Neural network-augmented differentiable finite element method for boundary value problems

• Published in: International journal of mechanical sciences Vol. 285; p. 109783
• Main Authors: Wang, Xi, Yin, Zhen-Yu, Wu, Wei, Zhu, He-Hua
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.01.2025
• Subjects: Differentiable numerical analysis; Finite element method; Fully connected neural network; Graph convolutional network; Multi-fidelity learning; Physics-informed neural network; Finite element method; Physics-informed neural network; Graph convolutional network; Differentiable numerical analysis; Fully connected neural network; Multi-fidelity learning

•A novel Neural Network-Augmented Differentiable Finite Element Method (NNDFEM) is proposed.•NNDFEM generalizes different kinds of neural networks in its backend and reveals their differences.•NNDFEM systematically unveils the role of 3D finite element approximation.•NNDFEM mitigates the overfitting...