KOMPASS brings print quality inspection to every printed pack, label and container. AI models trained on your artwork find smudges, ink spots, missing print and discolouration, even on curved or reflective surfaces, and signal the line before defective print reaches the market.
Presses, screens and coders repeat whatever they are doing. A smear, a dry nozzle or a damaged screen produces the same flaw on every pack until someone sees it.
Our AI Product Inspection system, KOMPASS™, brings AI print quality inspection to packs, labels and containers, using models trained on your good and defective prints. Every result is passed to the line controls with an image of the print.
Jidoka's demonstration shows in-process print inspection after screen printing. Bottles are rotated 360 degrees for image capture, images are stitched for inference, and defects such as smudge marks, ink spots, print shift, discolouration and illegible text are identified.
Round containers are rotated past the camera and flat packs are viewed directly. Lighting is set to keep reflections off the print.
The model checks the full printed surface and classifies smudges, spots, streaks, missing print and discolouration against your standard.
Defective items are rejected, recurring defects point to the press or screen and every result is stored with its image.
Send us good prints and examples of each defect you care about. We will assess the imaging and handling your print quality inspection needs.
Get In TouchEvery approach depends on a clear view of the print. The difference is how print quality inspection handles real variation and how many items it covers.
Marico needed 360-degree print inspection on screen-printed bottles in under half a second per bottle. KOMPASS automated the check that manual inspectors could not sustain at volume.
Marico's published results apply to its in-line print quality inspection of screen-printed bottles.
Print quality inspection with KOMPASS finds the visual print defects your quality team cares about: smudge marks, ink spots, streaks, voids, missing print, discolouration and illegible text. Each defect type is learned from real examples on your products. Defects are classified separately, so you can apply different limits to a tiny spot in a background area and a smear across a brand name. The same station can also confirm that every expected print element is present.
Classic systems compare each image with a reference image of a good print and count differing pixels. Reflections, small position changes and normal print variation all create differences, so rejections of good product climb on glossy or curved items. An AI model trained on acceptable variation learns which differences matter, which keeps sensitivity to real defects without rejecting good product.
Yes. Round containers are rotated in front of the camera or captured by several cameras, and the images are combined into a full view of the print. This 360-degree inspection is what screen-printed bottles need, because a defect can sit anywhere around the circumference. Glass and plastic containers both work, with lighting adjusted for each. The full view also shows where on the bottle a defect sits, which helps trace its cause.
It works on direct screen printing, pad printing, digital print, printed films and labels. Bottle decoration is a common application because a damaged screen or ink problem repeats on every bottle. The model is trained on the print methods and substrates you run, so the station knows what normal looks like for each. New products are added by training with examples rather than rewriting inspection rules.
As early as the station can see the print, ideally straight after the printing step. Misprint detection there limits how many items are affected before the press, screen or coder is corrected. Repeated failures can stop the line automatically, so a damaged screen is not left running for a full shift. Images of the first failures also show whether the fault is getting worse or staying stable.
By training on acceptable prints that include normal variation, by setting limits with your quality team and by validating on a labelled set before automatic rejection begins. Borderline items can go to a review lane at first. Tracking false rejects separately from missed defects shows whether the station needs more training or different lighting. Clear acceptance examples also help operators trust the station from the first week.
The defect type and its position on the print often point to the cause, such as lines from one ink station or spots from a damaged screen area. Where lines have several printing heads, the station can tag each result with the head or station that printed it, so maintenance knows exactly where to look. That turns each rejection into useful information instead of just waste.
Send good prints and examples of each defect, your acceptance standard, the print methods and substrates, and the line speed. Tell us where the station can sit after printing and where rejects can go. With that, we can recommend imaging, handling and training for print defect detection on your line. Defect examples from different printing heads help the model learn which faults matter most.
Share good prints, defect examples, print methods and line speed. We will set up print quality inspection at the right point after printing and give you timing and cost. Deployment is typically 6 to 8 weeks.
