Transdermal patch control with AI — one micron of thickness out is one dose released out.

The problem

Three plant-floor realities that almost never get cross-checked in time.

On a transdermal patch line — drug-in-adhesive matrix or reservoir, coating onto the liner, rotary die-cutting, foil pouching — the guarantee of the delivered dose rests on three pieces of data that, in most plants, each live on their own island:

1. Laminate thickness — microns of API-loaded coating on the liner. It shifts with web speed, oven temperature and blend viscosity. Verified by a beta or laser gauge, but rarely correlated live with visual data and with die-cutting.
2. API content in the matrix — the percentage of active ingredient in the blend and the uniformity of the coating. Verified by sampling + IVRT (in vitro release test) at release, not in line.
3. Die-cutting and pouch sealing — the dimensions of the cut patch and the integrity of the foil seal. What the operator checks by sampling at the start and end of the batch.

The quality manager knows everything lines up for 99% of the roll. That 1% is what triggers the OOS investigation off the back of an IVRT result — and the suspicion that falls on the whole batch. Caught late it is a deviation; caught in line, it is nothing.

How it fits the IRIS system

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

The transdermal patch problem is not a lack of information: it is information living on islands that, at the critical moment (thickness gauge drift, a change of adhesive batch, a die-cutting head wearing out), does not reach the person deciding in time. iLEAN acts as the putty that fills those gaps, without asking you to change the coating line or the die-cutter.

Edge sees the matrix, the die-cutting and the pouch seal. Connect reads the thickness gauge wherever it lives — modern PLC, old panel, a forgotten spreadsheet. The agent cross-checks against the recipe and the IVRT specification and, if something does not add up, holds the roll. The person signs — never the other way round.

The three iLEAN pieces applied to transdermal patches:

• Edge — terminals with machine vision (CNN) over the coating line (matrix appearance: bubbles, craters, particles), the die-cutter (cut dimensions and cleanliness) and the pouch packer (foil seal line). It fires actuators (roll marker, pouch rejector, stack light) before cartoning. It works with no network. If the plant loses its internet connection, Edge keeps verifying.
• Connect — captures the thickness gauge whether it comes from the PLC over OPC-UA, from an old terminal via OCR, or from the formulation lead's setpoint spreadsheet. It also captures web speed, oven temperature, blend viscosity and whatever arrives from outside (an email from the R&D lead with a formulation change, a WhatsApp message from the supplier flagging a deviation in an adhesive batch).
• Agent — cross-checks thickness, appearance, die-cutting, sealing, the MES recipe and the IVRT/GMP specification. If there is a deviation it does not send an email at 10pm: it holds the affected stretch of roll and alerts the quality manager through whatever channel they use. The person validates and signs; the line does not restart on its own. And it assembles the batch record with every reading, ready for the auditor.

See the full IRIS architecture →

Before and after

Transdermal patches controlled by manual IPC vs. cross-checked with iLEAN

Aspect	Manual control + sampling-based IPC	With iLEAN Edge + Connect + Agent
Laminate thickness	In-line gauge, not correlated live	Gauge + vision + die-cutting correlated
Visual matrix defects	Eyeball sampling at start/end	Continuous CNN vision across the whole web
Die-cutting	Calipers every N pieces, the rest assumed	Dimensions and cleanliness verified piece by piece
Pouch sealing	Peel test by sampling	Seal line verified pouch by pouch
Operation with no network	n/a	Edge keeps running on the panel's own power
File for the FDA/EMA auditor	Rebuild it from paper, days of work	Automatic batch dossier with images and traces

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.

• Pharma plant with a coating line + rotary die-cutter + foil pouch packer for transdermal patches (drug-in-adhesive matrix or reservoir), typical sector web speed, multi-SKU with die changeovers.
• Edge pilot on the line (cameras over the coater, die-cutter and pouch packer + actuators + integration with the thickness gauge and the MES). First value expected within a few weeks.
• Reduction of roll stretches with thickness deviations or visual defects of ≥ 30% in the first pilot batch, depending on the initial severity and the spread of the coater.
• Indicative payback between 4 and 9 months, depending on the documented frequency of IVRT-driven OOS results and the scrap cost of API-loaded adhesive matrix (expensive).
• The hard lever is a single OOS deviation avoided: investigation, rework of the affected roll, risk of a scrapped batch, auditor attention.

And the quality manager's reasonable doubt

“What if the AI gets it wrong and lets through a stretch of roll with thickness out of range?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI simply recontextualizes a piece of data from one system into another (reading the gauge and comparing it with the specification), the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN holds the stretch and the person signs. The three safety rings exist precisely for this.

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

← MDI inhaler control with AI · Nasal spray control with AI → · Pharma blister control with AI →