Pharmaceutical · appearance inspection line

Tablet and capsule inspection line — teach normal images, catch defects, and track defect kinds and NG rate

Three cameras judge a tablet line, a capsule line and a backlit capsule line, one image every 2.5 seconds. A reference image difference taught with 10 normal images flags whatever differs from normal, while colour, print, diameter, length and brightness are checked by tools that measure them. An NG paints the defect red on the result image, and one screen shows the defect kind, measured values, counts, NG rate and the recent NG list. Consecutive NGs and the NG rate raise alarms.

Example setup

This is an example setup, not a real plant or customer installation. The tablet and capsule images and their markings (XQ embossing, DEMO print) are computer-drawn and made up, with no relation to real medicines or brands. The 120/120 match is on the sample's own drawn images, not on real line variation, and it is not a validated inspection standard. The inspection room is computer-rendered too. The reference image difference tool arrives in the LinkVion release after 0.1.0.24.

3lines — tablet, capsule and backlit capsule, three cameras (one image per 2.5 s)
3 · 3vision jobs and recipes (one per line)
40 · 1 · 6tags, script and alarms
120/120verdicts and defect kinds matched on the sample images (90 normal, 30 defective, drawn)

Running it

Three lines run; an NG brings the defect mark and an alarm

The inspection room is computer-rendered, and the panel PC shows a real-time recording of the run screen with the same sample running in LinkVion. The screen text is small, so the card of the line with the latest NG (top left) and the alarm banners from the bottom of the screen (bottom left) are cropped from the recording and shown close up.

Example setup

Real time: each line judges one image every 2.5 s → an NG gets a red defect mark and its kind on the result image → alarm banners for 3 NGs in a row and over 25 % NG in the last 20 → the recent NG list with time, line and defect kind → alarms stacking per line · 55 s

The room and the tablets and capsules on the belts are computer-rendered for illustration; the products on the belts are not what the cameras actually saw. The run screen was recorded in Korean (a button at its top right switches it to English). You cannot run it on this page.

On the screen

How the inspection works

  • Three vision jobs, three recipes
  • Reference image difference (taught with 10 normal images)
  • Colour, print, diameter, length and brightness tools
  • Defects marked on the result image
  • Consecutive-NG and NG-rate alarms
  • Script sorts defect kinds and collects recent NGs
  • Korean and English screens
Three lines, one product each
The tablet line views a white round tablet (XQ embossing) under top light, the capsule line an ivory and red capsule (DEMO print) under top light, and the backlit line a powder-filled capsule over a transmitted light. Three folder replay cameras play 40 images per line (30 normal, 10 defective) at 2.5 s intervals.
Reference image difference
Taught with 10 normal images, it learns the small variation of a good product and flags anything outside it, without knowing in advance where a particle will sit or what colour it is. For the printed capsule the region is split into the print band, the body dome and the rest, and only the print band gets a wider position tolerance.
Measurable values get measuring tools
Tablet discolouration uses colour identification, edge chips a broken-area count, capsule print smears and gaps print verification, a 0.2 mm dent four body diameters, empty or underfilled capsules inner brightness, length defects overall length, and double caps the cap diameter. A script reads which tool failed and sorts the defect kind.
Alarms and recent NGs
Each line raises an alarm for 3 NGs in a row (cleared after 5 OKs) and for over 25 % NG in the last 20 images. The recent NG list at the bottom keeps up to 8 rows of time, line and defect kind, and each line shows OK, NG, consecutive NG, today's NG rate and the last-20 NG rate.

Limits

What this case does not show

  • Not a real line or real cameras. The room and the products on the belts are computer-rendered, with a recording of the same sample running in the product laid over the panel PC.
  • The tablet and capsule images are drawn and the markings (XQ, DEMO) are made up. 120/120 is the result on these images, and the recipe thresholds are not real inspection standards. Real products need re-teaching and validation with real normal images.
  • The reference image difference tool is available from the LinkVion release after 0.1.0.24. The sample needs the vision inspection feature to run.
  • Cameras created at the same time make the three lines show NGs and alarms together. There is no live version on this page, only the video.