Hot stamping press — the die does not break on a cold blank, because someone looked at the hydraulic pressure in time.

The problem

The blank arrives cold, pressure rises, the die breaks — and everyone was looking at their own screen.

In hot stamping the window is narrow and very unforgiving: the boron-steel blank leaves the furnace above 900 ºC, travels to the die and must be struck before it cools below the austenitisation range. If anything along the way stretches out — a micro-stoppage downstream, a delay in transfer, a furnace zone lagging behind — the blank enters cold. The material does not flow. The hydraulic cylinder pushes against a blank that no longer wants to deform. Pressure rises, and the die picks up the bill:

1. The hydraulic pressure curve of the cycle starts coming out steeper and with a higher peak — but the PLC only trips when it crosses the absolute maximum, and by then the die has already run several cold strokes.
2. Furnace temperature and the entry pyrometer live in another system, on another panel, watched by another person.
3. The real transfer time (when the gripper picks and places) is never cross-checked against the pressure of the stroke anywhere.

The classic setup (PLC + alarm + the die shop lead's eye) works 99% of the time. That 1% is the die breakage — direct cost of the die, line stoppage, a replacement against the clock, and an OEM customer who has to be told why the B-pillar for their next batch is late. It is not a lack of skill; it is information living in islands at the critical moment.

How it fits into the IRIS system

iLEAN Edge does not add another screen — it seals the gap between the press, the furnace and the real stroke rate.

The piece that solves the pain is iLEAN Edge: a physical terminal on the shop floor, next to the press cabinet, that watches hydraulic pressure the way a veteran die shop lead would if they could keep an eye on the actuator curve, the furnace thermocouple and the transfer stopwatch all at once.

Edge sees the hydraulic pressure curve of the cycle. Connect reads furnace temperature, the die plan and the recipe wherever they live. The agent cross-checks furnace + transfer + pressure and proposes to the die shop lead that the next blank be held. The person signs — the die does not break on a cold blank.

The three iLEAN pieces applied to hydraulic pressure on the hot stamping press:

• Edge — terminal at the cabinet, reading pressure in the main hydraulic circuit of the actuator (analogue transducer) and, depending on the press, cushion pressure, closing time and slide speed. The CNN learns the normal shape of the pressure curve per part and per die — not an absolute maximum, but a cycle signature. It catches the steep rise before the ceiling. It works without a network: if connectivity with the MES drops, Edge keeps watching the curve and warning the local panel. What is critical cannot depend on the WiFi.
• Connect — captures furnace temperature by zone and the die-entry pyrometer from the SCADA, the die plan from the MES, and the OEM customer recipe wherever it comes from (including the quality spreadsheet with austenitisation ranges per part). It also captures the steel supplier's email about the change of coil supplier, and the shift lead's note on the reason for the 14:20 micro-stoppage downstream.
• Agent — cross-checks the deviating pressure signature against real furnace temperature, transfer time and die history. When it sees the full picture (furnace lagging + slower transfer + steeper pressure curve), it proposes to the die shop lead: hold the next blank, drop the stroke rate by 4, or switch to a more forgiving part while the furnace restabilises. The lead signs. It does not stop the press on its own — except at the autonomy level the leadership team configures.

See the full IRIS architecture →

Before and after

Hot stamping press with PLC and alarm vs. press with iLEAN Edge on hydraulic pressure

Aspect	PLC + maximum alarm	With iLEAN Edge on pressure + furnace + transfer
Signal that triggers the reaction	Absolute pressure maximum, the same for every part	Learned pressure curve signature per part and per die
Cross-check with furnace temperature	Separate screen, separate person	Cross-checked within the same cycle, on the shop floor
Cross-check with real transfer time	Not measured in detail	Edge correlates it with the pressure of the stroke
Moment of the warning	Once there has already been a cold stroke	Before the next stroke
Operation without a network	Alarm yes, context no	Edge keeps watching the curve on the cabinet light
Dossier for the OEM audit	Reconstructed by hand	Per cycle, with pressure curve, furnace and human signature

Impact estimate

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

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

• Automotive Tier 1 with a hot stamping line (B-pillar, frame reinforcements, structural bumpers), roller hearth furnace + hydraulic press + robotised transfer, one or two documented die breakages per year.
• Edge pilot on the press (pressure transducer + integration with the PLC, the furnace SCADA and the MES). First expected value within a few weeks: detection of the deviating pressure signature before the absolute ceiling, in combination with real furnace temperature.
• Indicative payback between 4 and 9 months, depending on the frequency of documented incidents and the average cost of a die breakage + line stoppage + replacement.
• Expected reduction of cold-stroke events that compromise the die of ≥ 30% in the first year, depending on the maturity of furnace control. The hard lever is a single die breakage avoided — it pays for the pilot with room to spare.

And the die shop manager's reasonable doubt

«What if the AI detects a deviation that isn't there and makes me stop the press when it is running fine?» — the proposal does not execute itself. iLEAN works on an anchored task: it reads the pressure curve the transducer already produces and compares it with the learned signature of the die. On anchored tasks, the best models brought error below 1.5% ^[1]. And even so, the die shop lead sees the curve, sees the proposal, and signs. The press does not stop on its own — except where the leadership team configures it. And the three safety rings are there so the critical OT network stays isolated.

^[1] OpenAI paper «Why Language Models Hallucinate», 2025 — on the reliability of AI in anchored tasks.