Airbag assembly control with AI — a defect is not scrap, it is a recall of the whole series.

The problem

The airbag has zero tolerance, and the cost of failure is paid by the OEM with its own name on it.

An airbag module assembly station combines five realities that almost never live in the same system:

1. The inflator batch — it comes from the pyrotechnic supplier. If that batch carries a defect, every unit built with it is suspect.
2. The ultrasonic weld of the folded cushion — pressure, energy and time parameters on the welder. If it drifts outside the window, nobody sees it until deployment.
3. The torque and angle of the module bolts — the smart nutrunner records them, but that data does not always reach the batch file.
4. The module's laser marking — the unique serial number that will have to be matched to the VIN.
5. The model recipe — which module version goes on each vehicle variant, a data point that changes whenever the OEM updates a specification.

The quality manager knows this. The operator has a smart torque wrench and a pick-to-light screen. And even so, a badly tightened module or a poorly welded cushion can go through 8 more stations and reach the OEM. The later audit spots the pattern when there are already thousands of units in the field. Every airbag recall that makes the press started exactly like this.

How it fits the IRIS system

iLEAN does not add a sixth system — it seals the cracks between the five you already have.

The airbag control problem is not a lack of data: it is data in islands (ultrasonic welder, smart nutrunner, laser marker, supplier ERP, line MES) that at the critical moment — the exit from the station — have not been cross-referenced. iLEAN acts as the putty that fills those gaps, without asking you to change the vertical MES or the nutrunner.

Edge sees the assembly at every step. Connect reads the torque, the ultrasonic curve and the inflator batch wherever they live. The agent cross-references them with the model recipe and, if something does not add up, holds the part. The person signs — never the other way round.

The three iLEAN pieces applied to airbag assembly control:

• Edge — a terminal with machine vision (CNN) at every station: presence and orientation of the inflator, integrity of the cushion fold, alignment of the module in the housing, reading of the DataMatrix/laser marking. It fires the actuator (light, conveyor stop) in milliseconds if something does not add up. It works without a network: if the plant loses WiFi, Edge keeps reading and holding modules.
• Connect — captures the torque and angle from the smart nutrunner, the energy curve from the ultrasonic welder and the inflator batch from the supplier's delivery note — whether the data comes from a modern bus, from the nutrunner's old PC, or from the supplier's PDF. And it captures what arrives from outside (an email from the pyrotechnic supplier warning about a withdrawn batch) at second zero.
• Agent — cross-references the inflator batch, the recorded torque, the ultrasonic curve, the module's DataMatrix, the destination VIN and the model recipe. If something does not add up — torque outside the window on a single bolt, an inflator from a batch under suspicion — it holds the part and alerts the quality manager on their own channel. The person validates and signs; the line does not restart by itself.

See the full IRIS architecture →

Before and after

Manual assembly + later audit vs. in-line control with iLEAN

Aspect	Today (operator + nutrunner + later audit)	With iLEAN Edge + Connect + Agent
Inflator batch traceability	Supplier delivery note filed; cross-checked after the fact	Cross-checked with every module at second zero
Critical tightening torque	Recorded on the nutrunner PC, not linked to the VIN	Captured by Connect and linked to the module and the VIN
Ultrasonic weld of the cushion	History on the welder, local alarm if it drifts	Curve stored per part, anomaly caught before close-out
Defect detection	Dimensional audit at the end of the shift (1 in N)	At the station, before it leaves, module by module
Batch-to-VIN link	Rebuilt if there is an incident, taking weeks	Per-unit dossier, automatic, ready for PPAP/8D
Operation without a network	n/a	Edge keeps running on the panel's own power

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.

• Airbag module assembly line with critical angle tightening, ultrasonic welding of the cushion and laser marking. IATF 16949 standard + the OEM's specification book.
• Edge pilot at one assembly station (camera over the module + integration with the nutrunner and the ultrasonic welder). First value expected within a few weeks.
• Indicative payback between 4 and 9 months, depending on the frequency of non-conformities raised by the OEM over recent years and the average cost of an 8D + passive safety sorting.
• The hard lever is a single recall avoided: the cost of one workshop call-back in passive safety pays for the pilot and the rollout.

And the quality manager's reasonable doubt

“What if the AI gets it wrong and lets a defective module through?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI does nothing more than read a DataMatrix and cross-reference it with a batch and a VIN, the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN holds the part and the person signs. The three safety rings exist precisely for this. The automotive quality standard of the order of 25 PPM ^[2] no longer holds up with sampling: it holds up with vision on every part.

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

^[2] Symestic — automotive quality standard of the order of 25 PPM.