Anticipating product variation

You want to ensure your vision application is robust, so how do you figure out what it might have to deal with? I’m going to suggest three approaches.

1. Run a FMEA.

FMEA stands for Failure Mode Effect Analysis. It’s typically done as way of ensuring robust product design but I suggest it can be an effective tool for identifying what product variation could occur. It doesn’t need to be fancy: just gather a team who know the product and the process and brainstorm for thirty minutes. Unlike regular brainstorming though, don’t try to estimate likelihoods and discard the improbable event: try to design for them all.

2. Model variations.

A few years back National Instruments unveiled a tool they called “Model Vision System Variations.” The principle behind this was to automatically generate many subtle variations of the image that might be acquired and run these through the system in emulator mode. While the NI tool was aimed more at vision system issues like part motion or camera vibration and lighting changes the approach could be extended to product variation. Just open up PhotoShop and dream away.

3. Manage it.

A defining characteristic of a Black Swan event is its unpredictability. If you can predict that the shade of the part might vary you can plan for it, but then it’s not a Black Swan. So I suggest an alternative: manage your customer’s expectations.

Okay, this is really just good project management but it doesn’t seem to happen very often. What you should be doing is spelling out what the visions system will NOT be doing. Tell that it only measures the diameter of a hole and it does not check for scratches in the paint around the hole. I know none of us what to sound negative, but seriously, be very clear about the limitations of your system. Then, when a Black Swan event occurs, your customer will understand why the vision system didn’t deal with it.