Abstract:We propose a new approach for data-driven automated discovery of material laws, which we call EUCLID (Efficient Unsupervised Constitutive Law Identification and Discovery), and we apply it here to the discovery of plasticity models, including arbitrarily shaped yield surfaces and isotropic and/or kinematic hardening laws. The approach is unsupervised, i.e., it requires no stress data but only full-field displacement and global force data; it delivers interpretable models, i.e., models that are embodied by parsimonious mathematical expressions discovered through sparse regression of a potentially large catalogue of candidate functions; it is one-shot, i.e., discovery only needs one experiment. The material model library is constructed by expanding the yield function with a Fourier series, whereas isotropic and kinematic hardening are introduced by assuming a yield function dependency on internal history variables that evolve with the plastic deformation. For selecting the most relevant Fourier modes and identifying the hardening behavior, EUCLID employs physics knowledge, i.e., the optimization problem that governs the discovery enforces the equilibrium constraints in the bulk and at the loaded boundary of the domain. Sparsity promoting regularization is deployed to generate a set of solutions out of which a solution with low cost and high parsimony is automatically selected. Through virtual experiments, we demonstrate the ability of EUCLID to accurately discover several plastic yield surfaces and hardening mechanisms of different complexity.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to automatically discover the plastic models of materials in the absence of stress data, including yield surfaces of arbitrary shapes and isotropic and/or kinematic hardening laws. Specifically, the paper proposes a new method - EUCLID (Efficient Unsupervised Constitutive Law Identification and Discovery), which does not require stress data but only full - field displacement and global force data. Through this method, researchers hope to accurately discover plastic yield surfaces and hardening mechanisms of different complexity levels from a single experiment. ### Key points of the paper: 1. **Problem background**: - Current data - driven and machine - learning methods have made significant progress in material modeling, but these methods usually require a large amount of stress - strain pair data and are black - box in nature and difficult to interpret. - It is very difficult to obtain stress data in experiments, especially for complex loading histories and multi - dimensional stress states. - Traditional material models rely on constitutive laws assumed a priori, which may lead to modeling errors. 2. **Proposed method**: - **EUCLID**: An unsupervised discovery framework that does not depend on a priori - selected material models and only uses full - field displacement and global force data. - **Material model library**: Construct the yield function through Fourier series expansion, and introduce the cumulative plastic multiplier and back stress to describe isotropic and kinematic hardening behaviors. - **Optimization problem**: Based on the weak form of linear momentum balance, use physical constraints to select the most relevant Fourier modes and identify hardening mechanisms. - **Sparse regularization**: Promote sparse solutions through \( \ell_p \)-regularization, avoid over - fitting and generate concise material models. 3. **Objective**: - Purely automatically discover the fully general and evolving plastic yield surfaces of three - dimensional elastoplastic solids from the two - dimensional displacement field and reaction force measurement data of a single experiment. - Achieve the automatic discovery of material models without stress data, relying only on full - field displacement and global force data. ### Formula summary: - **Yield function**: \[ f(r, \alpha, \gamma) = \sqrt{\frac{3}{2}} r - H_{\text{iso}}(\gamma) \sum_{i = 0}^{n_f} \theta_i \cos(3i\alpha) \] where: - \( r=\sqrt{\frac{\pi_1^2+\pi_2^2}{1 + \pi_2^2}} \) - \( \alpha = \text{atan2}(\pi_2, \pi_1) \) - \( \pi_1=\sqrt{\frac{2}{3}}\sigma_1-\sqrt{\frac{1}{6}}\sigma_2-\sqrt{\frac{1}{6}}\sigma_3 \) - \( \pi_2=\sqrt{\frac{1}{2}}\sigma_2-\sqrt{\frac{1}{2}}\sigma_3 \) - **Hardening function**: \[ H_{\text{iso}}(\gamma)=1 + H_{\text{iso},1}\gamma + H_{\text{iso},2}(1 - \exp(-H_{\text{iso},3}\gamma)) \] \[ \dot{\sigma}_{\text{back}}=H_{\text{kin},1}\dot{\epsilon}_p - H_{\text{kin},2}\dot{\gamma}\sigma_{\text{back}} \] - **Cost function**: \[ C(\theta, H)=C_{\text{free}}(\theta, H)+\lambda_r C_{\text{disp}}(\theta)

Discovering plasticity models without stress data

Automated discovery of generalized standard material models with EUCLID

Bayesian-EUCLID: discovering hyperelastic material laws with uncertainties

Automated model discovery of finite strain elastoplasticity from uniaxial experiments

Automated identification of linear viscoelastic constitutive laws with EUCLID

NN-EUCLID: Deep-learning hyperelasticity without stress data

The Language of Hyperelastic Materials

Automated discovery of interpretable hyperelastic material models for human brain tissue with EUCLID

Optimal Data-Generation Strategy for Machine Learning Yield Functions in Anisotropic Plasticity

Physics-informed Data-driven Discovery of Constitutive Models with Application to Strain-Rate-sensitive Soft Materials

Relating plasticity to dislocation properties by data analysis: scaling vs. machine learning approaches

Modular machine learning-based elastoplasticity: generalization in the context of limited data

Learning Physics-Consistent Material Behavior Without Prior Knowledge

On automated model discovery and a universal material subroutine

Constitutive model characterization and discovery using physics-informed deep learning

Mechanistically Informed Data-Driven Modeling of Cyclic Plasticity Via Artificial Neural Networks

Distance Minimizing-Based Data-Driven Computational Plasticity Method with Fixed Dataset

A Generalized, Computationally Versatile Plasticity Model Framework - Part I: Theory and Verification Focusing on Tension‒compression Asymmetry

A machine learning based plasticity model using proper orthogonal decomposition

Extreme sparsification of physics-augmented neural networks for interpretable model discovery in mechanics

Calibrating constitutive models with full-field data via physics informed neural networks