Cyril-fr/blast-fragmentation-ai

AI-powered blast fragmentation prediction & optimisation system — Kuz-Ram, neural networks, NSGA-II, Ghana EPA compliance. Final-year project, Mining Engineering, UMaT. # AI-Powered Blast Fragmentation Prediction & Optimization System A system for predicting and optimizing rock fragmentation in open-pit mining operations, combining classical engineering models with machine learning. ## Features ### 🎯 Fragmentation Prediction - **Kuz-Ram Model** — industry-standard empirical model (Cunningham 1983/1987) - **Rosin-Rammler Distribution** — size distribution fitting - **Swebrec Model** — better fit for finer fractions (Ouchterlony 2005) - **Neural Network Predictor** — sklearn ensemble (RF + GB + MLP) with uncertainty ### 🔧 Parameter Optimization - **NSGA-II** — multi-objective (fragmentation / cost / vibration) with Pareto front - **Bayesian Optimization** — sample-efficient single-objective search - **Curve Matching** — find parameters that reproduce a target distribution - **Quick Optimize** — fast golden-section search for a target P80 ### 📊 Analysis & Data - Muckpile image analysis (OpenCV watershed segmentation) - Sieve-data upload with automatic percentile extraction - Prediction & measurement results persisted per blast design - Performance trends: target vs predicted vs actual P80 over time - Explosives catalog (seeded with common products incl. wet-hole emulsion blends) & rock-properties records seeded with Ghana Birimian/Tarkwaian presets - Kuz-Ram rock factor calculator from RMR / density / UCS ### 🛡️ Environmental Compliance - Sensitive-receptor register (villages, schools, structures) with Ghana EPA guideline limits (2 mm/s PPV, 117 dBL airblast) per receptor - Pre-blast assessment: MIC from charge geometry, USBM scaled-distance PPV and airblast prediction per receptor with pass/fail, maximum compliant MIC, Lundborg flyrock clearance radius, timing sanity checks (8 ms rule, burden relief, cutoff risk) - Measured seismograph records per blast; site-law calibration (log-log fit of K and β with R², provenance tracked) ### 👥 Accountability (multi-user) - Login with engineer / approver / admin roles - Blast lifecyc …