Laser_Ultrasound_Reconstruction_PINN

This repository contains the PyTorch implementation and datasets for the paper: "[Robust Sparse Wavefield Reconstruction and Defect Localization in Low-SNR Laser Ultrasonics Using Physics-Informed Neural Networks]".

We propose a Physics-Informed Neural Network (PINN) framework to reconstruct high-fidelity laser ultrasound signals from low-SNR measurements. The method is validated on both defect-free and defective samples.

📂 Repository Structure

.
├── data/
│   ├── intact/
│   │   ├── low_snr/                # Input: Noisy signals (Intact samples)
│   │   │   └── intact_low_snr.pkl
│   │   └── high_snr_gt/            # Ground Truth: Clean signals (Intact samples)
│   │       └── intact_high_snr_gt.pkl
│   ├── damaged/
│   │   ├── low_snr/                # Input: Noisy signals (Damaged samples)
│   │   │   └── damaged_low_snr.pkl
│   │   └── high_snr_gt/            # Ground Truth: Clean signals (Damaged samples)
│   │       └── damaged_high_snr_gt.pkl
├── models/                # Saved checkpoints and pre-trained models
├── src/                   # Source code for PINN, loss functions, etc.
├── train.py               # Script to train the PINN
├── evaluate.py            # Script to test and visualize results
├── requirements.txt       # Python dependencies
└── README.md