Machine Capability study in high precision manufacturing
- Thread starter finchcs88
- Start date
"Maybe it's a dumb question but I'd just have to increase my sample size then, right?"
Do you know the root cause of the outlier? Sounds like a special cause. That may help you zero in on a reasonable sampling plan.
Other than that, all you can do is increase sampling.
Do you know the root cause of the outlier? Sounds like a special cause. That may help you zero in on a reasonable sampling plan.
Other than that, all you can do is increase sampling.
So as of right now, the operators are not aware of why they are switching the boring bar every hundred parts, and the QE that I am filling in for also quit without much warning so I have nothing rolling over from his work. So far I have not found out the special cause that might be resulting in the outliers, but we definitely do have difficulties with machine warmup. We're running a CNC lathe with automated loading, and the runs tend to be about 400-500 pieces from what I've seen. The operators are checking every single spherical pocket with a go/no-go gauge and that is how they find the outliers. If it's undersized, they rework it, and because they'd rather rework than scrap the pieces they run the parts on the smaller side. The night shift will also clear their scrap pieces before I even see them. Other than that, I'm the one going in and measuring the spherical pockets using a CMM (resulting in the hours of measuring even for small sample sizes). I've done samplings of 5 every 50 and several runs of 30 consecutive pieces because I was told that 30 is statistically significant. (which I now know isn't necessary in precision machining).
I just read your book @bobdoering and I think I have an idea of what I should be looking for. However, I'm a little confused about the roundness section on page 23(?). Do I need to find a way to plot the roundness on top of the max and min graph? How does the roundness work into finding the slope of the tool wear?
I just read your book @bobdoering and I think I have an idea of what I should be looking for. However, I'm a little confused about the roundness section on page 23(?). Do I need to find a way to plot the roundness on top of the max and min graph? How does the roundness work into finding the slope of the tool wear?
Last edited:
John Predmore
Trusted Information Resource
Maybe you will find something useful in my story.
I recall one close-tolerance machining problem with a reject rate consistently around 200 ppm, which appeared to be random. What sample size would you choose to investigate a 200 ppm problem? How big a DOE study would you lay out, when the occurrence is 200 ppm?
The engineers did not make much progress until someone (me) arrived at 4:30 am to hand-collect the first 20 pieces off the machine. I knew to look for Red X variation where there are discontinuities in the process. It wasn't until I measured parts for my multivari study that we proved ALL out-of-tolerance parts were made in the first 5 minutes of a 16-hour run. This was not a case of tool wear, but expansion of the cutting tool before the cutting oil reached equilibrium temperature.
The cause was hidden from most observers because after machining, the first 20 parts were tumbled with hundreds of following parts in order to clean them, and then a few hundred parts were feeder-bowl fed to the assembly process, in guess what? random order.
The problem was easy to solve after we gained this insight. We looked at oil heaters to maintain 140°C, but it was cheaper (at a penny a piece) to throw away the first 20-30 pieces of every day's run. In your shop, you will have to come up with your own viable solution.
This is when I learned, randomness is more a statement of our ignorance rather than any useful information about the nature of the problem.
If your operators are reworking or scrapping the parts before you sample them your Cpk results are worthless. Also if you really want to understand the process you will have to let it run and not replace the boring bar until you have determined when it wears out to the point of making out of spec parts.
Similar threads
