Furnace Copilot
Your furnace already speaks. SmeltIQ learns its language.
Furnace Copilot brings these signals together to understand the condition of the heat in real time — helping operators act before deviation becomes loss.
Six live views of a furnace you could not see before.
Furnace Copilot is not analytics on yesterday's data. It is real-time models designed to amplify operator intelligence, minimise thermal inefficiency and prolong asset life.
All insights are presented live and can be shown on operator consoles, mobile HMI devices, or integrated directly into SCADA and DCS systems.
Live melt curve vs target deviation map
Where this heat is against the target trajectory, second by second.
Acoustic risk map by zone and heat cycle
Sound signatures that precede collapse, arcing instability and mechanical trouble.
Real-time combustion health index
One number for how well the furnace is burning right now — and which way it is heading.
Charge composition fingerprint from vision AI
What actually went into the furnace, read from the image — not from the plan.
Predicted tap time with confidence intervals
Prepare the ladle and the crane for when the heat will actually be ready.
Furnace fatigue index and remaining-useful-life scorecard
How hard the asset has been run, and how much life is left before intervention.
From camera to decision in five steps.
A high-resolution industrial camera near the furnace captures the bed continuously, in an environment where people cannot stand and watch. The rest is models and workflow.
Capture
Industrial camera captures the furnace bed — temperature variation, material movement, electrode condition — in extreme heat and dust.
Transmit
Images stream to Furnace Copilot through direct platform integration or secure API upload.
Analyse
AI detects bed collapse, electrode width, crust formation, bed condition and activity homogeneity in real time.
Show
Live furnace monitoring, trend analysis and alerts on an operator dashboard — or straight into your SCADA/DCS.
Predict
Historical data and predictions recommend maintenance timing, operating parameters and corrective action before downtime.
What the operator sees.
Clinker analysis, audio analysis and furnace-bed analysis on one console. Every detection carries a plain-language reason, so the shift can act on it without a data scientist in the room.
Camera · EAF bed · 12 fps
Combustion health index
Smoke intensity & brightness
Audio analysis
Last 8 heats · EAF-2
| Heat | Grade | kWh/t | Tap °C | Flags | Outcome |
|---|---|---|---|---|---|
| 4712 | 42CrMo4 | 468 | 1,635 | 1 | In spec |
| 4711 | 42CrMo4 | 474 | 1,631 | 0 | In spec |
| 4710 | 16MnCr5 | 497 | 1,628 | 2 | Re-blow |
| 4709 | 16MnCr5 | 481 | 1,642 | 0 | In spec |
| 4708 | C45 | 452 | 1,626 | 0 | In spec |
| 4707 | C45 | 449 | 1,624 | 0 | In spec |
| 4706 | 20MnV6 | 509 | 1,647 | 3 | Off-spec Mn |
| 4705 | 20MnV6 | 486 | 1,638 | 1 | In spec |
Select a heat to load its detection record.
Energy per tonne · last 8 heats
Detection record
Shift B · 06:00–14:00 · EAF-2
Energy by hour
Peak at 11:00 coincides with the 20MnV6 sequence and the 18-minute electrode stop.
Flags raised this shift
- Crust formation4 heats · all cleared before tap
- Electrode width drift2 heats · within envelope
- Off-spec Mn on 4706followed an operator override
- Acoustic collapse precursornone detected
Versus previous shift
- Mean kWh/t−9
- Off-spec heats−2
- Unplanned downtime+6 min
- Heats tapped+1
Live console: camera detections, combustion health index, smoke and brightness trend, and acoustic collapse detection. Select any detection to see the reason behind it.
It doesn't just detect deviations. It pre-empts them.
Operational impact reported for AI-based furnace monitoring, and the levers behind it.
Furnace steelmaking typically consumes 400–500 kWh per tonne. Audits report 420–775. The spread is decisions, not physics.
Manual tracking vs Furnace Copilot.
Round-the-clock monitoring by people is expensive, inconsistent and blind between rounds. A camera and a model never look away.
In our cost model, replacing three-person manual tracking with Furnace Copilot removes most labour and indirect supervision cost while raising real-time accuracy from the low teens to an estimated 90–95%.
| Manual tracking | Furnace Copilot | |
|---|---|---|
| Accuracy in real time | 13–15% | 90–95% (estimated) |
| Output | Similar smoke and bed patterns noted by eye | Graphs, anomalies, insights, alerts |
| Scalability | Limited to periodic images | Video and real-time data streams, across furnaces |
| Coverage | Rounds and shifts | 24/7, every furnace, no added headcount |
| Errors | Human fatigue and variation | Consistent detection with confidence scores |
Plugs into the systems you already run.
Furnace Copilot connects natively with multiple MES modules, feeds ERP for batch traceability, maintenance logging and cost variance, and pushes automated emissions telemetry to ESG frameworks.
Industrial systems
- SCADA / DCS integration
- Operator consoles and mobile HMI
- Industrial camera and acoustic feeds
MES and ERP
- Native connectors to multiple MES
- ERP PM / PP / CO modules
- Batch traceability and cost variance
ESG reporting
- SASB · CDP · UN SDGs
- ISO · SBTi
- Automated Scope 1 emissions telemetry
Questions we get asked
Furnace monitoring, answered.
What operations and maintenance teams ask about AI-based furnace monitoring and predictive maintenance.
What is Furnace Copilot?
Furnace Copilot is real-time furnace intelligence for electric arc furnaces, induction units, refining vessels and smelters. It reads industrial camera images, acoustic signals, temperature, power and process behaviour through AI models trained on furnace behaviour, and reports bed condition, crust formation, electrode width, activity homogeneity, combustion health and predicted tap time — so operators can act before a deviation becomes a loss.
How does camera-based furnace monitoring work?
A high-resolution industrial camera installed near the furnace captures the bed continuously in heat and dust conditions where a person cannot watch for long. Images stream to the platform by direct integration or secure API upload. Computer-vision models classify bed collapse indications, electrode width, crust formation, bed condition and activity homogeneity, each with a plain-language reason so the shift can act without a data scientist present.
Can AI predict furnace failures before downtime?
Predictive signals come from combining vision, acoustics and process history: acoustic signatures that precede bed collapse and arcing instability, fatigue accumulation on electrode arms and refractory, and drift in combustion health. These are surfaced as a ranked risk and remaining-useful-life view so maintenance is planned rather than reactive.
How much energy can furnace monitoring save?
Furnace steelmaking typically consumes about 400–500 kWh of electricity per tonne, but audits have reported a range of roughly 420 to 775 kWh per tonne — a spread driven by decisions rather than physics. A published stainless-steel case using AI tap-temperature prediction reported 13% better MAE and 17% better RMSE, with measurable electricity and operating-cost savings. Figures for your plant are established against your own baseline during a pilot.
Does it replace operators or the SCADA system?
Neither. Furnace Copilot adds a perception and prediction layer; insights appear on operator consoles, mobile HMI devices, or directly inside your existing SCADA or DCS. Control actions remain with your existing systems and your operators.
Don't just monitor the furnace. Understand it.
One camera, one furnace, one month of data. That is enough to show what Furnace Copilot would have caught.