AI fuel tank control — a leak is not an incident, it is a recall.

The problem

The tank that passes every test and still reaches the OEM defective.

A modern fuel tank is a multilayer HDPE part blown in a mold, with a parison weld bead, a welded insert for the pump, an EVAP valve, a rollover valve and a leak test at end of line. When an OEM rejects a batch, it is almost never because the helium test was skipped: it is because the test was done properly but the part had a microcrack right on the weld bead that the bench did not detect at the defined pressure.

The three realities of the tank live in islands:

1. The blow molder — temperature, pressure and cycle time curves, in the machine builder's own SCADA. The veteran on shift glances at it out of the corner of their eye.
2. The weld bead and the insert — visual inspection by eye, sometimes with a stand-alone camera without intelligent reading.
3. The leak test bench — a helium or pressure-decay log on an old PC, exported to CSV at the end of the shift.

The quality manager wants to cross-reference the three for every serial number. They cannot: there is no system that brings them together in real time. When the OEM returns a batch, everyone realizes the cause had been there since Thursday's night shift — but by then it was too late.

How it fits the IRIS system

iLEAN does not replace your blow molder or your helium bench — it seals the cracks between them.

The problem is not a lack of information: every machine has it. It is information in islands that is never cross-referenced in time. iLEAN acts as the putty between the blow molder, the weld bead inspection, the leak test bench and the ERP/MES — without asking you to change a single piece of equipment.

Edge sees the weld bead before the insert. Connect reads the blow molder curve whether it comes from the SCADA or from the old PC. The agent cross-references it with the helium log and, if something does not add up, holds the tank before the truck.

The three iLEAN pieces applied to fuel tank control:

• Edge — a terminal with machine vision (CNN) at the blow molder exit. It inspects the parting-line weld bead and the insert geometry in milliseconds and fires the light stack/ejector if it detects porosity, flash or displacement. It works without a network.
• Connect — captures the blow molding curve from the machine builder's SCADA (even if it is an old, isolated PC), the leak bench log and the HDPE batch from the ERP. And it captures whatever arrives from outside (a recipe change from the polymer supplier, an OEM alert about a PPAP) at second zero.
• Agent — cross-references the blow molding curve, the Edge inspection of the weld bead, the helium log and the raw material batch. If the combination enters a pre-failure pattern, it holds the part and alerts the shift lead with a specific suggestion — before the next tank comes out defective. The person validates and signs; the line does not restart on its own.

See the full IRIS architecture →

Before and after

Classic control vs. cross-referenced control with iLEAN

Aspect	Classic control	With iLEAN Edge + Connect + Agent
Blow molding curve	Machine builder's SCADA, watched by the operator	Captured at second zero, cross-referenced with the rest
Weld bead and insert	Visual inspection, sometimes a mute camera	CNN on every part, ejection in milliseconds
Leak test bench	CSV log at the end of the shift	Log integrated in real time with curve and image
Traceability by serial number	Rebuilt by hand on Friday	Automatic dossier per part (PPAP-ready)
Root cause of a micro-leak	A meeting the following week	Pre-failure pattern flagged to the shift lead
Operation without a network	n/a	Edge keeps inspecting on the panel's own light

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.

• Tier 1 plant with 2 multilayer HDPE blow molders, a helium leak bench and a pressure-decay test at end of line, under an OEM requirement of IATF 16949 / PPAP level 3.
• Edge pilot on one blow molder (cameras over the weld bead and the insert + actuator + integration with the blow molder SCADA and the leak bench). First value expected within a few weeks.
• Reduction in scrap from weld bead or insert defects of the order of 30% or more, depending on the starting PPM and the maturity of the control plan.
• Indicative payback between 4 and 9 months, with the cost of a single leak-driven recall (reverse logistics, rework, OEM penalty, brand damage) as the hard lever. One recall pays for the pilot.

And the quality manager's reasonable doubt

“What if the AI gets it wrong and lets a tank with a microcrack through?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely recontextualizes a piece of data from one system to another (reading the weld bead, comparing the curve, cross-referencing the helium log), the best models brought error below 1.5%^[2]. And even so, what is critical is never decided alone: iLEAN holds the tank and the person signs. The three safety rings exist precisely for this.

^[1] Automotive quality standard of the order of 25 PPM — Symestic.

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

Related: catalytic converter and exhaust pipe · radiator / heat exchanger · water pump.