A nice overview of the techniques Steve mentions are in this link:
http://www.spcforexcel.com/ezine/nov2006_2/nov_2006pf.htm
However, be careful when interpreting the results of the funnel experiment with grinder operators. They need to know that they are not going to get the distribution that you get in that experiment - their machines will generate a uniform distribution (rectangle). The moral of its story is not that you have to leave the process set at one setting, but
that you must not adjust the process until the control limits tell you it needs adjusted. If you can not make that clear after doing the funnel experiment (because of the difference between the distributions), then you may opt to drop it.
Here is the story I tell to precision machining people to get the point across:
I was sitting in a process planning meeting (
APQP) when the engineers around the table proclaimed that the grinder they had would not be capable of running a part to a specific diameter tolerance. I felt they were just shooting from the hip, and I was curious if it was true.
I went out to the grinder and spoke to the operator. He was doing SPC on his operation, plotting the outer diameters. I looked at the chart, and it was a classic normal control chart with points randomly jumping about the mean.
I asked how often he was plotting his data, he said every two hours, just like the control plan said. I asked him how often he was adjusting his process, he said every 15 minutes.
My head dropped in dismay...
I asked him to try something different. I asked him to adjust his grinder to the
lower control limit. I told him to
ignore the mean. Run the grinder, and do not adjust it until the diameter reached the
upper control limit. Then, adjust it back to the
lower control limit.
He did that. Do you know how long it took to reach the upper control limit?
A week.
So, clearly his adjusting every 15 minutes to try to keep the machine at the mean was
overadjustment. In fact,
the operator had become the process. That made the process "normal", and most operator processes are normal distributions. But the machine process was
not. It was a uniform distribution. It was masked by the unnecessary adjustments to the mean by the operator. CNC operators are notorious for overadjustment, because it is easy to push the buttons for an offset. I tell them if they want to push buttons, push the buttons on their calculator, not the machine.
Many quality professionals are fooled by seeing these supposedly 'normal' processes and their accompanying charts, and believe they really are. They use these charts to justify their claim that the process is indeed normal, in control and capable. Fools gold, my friends. It is usually garbage data.
X-bar-R charts encourage adjusting to the mean - and therefore encourage overadjustment in precision machining. That is
one reason why they are
the absolute worst chart for precision machining. For the uniform distribution,
the mean has no meaning!
If I walk up to a precision machining process and see an X-bar-R chart exhibiting random variation about the mean, my first assumption is the process is out of control!
More often than not, I am correct.
BTW, CMfgT, that is the text for "Story No. 1" in the presentation I sent to you.