AI Autopilot — Gold Grinding Circuit Optimization
Challenge
The grinding circuit set the pace of the whole plant — and manual control was limiting its throughput and consistency. Operators could not hold optimal setpoints 24/7 as rock type, feed blend and percent solids shifted underneath them.
Method
25-day controlled test (January 10 – February 5, 2024) on the same grinding circuit, alternating AI-guided and manual operation in 48-hour cycles to control for rock type, feed blend, percent solids and operating rates.
Only periods with ≥85% active operational hours entered the analysis — 265 operational hours per mode — ensuring a like-for-like comparison. The test followed a two-year collaboration between Brainiall and the producer, and was delivered in partnership with SoftwareOne on AWS cloud infrastructure.
Inside the circuit
AI Autopilot read the circuit’s live process data and recommended the next best setpoints in real time — an advisory layer on top of the existing control system, with operators in command and safety systems untouched throughout the test.
Results
- 435 t/h (AI) vs 412 t/h (manual) — +6% throughput
- 115,347 tons processed under AI vs 109,079 tons under manual, over matched 265-hour windows
- +3% on hard ore (Wi ≥ 14 kWh/t), above plant design rates
- Particle size (p80) held consistent, with lower operational variance — less equipment stress and more predictable downstream flotation recovery
- Specific energy consumption (kWh/t) essentially flat — lower energy intensity per ton
- ≈2,694 additional ounces/year — ≈$781K incremental annual profit (net of AISC)
Results may vary with ore grade and operating conditions.