpreted via analytical ideal models. Other methods rely on local measurements, i.e. full displacement fields obtained through digital image or volume correlation, along with reaction forces. Among these are Finite Element Model Updating (FEMU), the Equilibrium Gap Method (EGM), the Virtual Fields Method (VFM), and the constitutive equation gap method (CEGM). For comprehensive overviews we refer to Avril et al. 2008 ; Roux and Hild 2020 ; Fuhg et al. 2024 ; Römer et al. 2025 . The a priori selection of a functional form for a material model introduces a subjective bias and potentially leads to costly trial-and-error procedures until a suitable form is found. Recognizing this issue, recent research in the field has challenged the conventional paradigm with different ideas. Model-free approaches give up constitutive laws altogether and replace them with discrete material observations ( Kirchdoerfer and Ortiz 2016 ; Conti et al. 2018 ; Nguyen and Keip 2018 ; Carrara et al. 2020 ) . Other approaches substitute the interpretable functional form with a more flexible, but not interpretable ansatz based on neural networks or other parameterized architectures from machine learning ( Huang et al. 2020 ; Zhong et al. 2022 ; Thakolkaran et al. 2022 ; Xu et al. 2023 ; Fuhg et al. 2024 ) , often augmented through the enforcement of physics constraints ( Fuhg et al. 2024 ) . A third approach is known as EUCLID, which stems for Efficient Unsupervised Constitutive Law Identification and Discovery ( Flaschel et al. 2021 ) . This method unifies model selection and parameter identification (whereby the combination of the two tasks is denoted as model discovery ) by automatically selecting the most suitable material model from a wide library of possible candidates through sparse regression. This eliminates subjective bias and the need for a trial-and-error procedure, while preserving interpretability ( Flaschel et al. 2021 ; Flaschel et al. 2022 ; Marino et al. 2023 ; Flaschel et al. 2023a ; Flaschel 2023 ; Joshi et al. 2022 ) . Most recently, the automatic creation of the material model library through grammars has also been pursued ( Kissas et al. 2024 ) and alternatives to the LASSO algorithm for sparse regression have been explored ( Linka and Kuhl 2024 ; Flaschel et al. 2025 ; Urrea-Quintero et al. 2025 ) . In principle, the choice of the ansatz for model selection, i.e. for the potential functional form of the model (e.g. a single function, a neural network, or a material model library) and the choice of the objective function for the parameter identification (e.g. based on global or local data) are completely independent and can be combined in multiple ways. Early studies on EUCLID advocate the use of local data, i.e. displacement fields and the corresponding reaction forces, and adopt for parameter identification the same objective function of the EGM (in turn a special case of the VFM). Also, these early studies use data generated synthetically through finite element (FE) simulations with the addition of noise. A more recent contribution applies EUCLID to experimental data on human brain tissues ( Flaschel et al. 2023b ) . In this study, the objective function is based on global data obtained from uniaxial compression and torsion tests. In this work, we assess the performance of EUCLID to discover the hyperelastic constitutive law of natural rubber. An experimental campaign is performed on different specimen geometries, generating both global and local data. Our main objective is twofold: on the one hand, we aim to compare the conventional paradigm (manual model selection and parameter identification) with the new paradigm (material model discovery based on EUCLID) on new experimental data. On the other hand, we intend to assess the influence of using global vs. local data on the performance of both paradigms. By using different specimen shapes, we also assess the ability of the obtained models to generalize to unseen geometries and stress states as well as the accuracy in reproducing both local and global behaviors. Additionally, we quantify the experimental noise and investigate the coverage of the material state space achieved with the global vs. local datasets. The remainder of this paper is structured as follows. In Section 2 , we present an overview of EUCLID, outlining the structure of the material model library and the methodologies used for discovery from global and from local data. Section 3 describes the experimental campaign and the data processing. Section 4 discusses the performance of both the conventional paradigm and EUCLID based on the experimental data. Finally, the key findings and their implications are summarized in section 5 . 2 Overview of EUCLID This section provides a brief overview of EUCLID as proposed in Flaschel et al. 2021 and Flaschel et al. 2023a . 2.1 Problem setting and model library Let us consider the undeformed domain \Omega\subset\mathbb{R}^{2} with boundary \partial\Omega , representative of a specimen made

Read the complete text