LED headlamp control with AI — the OEM does not send back one unit, it sends back the whole batch.

The problem

An LED headlamp is precision optics, and almost all of the control lives in sampling.

A modern LED headlamp (full-LED, matrix, adaptive) combines five critical parameters that are almost never inspected on 100% of units in line:

1. LED binning — the supplier ships LEDs sorted by intensity, color and forward voltage. Mix bins inside the same module and the beam loses uniformity.
2. Reflow soldering of the SMDs — poor thermal contact with the heat sink and the LED drifts in color or dies on the first temperature cycle.
3. Injection of the polycarbonate lens — micro-cracks, bubbles and mold marks that a tired eye at 6 p.m. does not see.
4. Alignment of the collimating optics — tenths of a millimeter decide whether the beam falls inside the photometric envelope of the UNECE regulation or outside it.
5. Housing sealing — moisture inside the headlamp = condensation months later, warranty claim open.

The quality manager knows it. The homologation bench is slow and sample-based. Human visual inspection is subjective and gets tired. The classic system works for the vast majority of headlamps — and the OEM rejection lands on the one that slips through. One rejected batch costs the equivalent of weeks of line output.

How it fits the IRIS system

iLEAN does not replace the homologation chamber — it puts a 100% in-line control next to it.

The LED headlamp problem is not a lack of equipment: it is information split into islands (supplier binning in a PDF, reflow oven parameters on its own PC, lens dimensional data in another system, a human eye at end of line). At the critical moment — right before final assembly — those five sources are never cross-referenced. iLEAN acts as the putty that fills those gaps without asking you to change the homologation chamber or the injection press.

Edge sees the lit headlamp inside the light booth. Connect reads the supplier binning, the oven parameters and the model recipe. The agent cross-references them with the UNECE regulation and holds. The person signs — never the other way round.

The three iLEAN pieces applied to LED headlamp control:

• Edge — a terminal with multispectral vision (CNN) looking at the lit headlamp inside a compact controlled light booth. It measures uniformity, scatter, color per LED, micro-cracks in the lens and thermal hot spots in a single pass. It fires the actuator (stack light, diverter) in milliseconds. It works with no network: if the plant loses WiFi, Edge keeps measuring and holding units.
• Connect — captures the declared binning of the LED batch from the supplier's CoC (a PDF by email), the reflow oven parameters (temperature profile per board), the dimensional data from the lens injection press and the model recipe from the ERP/MES. Whatever arrives from outside (a supplier email warning that a bin is unavailable) enters at second zero.
• Agent — cross-references declared binning against measured binning, the oven profile against the LED's thermal signature, the lens micro-cracks against the mold signature, and the photometric result against the model's UNECE envelope. If something does not add up, it holds the unit before final assembly and alerts the quality manager on their own channel. The person validates and signs; the line does not restart on its own.

See the full IRIS architecture →

Before and after

Visual inspection + sampling vs. 100% in-line control with iLEAN

Aspect	Today (operator + homologation sampling)	With iLEAN Edge + Connect + Agent
Photometric control coverage	1-in-N sampling on the homologation bench	100% in an in-line Edge light booth
LED binning	Trust in the supplier's CoC, no verification	Declared binning cross-checked with measured spectrum per unit
Micro-cracks in the lens	Tired human eye at the end of the shift	CNN vision on the lens, sub-millimeter marks detectable
Thermal hot spots	Not detected until the accelerated aging bench	Thermal signature in line with the first voltage applied
When the rejection is caught	The OEM rejects the batch at goods-in, full sorting	Before final assembly, unit by unit
File for PPAP/8D	Rebuilt by hand, weeks	Per-unit dossier, automatic, with a photo of the beam

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.

• Full-LED headlamp line with a reflow oven for SMDs, a polycarbonate lens injection press and a photometric homologation bench at the end.
• Edge pilot on one line (compact light booth + multispectral camera + integration with the supplier's binning CoC). First value expected within a few weeks.
• Indicative payback between 4 and 9 months, depending on the frequency of OEM rejections over recent years and the average cost of sorting plus delivery delay.
• The hard lever is a single batch kept out of sorting: the cost of one OEM rejection for photometric deviation pays for the pilot and the rollout.

And the quality manager's reasonable doubt

“What if the AI gets it wrong and lets an out-of-envelope unit through?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely compares a measured spectrum against a declared envelope, the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN holds the unit and the person signs. The automotive quality standard, in the order of 25 PPM ^[2], cannot be held with sampling: it is held 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 in the order of 25 PPM.