Reproduce “Capacitive Touch Sensor Modeling With a Physics-informed Neural Network and Maxwell's Equations” on OpenResearch

About this paper

Ganyong Mo, Krishna Kumar Narayanan, David Castells-Rufas, Jordi Carrabina

This work introduces a Physics-Informed Neural Network (PINN) for modeling capacitive touch sensors, accurately predicting electric field distributions and potentials by incorporating Maxwell's equations into the training process. The PINN achieves a 10x speedup in inference time compared to traditional Finite Element Method (FEM) simulations, providing accurate results with an average NRMSE of 6.4% across various finger proximities.

Key insights

• Physics-Informed Neural Networks (PINNs) can accurately model complex electromagnetic fields by integrating Maxwell's equations directly into the neural network's loss function.
• A combinatorial loss function, balancing data fidelity, physical laws, and boundary conditions, is effective for training PINNs to ensure both accuracy and physical consistency.
• PINNs offer continuous solutions across the domain and inherent differentiability, enabling efficient parametric analysis and gradient-based optimization for sensor design.

Summary by alphaXiv.