Blue veining in Stilton PDO — verifying the piercing before the cheese goes into the cave.
The blue veins of Stilton PDO are born from hundreds of manual punctures that carry oxygen to the Penicillium roqueforti. If the piercing is sparse, badly spread or too shallow, the veining comes out weak — and that only shows at the cut, weeks later. iLEAN Vision reads every cylinder with a CNN as it comes off the piercing bench, counts and maps the holes, and holds the cylinder if the pattern falls short. The line does not restart on its own — the person signs.
A piercing mistake shows up when the cheese is cut — and by then it is too late.
Stilton PDO is seeded with Penicillium roqueforti in the vat, but the mold needs air inside the cylinder to spread and form the blue veins. Because the cylinder closes up during pressing, the mold stays dormant. The operator pierces each cylinder by hand with long needles — several hundred punctures spread across the whole surface — to create the channels along which the Penicillium travels.
Three ways the piercing can go wrong:
- Sparse piercing — few veins, veins that never reach the center of the cylinder.
- Badly distributed — veins concentrated in one area, other areas left white.
- Shallow punctures — surface veining with no penetration.
Control today: the supervisor walks past the piercing bench and looks; sampling plus cutting a cylinder at the end of ripening, weeks later. By the time the defect is visible, the cylinder has been in the cave for weeks. The difference between a well-veined PDO cylinder and a second-grade one is several euros per kilo, in a long-aged cheese with tight margins.
iLEAN Vision does not pierce — it verifies that the piercing meets the PDO pattern before the cave.
The operators keep doing the piercing — that is what tradition and the PDO specification require. What changes is the feedback: instead of discovering an insufficient pattern at the cut, weeks later, iLEAN Vision reads every cylinder as soon as it leaves the piercing bench and warns before it goes into the cave. iLEAN acts as the putty between the supervisor's eye (which cannot be on every cylinder) and the sample cut (which arrives too late).
Edge looks at every cylinder after piercing. The CNN counts holes and maps distribution. If the pattern falls short, the cylinder goes back to the bench. The final cave receives only cylinders with a validated pattern — the person signs.
The iLEAN pieces applied to veining control in Stilton PDO:
- Edge — vision over the belt that feeds the cave. A camera with calibrated lighting reads the cylinder right as it comes off the piercing bench. A CNN trained on the PDO pattern agreed with the quality manager counts visible holes, computes density per quadrant and detects low-coverage zones. It works without a network: if the plant loses WiFi, Edge keeps reading and holding.
- Connect — capture of the piercer's log. If the plant uses paper notebooks or Excel sheets to record shifts, milk batches and the assigned operator, Connect digitizes them and cross-references them with the vision results. A weak pattern that keeps repeating on a particular shift: the system detects it and alerts the supervisor.
- Agent. Cross-references vision + log + the history of ripening results at the cut. It learns which piercing patterns give good veining in this cheese, in this plant — and builds quantitative feedback for the operator ("your piercing left the top half covered and the bottom half weak"). The cylinder is held; the person validates. The system does not send anything to the cave on its own.
Piercing control by sampling vs. verification with iLEAN Vision
| Aspect | Classic control (supervisor + sampling) | With iLEAN Vision (Edge + agent) |
|---|---|---|
| Control coverage | One sample per shift + the supervisor's eye | Every cylinder, as it leaves the piercing bench |
| Feedback to the operator | "Today came out fine" / "today was so-so" | "Top half covered, bottom half weak — go over zone X" |
| Detecting an insufficient pattern | At the final cut, weeks later | Before the cylinder enters the cave |
| Drift by shift or by operator | Hard to isolate | Cross-referenced with the log and flagged to the supervisor |
| Operation without network | n/a | Edge keeps reading and holding on the panel's own power |
| The veteran's knowledge | It leaves with them | Captured as a reusable PDO pattern |
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 piercing line. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.
- PDO dairy with one piercing line, several thousand cylinders/week, occasional documented drift between shifts.
- iLEAN Vision pilot over the belt that feeds the cave: camera + calibrated lighting + hold actuator, integrated with the shift log. First value in a few weeks (detection above the threshold agreed with quality).
- Indicative payback between 4 and 9 months, depending on how many cylinders drop out of PDO grade each year because of weak veining and on the price gap between PDO and second grade.
- Expected reduction of ≥ 30% in the risk of a grade downgrade caused by insufficient piercing, based on your current pattern of internal rejections.
- The hard lever is every cylinder that stays in PDO instead of dropping to second grade — plus capturing the veteran piercer's knowledge as a reusable reference.
And the master cheesemaker's reasonable doubt
“What if the vision gets it wrong and holds a cylinder that was fine?” — hallucination is a problem of free generation, not of anchored tasks. Counting visible holes and mapping their distribution is an anchored task: the best models brought error below 1.5% [1]. And even so, a held cylinder is never scrapped on its own: it goes back to the piercing bench and the person decides. The three safety rings exist precisely for this.
[1] OpenAI paper “Why Language Models Hallucinate”, 2025 — on the reliability of AI in anchored tasks.
What people ask about blue veining control in Stilton PDO
Why is Stilton PDO pierced?
Stilton PDO needs oxygen inside the cylinder so that the Penicillium roqueforti (seeded in the vat) can develop its characteristic blue veins. Because the cylinder closes up during pressing, the mold stays dormant. To wake it up the cheese goes through manual piercing — several hundred needle punctures per cylinder with long needles, spread across the surface. Every hole is an air channel along which the Penicillium spreads. If the piercing is sparse, badly distributed or too shallow, the veining comes out weak or does not come out at all — and the cylinder loses its PDO grade.
What is the operational problem with manual piercing?
It is repetitive, tiring work, done by several different operators across different shifts. The quality of the piercing depends on who did it, at what point in the shift and with which tool. The typical quality control is sampling + cutting a cheese at the end of ripening — weeks later, when the cylinder can no longer be undone. Every Stilton PDO cylinder that ripens badly because of insufficient piercing is product that drops a grade, in a long-aged cheese with tight margins.
How does iLEAN Vision verify the piercing before the final cave?
A camera with convolutional neural networks (CNN) reads the cylinder right as it comes off the piercing bench, on the belt or on the table where it passes to the cave. It counts the visible holes on each surface, maps their spatial distribution (density per quadrant, average distance between punctures) and detects zones with insufficient coverage. If the pattern does not reach the PDO threshold the system has learned, the cylinder is held and goes back to the piercing bench. The line does not restart on its own — the person validates.
Doesn't this contradict the craft character of Stilton PDO?
No. The system does not pierce — the operators still do that, as tradition and the PDO specification require. What iLEAN Vision does is exactly what the quality supervisor already does when they walk past the bench and look: count and check the distribution. Only it does it on every cylinder, not on a sample. It assists and simplifies the operator's work, it does not replace it. The knowledge of the veteran who knows “this pattern is going to turn out right” is captured as a reusable reference.
How much does AI vision on Stilton piercing cost and how long does it take to deploy?
A typical pilot covers one piercing line: camera + calibrated lighting + a hold actuator (stack light or diverter), plus the annual license for the agent that learns the PDO pattern. First value in a few weeks (detection above the threshold agreed with the quality manager). Indicative payback of several months, depending on how often cylinders currently drop a grade because of weak veining and on the price gap between PDO and second grade. We ask for your plant's data and send you the estimated ROI in 48h.
Tell us your case and in 48h we'll send you the estimated ROI of Stilton PDO piercing control for your plant.
We work on your piercing line's real data, not ours. Diagnostic with no commitment.
Request estimated ROI in 48h See food industry