My very first interface with machine-vision systems was working with a Canadian company that wanted to use a PLC to automate television CRT tuning using four high-resolution analog cameras. The cameras were on a gantry system that would be placed in front of the CRT, and a pattern generator would provide the signal to display. The PLC was to automate the setting of the RGB guns and convergence to provide for a color-accurate display that was sharp.
It didn’t work for reasons that have slipped past my recall, but it had something to do with the lack of data coming to or from the cameras to allow the PLC to perform its duties. The project was in the 1980s while I was working for Rockwell Automation.
Machine vision has come a long way since then. My last interface was with a Keyence image color camera, which was detecting whether or not a tube had been inserted into a bag for sealing. Detecting the orientation of the tube meant two things. First, was the tube inserted far enough into the bag? And, second, the tube could vary in angle by 45°. The camera had to be very fast in determining the result.
And it was and worked flawlessly. It had to evaluate color, as well. Lots to do, and it did it well. We have come a long way.
In the beginning, machine-vision systems were disguised as barcode readers as such. Reading barcodes and QR codes is child’s play today.
Heuft makes bottle inspectors. I saw it in action at a local brewery while it was scanning bottles at 120 bottles per minute for foreign objects that had been left in the bottle after washing. It was fascinating to watch and to understand the technology behind it.
It interfaced with a PLC for bottle rejection and control. It simply compared a black-and-white image to a stock pattern, and, if there was a difference, then it rejected the bottle.
What a difference 20 years makes in vision.
Black-and-white vision can do pattern recognition easily enough, and today’s cameras can go far beyond that by using color for part sortation and final inspection among other tasks. Some cameras can detect colors in 24-bit resolution, which translates into 16 million different color variations, and that is probably better than any human can detect. The nuances in color are lost with the human eye, but not to technology.
Cognex, a vision vendor, can detect color variations down to the pixel level using its color model tool. Imagine producing cosmetics and scanning for imperfections in the finished product based on color tones. The surface of the product has to be uniform to within specifications, and the vision system can perform this task.
Most machines have network connectivity these days, and IP cameras can play an important role in machine vision. An IP camera can be placed strategically on the machine to allow the operator to see certain operational locations that would be unsafe for them see natively.
Machine vision covers a wide assortment of options from a task perspective. I am sure that if you take a look at a modern day assembly system you could identify where a vision system could be employed.
If you think it, it can be done. A vision system can detect anything with a very high degree of resolution and accuracy. The ability to interface with control systems has been greatly enhanced, as well as the ability to pre-process information and deliver that information directly to HMI/SCADA systems.
The ability to be networked has expanded the application base, as well as the capabilities such as machine learning, operational learning and cloud interfacing. It truly is a different world today.
I learned that the automotive industry uses color-coded glues in manufacturing processes. There is no better way to have quality control of the glue application than to have a vision system monitoring the application from the robot, as well as final inspection once the robot is done. Redundant quality checks are paramount with these subassemblies.
The ability to see using technology has been revolutionary. The incurred costs are minuscule compared to waste and lost production due to a failed process, which can be nipped in the bud using machine vision.
