KOMPASS brings OCR inspection to every pack, part and label on the line. AI models read the printed text, even when print quality varies, check it against what should be there and send a result to the line controls before an unreadable or wrong code leaves the plant.
Dates, codes and product text are printed thousands of times a shift. Without OCR inspection, inspectors can confirm that something is printed, but not read every character on every pack at line speed.
Our AI Product Inspection system, KOMPASS™, brings AI-based OCR inspection to packs, parts and labels: models trained on your print read the text, then check it against the expected content. Each decision reaches the line controls with an image of the print.
Jidoka's demonstration shows KOMPASS using AI-driven OCR to read batch codes, dates and serial numbers in real time. A print-synchronised trigger captures each pack, and mismatches or unreadable print can trigger automated rejection through the line PLC.
A trigger tied to the printer or conveyor captures each item with the text in the same place. Lighting is set for the substrate and print method.
The model locates the text region and reads every character, including faint, smudged or skewed print that rule-based tools reject.
The reading is checked against the expected text. Failures are rejected or held, and each result is stored with its image.
Send us examples of good, faint and wrong prints and the text that should appear. We will assess the imaging your OCR inspection needs.
Get In TouchEach method needs a clear view of the print. The difference is how each copes with real print variation, and whether OCR inspection checks the content at all.
In Diageo's labelled-bottle line, KOMPASS reads label text and checks orientation and condition in-process, ejecting non-conforming bottles automatically.
Diageo's results cover its complete labelled-bottle inspection, where OCR inspection of label text is one part.
OCR inspection built on AI is trained with images of your own print, so it copes with smudges, faint characters, skew and busy backgrounds that break template matching. Deep learning OCR does not depend on perfect character segmentation, where each letter must be cleanly separated before it can be read. That matters on fast lines where print quality changes as printers warm up, ribbons wear and nozzles partially block. It also means new fonts or layouts are added by training with examples, not by rewriting rules.
The same platform reads continuous inkjet codes, laser-marked text, thermal transfer overprinting and printed labels. Each method looks different. Inkjet print is dot-based and can bleed on porous board, while ribbon-based print is crisp until the ribbon wears. Training includes the print methods and substrates you actually run, so the model learns what normal variation looks like for each one before it starts making decisions.
Both. Industrial OCR returns the characters it reads, and KOMPASS then compares them with the expected string for the running product, such as a date, batch code or product name. A pack fails if the text is missing, unreadable or different from what should be printed. The comparison can come from a recipe, the printer or a line system, so changeovers update it automatically.
Yes, when the trigger, exposure and lighting suit the line. Expiry dates are one of the most common applications because a wrong or illegible date causes retailer rejections. On packaging lines the camera must capture each item with enough light and a short exposure to avoid blur. Throughput is confirmed against the full station cycle, including rejection, rather than reading speed alone.
Track the read rate, the share of items read correctly, together with false rejects and missed errors on a labelled set that represents production. Include faint, smudged and deliberately wrong prints. Test across shifts, printers and substrates, then repeat after any change to the printer or packaging. Low-confidence reads should go to review rather than being forced into a pass or fail decision. Agree the acceptance targets before go-live so everyone judges the station against the same numbers.
Low print contrast is a common cause. Dark ink on a dark or patterned background, or faint print on a glossy surface, can leave characters that a person reads with effort but a camera barely separates from the background. Lighting angle, colour and polarisation are adjusted during the trial. Where contrast is too low for any reliable reading, the result also tells you the printer needs attention.
The station sends the configured response to the line, such as a reject, a hold or a stop after repeated failures. Each failure is stored with its image, the text read and the text expected. Reviewing those images shows whether the cause is the printer, the packaging material or the data sent to the printer, which shortens the time needed to correct the problem. Repeated failures from one printer can also stop the line before a whole pallet is affected.
Send samples of good prints, faint and smudged prints and known wrong codes, along with the text that should appear and how it changes between products. Add the line speed, packaging materials and where a reject can happen. With that, we can assess lighting, camera position and training needs, and recommend how OCR machine vision should check printed text on your line.
Share your print samples, the text that should appear, your materials and line speed. We will scope OCR inspection for your checkpoint and explain timing and cost. KOMPASS deployments generally take 6 to 8 weeks.
