📋 Case Study

Iron Ore Mine Sensor Fusion for Banded Iron Formation (BIF) Delineation

Conventional geophysics failed to resolve thin hematite bands (<2m) within jaspilite, causing 22% grade variance in ROM feed

🏗️ Project Overview

Pilbara open pit, Australia

🎯 Challenge

Conventional geophysics failed to resolve thin hematite bands (<2m) within jaspilite, causing 22% grade variance in ROM feed

🔧 Design Approach

Multi-scale fusion: airborne magnetics (200m), ground EM (10m), and drill-core hyperspectral (0.5m); ensemble U-Net segmentation trained on 12,000 core images

📐 Design Diagram

AI-generated project design illustration

📐 Key Calculations

Band Detection Precision

TP/(TP+FP)

Result: 94.1%

Critical for ROM blending accuracy

Feed Grade Std Dev Reduction

σₚᵣₑ − σₚₒₛₜ

Result: 1.42% → 0.67% Fe

Reduces downstream processing variability

📊 Results

ROM feed grade variance halved, crusher throughput increased 7.2% via optimized blend sequencing, $14.2M annual savings in penalty payments

💡 Lessons Learned

•Core image labeling required stratigraphic expert consensus
•Airborne-to-ground resolution gap bridged via generative adversarial upscaling
•Model versioning tied to blast round numbers for traceability

✅ Key Takeaways

1Core image labeling required stratigraphic expert consensus
2Airborne-to-ground resolution gap bridged via generative adversarial upscaling
3Model versioning tied to blast round numbers for traceability