Automotive wiring with AI vision — the intermittent field fault is born in a connector position the tester cannot see.

The problem

The electrical test does not see the defect that costs the most in the field.

A car wire harness is hundreds of wires, dozens of connectors, a dozen variants per model. The traditional filter is the electrical continuity test at the end of the board: if electricity gets from A to B, the harness passes.

1. What the tester sees: continuity, short circuits, out-of-range resistance.
2. What the tester does not see: that the yellow-red wire is in cavity 7 of the connector and not in cavity 8. Continuity may be perfectly correct — but the assembled car does strange things months later.
3. What the tester does not see: that the terminal is properly crimped and not loose (continuity is fine when new, and the contact works itself loose with vehicle vibration → intermittent field fault).
4. What the tester does not see: that the sleeve is present, that the textile wrap is where it should be, that the fuse box carries the right markings for the variant of the OEM order.

The position/crimp/variant defect is not discovered in the plant: it is discovered months later as an intermittent breakdown in the field, with warranty cost, a call from the OEM's SQA and reputational damage. The veteran operator at the end of the board catches plenty of defects — but not the same number every day, and not the same in every shift.

How it fits the IRIS system

iLEAN does not replace the electrical test — it adds the filter the test cannot provide.

The wiring problem is not a lack of information: it is information split into islands (the BOM in the PLM, the order in the MES, the crimp curve inside the Schleuniger, the image of the board nowhere at all). iLEAN is the putty that seals those joints without asking you to change the board or the tester.

Edge sees the board with several cameras. Connect reads the BOM, the OEM order, the crimper. The agent cross-references position ↔ BOM ↔ variant and holds the harness before the final test. The person validates.

The three iLEAN pieces applied to automotive wiring:

• Edge — several cameras around the assembly board, with controlled lighting. CNNs trained to recognize the wire color in every cavity of every connector, the presence of the sleeve, the fuse box markings, the completeness of the textile wrap. They compare against the pattern of the specific BOM of the variant in production. Verdict in milliseconds. It works without a network.
• Connect — captures the BOM from the PLM (Teamcenter, Windchill, ENOVIA) and the work order from the MES with the exact variant. Captures the crimp curve of every terminal on modern machines (Schleuniger, Komax) through their native interface, or reads the legacy local PC on older machines. Captures the OEM SQA notices with spec changes at second zero.
• Agent — cross-references the image of the board with the variant's BOM, with the crimp curve of the terminal batch and with the defect history of the sub-assembly supplier. When it detects a wrong position, a crimp outside the envelope or a mixed-up variant, it holds the harness before the final test. It does not move the crimping press on its own — the person validates and signs.

See the full IRIS architecture →

Before and after

Electrical test + visual inspection vs. AI vision + BOM cross-check

Aspect	Continuity tester + inspector	With iLEAN Edge + Connect + Agent
Electrical continuity	Validated 100%	Validated 100% (unchanged)
Wire position in the cavity	Human eye against a mock-up only	Multi-camera vision against the specific BOM
Crimp quality	OK/NOK verdict from the machine	Learned envelope + cross-check with terminal batch
Wrong variant from a mix-up	Caught on the OEM's final line (bad)	Caught on the board, before the final test
Intermittent field fault	SQA call months later	Filtered at source, before shipping
File for IATF 16949 / OEM	Rebuild by harness number by hand	Dossier per harness number with images + crimp curve

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.

• Harness plant supplying 1-2 OEMs, a mix of models with 5-15 variants each, assembly board + Schleuniger/Komax crimpers + 100% electrical test + final visual inspection.
• Edge pilot on 1-2 critical boards (the ones with the most variants and the most mixing) + Connect integration with the PLM, the crimpers and the MES. First value expected within a few weeks.
• Indicative payback between 4 and 9 months, depending on the current frequency of field claims for intermittent faults and the average warranty cost.
• Hard levers: intermittent field fault avoided (the most expensive one) + harness rework avoided + automatic dossier per harness number for IATF 16949 audits and OEM SQA.
• The automotive standard of the order of 25 PPM ^[1] does not hold up in wiring with electrical testing alone — the position defect passes the test.

And the quality manager's reasonable doubt

“What if the AI mistakes a glare on the connector for a misplaced wire and rejects a good harness?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI compares an image against the specific BOM of the variant in production, the best models brought error below 1.5% ^[2]. And even so, what is critical is never decided alone: the system holds the harness and the person validates. iLEAN's three safety rings exist precisely for this.

^[1] Symestic — automotive quality standard of ~25 PPM.

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

Related processes: automotive PCB defects · assembly poka-yoke with vision · coil traceability in stamping.