Perfume cap crimping and torque with AI — the fragrance that evaporates escapes through a tenth of a newton.

The problem

The fragrance does not vanish all at once — it escapes by tenths of a newton, slowly.

A premium perfume is a delicate product in a decorative box. What the customer pays for is not the liquid: it is the balance of top, heart and base notes in the formula. And that balance depends on an airtight closure. A slightly loose collar crimp, a cap torque a few tenths below specification, an asymmetric decorative seating — the unit looks perfect at the end of the line and starts losing its most expensive volatile compounds within a few weeks.

The operational problem, which lives on three familiar islands:

1. The closure geometry — collar, cap seating, visible symmetry. Inspected by sampling, usually at the end of the shift.
2. The applied torque — set by the capper. If it is modern, it records it. If it is old, it depends on manual calibration and on the operator's feel.
3. The return — arrives weeks later, with no way left to trace which run the bottle belonged to. Customer service logs the reason ("the perfume smells like alcohol," "it's not the fragrance I remember"), and nobody closes the loop back to the line.

The classic system (sampling + calibrated capper + visual inspection) works 99% of the time. That 1% is what reaches the customer with less fragrance than they paid for — and the damage is not just the return: it is the internal note in the brand's dossier.

How it fits the IRIS system

iLEAN does not replace your capper — it closes the loop between the crimp, the torque and the return.

The problem is not a lack of machines: perfumery cappers and crimpers are excellent. The problem is information living on islands that, at the critical moment (the out-of-spec unit that slips through sampling), does not reach the decision-maker in time. iLEAN acts as the putty that fills the gaps between the capper, the visual inspection and the returns desk.

Edge sees the collar. The capper reports the torque. The agent cross-references historical returns. The person signs every rejection.

The iLEAN pieces applied to perfume cap crimping and torque:

• iLEAN Vision (vision-equipped Edge) — a terminal with convolutional neural networks (CNN) on the collar and the cap right after closure. It measures crimp geometry, cap seating, visible symmetry. If something is out of tolerance, it pulls the unit before palletizing. Reaction time in milliseconds. Works without a network — the non-negotiable rule: what is critical cannot depend on WiFi.
• Connect — captures the torque applied by the capper (via modern API, via an added sensor for old cappers, or via inference from power draw if there is no other option). It also captures lost-fragrance returns from after-sales service, whether they arrive by email, web form or transcribed call. Without asking anyone to forward anything.
• Agent — cross-references crimp geometry + torque + historical returns + cap supplier changes. When a correlation shows up (lost-fragrance returns rise in the week the capper was calibrated below the optimal range), the agent says so. Maintenance and quality act on data, not on a hunch.
• Three safety rings — the critical closure parameters live in ring 1. What comes from outside (returns, supplier changes) passes through ring 2. It fits the AI Act and regulatory traceability for IT/CAIO.

See the full IRIS architecture →

Before and after

Sampling + calibrated capper vs. in-line control with iLEAN Vision

Aspect	Sampling + visual inspection	With iLEAN Vision + Connect + Agent
Collar geometry verification	Sampling every N units	100% in line, in milliseconds
Cap torque reading	Capper's system, isolated	Cross-checked at second zero with the visible closure
Heavy decorative / magnetic cap	Hard to inspect by eye	CNN trained on real samples of the SKU
Return for lost fragrance	No traceability back to the batch	Cross-referenced with the shift and the applied torque
Operation without a network	n/a	Edge keeps running on the cabinet's own power
File for after-sales and brand	Rebuilt by hand, weeks	Per-batch dossier, automatic, with geometry + torque

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.

• Premium perfume capping line, multi-SKU, bottles with heavy decorative or magnetic caps, a quantifiable history of returns for lost fragrance.
• iLEAN Vision pilot on one line (camera on the collar + torque reading + actuator + integration with the MES and with the returns system). First value expected within a few weeks: the first shift with 100% verification of geometry and torque.
• Indicative payback between 4 and 9 months, depending on the annual volume of lost-fragrance returns and the average cost of replacement + brand damage.
• Hard lever: a single avoided wave of fragrance returns pays for the pilot. After-sales stops being a black box and becomes a source of line improvement.
• Expected reduction in units with out-of-tolerance closures ≥ 30% as soon as the model has picked up its first full shift.

And the brand director's reasonable doubt

"What if the AI rejects good units with a perfectly closed cap?" — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely measures a geometry and compares it against a known tolerance, the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: doubtful units go through a validation screen before the actuator. The three safety rings exist precisely for this.

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