Question on determining defective units - I am not recording fixture to part rejected

[rmundroff]

I have a process that has 100's of fixture
for a lot I use some fixtures say 10-25 fixtures
for a lot I know # good pcs from lot, number of bad pcs, (lot size) as well as fixtures used HOWEVER I am not recording fixture to part rejected.

trying to find method to track this down for old data [Yes - I know I need to fix my data collection going foreword]

but if I have a lot of 10 and 2 were bad and I weight all 10 fixtures with a .20 POTENTIAL defect
then for a second lot of 15 with 6 defective I take those fixtures and weight them with a .40 potential defective
then after lots of lots 300+ I sum all the potential defective by a fixture and divide by number times fixture used
will this give me a fair good indication of the problem and not problem fixtures
If the distribution of the final proportion all the fixtures 100 fixture & 100 proportions (average potential defective) is normally distributed

this seems to be a good method, I ran a model and it was significant
so I think this would work

thought , comments, etc.
or if there is a standard or statistical test for this method I would appreciate the name so I can look it up

Thanks in advance,

 

[Bev D]

In theory, practice is just like theory, but in practice, practice is nothing like theory.

So in theory if the fixture has nothing to do with the number of defective units, then the number of defects will be a simple percentage of the lot size. If you were to plot percent defective per lot on a SPC chart you would see where the defect rate is statistically different. This might be because a few fixtures are ‘bad’ OR it might be due to some other causal mechanism acting up. You’ll never know.

Yes you should record the fixture(s) used per part I suppose. But why don’t you just directly test the parts on different fixtures? If the failure repeats on other fixtures, it’s not the fixture. If the failure only occurs on a specific fixture and not on others then it’s the fixture. You dont’ need fancy math...

 

[GRP]

But why don’t you just directly test the parts on different fixtures? If the failure repeats on other fixtures, it’s not the fixture. If the failure only occurs on a specific fixture and not on others then it’s the fixture.

Following up with the theory-practice dichotomy, I bet the part wont fail on one fixture but on some of them. And then you test a different part and the profile of failure among the fixtures is different. And then you try to understand why on some but not on others, and then you start measuring...

...and attribute gauging is quick and nice, in theory. Determining if the attribute gauging is doing a proper job, that´s hard work!

 

[Bev D]

Yeah - if the failure is intermittent or the defect is marginal then you get inconsistent results but the right diagnostic study can answer that question fairly easily. (This is where simple probability math can actually help) Not sure that the fixture is the measuring device, an assembly aid or an inspection holding fixture. to give better advice the OP would have to describe the defect, the role of the fixture and the measurement system itself.

 

[rmundroff]

Fixture is an assembly aid believed to cause the defects if worn

we have the % defects for 4 years
we are just trying to lok a select fixtures (the ones most likely to cause the defects of the 100's we have to prove that the Fixture DOES cause the defects if worn

 

[Bev D]

Looking at the data won’t really help you at this point. Looking at the fixtures and the parts will. Physics and geometry are your friend. So...get up from your desk and go out on the manufacturing floor.
First does the wear on the fixtures match the defect? Does it make physical sense that the wear can could cause the defect?
Next select a few really good/new fixtures and a few oldest/most worn fixtures (3 each will do it). (If you want to geek out on math look up Tukey-Duckworth). Then make some parts using the fixtures and test/inspect for the defect. You should use different operators for each fixture, but you could use randomly selected material/components from the Same lot or batch IF your trended defect rates dont’ Line up with material/component batch/lot changes...How many parts do you need to make? Well, that depends on the defect rate...you might start with the best defect rate you’d seen in the last year. If the fixtures do cause the defect, then you probably won’t need to make too many. If you selected the fixtures correctly, then teh defect rate for the 3 ‘bad’ fixtures should be substantially worse than the defect rates for the 3 ‘good’ fixtures (which should be near or at zero if the fixture is the only causal factor). If the defect rates are similar then it’s not the fixtures...
Of course it’s possible that there is a material/component and/or operator interaction. If you don't get clear results you can post your results here...and we can help you.