metadata
license: mit
language:
- en
tags:
- pytorch
- physics-informed-neural-network
- rock-mechanics
- crack-prediction
Transformer-PINN Rock Crack Prediction
Code for paper: Mechanism of micro-damage evolution in rocks under multiple coupled cyclic stresses
Requirements
Python 3.8+, PyTorch 2.0+, numpy, pandas, scikit-learn, matplotlib
Install: pip install -r requirements.txt
Input
5D feature vector: pH (1-7), FN (5-40), FT (10-40), T (25-900), phase (0 or 1)
Output
72D angular distribution representing crack count per 5-degree bin
Usage
Training:
python train.py
Inference:
python inference.py
Files
model.py - network architecture
data_loader.py - data processing
train.py - training script
inference.py - prediction script
Physics
Mogi-Coulomb yield criterion
Weibull strength distribution
Energy-based damage evolution
D0 = 1 - (1-Dft)(1-Dch)(1-Dth)
lambda = 1 - D0