Anticipating the unpredictable

I’ve been reading about Black Swans recently. Now this doesn’t mean I’ve developed an interest in ballet because the Black Swans I’m interested are those written about by Nassim Nicholas Taleb. NNT, as he likes to refer to himself, defines a Black Swan (and I’m paraphrasing,) as a significant or momentous event that no one saw coming. 9/11 was a classic Black Swan event for everyone except the hijackers.

So what does this have to do with machine vision?

I suggest it has everything to do with machine vision. In a vision inspection application a Black Swan can be either that bad part that gets through because we weren’t looking for it, OR the good part that gets rejected because in some way it is different to what we designed the system for. An example might be change in shade or color resulting from a small alteration in a coating process. Product function is not affected but an edge tool might fail, resulting in what we call a False Reject.

Now human inspection doesn’t suffer from Black Swans. This is because a human has the intelligence to recognize when something looks different and ask for guidance. In contrast, a vision system has no intelligence. One day perhaps we’ll have software so sophisticated that it will be able to replicate the human and ask a question, but until then, how do we anticipate unpredictable changes in the targets of our inspection systems?

You’ll be pleased to know that I do have some ideas, and I’ll share them with you in my next post.