Prefilled syringe inspection with AI — every unit leaves with its own file, not just its batch number.

The problem

The inspection machine fires a false alarm — and the operator loosens the thresholds.

A prefilled syringe line has the equipment. The problem is that everyday reality forces people to cheat with it:

1. The automatic visual inspection machine fires hundreds of false alarms per shift: glass reflections, micro-bubbles that are not defects, label positions inside tolerance that the model reads badly. The operator, under pressure on OEE, loosens the thresholds — and then a real particle slips through.
2. The FMD serialization system stores the serial number, batch, expiry date and parent-child aggregation. It works. But that data is never cross-checked with the inspection machine: if one specific serialized unit triggered a cosmetic alarm that the operator reclassified, that stays in another log.
3. The batch's Annex 1 file is rebuilt by hand: BMS, interventions, filter integrity, AQL, inspection machine data, aggregator data. Every batch, one person, several days.

The demanding standard in control-driven industries has historically hovered around 25 parts per million in automotive ^[1]; parenterals demand more. The problem is not a lack of equipment — it is that each machine only tells its own part of the story and nobody puts together the full history of the serialized unit. And that is why there is always a column missing when the inspector arrives.

How it fits the IRIS system

iLEAN does not replace the inspection machine or the FMD system — it cross-checks them.

The problem is not a lack of equipment — it is information living on islands. The automatic inspection machine does its job well, the serialization system does too, so does the BMS. iLEAN acts as the putty that joins the three stories into a single file per unit and per batch, without asking you to throw anything away.

Edge adds vision where the inspection machine fires false alarms. Connect cross-checks with FMD and the MES. The agent assembles the dossier per batch — and per unit when that is what you need. The QP signs with the evidence in hand.

The three iLEAN pieces applied to prefilled syringes:

• Edge — a terminal with machine vision (CNN) in line, next to the inspection machine or as a second step. It adds a layer of judgment over the false alarms (discarding harmless reflections and micro-bubbles) and cuts the noise so the operator does not loosen the thresholds. It also detects plunger and cap defects. It works with no network. What is critical does not depend on WiFi.
• Connect — captures data from the pharma MES (PAS-X, Werum, Korber), from the serialization system (Tracelink, Antares, SAP ATTP), from the filling SCADA and from the current inspection machine. It also captures excipient sheets and CoAs. Through whatever route each piece of equipment offers: direct integration, local reading, a camera on the panel.
• Agent — cross-checks serial numbers + inspection machine alarms + process data + environment + interventions. When a unit has an atypical history, it flags it. At batch closure, the QP receives the Annex 1 file already assembled. The three safety rings guarantee that release remains a human decision.

See the full IRIS architecture →

Before and after

PFS line with an isolated inspection machine vs. a line cross-checked with iLEAN

Aspect	Inspection machine + FMD in isolation	With iLEAN Edge + Connect + Agent
Inspection machine false alarms	Operator relaxes the thresholds	Edge's second vision filters; the threshold holds
Cross-check with FMD serialization	Two separate logs	A history per serialized unit
Grade A environmental data	BMS on the side, hourly CSVs	Cross-checked with the batch and the unit
Annex 1 file per batch	A dedicated person for several days	Dossier assembled at batch closure
Deep audit capability	Manual rebuild	History already cross-checked and searchable
Operation with no network	n/a	Edge keeps inspecting on the panel's own power

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 QP and the committee have an order of magnitude; we refine it during the diagnostic.

• Plant with a PFS line (biologic or small molecule), an automatic inspection machine + FMD deployed, and a manual Annex 1 file.
• Pilot with iLEAN Connect (MES + serialization + BMS) + Edge (second vision over the inspection machine). First value expected within a few weeks: the first batch dossier assembled automatically.
• Reduction of false alarms requiring manual rework of ≥ 30% against the documented history.
• Indicative payback between 4 and 9 months. The hard levers: recurring QA time spent assembling files + one Annex 1 deviation avoided on re-inspection + less inspection machine rework.

And the CAIO's reasonable doubt

“What if the AI hallucinates while filtering false alarms?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI judges a real image against real reference patterns (Edge's second vision), the best models brought error below 1.5% ^[2]. And even so, Edge does not discard potential defects: the flow is filter → the inspection machine keeps running, but the noise drops → the operator keeps the thresholds. What is critical — a confirmed defect, batch release — remains a human decision.

^[1] Quality standard of ~25 PPM in automotive — source Symestic.

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

Other related topics: Autoinjector pen with AI · Aseptic eye drops with AI · Visual inspection of injectables.