High speed, high precision gauging by machine vision

I am often (well about once a month,) asked if I could put together a vision system capable of gauging to single micron levels. I usually respond with a very cagey, “It depends,” because this kind of measurement really pushes the limits, not only of vision algorithms but of optics, lighting and of course camera sensors. Think about it: if the pixels in the camera measure seven microns, how are you going to gauge a part repeatably to one or two microns?

So when I came across “The Vision to Innovate,” an article on ControlDesign.com (April 12^th, 2011,) about a system that not only measures in single microns but does it at 20 parts per second, I had to read it very carefully.

This article, one of the best machine vision application case studies I’ve read in a very long while, was written by Markus Tarin, President of California-based MoviMed. Markus takes the reader through a thorough discussion of the application needs and an overview of some of the technical hurdles before explaining exactly how they did the job.

I don’t want to steal Markus’s thunder, (or ControlDesign’s page views,) by giving away all the details here, but let’s just say telecentric lenses, collimated blue backlights, a real time operating system and subpixel interpolation were all involved.

It would have been useful to know exactly how the system was run-off or qualified, because in my experience that’s no mean feat itself. Measurement uncertainty becomes such a big issue at this level of precision that if variation is found how you’re faced with the challenge of determining the source. Personally I like the AIAG’s GR&R procedure, though that’s a discussion for a separate blog post.

So, “The Vision to Innovate” is a great article about a very challenging machine vision application. I suggest you read it carefully.