Muffle furnace temperature in steel tempering — hardness is not prayed for, it is measured and signed.

The problem

The curve the SCADA draws and the curve the steel actually suffers are not the same one.

A batch of tempered steel comes out right or comes out brittle depending on a real curve that crosses three things:

1. The muffle furnace temperature by zone — not the PLC setpoint, the real one. Thermocouples drift, the load stacks heat unevenly, and the cold zone at the back never hears about the setpoint at the front.
2. The soak time — real soak, not theoretical soak. If the door stays open five minutes halfway through the curve, effective soak time is no longer the programmed one.
3. The heating and cooling ramp rate — what decides whether martensite forms as it should and whether tempering reaches the target Rockwell hardness, or whether the batch comes out brittle or soft.

The veteran in the heat treatment shop knows how to read that gap and compensates for it by eye. But his judgement lives in no system: it lives in his head. When he retires it leaves with him, and the muffle furnace gives no warning. The day the Rockwell test lands outside the window — because someone loaded the furnace differently, or a thermocouple drifted out of calibration with nobody noticing — the batch is scrap or rework, and the automotive customer applies the 25 PPM standard ^[1]. There is no margin.

How it fits into the IRIS system

iLEAN Edge doesn't touch the PLC — it seals the crack between the drawn curve and the real curve.

The problem in tempering is not missing data: what's missing is data cross-checked against the batch recipe and available at second zero. The SCADA draws the theoretical curve, the thermocouples measure, the Rockwell tests are done by sampling and the recipe lives in a spreadsheet or in a vertical MES. Four islands. iLEAN Edge is the filler that closes the gap between all of them, without asking you to change furnaces, PLC or SCADA.

Edge measures the real curve by zone. Connect captures the load recipe wherever it comes from — the ERP, the MES or the heat treatment master's spreadsheet. The agent cross-checks, compares against the process window and, if something drifts out, holds the batch in the record. The person signs — never the other way round.

The three iLEAN pieces applied to the tempering muffle furnace:

• Edge — a terminal with its own thermocouples in the furnace's critical zones, running in parallel to the PLC loop. It measures temperature, computes ramp rate and effective soak time, and fires an alert to the local panel in milliseconds if the real curve drifts away from the recipe. It works with no network.
• Connect — captures the load recipe (steel grade, target hardness, theoretical curve, process window) wherever it comes from: the ERP, the vertical heat treatment MES or the master's spreadsheet. And it also captures what arrives from outside: a new specification from the automotive customer, a change in the aerospace standard. Second zero.
• Agent — cross-checks real curve, recipe, Rockwell tests and customer specification. It generates the record per load (theoretical vs. real curve, door events, linked hardness, signature) ready for an IATF 16949 or Nadcap audit. If the curve doesn't add up, it holds the batch and alerts the supervisor. The person validates.

See the full IRIS architecture →

Before and after

Muffle furnace with a logbook vs. muffle furnace with an auditable iLEAN curve

Aspect	Classic tempering + logbook + SCADA	With iLEAN Edge + Connect + Agent
Tempering curve	The one the SCADA draws — setpoint, not reality	Real curve by zone, measured in parallel
Door events or thermal anomalies	If the operator notices, they write it down by hand	Automatic timestamped log in the record
Cross-check against the batch recipe	The master keeps it in his head or in a spreadsheet	Recipe linked to the batch from the ERP/MES
Traceability for IATF 16949 / Nadcap	Manual reconstruction weeks later	Record per load ready in seconds
Operation with no network	SCADA depends on the server	Edge keeps measuring on cabinet power alone
The veteran's know-how	Leaves with him the day he retires	Captured as a pattern in the system

Impact estimate

Impact estimate for your plant — to be validated with your own numbers.

The block below is an estimate to be validated with your plant's specific data. We lay it out so your committee has an order of magnitude; we refine it during the diagnostic.

• Heat treatment shop with 2 to 4 muffle furnaces, batches of tool steel and machined parts for automotive or aerospace, Rockwell testing today by sampling.
• Edge pilot on one muffle furnace (thermocouples in the critical zones + integration with the batch recipe + alert actuator). First value expected within a few weeks.
• Expected reduction in loads rejected for an out-of-window curve of ≥ 30% against the baseline we will measure during the immersion.
• Indicative payback between 4 and 9 months, depending on the historical frequency of loads with out-of-tolerance hardness and the average cost of rework or return.
• The hard lever is a single rejected load avoided: steel re-tempered, extra tests, customer return. With demanding customers, one load pays for the pilot.

And the quality manager's reasonable doubt

«What if the AI gets it wrong and lets through a load with hardness out of tolerance?» — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely compares a measured value against a process window (reading the real curve and checking it against the recipe), the best models brought error below 1.5% ^[2]. And even then, nothing critical is decided alone: iLEAN holds the batch in the record and the person signs. The three safety rings exist precisely for this.

^[1] Automotive quality standard in the order of 25 PPM (parts per million). Source: Symestic.

^[2] OpenAI paper «Why Language Models Hallucinate», 2025 — on the reliability of AI in anchored tasks.