Freezes the pack mid-flight
A 40 µs strobe at 0.75 m/s limits motion blur to about 0.03 mm. A red band-pass filter keeps room light out of the image.
Example engagement
Finding leaking seals at 180 packs per minute
A representative snack line where nitrogen-flushed pouches go stale because nobody can check seals fast enough, and how a Lensnode V, low-angle red light and a Gatepoint G2 would close the gap.
Example engagement — a representative scenario built from typical deployment parameters. Figures are modeled, not reported results from a named client.
Seal inspection at a glance
- Industry
- Food & beverage
- Services
- AI Visual Inspection Systems
- Perflens Care Essentials
- Hardware
- Lensnode V camera with low-angle red strobe lights
- Gatepoint G2 edge gateway
- Air-blast reject with a confirm sensor
- Timeline
- 3-week Feasibility Sprint, 6-week Proof of Value, 8-week deployment
Modeled results
~90%
fewer leaking-seal escapes (modeled)
<0.3%
false-reject rate on good packs (modeled)
100%
of packs imaged at both end seals
The situation
A fast bagger and a slow-burning complaint trend
Picture a snack plant packing single-serve bags of kettle chips on a high-speed vertical form-fill-seal bagger. Each pouch gets a fin seal down the back, a crimped jaw seal at each end and a nitrogen flush to keep oxygen away from the oil in the chips. The bagger runs at 180 packs per minute, three shifts a day.
A leaking seal doesn't show up at the plant. It shows up weeks later as a stale bag in someone's pantry, then as a complaint nobody can trace to a shift. The causes are familiar to anyone who has run a bagger: a chip fragment or a pinch of seasoning caught in the jaws, a pleat folded into the seal when the film tracks off center, or jaws running a few degrees cold after a changeover.
The plant's defense is an hourly vacuum-chamber leak test: a few packs submerged and watched for bubbles. It is a good test of the packs it checks. But a jaw problem that starts at 2:10 a.m. runs until the 3:00 a.m. pull, and 50 minutes at 180 packs per minute is about 9,000 packs.
Constraints
What the design had to live with
- Line speed
- 180 packs per minute; the takeaway belt runs at about 0.75 m/s and packs land at uneven spacing
- Packs and film
- Pillow pouches in metallized film with full-coverage print; 6 SKUs in two pouch lengths
- Environment
- Dry packaging room with salt and seasoning dust; daily dry clean, weekly low-pressure wet sanitation
- Space
- About 600 mm of open conveyor between the bagger discharge and the checkweigher
- Food safety
- Shatter-resistant windows on any optics; every bracket and fastener logged in the plant's foreign-material program
- IT/OT rules
- Nothing on the corporate network and no internet from the packaging floor; new equipment goes on an isolated OT VLAN
- Shifts
- Three shifts, five days, 4–6 SKU or film changeovers a week, weekend sanitation
Approach
From a box of leakers to a running station
-
Feasibility Sprint
The plant sends about 200 good packs and 60 leakers, each marked with its known cause. In our optics lab, low-angle red light wins: on metallized film the flat areas mirror the light away from the lens and go dark, while crimp ridges, pleats and trapped crumbs scatter it back. The printed graphics almost vanish. The go/no-go report finds a visible cause on about 9 in 10 sample leakers.
-
Proof of Value
A loaner Lensnode V and Gatepoint G2 judge every pack in shadow mode while flagged and passed packs go to the vacuum test. Success criteria, agreed in writing up front: catch at least 85% of the leakers the test finds, with false rejects under 0.5%.
-
Deployment
A stainless bracket and light shroud, encoder tracking and an air-blast reject with a confirm sensor; the bagger stops if a reject fails to confirm twice in a row. Factory acceptance runs on the plant's own film, then site acceptance against the Proof of Value baseline.
-
Care
Perflens Care Essentials tracks confidence and reject rates by SKU, retrains twice a year and flags drift when a new film lot changes how the crimp reflects light.
What the system does
Two seals per pack, judged in under 40 ms
Images both end seals
Two triggered exposures per pack, leading and trailing seal, using region-of-interest readout so the camera never limits line speed.
Separates causes
Product in seal, pleats and faint crimp are classified separately; an anomaly score catches disruptions the model was never trained on.
Rejects the right pack
Encoder tracking times the air blast, and the confirm sensor checks every flagged pack left the belt. See reject integration.
Tells the operator why
When product-in-seal events climb above the shift baseline, the stack light and a Console alert point to the jaws rather than the film.
Keeps every reject image
Perflens Console stores each reject with its SKU, cause and model version, so quality can answer a complaint with evidence.
Modeled results
What a modeled year on this line looks like
~90%
fewer leaking-seal escapes reaching customers
<0.3%
of good packs falsely rejected
<2 min
from a jaw fault to an operator alert, against up to an hour between pulls
100%
of packs imaged, instead of a few per hour
Modeled for one bagger at 180 packs per minute, three shifts, five days a week, with leak rates and causes typical of vacuum-test logs on this kind of line. Assumes about 9 in 10 leaking seals have a visible cause in the end-seal area; film pinholes and fin-seal leaks are invisible to this station. The false-reject figure assumes stable film supply; a new supplier is treated as a retraining event.
Phase two
What we would watch next on this line
- The fin seal. A second Lensnode V looking up through a conveyor gap would see the back seal on every pack, closing phase one's main blind spot.
- Jaw temperature. Reading jaw temperatures from the bagger PLC over OPC UA would link faint-crimp events to the heater that caused them.
- Fill weight and seal together. Correlating checkweigher rejects with product-in-seal events shows whether overfilled packs push chips into the jaws.
- The other baggers. The same model family, retrained per film and pouch size, becomes the template for the remaining lines, with synthetic defect data covering rare causes once real examples exist to validate against.
Related services
The services behind this scenario
AI Visual Inspection Systems
Turnkey cells with the optics, lighting, edge compute and reject integration this station relies on.
Read more
Perflens Care
Drift monitoring and scheduled retraining that keep a seal model accurate through film lots and SKU changes.
Read more
Food & Beverage
Inspection and monitoring designed around sanitation, foreign-material rules and fast packaging.
Read moreFAQ
Questions we hear often
Can a camera see a leak channel that small?
Usually not directly; a channel through a crimp seal can be a fraction of a millimeter wide and hidden inside the fold. A camera reliably sees the cause, such as a crumb in the seal, a pleat or a faint crimp. That is why the vacuum test stays as an audit, and why the modeled result is about 90%, not 100%.
What would a station like this cost?
Inspection-as-a-Service starts from $2,950 per station per month on a 36-month term, including hardware, Console, Care Essentials and hardware replacement. As a capital purchase, single-station systems typically run $48,000–$140,000 installed. See pricing.
Model your own seal line
Send a box of good packs and a bag of leakers. A three-week Feasibility Sprint shows which leak causes a camera can see on your film, and what catching them is worth.