LPBF Clustering and Explainable AI

This repository contains the source code for:

1. Generating K-Means Clusters using the proposed multi-step K-Means method.
2. Developing the melt pool shape prediction neural network.
3. Interpreting the trained neural network with Layer-wise Relevance Propagation.

Datasets

The datasets used in this work are produced by the Additive Manufacturing Metrology Testbed at the National Institute of Standards and Technology (NIST). The 3D Scan Strategies is publicly available.

Requirements

• Python 3.6+
• PyTorch
• Jupyter Notebook
• scikit-learn, numpy, matplotlib, pandas, OpenCV

Description of Files

• src/K_Means.ipynb: Performs the multistep K-Means process and provides visualization of the process.
• src/MeltpoolShapePrediction.ipynb: Generates proposed neural network architecture and deep learning training routine.
• src/LRP.ipynb: Interprets the trained neural network with LRP and analyzes the results, produces visualization of the LRP results.

Contributors

• Matthew Sato
• Stanford Engineering Informatics Group