PINN for Shell Structures - JAX

This repository provides unofficial, partial JAX-implementation for 'Physics-Informed Neural Networks for shell structures'.

By re-writing the original code, we could obtain substantial speed-ups. Please consult with result_jax_x64.txt and result_torch_x64.txt for a performance comparison.

How to run

Training a neural network in double-precision: JAX_ENABLE_X64=True python main.py.

Measuring relative $L^2$ error: python err_measure.py --method=lbfgs

Requirements

Please consult with requirements.txt.

Current Implementation

1. Fully clamped hyperbolic paraboloid
   • data

Reference

• Physics-Informed Neural Networks for shell structures
• JAX
• Equinox