A Spectral-based Physics-informed Finite Operator Learning for Prediction of Mechanical Behavior of Microstructures
arXiv:2410.19027 · doi:10.1016/j.jmps.2025.106219
Abstract
A novel physics-informed operator learning technique based on spectral methods is introduced to model the complex behavior of heterogeneous materials. The Lippmann-Schwinger operator in Fourier space is employed to construct physical constraints with minimal computational overhead, effectively eliminating the need for automatic differentiation. The introduced methodology accelerates the training process by enabling gradient construction on a fixed, finite discretization in Fourier space. Later, the spectral physics-informed finite operator learning (SPiFOL) framework is built based on this discretization and trained to map the arbitrary shape of microstructures to their mechanical responses (strain fields) without relying on labeled data. The training is done by minimizing equilibrium in Fourier space concerning the macroscopic loading condition, which also guarantees the periodicity. SPiFOL, as a physics-informed operator learning method, enables rapid predictions through forward inference after training. To ensure accuracy, we incorporate physical constraints and diversify the training data. However, performance may still degrade for out-of-distribution microstructures. SPiFOL is further enhanced by integrating a Fourier Neural Operator (FNO). Compared to the standard data-driven FNO, SPiFOL shows higher accuracy in predicting stress fields and provides nearly resolution-independent results. Additionally, its zero-shot super-resolution capabilities are explored in heterogeneous domains. Finally, SPiFOL is extended to handle 3D problems and further adapted to finite elasticity, demonstrating the robustness of the framework in handling nonlinear mechanical behavior. The framework shows great potential for efficient and scalable prediction of mechanical responses in complex material systems while also reducing the training time required for training physics-informed neural operators.
References in corpus (25)
- DeepONet: Learning nonlinear operators for identifying differential equations based on the universal approximation theorem of operators
- A numerical method for computing the overall response of nonlinear composites with complex microstructure
- Neural Tangent Kernel: Convergence and Generalization in Neural Networks
- Physics-informed neural networks for inverse problems in nano-optics and metamaterials
- A physics-informed variational DeepONet for predicting the crack path in brittle materials
- CAN-PINN: A Fast Physics-Informed Neural Network Based on Coupled-Automatic-Numerical Differentiation Method
- PhyCRNet: Physics-informed Convolutional-Recurrent Network for Solving Spatiotemporal PDEs
- Finite strain FFT-based non-linear solvers made simple
- Novel DeepONet architecture to predict stresses in elastoplastic structures with variable complex geometries and loads
- Geom-DeepONet: A Point-cloud-based Deep Operator Network for Field Predictions on 3D Parameterized Geometries
- Mixed formulation of physics-informed neural networks for thermo-mechanically coupled systems and heterogeneous domains
- A Survey of Complex-Valued Neural Networks
- Applications of physics informed neural operators
- Solving the Discretised Neutron Diffusion Equations using Neural Networks
- A Mathematical Guide to Operator Learning
- On the Effectiveness of Neural Operators at Zero-Shot Weather Downscaling
- Operator Learning: Algorithms and Analysis
- Learning the Stress-Strain Fields in Digital Composites using Fourier Neural Operator
- Continuum Attention for Neural Operators
- Comparison of two artificial neural networks trained for the surrogate modeling of stress in materially heterogeneous elastoplastic solids
- Improved Operator Learning by Orthogonal Attention
- Finite Operator Learning: Bridging Neural Operators and Numerical Methods for Efficient Parametric Solution and Optimization of PDEs
- An Implicit Adaptive Fourier Neural Operator for Long-term Predictions of Three-dimensional Turbulence
- A physics-encoded Fourier neural operator approach for surrogate modeling of divergence-free stress fields in solids
- A Physics-Informed Meta-Learning Framework for the Continuous Solution of Parametric PDEs on Arbitrary Geometries