Cutting the energy consumption of an industrial process with AI
A model with the process physics built in calculates, for every shift, what power, temperature or flow rate a furnace, a dryer or a thermal line needs to run at to meet quality with the least gas or electricity. The savings come from operating closer to that minimum, without changing any equipment.
Where the unnecessary energy goes
In a reheating furnace, some of the gas is lost to flue gases, to the walls and to the times when the furnace is hot and the line is stopped. In a dryer, the air comes out hotter and drier than the product actually needs. The operator knows this and corrects it by hand when they can. The problem is that the minimum-consumption point shifts with every format, every alloy, every ambient temperature and every energy price, and nobody can recalculate it every hour.
The model can. It combines the equipment's energy balance (what goes in, what comes out as product, what's lost) with SCADA or PLC data from recent months, and learns where the margin is under each operating condition.
What the model does every shift
Reads what already exists
Zone temperatures, gas flow or electrical power, line speed, format and planned production. No new sensors, no civil works.
Calculates the setpoint
The power or temperature that meets the required outlet temperature with the lowest consumption for the product coming up, and how much it saves compared with the current setpoint.
Explains why
Every recommendation carries the variable driving it and the physical reason behind it. The shift supervisor decides whether to apply it; if the plant and the model disagree, the plant wins and the model gets readjusted.
Two cases with real plant data
Reinosa Forgings & Castings. In September 2025, GASAI's final phase was completed — the system that fine-tunes gas consumption in heat-treatment furnaces on top of the existing SCADA. It was validated across five furnaces on one of the main forging lines, where bars up to 8 metres long and 500 mm in diameter are produced, as part of the Government of Cantabria's INNOVA programme. Javier Cordón, the plant's Technology Director: “GASAI won't just let us cut energy consumption and emissions — it will also improve our precision, reliability and operational control.” Read the full case study.
The minimum power needed to keep the metal at temperature used to be set each shift based on the operator's experience. The model, built with the furnace's thermal balance and a year of operating data, calculated a reproducible, safe minimum and brought it down from 100 kW to 80 kW. Measured with the plant's own meters over a full year, the furnace's energy consumption fell by 20 to 30%. The operator keeps control of the equipment.
How the savings are demonstrated
A baseline is set using specific consumption (kWh or Nm³ of gas per tonne) from previous months, corrected for production and format, and compared against specific consumption after go-live, measured with the plant's own meters. It's the same method required by the INNOVAE grant to justify the 20% savings, and the one recognised by an ISO 50001 auditor.
The plant manager receives a per-shift report with three figures: actual specific consumption, the specific consumption the model calculated as achievable, and the difference in euros at that day's energy price. Below that, the three setpoints that would have saved the most, and why.
From the first meeting to operation
We pick the process with the highest consumption and the most margin, and review what signals exist.
Historical SCADA or historian data, a plant visit and an estimate of the savings margin based on your own data.
On the existing signals, without touching the control system. For the first few weeks the model only recommends; after that, if the plant decides to, it writes setpoints within limits set by engineering.
Specific consumption measured against the baseline every month. The model is retrained when the feedstock, the format or the equipment changes.
Related: Method Predictive models Planning with AI Glossary Industrial AI in Cantabria
Common questions before the diagnosis
How much can a model like this cut consumption?
It depends on the margin between how the equipment is operated today and the minimum the process physics allows. In a forging furnace validated across five furnaces on a main line, the first tests confirmed strong potential for gas savings. In an induction furnace at an automotive foundry, measured over a year on-site with the plant's own meters, the furnace's energy consumption fell by 20 to 30%. The prior diagnosis estimates your own equipment's margin using your own data, not someone else's.
Do I need to change the SCADA, the MES, or install new sensors?
No. The model reads the signals that already exist (zone temperatures, gas flow or power, line speed, planned production) over OPC UA, Modbus or historian exports. The setpoints it calculates are handed to the control room; the control system remains in charge.
When do you see the first result?
The diagnosis, with historical data and a plant visit, takes two to three weeks and already tells you how much margin there is. The model goes into operation on the existing signals within three to four months, and savings are measured against the baseline from the first month of operation.
Does the data leave the plant?
It can stay inside. The model is deployed on a plant server or in a private cloud, depending on each client's systems policy.
How is this different from an energy dashboard?
A dashboard shows what was consumed. The model calculates what should have been consumed for the same product and explains the difference: which variable, in which shift, how much. And it proposes the setpoint for the next shift.
Can it be used to justify the INNOVAE grant?
Yes. The baseline with specific consumption corrected for production, together with the follow-up measurement, is the method the call for applications requires to certify the 20% final-energy savings, and the one recognised by an ISO 50001 auditor.
45-minute exploratory meeting
We pick a process, review what signals exist, and decide whether it's worth a diagnosis. No cost, no obligation.
Request the meetingSee the method