Doubt about the correct control chart

[Miner]

Now, my next question is... When do I need to react with a corrective action? At every single point out of the 3-sigma limit?

With a frequency of once per month that is the only one that makes sense. It would take 6 months to recognize a trend and 8-9 months to recognize a run. Given the importance of reflow temperature, I would at least consider and discuss the possibility of weekly checks.

 

[Bev D]

I agree that monthly is really long…however I wouldn’t rule out any rules. Including the 2 out of 3 rule. Remember that if there is a small shift yes it will take 8 months to ‘detect’ 7 points above or below the average but a single point won’t have occurred in that time period.

I suspect that the data points won’t show much variation from month to month so it is likely that the most possible OOC occurence will be a single point out but ya never know. It doesn’t cost any more to use all of the rules than to use only one or two…

 

[Steve Prevette]

The short answer is yes. Usually people use a suite of rules - I use Acheson Duncan's which is
Points outside the UCL / LCL
Seven in a row the same side of the average
Seven in a row all increasing or decreasing
Two of three at two standard deviations from the center line
Four of Five at one standard deviations from the center line.

Now by reaction - we mean - you have received an unusual result, you should investigate what was different about the operation at that time. You will be looking for a "special cause" for the change. Ask "What Happened?"

Now, you may be stable with no such signals. But if you are stable WHERE YOU DO NOT WANT TO BE you will need to change the system / process to achieve the capability you need. No amount of examination of recent results will help you - you need to look at the long run performance in this case.

 

[buralbe]

Very good discussion and great comments!

I agree that we should try to run the oven profile more often. I will make a proposal internally.
We will use an I-MR control chart.

 

[DNARYA]

I agree with Beth above - great post. I will say my personal preference is I-mR, which Dr. Don Wheeler is a big proponent of. But since you do have three readings of the same parameter - it is a great opportunity to see if they are consistent between the three readings and detect if one is drifting.

yes I MR CHART IS BEST AS OVEN TEMP HAVE JUST READING IN ONE TAKE , SAMPLE IS ONE