IATF 7.1.5.1.1 Measurement system analysis (Visual Inspection)

Unless the defects are part specific - like paint build-up due to the component shape, for example - then I can't see why it would be. You have controlled/specified visual inspection viewing criteria? Like distance? Lighting intensity? Eyesight checks?

 

Exactly so.

I'd suggest that you'll need to get control of these things.

Many moons ago, I ran the RI function and we had a LOT components with painted finishes. To ensure my 4 inspectors were somewhat "calibrated" on what to look for, we set up a viewing environment, checked the lighting level (checked with lux meter), wrote an instruction (view at arm's length, not longer than 2 minutes etc) and then gave them examples with defects (or not) to see what they detected - and to observe them doing it correctly. They would often lift the parts closer if they "thought" there was a defect and we could smack 'em on the head to stop them doing that

It worked well. Once in a while, we'd introduce some marginal defects to verify what they were actually detecting.

 

Well yes, that's right. It's your choice. It shouldn't be on any auditor to decide. Now, if you aren't controlling the inspections (along the lines of my previous post) then it's hard to demonstrate you have control - and training isn't alone, going to work. Now, add to that, the fact that (some) IATF auditors are less experienced than you are and we get a trainwreck - which is why we're here to help.

I learned my approach from someone else who had the same challenges, in a different industry - but comparable "visual" qualities (actually the hi-fi sound equipment market). I'm not smart enough to create my own, but I know a good process when I see one. So I copied it.

When you know what works, you are better able to stand your ground against clueless auditors...

 

As I responded to your same question at the Cove:

If you have a visual inspection listed in your control plan you must perform a statistical study. This is exactly what the standard says as you quoted it... Since visual inspection typically only involves categorical (aka attribute) data then you must perform an “attribute study”. Any statistical study of inspections, measurements or tests is called a MSA (or colloquially a Gauge R&R). SO, you must perform an attribute MSA. But more importantly you should regardless of the requirements of the standard. Visual inspection is notoriously variable and ineffective so if you are relying on this it is in your best interest to perform the study, understand the effectiveness and improve it where necessary. OR you will have heck to pay in dealing with the inevitable inspection miss that makes it to your Customer….that will be far more painful than executing standard ‘good quality engineering practice’…

As to your last question about every part and defect, the answer is basically yes. UNLESS you can reasonably justify that many parts and defects are so similar and they are inspected under the same lighting, training and time to inspect conditions that they constitute a family of parts.

In this case the standard is perfectly clear - regardless of how an incompetent auditor may ‘interpret’ it.

 

First: the statement that visual inspection is only 80% effective is a zombie myth. There is no data or subjective research to substantiate this made up claim. It persists despite all evidence to the contrary. Please erase it from your memory. Now remember that even with sampling plans for attributes data are based on assumption that there will be some level of defects that the plan will accept and pass. This level is based on the severity of the defect and what number of defects are acceptable given the cost and severity…this is determined by you or your Customer. So for an attribute MSA you establish the same type of level. Based on the severity and cost, how many defects can you - and your Customer - tolerate that are incorrectly assessed as acceptable and how many are incorrectly assessed as rejectable. In other words what false acceptance and false rejection rate can you tolerate. This requires a bit of math but doesn’t all engineering, science, accounting, etc.?

 

Attached is a guidance I wrote for my last organization on MSA. In it is a section on how to conduct attribute (categorical) MSAs and how to calculate false acceptance and false rejection rates. You might find it helpful. It also has references for further study...

 

The effectiveness of visual inspection will vary a lot depending on a number of factors. Some highly influential factors include the type of defect (i.e., how easy it is to see), the operational definition of the defect (i.e., how easy it is to correctly classify), the relative frequency of occurrence (i.e., rare defects are easier to miss), etc. While 80% might be an overall average, the variation of individual scenarios around this 80% is very wide.

 

Helpful for what? As Miner has said the variation of visual inspection is quite wide. So wide that it cannot possibly help you ‘know’ is your system is good, bad or indifferent. It might be helpful to spur the recalcitrant on to performing an actual study, but other than that I am at a loss as to how it would help anyone.