Abstract:Accurately modeling the mechanical behavior of materials is crucial for numerous engineering applications. The quality of these models depends directly on the accuracy of the constitutive law that defines the stress-strain relation. Discovering these constitutive material laws remains a significant challenge, in particular when only material deformation data is available. To address this challenge, unsupervised machine learning methods have been proposed. However, existing approaches have several limitations: they either fail to ensure that the learned constitutive relations are consistent with physical principles, or they rely on a predefined library of constitutive relations or manually crafted input features. These dependencies require significant expertise and specialized domain knowledge. Here, we introduce a machine learning approach called uLED, which overcomes the limitations by using the input convex neural network (ICNN) as the surrogate constitutive model. We improve the optimization strategy for training ICNN, allowing it to be trained end-to-end using direct strain invariants as input across various materials. Furthermore, we utilize the nodal force equilibrium at the internal domain as the training objective, which enables us to learn the constitutive relation solely from temporal displacement recordings. We validate the effectiveness of the proposed method on a diverse range of material laws. We demonstrate that it is robust to a significant level of noise and that it converges to the ground truth with increasing data resolution. We also show that the model can be effectively trained using a displacement field from a subdomain of the test specimen and that the learned constitutive relation from one material sample is transferable to other samples with different geometries. The developed methodology provides an effective tool for discovering constitutive relations.

What problem does this paper attempt to address?

The paper attempts to address the problem of learning physically consistent material behavior relationships (i.e., stress-strain relationships) from material deformation data without prior knowledge. Specifically, existing methods have the following limitations when learning the constitutive relationships of materials: 1. **Physical Consistency**: Many existing machine learning methods cannot ensure that the learned constitutive relationships are consistent with physical principles. 2. **Dependence on Predefined Libraries**: Some methods rely on predefined libraries of constitutive relationships or manually designed input features, which require extensive expertise and domain knowledge. 3. **Data Limitations**: In practical applications, especially in situ measurements, strain data is often available while stress data cannot be directly obtained. To overcome these limitations, the paper proposes a new unsupervised machine learning method called uLED (unsupervised learning of hyper elastic constitutive relations in dynamic environments). This method uses Input Convex Neural Networks (ICNN) as a surrogate constitutive model and learns the constitutive relationships through displacement records in dynamic environments. The main features of uLED include: - **No Stress Data Required**: Training is conducted using only displacement data, without the need for stress data. - **No Predefined Libraries**: It does not rely on predefined libraries of constitutive relationships or manually designed input features. - **Physical Consistency**: An optimization strategy ensures that the learned constitutive relationships are consistent with physical principles. - **Dynamic Formulation**: It is suitable for strain rate-dependent materials and situations where constitutive relationships need to be inferred from in situ measurements. With these improvements, uLED can effectively learn and generalize constitutive relationships under different materials and geometries, and it is robust to noise.

Learning Physics-Consistent Material Behavior Without Prior Knowledge

Learning the physics-consistent material behavior from experimentally measurable data via PDE-constrained optimization

Learning solution of nonlinear constitutive material models using physics-informed neural networks: COMM-PINN

NN-EUCLID: Deep-learning hyperelasticity without stress data

Automated model discovery of finite strain elastoplasticity from uniaxial experiments

Learning solutions of thermodynamics-based nonlinear constitutive material models using physics-informed neural networks

Constitutive model characterization and discovery using physics-informed deep learning

Equilibrium-based Convolution Neural Networks for Constitutive Modeling of Hyperelastic Materials

Neural Material: Learning Elastic Constitutive Material and Damping Models from Sparse Data

Learning Invariant Representation of Multiscale Hyperelastic Constitutive Law from Sparse Experimental Data

Inverse Modeling of Viscoelasticity Materials using Physics Constrained Learning

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

Thermodynamically Consistent Machine-Learned Internal State Variable Approach for Data-Driven Modeling of Path-Dependent Materials

Unbiased construction of constitutive relations for soft materials from experiments via rheology-informed neural networks

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

A generic physics-informed neural network-based constitutive model for soft biological tissues

Machine learning of evolving physics-based material models for multiscale solid mechanics

A Physics-informed Assembly of Feed-Forward Neural Network Engines to Predict Inelasticity in Cross-Linked Polymers

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

A Physics Preserving Neural Network Based Approach for Constitutive Modeling of Isotropic Fibrous Materials

Learning Elastic Constitutive Material and Damping Models.