Machine vision camera applications abandoned instead of corrected. Finally, Herbert Ho from Sensorplex clears the view.
The next time you wander across the shopfloor, take notice of the machine vision cameras installed, and you may notice that many of these are often offline or turned off, allowing all products good and bad, to pass through the system. If you ask the questions, apart from rude comments, the reason provided will be that they are too sensitive and constantly reject good parts!
Vision systems are often selected after a static on-bench demonstration, where the product is hand-oriented. This usually produces good results but does not represent what happens in production when the product is often moving.
Most conveyors have guide rails to contain the product, which often acts as visual obstacles, and the cameras cannot be installed as demonstrated. One good example is detecting a feature on a bottle. The demonstrator can orientate the product to suit the camera on the bench, but when the bottle is on a conveyor, the whole scenario changes. We have an unstable product on a moving platform. The camera has to deal with a product that can wobble and twist. As the camera is usually mounted perpendicular to the direction of travel, the product can come closer or further away depending on how loose the guide rail is, creating focusing issues.
A very popular vision application is the monitoring of batch or expiry codes. The technique commonly used is Optical Character Recognition (OCR). Too often, little attention is paid to the mounting of the printhead. Continuous inkjet printing is a marvellous technology but can be very sensitive to how the printer head is mounted. The slightest change in the position of the printhead can cause significant changes in the printed text. 1-2mm closer to the target, the characters decrease in size and increase in boldness. 1-2mm further, the texts become larger and reduces density. The slightest rotation of the printhead can skew the characters while tilting causes characters to be narrow at the bottom and wide at the top!
Similarly, when the conveyor belt speed increases, the printed text shrinks! The machine vision camera is ‘rule-based’, and when the text changes, it gets confused. But they all have tolerance settings. Yes – they do, but when dealing with characters like a ‘6’ or an ‘8’, increasing the tolerances of acceptance often leads to the system being unable to differentiate between such characters. As a result, the system may accept an expiry date of 2018 when printed as 2016. The consequences can be costly for a product such as milk, but they can be life-threatening in the pharmaceutical industry. Hence, the famous quote, “I can read it but the …. camera cannot” – is dangerous but too often the case.
Unlike a factory’s machinery, machine vision cameras are very different from photo-electric sensors, motor drives, gearboxes, etc. These are maintained by fitters and electricians, who are also often called upon to support machine vision cameras. Apart from learning about cameras, they also have to learn about lighting, which is a vital part of getting machine vision cameras to work correctly.
Sensorplex recognised the issues early and set out to solve the problem. To remove the task of the in-house maintenance crew looking after the cameras, Sensoplex developed the Sensorplex PVS package, which takes over all the manual functions of setting and maintaining cameras on the shop floor. As remote support is standard, the vision systems can be maintained remotely. The computer stores and loads the camera parameters while logging event details. The PLC monitors the triggering and timing functions. All systems are supplier turnkey, meaning they have to be proven on the line and not only on the bench. This is where over 20 years of experience in industrial supply sensors come into play. Sensorplex engineers understand what happens on the production line.
The first Sensorplex PVS was installed in 2008 and is still running, monitoring over a million milk bottles each week to ensure that all products go out with the correct labels and printed date code.