Spectral Normalization and Voigt-Reuss net: A universal approach to microstructure-property forecasting with physical guarantees

TL;DR

The paper addresses the need for rapid, physically consistent forecasting of effective properties in heterogeneous materials by embedding Voigt-like upper and Reuss-like lower bounds into machine-learned surrogates. It introduces spectral normalization, a model-agnostic framework that enforces Löwner bounds for symmetric positive definite tensors, instantiated as the Voigt-Reuss net (VR net) using an orthogonal-parameterized decomposition of the bound gap. The approach is validated on large 2D and 3D thermal homogenization datasets generated with Fourier-Accelerated Nodal Solvers, showing improved accuracy, robustness, and guaranteed adherence to bounds, especially in data-scarce regimes. The method is architecture- and input-agnostic, enabling seamless integration into digital twins and microstructure-design workflows, with broad applicability to conductivity, permeability, elasticity, and beyond.

Abstract

Heterogeneous materials are crucial to producing lightweight components, functional components, and structures composed of them. A crucial step in the design process is the rapid evaluation of their effective mechanical, thermal, or, in general, constitutive properties. The established procedure is to use forward models that accept microstructure geometry and local constitutive properties as inputs. The classical simulation-based approach, which uses, e.g., finite elements and FFT-based solvers, can require substantial computational resources. At the same time, simulation-based models struggle to provide gradients with respect to the microstructure and the constitutive parameters. Such gradients are, however, of paramount importance for microstructure design and for inverting the microstructure-property mapping. Machine learning surrogates can excel in these situations. However, they can lead to unphysical predictions that violate essential bounds on the constitutive response, such as the upper (Voigt-like) or the lower (Reuss-like) bound in linear elasticity. Therefore, we propose a novel spectral normalization scheme that a priori enforces these bounds. The approach is fully agnostic with respect to the chosen microstructural features and the utilized surrogate model. All of these will automatically and strictly predict outputs that obey the upper and lower bounds by construction. The technique can be used for any constitutive tensor that is symmetric and where upper and lower bounds (in the Löwner sense) exist, i.e., for permeability, thermal conductivity, linear elasticity, and many more. We demonstrate the use of spectral normalization in the Voigt-Reuss net using a simple neural network. Numerical examples on truly extensive datasets illustrate the improved accuracy, robustness, and independence of the type of input features in comparison to much-used neural networks.

Figures (17)

• Figure 1: Schematic representation of the one-dimensional laminate structure and its effective Young's moduli.
• Figure 2: General machine-learning model employing spectral normalization to enforce two-sided Löwner bounds.
• Figure 3: Feed forward neural network architecture for predicting effective property tensor with guaranteed Voigt-Reuss bounds. The feature vector contains the material parameters as well as morphological descriptors representing the microstructure. The output contains normalized eigenvalues $\ul{\widehat{\xi}}_\lambda$ and parameterized orthogonal matrix representation $\ul{\widehat{\xi}}_\mathrm{q}$ that maps to a physically valid effective tensor $\ul{\ul{\widehat{Y}}}$ via $\widehat{\ul{\ul{\widetilde{Y}}}}$.
• Figure 4: Workflow for constructing the effective conductivity tensor $\overline{\hbox{\boldmath $\kappa$}}$ from high-fidelity simulations.
• Figure 5: Some exemplary 2D microstructures used for training and validation.

Theorems & Definitions (4)

• remark 1
• remark 2
• remark 3
• remark 4