Furnace Twin: guides, playbooks and tools.

By the Autowhat AI team · Updated 7 Oct 2026

• Blog posts

• Try the Furnace Recipe Change on the Twin FirstMost furnace improvement ideas are never tried because every trial risks a heat. Furnace Twin, a physics model calibrated on every heat, lets teams compare reductant, basicity, temperature and feed changes side by side, then send only the best scenario to a two-heat trial. Trials become tests of a prediction, not gambles.
• Furnace Answers Should Cite Their SourcesA fluent assistant that gives an unsourced answer about a smelting furnace is a risk. Furnace Twin answers plain-language questions from plant records, the calibrated physics model and a literature library, cites each source and uses a local language model. Good citations need good records, so shift handovers are captured by voice.

• Guides

• Mass and Energy Balance for a Smelting FurnaceA smelting furnace mass balance traces ore, reductant, flux and electrode to metal, slag, gas and dust element by element; the energy balance sets electrical input against reactions, heating and losses. Built from per-heat data and calibrated on plant history, together they show where the energy goes.
• Specific Energy Consumption in Smelting FurnacesSpecific energy consumption is the electricity used per tonne of metal or alloy tapped. In a smelting furnace it moves with charge chemistry, reductant ratio, slag basicity, electrode condition, burden size and moisture, tapping practice and downtime. Tracking it per heat, not monthly, links overruns to their cause.
• Slag Basicity in Smelting: What It Is and Why It MattersSlag basicity is the ratio of basic to acidic oxides in slag, most simply CaO/SiO2. In a smelting furnace it affects slag melting behaviour, viscosity, metal recovery, electrical behaviour and refractory wear. Control it by calculating flux from assays and checking tapped slag every heat.

• Playbooks

• Capturing Furnace Operator Knowledge Before It LeavesTo keep furnace know-how in the plant, capture data every heat, turn shift handover into structured intake, write down the reductant and basicity rules operators use, gather past reports into one library, answer questions with cited sources and test recipe changes on a calibrated model first.

• Comparisons

• Physics-Based vs Data-Only Furnace ModelsData-only machine learning fits past furnace data but rarely explains why and is unreliable outside past practice. Physics-based models explain but drift from the real furnace. A hybrid, physics calibrated on every heat, gives explainable estimates and safer scenarios for new recipes.

The product behind these guides. Furnace Twin — Ask the furnace, see the proof. Book a 30-minute working session.

More products and guides: all guides · all products · glossary.