EN 14683 surgical mask control with AI — basis weight, BFE and welding reconciled before packing.

The problem

Three parameters that almost never come together in time.

A surgical mask conforming to EN 14683 comes out right when three things line up, and only if all three do:

1. Basis weight of the filtering meltblown — the melt-spun polypropylene filtering layer is what delivers the BFE. Its basis weight varies with the raw material batch, with the extruder temperature and with die wear.
2. Bacterial filtration efficiency (BFE) — it is sampled from every batch and sent to the lab. The result arrives when the batch is already closed.
3. Ultrasonic welding — power, amplitude, time and horn pressure. It drifts with wear, with ambient temperature and with every roll change.

The operator watches what he can. The quality manager reconciles the data when the batch closes. It works 99% of the time. That 1% is a batch that never gets released, a complaint from a healthcare customer, or worse, a mask in an operating room that does not filter what it claims to filter. And in a medical device, that is no longer a quality cost — it is manufacturer liability.

How it fits the IRIS system

iLEAN Edge does not add a fourth system — it seals the cracks between the three you already have.

The problem with surgical mask control is not a lack of sensors: it is that basis weight, welding and BFE information lives in islands and nobody cross-references it with the batch recipe in real time. iLEAN acts as the putty that fills those gaps, without asking you to change the extruder, the die-cutter or the LIMS.

Edge reads the in-line sensors. Connect picks up the BFE from the lab. The agent cross-references everything with the recipe and holds the batch before packing if something does not add up. The person signs — never the other way round.

The iLEAN pieces applied to EN 14683 surgical mask control:

• Edge — a terminal with vision + sensor reading on the line itself. It connects to the continuous basis weight gauge on the meltblown, to the ultrasonic welding parameters (power, amplitude, time) and to the camera watching the bonding of the layers. It cross-references the data in milliseconds against the recipe. If there is a combined deviation, it fires a stack light + actuator before packing. It works with no network: the basic cycle does not depend on WiFi.
• Connect — captures the BFE result the moment it is issued, whether it comes from the LIMS, from a signed lab PDF or from a sheet kept by the quality control technician. It also captures raw material changes, horn maintenance reports and supplier notices.
• Agent — lives in Central, cross-references basis weight + welding + BFE + batch recipe against the EN 14683 limits (BFE ≥ 95% type I, ≥ 98% type II/IIR, differential pressure, type IIR splash resistance). When the lab BFE arrives, it reassesses the release decision and alerts the manager if the combination of batch samples lands in the risk zone.

And the safety boundary: the agent holds and proposes; the quality manager signs the release or the rejection of the batch. The three rings of the IRIS system guarantee that no medical device reaches the market without human validation.

See the full IRIS architecture →

Before and after

Sampling control and end-of-batch review vs. cross-referenced in-line control with iLEAN

Aspect	Sampling control + end-of-batch review	With iLEAN Edge + Connect + agent
Meltblown basis weight	Checked every N minutes, by sample	Continuous, cross-referenced with the batch recipe
Welding parameters	Validated at the start of the shift	Monitored cycle by cycle, drift detected
Lab BFE	Arrives when the batch is already packed	Captured by Connect and linked to the batch automatically
Release decision	Review the batch paperwork at closing	Batch pre-flagged by the agent, the person validates
Line with no network	n/a	Edge keeps cross-referencing sensors and holding when required
File for the Notified Body	Rebuilt by hand, batch by batch	Complete per-batch dossier, with photo and sensor charts

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.

• EN 14683 surgical mask plant with 1-2 high-cadence lines, type IIR for a healthcare customer, BFE sampled per batch.
• Edge pilot on one line (basis weight sensor reading + welding parameters + camera + integration with the MES and LIMS for BFE). First value in a few weeks.
• Indicative payback between 4 and 9 months, depending on how often batches went unreleased or were reprocessed in recent years. Expected reduction in borderline batches ≥ 30%.
• The hard lever is a single batch recalled from the market or not released on time: product, reverse logistics, brand damage with the healthcare customer, and MDR compliance. One recall pays for the whole pilot.

And the technical director's reasonable doubt

“What if the agent decides to release a batch it shouldn't?” — it does not decide. It proposes. Hallucination is a problem of free generation, not of anchored tasks: the agent does nothing more than cross-reference a sensor reading with a recipe and a lab result (data that already exists). In anchored tasks, the best models brought error below 1.5% ^[1]. And even so, no batch ships without the person signing.

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