Tuesday, July 26, 2011

Don’t move the camera (but if you do have to …)

I recently saw a vision application that was keeping the engineers up at night. The problem was that inspection results were not repeatable. A quick look at the system in operation revealed that the camera was being raised and lowered on each cycle to get the part in and out. Most machine vision pros will tell you to never move the camera, so I thought readers might find it instructive to expand on this.

It doesn’t take a lot of engineering know-how to grasp that as after the camera had moved into position it’s mounting bracket would vibrate for a couple of seconds. As a result, the images were a little blurred and therefore edge locations were detected inconsistently. The engineering team had figured out that a dwell of two to three seconds would allow the system to settle before grabbing the image, but unfortunately the production rate wouldn’t allow that, so what should they do?

Well I resisted the temptation to say “I wouldn’t start from here if I were you,” although it seemed pretty obvious that the mechanical design was lacking. But the system had been built and paid for, so they needed some simple ideas. Here’s what we came up with:

  1. Shorten the exposure time as much as possible, which means opening the aperture, cranking up the gain or my favorite, throwing more light on the target.
  2. Increase the mass of the camera bracket to more rapidly attenuate the vibration.

So what did they do? Well a bit of everything really. The added in the longest dwell they could get away with without slowing down the system, added some mass to dampen the vibrations more quickly, and bought a halogen light so the exposure time could be reduced. Taken in concert, these changes helped the system meet the repeatability goals.

Lesson learned, I hope.

No comments: