Guides
The camera, light, lens, and software that check product on a line, and how each picture becomes a pass or a reject.
A machine vision system takes a picture of each unit as it passes a fixed point on the line, measures the feature it was set up to measure, and tells the line what to do with that unit. The question is narrow and decided ahead of time: is the cap seated, can the lot code be read, are there twelve pieces in the tray. Because the question stays fixed, the system answers it the same way on every unit and every shift.
Engineered lighting decides more results than any other part. It makes the fault stand out, and a shield or enclosure keeps the picture the same from the first shift to the third, whatever the room lights or a dock door are doing. See Lighting and optics for inspection.
The lens focuses the scene onto the sensor. With the sensor size, it sets how far the camera sits from the product, how wide the view is, and how much depth stays sharp. The imaging calculator works out the focal length and the size of one pixel on the part.
A grid of light-sensitive pixels, plus electronics that set the exposure and send the picture out. More pixels across the same view means finer detail on the part.
Runs the inspection steps on every picture. It can sit inside the camera, in a vision controller, or in an industrial PC.
A trigger, usually a photo eye or an encoder count, tells the camera the unit is in place. The result goes to the PLC and the reject. The picture and the result go to storage for quality records.
The trigger fires. The light flashes for a short exposure that freezes motion, and the camera sends the frame.
Filters even out brightness, reduce noise, or sharpen edges so the feature reads clearly. Many checks skip this step.
The software finds the product and the area to check, so the tools land in the right place when a pack arrives a little off center or turned.
Tools turn pixels into numbers: a fill height in millimeters, a piece count, the text of a lot code, the size of a dark spot on a seal.
Each number is compared with limits set from real good and bad samples. A fail sends a reject signal timed to the moment that unit reaches the reject, and the image and reason are saved.
Setting the limits takes most of the commissioning time. Limits that are too tight cause false rejects and slow the line. Limits that are too loose let faults ship. Both are tuned with real product from the line.
Each pixel holds one brightness number, usually 0 for black to 255 for white. Most line checks use it: presence, position, fill, print, and codes.
Each pixel holds red, green, and blue values. Use it when color is the fault: the wrong flavor sleeve, a missing colored tablet, a burned edge on a baked product.
Software turns a gray-scale picture into pure black and white at a chosen cutoff. Counts and outline checks run fast on it while the contrast between product and background stays strong.
Each point holds a height or distance in place of a brightness. Use it when the fault is shape: a dented can, a cap riding high, a crushed corner. Some 3D cameras return a grid; others return a cloud of X, Y, Z points.
When rule-based tools can do the work with reasonable effort, they are usually simpler to run and maintain. Many stations use both: rules to locate the product and read codes, and a trained network to judge appearance.
