Most good ideas for running a smelting furnace are never tried, because the only place to try them is the real furnace. We think that is the most expensive habit in the plant.
Every furnace team has a list of ideas nobody has tested. A slightly different reductant ratio. A change in flux to move slag basicity. A new ore blend that is cheaper per tonne. The ideas are often sensible. They stay on the list because the only way to find out is to run them on the furnace, and a bad heat costs power, metal and sometimes a few days of recovery. So the safe choice, every time, is to change nothing.
A bakery that wants to change its bread recipe, say a little less yeast and a longer proof, does not switch the whole morning’s production on day one. It bakes a test batch, compares it with the usual loaf, and only then changes the full run. The test batch is cheap and the full run is expensive, so the test comes first.
A smelting furnace has no cheap test batch. Every heat is a full run. That is the gap a twin is meant to fill: a place to try the change where a failure costs nothing but computing time.
A scenario tool is only useful if its answers are close to what the furnace would do. Furnace Twin is a physics model of your furnace, covering mass and energy balance, slag chemistry and kinetics, calibrated on plant data and corrected with plant history so estimates match the furnace you actually run. It is calibrated on every heat, and every model version is evaluated and logged. Estimates come with a range, so you can see how sure the twin is.
The physics matters here. A model that has only learned patterns from past heats can tell you what usually happened. It struggles to say what will happen when you do something you have never done, which is exactly what a trial is. A physics model can reason about a new reductant ratio because it traces every tonne of ore, reductant and flux to metal, slag and gas.
On the scenario engine, reductant, basicity, temperature and feed changes are compared side by side before a trial. A sensible comparison looks something like this:
| Question | What to compare across scenarios |
|---|---|
| Will it use more or less power? | Estimated specific energy per heat |
| Will the metal still meet grade? | Estimated grade and recovery |
| What happens to the slag? | Basicity, viscosity and co-reduction of iron or chrome |
| How sure is the model? | The range on each estimate, and how the case differs from calibration data |
The point is not to let the model decide. It is to narrow a long list of ideas down to the one or two worth risking a real heat on.
In Furnace Twin, only the best scenario goes to a two-heat trial. That keeps the real-furnace risk small and bounded, and it means the trial is a test of a prediction, not a leap in the dark. After the trial, the new heats re-calibrate the model, so the next scenario is estimated on a twin that has seen the change.
When trying an idea is cheap, people start suggesting more of them. A junior engineer’s question about basicity becomes a scenario rather than a risk someone has to approve. Over time the list of untested ideas gets shorter, and the reasons for the recipe become written evidence rather than habit.
For the physics underneath, read mass and energy balance for a smelting furnace and slag basicity in smelting. More is on the Furnace Twin guides page. Any recipe change on a live furnace should still follow your plant’s safety and process approval procedures.
No. It narrows the list. In Furnace Twin, scenarios are compared on the model first and only the best goes to a two-heat trial on the real furnace.
A trial is something the furnace has not done before. A physics model traces mass, energy and slag, so it can reason about a new recipe, while a data-only model mainly knows what usually happened.
The new heats re-calibrate the model, and every model version is evaluated and logged.
See it on your own data. Furnace Twin — Ask the furnace, see the proof. Book a 30-minute working session with an engineer.
