Grinding alone consumes up to 70% of a mineral processing plant's energy budget, and every kilowatt-hour per tonne saved at the mill drops straight to the operating margin. The catch is that grinding is also where ore competency, mill load, motor draw, and downstream flotation chemistry all interact at once — too many variables for an operator to optimize manually, too few for fixed SCADA setpoints to capture, and too dynamic for the monthly metallurgical review to catch in time. An iFactory closed-loop module runs a plant-wide, on-prem AI model that reads thousands of sensor tags every second, learns the specific ore body and the specific circuit, and closes the loop from sensor reading to setpoint recommendation in real time. Deployment is eight to twelve weeks against existing PLCs and historians, the model runs locally for 99.9% uptime regardless of mine site connectivity, and operators get a recommendation per shift instead of a report per month.
iFactory AI · Mining & Mineral Processing
Predictive Closed-Loop Manufacturing for Mining
AI-native, on-prem, mine-site ready. Closes the loop from crusher to tailings, learns your specific ore body, runs at 99.9% uptime, and deploys in 8 to 12 weeks against the PLCs and historians already on site.
70%
grinding's share of processing energy
99.9%
uptime SLA, on-prem
8-12 wks
to first closed-loop recommendation
Site-local
no cloud dependence
What "Closed-Loop Manufacturing" Means at a Mine Site
In a mineral processing plant, closed-loop manufacturing means the AI doesn't stop at telling the operator something is off — it tells the operator exactly which setpoint to change, projects what the change will produce, and then verifies the outcome against the prediction once the move is made. The loop closes because the verification feeds back into the model, which gets sharper with every cycle. Open-loop monitoring tells you about a problem after the fact. Closed-loop control prevents it, then proves it prevented it. On a mine site with declining ore grades and rising energy costs, that distinction is the entire ROI conversation.
The Ore Processing Loop — Where AI Closes Each Step
The closed-loop AI sits across every stage of the ore processing circuit, not just one. Optimizing a single crusher or mill in isolation usually pushes the bottleneck downstream — starving cyclones, choking flotation cells, or destabilizing the thickener. The plant-wide model treats every unit as part of one dynamic network, so the recommendation at the crusher already accounts for what it will do to flotation thirty minutes later.
1
Crushing
Primary and secondary crushers reduce ROM ore to a feed size for the mill circuit.
AI closes
Closed side setting, throughput vs power draw, recirculating load minimization, liner wear prediction.
→
2
Grinding
SAG and ball mills produce the slurry feed for separation. Single largest energy line.
AI closes
Mill load, motor power-draw limits, ore competency adaptation, kWh per tonne minimization, liner scheduling.
→
3
Flotation
Reagents and air separate valuable minerals from gangue into concentrate.
AI closes
Reagent dosing, froth depth, air rate, recovery vs grade tradeoff, predictive concentrate grade.
→
4
Thickening
Concentrate and tailings streams dewatered. Water recovered, solids prepared for disposal.
AI closes
Flocculant dosing, rake torque, underflow density, overflow clarity, water recovery rate.
→
5
Tailings
Tailings pumped to storage facility. Critical environmental and stability monitoring.
AI closes
Pump efficiency, pipe wear prediction, dust monitoring, dam pressure signals, regulatory reporting.
The Closed-Loop Cycle — Six Stages, Continuous
The cycle runs continuously, twenty-four hours a day, on every signal coming off the plant. Each stage feeds the next, and every cycle makes the model a little sharper at predicting the specific ore body, the specific equipment, and the specific operating conditions on this site.
1
Sense
Read thousands of tags continuously from PLCs, DCS, and instrumentation — power, flow, pressure, density, vibration, particle size.
→
2
Model
AI learns this ore body and this equipment — how mill load, ore competency, and motor draw interact in real time.
→
3
Predict
Forecast next-step state under current conditions — what happens to grade, recovery, kWh per tonne if nothing changes.
↓
6
Verify
Measure the outcome against the prediction. Feed the result back into the model so the next cycle is sharper.
←
5
Act
Operator approves the recommendation with one tap, or AI auto-applies for pre-approved low-risk moves.
←
4
Recommend
Generate the specific setpoint change — flocculant dose, mill power limit, froth depth — and project the lift.
Want this loop scoped against your circuit? Talk to a mining specialist and we'll walk through the signals already on your PLCs.
iFactory AI vs SAP PCo — What's Actually Different
SAP Plant Connectivity is a data acquisition layer — it pulls PLC and SCADA data into SAP DM or MII. It is not an AI platform, and it was not designed for closed-loop control. iFactory AI sits in a different category: it ingests the same data, runs mining-specific AI models on it, and produces the recommendations and verifications that close the loop. The comparison below is the one most plants run before they commit.
Core Architecture
SAP PCo
Connectivity middleware. Ships PLC and SCADA data to SAP DM or MII. No AI, no models, no recommendations.
iFactory AI
AI-native platform. Connectivity, ore-body-specific models, predictions, recommendations, and verification in one stack.
Deployment Model
SAP PCo
Tied to SAP DM/MII stack. Cloud-dependent for the data manufacturing layer. Requires SAP licensing across the chain.
iFactory AI
Fully on-prem appliance. No cloud dependence. Runs alongside any MES or none at all. Independent of SAP licensing.
Mining Specificity
SAP PCo
Industry-agnostic data acquisition. No mining domain models, no metallurgical knowledge, no closed-loop control logic.
iFactory AI
Mining-tuned models for crusher, mill, flotation, thickener, tailings. Domain logic built in, learns the ore body.
Time to Value
SAP PCo
12 to 18 months including the SAP MES layer above it. PCo alone is faster, but on its own it produces no value.
iFactory AI
8 to 12 weeks to first closed-loop recommendation. The model and the connectivity ship together.
Site Connectivity Needs
SAP PCo
Persistent network back to SAP cloud or central data centre required for the DM functions to operate.
iFactory AI
Runs fully on the on-prem appliance. Site connectivity loss does not break the closed loop or the recommendations.
Licensing Model
SAP PCo
Per-tag and per-instance licensing. Costs scale with the number of signals brought into the SAP environment.
iFactory AI
Per-plant license. Adding more signals does not increase the cost. Predictable for budgeting across a portfolio.
Why On-Prem AI Fits Mining
Most general-purpose AI platforms assume reliable cloud connectivity, central data lakes, and persistent network back to a corporate data centre. A mine site reliably has none of those. Four constraints make on-prem the default architecture for mining, not the exception.
Remote Site Connectivity
Most mines run on intermittent satellite or microwave links. A closed-loop AI that needs cloud round-trips would be down more than it's up. On-prem runs locally regardless of what the site's link is doing.
Real-Time Control Latency
Setpoint changes on a flotation cell or mill have to react inside the process timescale. Even good cloud latency is too slow for the recommend-and-verify loop. On-prem keeps the loop inside the plant.
Cybersecurity & OT Posture
Mining has seen sharp increases in OT-targeted cyber activity. On-prem AI air-gapped or DMZ-segregated from corporate IT is the architecture most plant cyber teams already prefer for any new system.
Data Sovereignty
Some jurisdictions require operational data to remain on the site or inside the country. On-prem deployment clears the sovereignty box without negotiating cloud region exceptions or data residency riders.
Want an on-prem appliance specced against your site's connectivity profile? Book a demo and we'll size it to your circuit.
What Changes at the Operator's Console
The biggest visible change is at the operator's console, where the loop actually lands. The console transitions from a passive readout of plant state to a stream of specific, time-stamped recommendations the operator can act on inside the shift. Five practical changes are what operators describe after the first month.
Before iFactory
Open-loop monitoring
Operator reads SCADA, calls metallurgist on the radio, waits
Setpoint changes happen at shift handover or after lab assays
Recovery losses surface in the daily metallurgical review
Reagent dosing follows fixed recipes, not real ore conditions
Tail-of-shift "what went wrong" reconstruction every morning
After iFactory
Closed-loop with recommendations
Operator sees a specific recommendation with projected lift
Setpoint changes inside the shift, not after the assay
Recovery anomalies surface live, not in the next-day review
Reagent dosing tracks real-time ore competency and flotation kinetics
Outcome verified against prediction before the next move
8 to 12 Week Deployment Shape
The deployment runs in parallel with normal plant operations. No production stops, no PLC code changes, no rip-and-replace of the historian. Four phases cover the practical stand-up on almost every mine site, and the schedule scales with the number of circuits, not the age of the equipment.
1
Connect & Capture (Wk 1-2)
Read-only OPC and Modbus connections to existing PLCs and historian. On-prem appliance commissioned in the plant network. No PLC code touched, no control-loop disturbance.
2
Baseline & Train (Wk 3-6)
AI learns the ore body and the equipment behavior on this site. Live dashboards go up alongside existing reports. Metallurgical team validates the baseline against historical performance.
3
Pilot Recommendations (Wk 7-10)
First recommendations generated for operator and metallurgist review. Acceptance rate, projected versus realized lift, and operator feedback all tracked and fed back into the model.
4
Closed Loop Live (Wk 11-12)
Recommendations running in production on the operator console. Low-risk moves eligible for auto-apply. Verification cycle running on every recommendation, model sharpening on every shift.
Want this 12-week plan scoped against your circuit and your operator workflow? Get a quote and we'll size it to your plant.
Frequently Asked Questions
Can iFactory AI run alongside SAP if we already have it?
Yes. iFactory is independent of the SAP stack, so it can run alongside SAP DM, MII, or PCo without conflict. Most mining operators with existing SAP investments keep SAP for the ERP and master data side, and run iFactory AI for the closed-loop optimization layer that SAP PCo was never built to deliver. The two systems can share data through standard interfaces.
What if our PLCs and historian are twenty years old?
The connector handles legacy controls across Allen-Bradley, Siemens, ABB 800xA, Honeywell, and Yokogawa families. Read-only OPC and Modbus is the default, with native protocols where available. The AI does not need modern PLC features — it needs tag-level signals, which almost any plant of any era already produces. Legacy controls are the rule, not the exception, in mining deployments.
Does the AI replace our metallurgists and operators?
No. It augments them. Every recommendation is reviewed by the operator and, for the larger moves, the metallurgist on shift — the AI surfaces the opportunity and quantifies the projected lift, the humans validate the operational feasibility. The combination produces more closed-loop moves per shift than either could alone, with verification on every one.
What happens when site connectivity drops?
Nothing changes at the plant. The on-prem appliance keeps reading sensor tags, running the model, generating recommendations, and verifying outcomes. Site-to-corporate reporting catches up when connectivity restores, but the closed loop itself does not depend on the link. This is the central reason on-prem is the default architecture for mining sites.
How quickly do we see recovery or throughput improvement?
First measurable lift typically by week ten, when pilot recommendations have been accepted long enough to show up in the metallurgical accounting. Recovery uplift, kWh per tonne reduction, and steadier concentrate grade are the three most common early-stage metrics. The verification cycle is what proves the lift is from the recommendations rather than ore-body variation.
AI-native. On-prem. Mine-site ready.
See Closed-Loop Recommendations Running on Your Circuit
Bring one circuit — preferably the one where grinding energy or flotation recovery has been your biggest miss. We'll connect read-only to the PLCs and historian already on it, stand up the AI baseline inside six weeks, and have the first closed-loop recommendation on your operator's console by week ten. The on-prem appliance runs at 99.9% uptime regardless of your site link, and the licensing is per-plant — no per-tag surprises as you add signals.
10 wks
first recommendation, your circuit
Per-plant
license, not per-tag
Read-only
no PLC code touched
