Abstract:Traditional electrostatic simulation are meshed-based methods which convert partial differential equations into an algebraic system of equations and their solutions are approximated through numerical methods. These methods are time consuming and any changes in their initial or boundary conditions will require solving the numerical problem again. Newer computational methods such as the physics informed neural net (PINN) similarly require re-training when boundary conditions changes. In this work, we propose an end-to-end deep learning approach to model parameter changes to the boundary conditions. The proposed method is demonstrated on the test problem of a long air-filled capacitor structure. The proposed approach is compared to plain vanilla deep learning (NN) and PINN. It is shown that our method can significantly outperform both NN and PINN under dynamic boundary condition as well as retaining its full capability as a forward model.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the inefficiency caused by the need to recalculate traditional electrostatic field analysis methods when boundary conditions change. Specifically, traditional grid - based methods (such as the finite - difference method) and the relatively new physics - informed neural network (PINN) both need to be re - solved or re - trained when boundary conditions change, which is very time - consuming and inefficient. ### Problem Description 1. **Limitations of Traditional Methods**: - Traditional electrostatic simulation methods (such as the finite - difference method, FDM) convert partial differential equations into algebraic equation systems and approximate solutions through numerical methods. These methods have long calculation times and need to be re - solved when initial or boundary conditions change. - Although the physics - informed neural network (PINN) can simplify the solution process to a certain extent, it still needs to be re - trained when boundary conditions change. 2. **Research Objectives**: - Propose an end - to - end deep - learning method that can avoid recalculation or re - training when boundary conditions change, thereby significantly improving computational efficiency. - This method is particularly suitable for the problem of electric field distribution in capacitive structures and aims to quickly reconstruct the corresponding electric field distribution by directly inputting boundary parameters. ### Solution The paper proposes a new method named "Boundary - Decoder Network", and its main features are as follows: - **End - to - End Deep - Learning Model**: This model can directly accept changes in boundary conditions as input and reconstruct the corresponding electric field distribution online. - **Efficient Representation**: Regression modeling, inverse prediction, and latent - space representation are all processed by a boundary network. - **Semi - Supervised Model**: The latent space is jointly trained by an encoder - decoder (unsupervised) and a boundary - decoder network (supervised). ### Experimental Results The experimental results show that this method is significantly superior to traditional neural networks (NN) and physics - informed neural networks (PINN) under dynamic boundary conditions and performs well in both the original space and the latent space. Specifically: - **Error Comparison**: Under different boundary - condition parameters \( d \), the squared error of the proposed method is significantly lower than that of other methods. - **Generalization Ability**: Even under unseen boundary conditions, this method can maintain high accuracy and stability. ### Conclusion The boundary - decoder network proposed in the paper provides an efficient and flexible method to handle the problem of boundary - condition changes in electrostatic field analysis, especially performing excellently under dynamic boundary conditions and being able to significantly reduce calculation time and resource consumption. ### Formula Representation The main formulas involved in the paper include: - Boundary - Condition Definition: \[ V=\delta V / \delta y, d \quad \text{for} \quad 2 \leq x \leq a / 2, y = 0 \] \[ V = V_0, \quad 1 \leq x \leq d / 2, y = 1 \] - Loss Function: \[ \text{enc - dec loss}=\min_w \| T - \hat{\gamma} \|^2 \] \[ \text{bou - dec loss}=\min_w \| T - \hat{\gamma} \|^2-\lambda \| \gamma - \hat{\gamma} \|^2 \] These formulas show the training objectives and optimization processes of the model.

Boundary-Decoder network for inverse prediction of capacitor electrostatic analysis

A Physics-Informed Neural Networks Algorithm for Simulating Semiconductor Devices

Physics-Informed Neural Networks for Solving 2-D Magnetostatic Fields

A Spatial-Temporally Adaptive PINN Framework for 3D Bi-Ventricular Electrophysiological Simulations and Parameter Inference

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

Physics-Informed Neural Networks for Electrical Circuit Analysis: Applications in Dielectric Material Modeling

Physics-Informed Neural Networks for Solving Two-Dimensional Magneto-Static Fields

Extremization to Fine Tune Physics Informed Neural Networks for Solving Boundary Value Problems

BINN: A deep learning approach for computational mechanics problems based on boundary integral equations

Physics informed deep learning for computational elastodynamics without labeled data

FDM data driven U-Net as a 2D Laplace PINN solver

Domain-decoupled Physics-informed Neural Networks with Closed-form Gradients for Fast Model Learning of Dynamical Systems

Physics-informed machine learning of redox flow battery based on a two-dimensional unit cell model

A Physics Informed Neural Network for Time-Dependent Nonlinear and Higher Order Partial Differential Equations

Physics-Informed Fully Convolutional Networks for Forward Prediction of Temperature Field and Inverse Estimation of Thermal Diffusivity

Transfer learning based physics-informed neural networks for solving inverse problems in engineering structures under different loading scenarios

Theory-guided physics-informed neural networks for boundary layer problems with singular perturbation

Approximating electromagnetic fields in discontinuous media using a single physics-informed neural network

Boundary integrated neural networks (BINNs) for 2D elastostatic and piezoelectric problems: Theory and MATLAB code

Exact Dirichlet boundary Physics-informed Neural Network EPINN for solid mechanics

PhyCRNet: Physics-informed convolutional-recurrent network for solving spatiotemporal PDEs