Justifying vision systems for automated inspection

It’s extremely difficult to save labor with machine vision. Most vision projects with a labor-saving goal set out to automate what an end of line inspector does. But if you’ve ever studied what these ladies, (they’re almost always ladies, generally of a ‘certain age’ and a cuddly frame,) you’ll know they do far more than look at parts.

Handling is a big part of their workload, but there’s often some packing and labeling involved too. Sometimes the job entails doing some rework or rectification if a defective item can be salvaged. Try getting your vision system to do all that!

So I keep returning to the value of not shipping nonconforming product, and this is where I’m at:

Let’s say there’s a certain probability of manufacturing an item that doesn’t meet spec. Call it P.

And there’s a probability that our lovely end-of-line inspector misses the defective item. We could call that M.

Then, if we make Q pieces per year, the number of bad ones reaching our customer, N = P x M X Q.

Now I don’t know the total impact of those N bad items reaching our customer, but I can probably estimate some of the direct costs. For example, we expend labor handling the returned item, completing paperwork, and disposing of the junk product (or perhaps reworking.) It’s also almost certain that our customer will expect a credit for the item, and may expect us to pay for the return shipping too.

In some industries it’s normal for a whole shipment to be quarantined if a single defective item is found, and usually the manufacturer is expected to sort it, at considerable expense. And some customers, if sufficiently irritated, will insist on third party inspection, again at the manufacturer’s expense.

So it seems the cost of shipping a bad part can be very high. And that’s without trying to cost the damage to your reputation.

Next time you’re looking to justify an inspection system, try approaching it this way. I’d be very interested in hearing, confidentially, how it works out.