Suspension damper control with AI — the tolerance does not show on the bench, it shows in the curve.

The problem

The bench says OK, the driver says the car floats through corners.

A damper combines five sources of variability that a generic end-of-line bench cannot separate:

1. The chrome-plated rod — micro-scratches, straightness, plating finish. A sub-millimeter defect destroys the seal after a few thousand kilometers and opens a warranty claim.
2. The internal valve assembly — it defines the model's force-velocity curve. It changes with the vehicle version (standard, Sport, off-road) and is assembled at high cadence.
3. The hydraulic oil and gas charge — viscosity and pressure specific to each model, captured (or not) at the filling station.
4. The gaskets and seals — different supplier, different batch, prone to contamination or misplacement.
5. The target model recipe — which curve each VIN requires. The generic bench measures a common envelope, not the model-specific curve.

The quality manager knows it. The bench gives a go/no-go against an envelope, not against the model-specific curve. A unit can pass the go/no-go and still fail the ride comfort the OEM demands. The comfort-related return arrives months after manufacturing, when it can no longer be tied back to the batch.

How it fits the IRIS system

iLEAN does not replace the bench — it puts the bench curve next to the model recipe.

The damper problem is not a missing test bench: it is information in islands (the curve on the bench PC, the rod image in another system, the oil batch on a delivery note, the model recipe in the MES). At the critical moment — when the unit ships — those five sources are never cross-referenced. iLEAN acts as the putty that fills those gaps without touching the bench or the filling station.

Edge sees the rod, the seal and the gasket. Connect reads the full bench curve and the model recipe. The agent cross-references the envelope expected for that model and holds the unit. The person signs — never the other way round.

The three iLEAN pieces applied to damper control:

• Edge — a terminal with machine vision (CNN) over the critical stations: the chrome-plated rod (micro-scratches, straightness), gasket and seal assembly before closing, the oil trace after the first bench cycle (an incipient leak), and the damper's laser marking. It triggers the actuator (stack light, diverter) in milliseconds. It works without a network: if the plant loses WiFi, Edge keeps reading and holding units.
• Connect — captures the complete force-velocity curve from the dynamic test bench (not just the go/no-go), whether it comes from the bench's modern bus or from an old PC. It captures the oil and gas batch from the supplier's delivery note (a PDF if that is what there is) and the model recipe from the ERP/MES. Whatever arrives from outside (an email from the oil supplier flagging a viscosity change) enters at second zero.
• Agent — cross-references the measured curve with the curve expected for the target VIN, the rod image, the oil batch and the model recipe. If the measured curve falls inside the generic envelope but outside the model envelope, it holds the unit and alerts the quality manager. The person validates and signs; the line does not restart on its own.

See the full IRIS architecture →

Before and after

Generic go/no-go bench vs. in-line control with iLEAN

Aspect	Today (generic end-of-line bench)	With iLEAN Edge + Connect + Agent
Acceptance criterion	An envelope shared across several models	The curve specific to the target VIN
Rod defects	Visual inspection of the plating, by sampling	CNN vision on 100%, before seal assembly
Leaks after the first cycle	Caught by a visible trace at the next station	Edge looks for the oil trace at second zero
Oil / gas batch	Trust in the delivery note, no cross-reference	Batch linked to every unit built
Ride-comfort returns	No traceability — scattered returns	A detectable pattern: which batch, which model, which VIN
File for PPAP/8D	Rebuilt if an incident hits, weeks of work	Per-unit dossier, automatic, with curve and photo

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.

• Multi-model damper assembly line, with a rod plating/inspection station, gasket assembly, oil and gas filling and a force-velocity dynamic test bench at the end.
• Edge pilot on one line (camera over the rod + camera over seal assembly + integration with the full bench curve). First value expected within a few weeks.
• Indicative payback between 4 and 9 months, depending on the frequency of ride-comfort returns and OEM rejects for curves outside the envelope over the last few years.
• The hard lever is the reduction in ride-comfort returns, which today arrive months later with no traceability, plus a single batch avoided in OEM sorting.

And the quality manager's reasonable doubt

"What if the AI accepts a curve that sits right on the edge of the envelope?" — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI simply compares a measured curve against an envelope declared per model, 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. An automotive quality standard on the order of 25 PPM ^[2] cannot be sustained with sampling on a generic bench: it is sustained by comparing against the model envelope, unit by unit.

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

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