I work at a manufacturing facility making wheel hubs and ring gears. Occasionaly as I'm sure all do, we have to sort product. We recently had a situation in which we got extremely lucky, we had a hub with no spline pass through all our systems until someone caught it in packaging. The reasons it passed all the (automatic) inspections is because the operators were performing a sort on the line and pulling pieces from all points during the process. I want to write a standardized sorting procedure, i.e. isolation, containment, tagging, etc. I have something basic I wrote but I was wondering if anyone has any "lessons learned", either general comments or some example of a similar procedure you have or have seen.
Send to friendSign In to get started!
Comments
qdigest 10/20/2004
Mkomarmy provided you with a good explanation for your question, I just wanted to make one other comment on item 2 for further clarification.
I agree that Shewhart statement about control limits at +/- 3 sds is correct. A common misinterpretation that people make is to use +/-3 sds of the individual data values. His intention was to use the sd calculated from within subgroup variation (Rbar/d2). Using the sd from all the individual data points will widen the spread between the control limits.
mnanni 10/19/2004
Hi !!! I have the same problem for % Yield. I don´t know what control chart do I have to use.
Last week, a friend told me: For % Yield you have to use a p chart (or np chart).
But I don´t know if he is correct. Because I used it and ALL points were out of control. This is o.k. because my "n" es too big, and that produce narrow control limits. (My "n" es huge, because my company produces small pieces, like chips, and every lot has aprox 300,000=n, and p=0.6%)
Then, I tried the individuals-Moving Range chart and all points are in-control. Oh! I was happy, but when I did the normality test, my "Y" (% Yield) is NON-Normal... then I don´t if the individuals-moving range chart is o.k. for this problem, because my data is non-normal.
Please, if someone can help us.
Thanks!!!
mkomarmy 8/12/2004
1)It is good that you are able to record data for such long time periods. With this amount of data you should be able to determine whether your process is stable or not. I am assuming that you are either sampling from each batch or measuring them 100% to obtain the parameters listed. If that is the case, you could use an individual and moving range chart, and calculate UCL and LCL for X using : Mean(X) +/- E2*R-bar, and the UCL and LCL for the moving range : D4*R-bar and D3*R-bar. The limits and means should be calculated on the first 100 data points, and the process evaluated for stability (runs, out of control points, etc.). If stable, the limits should be fixed, and ongoing production monitored to those limits to detect changes in targeting or variability. The calculation of limits should not be updated with the addition of new data, unless a known process change has taken place.
2)Shewhart is correct. If a process is in control and only common cause variation is present, it should be a rare occurrence for a sample to fall outside the limits (99.7% should fall within them, only .3% outside them). If a sample is outside the limits, there is a much higher probability that something has happenned to the process (a special cause) that needs to be corrected.
If the process has not changed, there is no need to recalculate. Assuming the limits are set properly and the process remains in control, the performance will remain the same as in previous years. If the process has been improved as evidenced by a mean shift or reduction in variation as evidenced on the control chart, the limits should be recalculated only from the point the change occurred. The population is different at that point and including the previous data will not indicate the current output of the process. The same is true of a process that has degraded (increase in common cause variation) but is still stable. New limits should be calculated using the data from the new population only.