Copper — sulfides & oxides
From blasting to cathode: optimization of fragmentation, ore blending, flotation or SX-EW. The twin connects geometallurgy with the plant to maximize recovery and throughput at minimum energy.
Simulation + Real-time optimization · Copper & Lithium
The Waze of mining: your mine plan, recalculated in real time.
Waze doesn't tell you your route was wrong: it recalculates it every second with live data. We do the same with your mine's extraction plan — a digital twin that simulates the entire operation and an optimizer that corrects decisions online, bench by bench, shift by shift, to raise the value of every tonne of metal.
An alliance of Honeywell (industrial OPC layer, process control) + NVIDIA (GPU-accelerated computing) + TRIBUCORP (AI engineering, NPN partner and Dell Services).
US$80M
in additional annual profit achieved by Codelco at Chuquicamata using ML on extracted-ore data (≈ 8,000 t Cu/year). Source: Codelco / Bloomberg.
+Recovery
BHP Escondida raises copper recovery with AI recommendations on real-time plant data, including hourly predictions for flotation. Source: BHP.
+4.4% / −7.6%
more throughput and less energy in SAG milling with ML models, in a Chilean copper mining case. Source: peer-reviewed study (MDPI, 2023).
−10% diesel
in Komatsu FrontRunner autonomous fleets with real-time optimization, operating 40 Mt/year across 10 mines. Source: Komatsu / VMR.
“Every tonne moved on a six-month-old plan is margin left buried in the pit.”
Chuquicamata mine, Chile — the world's largest open pit
01 · TWO PHASES
First the map. Then the living route.
Waze needs two things: a faithful map of the streets and real-time data from drivers. At the mine, the map is the digital twin built from your historical data; the live data arrives via OPC from the industrial layer Honeywell already operates.
Phase 1 · The map
Simulation — a digital twin of the operation
With drillhole data, the block model, the mine plan and operating history (dispatch, blasting, crushing, plant) we build a simulator of the entire mine, calibrated against real results, running on GPUs.
- Ingestion and curation of years of historical data with RAPIDS (cuDF/cuML) and NeMo Curator
- Hybrid models: process physics + ML trained on Tensor Cores
- Thousands of "what if?" scenarios per hour: grades, fragmentation, fleet availability, water and energy
- Validation against real KPIs: the twin must reproduce your history before predicting your future
Phase 2 · The living route
Optimization — online decisions on the plan
The optimizer compares in real time what the mine is doing (OPC data) against what the plan expected, and recalculates: ore blending, plant setpoints, fleet assignment, bench sequencing. Recommendations arrive on the control boards your operators already use.
- Live industrial data via OPC UA from DCS/SCADA (Experion and others), without replacing your control layer
- Accelerated combinatorial optimization (NVIDIA cuOpt) and predictive control on the twin
- Explainable recommendations on your current boards: the operator decides, the system learns
- The Waze loop: execute → measure → compare with simulation → recalculate → improve the map
02 · THE FULL OPERATION
From drillhole to metal: every link generates data. Every link can be optimized.
The step-by-step of an open-pit copper operation (with its lithium variant), and the signal each stage feeds into the digital twin.
Exploration & drilling
Diamond/reverse-circulation drilling campaigns; grade assays and geometallurgy.
→ geological database, grades, hardness
Block model
The orebody is discretized into blocks with grade, tonnage, hardness and estimated recovery.
→ the twin's "base map"
Mine plan
Pit design, phases, bench sequencing, cut-off grade, LOM production schedule.
→ the "planned route"
Drilling & blasting
Drill patterns and blasts that fragment the rock; fragmentation defines everything downstream.
→ pattern design, vibration, P80
Loading & hauling
Shovels and haul trucks move ore and waste; dispatch assigns routes and destinations.
→ fleet GPS, cycles, tonnage, diesel
Crushing
Primary/secondary size reduction; the typical bottleneck of the operation.
→ OPC: power, level, particle size
Grinding
SAG and ball mills bring ore to liberation size; the largest energy consumer on site.
→ OPC: kWh/t, load, pressure, P80
Concentration / extraction
Cu sulfides: flotation → concentrate. Cu oxides: leaching + SX-EW → cathodes. Li brine: ponds/DLE → carbonate.
→ OPC: recovery, grades, reagents
Smelting & refining
Concentrate is smelted and refined; final metal quality and value are consolidated here.
→ realized value per tonne
The mine-to-mill key: blasting decisions (stage 4) determine grinding energy (stage 7) and recovery (stage 8). Today those decisions are made in silos, weeks apart. The Waze of mining connects them in a single, online optimization loop.
03 · TWO CRITICAL METALS
Copper and lithium: the metals of the energy transition, optimized tonne by tonne.
The same platform adapts to the two production circuits that dominate South American mining.
Lithium — brines & DLE
On the salt flats, the twin models evaporation ponds, brine chemistry and direct lithium extraction (DLE): water balance, residence times and carbonate yield, with climate variability built into the model.
04 · TECHNOLOGY
Your industrial layer + NVIDIA accelerated computing. Nothing is replaced; everything is amplified.
The architecture respects the existing automation pyramid: the DCS keeps controlling, the boards remain yours. A layer of intelligence is added on top of the data already flowing through OPC.
Industrial layer (OT)
Sensors, PLC/DCS (Experion and others), SCADA, historians. The OPC UA protocol exposes process data in a secure, standard way.
OPC UA · MQTT
Honeywell Experion / Forge
Honeywell Experion / Forge
Ingestion & curation
GPU pipelines that unify history (plans, dispatch, lab) with OPC streaming; cleaning, time alignment and features at scale.
NVIDIA RAPIDS (cuDF · cuML)
NeMo Curator
NeMo Curator
Simulator (digital twin)
Physical models + neural networks trained on Tensor Cores; 3D visualization of the operation and massive parallel scenarios.
CUDA · Tensor Cores · PhysicsNeMo
Omniverse / OpenUSD
Omniverse / OpenUSD
Online optimizer
Combinatorial optimization of fleet, blending and setpoints; millisecond inference; recommendations with confidence intervals.
NVIDIA cuOpt · Dynamo-Triton
NIM microservices
NIM microservices
Your current control boards
Recommendations appear where the operator already works: HMIs, integrated operations rooms, corporate dashboards. Human in command.
OPC write / API
HMI · IROC rooms
HMI · IROC rooms
“A 400-tonne haul truck burns the same diesel on the right route or the wrong one. The difference is who recalculates.”
Caterpillar 797 at Minera Escondida, Chile
05 · HOW WE START
A 16-week POC at a selected mine.
Together we choose a bounded, high-value circuit (typically blasting→crushing→grinding, or the flotation circuit) and demonstrate the impact with the real OPC data of your operation.
WEEKS 1–3
Data diagnostic
Inventory of sources: OPC historian, dispatch, laboratory, mine plan. Agreement on KPIs and the economic baseline of the chosen circuit.
WEEKS 4–8
Calibrated twin
GPU curation of historical data and construction of the simulator. Exit criterion: reproduce operating history within agreed tolerances.
WEEKS 9–13
Shadow-mode optimizer
The optimizer runs in parallel with the operation ("shadow mode"): it recommends without acting. We measure the delta between recommended and executed.
WEEKS 14–16
Business case
Quantification of capturable value per tonne, scaling architecture and roadmap to move to assisted loop and then production.
What the POC must demonstrate, in your mine's numbers:
- Δ metallurgical recovery (%) in the pilot circuit
- Δ throughput (t/h) with no additional CAPEX
- Δ energy (kWh/t) and water (m³/t) consumption
- Δ US$/tonne of metal — the KPI that governs everything
- Twin accuracy vs. reality (backtesting error)
- Adoption: % of recommendations accepted by operators
06 · FREQUENTLY ASKED QUESTIONS
What operations and IT teams ask us first.
What is the "Waze of mining"?
It is a real-time simulation and optimization platform for copper and lithium mining operations. Just as Waze recalculates your route with live traffic data, the platform builds a digital twin of the mine from your historical data and re-optimizes mine-plan decisions online — ore blending, plant setpoints, fleet assignment and bench sequencing — using the OPC data your operation already generates.
What is a digital twin of a mine and how is it built?
A digital twin is a calibrated simulator of the entire operation: from the block model to the plant. It is built from drillhole data, the mine plan and years of dispatch, blasting, crushing and plant history, processed on NVIDIA GPUs with RAPIDS and hybrid models (process physics + machine learning). Before predicting the future, the twin must reproduce real operating history within agreed tolerances.
Do we have to replace our DCS, SCADA or existing control boards?
No. The architecture respects the existing automation pyramid: the DCS (Experion or other) keeps controlling and operators keep using their boards. The platform reads process data via OPC UA and returns explainable recommendations to your current HMIs and dashboards. The operator always decides; the system learns.
How long does the proof of concept (POC) take and what does it deliver?
The POC runs for 16 weeks on a bounded, high-value circuit (for example blasting→crushing→grinding, or flotation). It delivers a calibrated digital twin, an optimizer running in shadow mode and a quantified business case: deltas in recovery, throughput, energy, water and US$/tonne, plus the scaling roadmap.
What data is needed to start?
What your operation already generates: the DCS/SCADA OPC historian, fleet dispatch data, laboratory results, the block model and the mine plan. The first 3 weeks of the POC are precisely a diagnostic of these sources; no new instrumentation is required to begin.
Does it work for both copper and lithium?
Yes. For copper it covers sulfides (flotation → concentrate) and oxides (leaching + SX-EW → cathodes). For lithium it models evaporation ponds and direct extraction (DLE), incorporating water balance and climate variability. The mine-to-mill optimization chain is the same: connecting decisions that today are made in silos.
What results has this approach achieved in the industry?
Public cases from major miners back it up: Codelco reported ~US$80M in additional annual profit at Chuquicamata using ML on ore data; BHP Escondida raises copper recovery with real-time AI; peer-reviewed studies show +4.4% throughput and −7.6% energy in SAG milling; and Komatsu autonomous fleets cut diesel consumption by ~10%.
The mine that recalculates wins. The one following a 6-month-old plan pays for it.
Codelco, BHP and the majors have already proven it with in-house platforms. Our alliance puts that same capability — Honeywell industrial layer, NVIDIA computing, TRIBUCORP engineering — within reach of any copper or lithium operation, as a service.
Book a free diagnostic