Complementary physical and data-driven-based approaches for crack detection in metal powder compaction

This repository contains force sensor data and Jupyter notebook analysis code for the manuscript "Complementary physical and data-driven-based approaches for crack detection in metal powder compaction" submited to Materials Letters (2026). Contents: - Force sensor measurements from 400 powder compaction specimens (190 MB JSON line file) - Jupyter notebook implementing five ensemble machine learning classifiers - Complete documentation for reproducibility - Dependency specifications (requirements.txt) Key Features: - 6 force sensors with 700 measurements per compaction cycle - Feature engineering: statistical features from force differentials - Models: Random Forest, Gradient Boosting, AdaBoost, Bagging, Extra Trees - Performance: >90% accuracy in crack detection - FAIR compliant: Findable, Accessible, Interoperable, Reusable The Jupyter notebook is fully documented and can be executed with minimal setup (pip install -r requirements.txt).