Control Charts; Confused
Six Sigma – iSixSigma › Forums › Old Forums › General › Control Charts; Confused
- This topic has 18 replies, 8 voices, and was last updated 13 years, 5 months ago by
Vallee.
-
AuthorPosts
-
December 17, 2008 at 5:14 pm #51527
Kevin DuCasseParticipant@Kevin-DuCasseInclude @Kevin-DuCasse in your post and this person will
be notified via email.I have the responsibility of setting up a metric for final electrical test and I was trying to use an attribute chart to track the defects per unit. The parts that we are testing can fail for one or more tests througout the test process and we would like to capture each event.
Which would be best to use, the C, U, P, or Np? Would DPMO be a better choice?
0December 17, 2008 at 5:57 pm #178814
Ken FeldmanParticipant@DarthInclude @Darth in your post and this person will
be notified via email.U chart is chart of choice if the number of parts being tested varies from time period to time period. Otherwise a C chart is fine.
0December 17, 2008 at 6:04 pm #178815
Gary ConeParticipant@garyaconeInclude @garyacone in your post and this person will
be notified via email.For anyone to give you a definitive answer to this, we would need to know a lot more about your process and product.
A couple of things to think about –
1) Are you high volume, low mix or the opposite?
2) Is your defect distribution predictible? For example, if you have a DPU of 1, you should expect 37% of your units to have no defects, 37% to have one defect, 18% to have two defects, and so on.
If 1 and 2 are true, go with an np chart. Better yet use Lean concepts to limit space in the repair area where if your limited space is ever full, shut down the line because you are having an unexpected number of defects.0December 17, 2008 at 6:09 pm #178818
Ken FeldmanParticipant@DarthInclude @Darth in your post and this person will
be notified via email.Hey, stop making it complicated. You are starting to sound like that idiot Stan. A C or U chart is fine.
0December 17, 2008 at 6:12 pm #178819
Gary ConeParticipant@garyaconeInclude @garyacone in your post and this person will
be notified via email.what would you be trying to detect with a u or c chart?
0December 17, 2008 at 6:17 pm #178820
Ken FeldmanParticipant@DarthInclude @Darth in your post and this person will
be notified via email.According to the poster: “track the defects per unit”That’s sort of the definition of the C or U chart. Of course this assumes that the defect occurrence is not frequent and that he is seeking to monitor change over time. Also assumes that he is testing some number of units per time period and wishes to see if the process is getting better or worse or staying the same.
0December 17, 2008 at 6:35 pm #178822
Kevin DuCasseParticipant@Kevin-DuCasseInclude @Kevin-DuCasse in your post and this person will
be notified via email.The process is low volume, low mix. We have four different products that we produce from the same type of process.
The defect distribution is not predictable. I was leaning heavily on the use of the c chart, but I believe it is to be used if the defects are rare. So I turned my attention to the u chart.
0December 17, 2008 at 6:56 pm #178824
Gary ConeParticipant@garyaconeInclude @garyacone in your post and this person will
be notified via email.You need to ask why you want a control chart. Conrol charts don’t track data, data collection systems do that.
If you know your defect distribution is not predictable, whay aren’t you working on that instead? When defect distribution is not predictable, solutions are usually pretty easy and straight forward.0December 17, 2008 at 7:04 pm #178825
Kevin DuCasseParticipant@Kevin-DuCasseInclude @Kevin-DuCasse in your post and this person will
be notified via email.Thank you for your input. I had concluded the same. I do not need a control chart at all. Please explain your last statement that “when defect distribution is not predictable, solutions are usually pretty easy and straight forward.”
0December 17, 2008 at 7:13 pm #178826
Gary ConeParticipant@garyaconeInclude @garyacone in your post and this person will
be notified via email.As explained in ever SPC book in existance, c and u charts have extremely limited usefulness. In electronics, the number of defects found in any given unit is going to very small as compared to say, defects per 100′ of a seemless tube, or defects per 1000 sq ft of float glass (good applications of c and u).
Why would you need a chart when you can tell the folks that the odds of having x or more defects in any given unit is very small and when you have one, set it aside for detailed analysis. You don’t need a chart, you need a simple rule.
The other side of this is he already knows his defect distribution is unpredictable, and any reasonable data collection system will have enough information in it to go solve that problem. In electronics, this unpredictability is often caused by the repair techs skill or a flawed fault tree or a MSA problem.
c ad u assumes that the defects follow a poisson distribution, he already knows that is not true, so he should make it true before worring about a chart.0December 17, 2008 at 7:15 pm #178827
Gary ConeParticipant@garyaconeInclude @garyacone in your post and this person will
be notified via email.Go get data on DPU and let me know what it is for any or all of the 4 products and I’ll walk you through it,
This should have been in your Six Sigma training. I am sure it is in Darth’s.0December 17, 2008 at 8:24 pm #178832DPMO is rubbish. Forget it.
0December 17, 2008 at 8:44 pm #178833hahahaha, you make funny!
0December 18, 2008 at 1:51 pm #178858I’m surprised no one has suggested that you simply gather the failure information over a defined period of time and prepare a pareto analysis of the data. Using a pareto chart has worked best for me when several different failure modes are present in one system test.
0December 18, 2008 at 2:08 pm #178860Making some big assumptions here Gary.
0December 18, 2008 at 2:10 pm #178861Trudy,
While a pareto chart can point you in a direction there is no statistical value behind the chart.
Many false trails can be obtained by using an incorrect chart.
Very obvious trends can be observed and in some cases that may be enough but usually you want to know more than what a pareto chart will offer.0December 18, 2008 at 2:11 pm #178862
Gary ConeParticipant@garyaconeInclude @garyacone in your post and this person will
be notified via email.This is kind of like telling someone to make a pareto. No big
assumption.0December 18, 2008 at 3:29 pm #178864Ron,
I agree. Pareto analysis may help take the confusion out of what type of chart to use by allowing the focus to be on a single failure mode. I’ve never applied a control chart for systems using multiple failure modes.
Trudy0December 18, 2008 at 4:30 pm #178866
ValleeParticipant@HF-Chris-ValleeInclude @HF-Chris-Vallee in your post and this person will
be notified via email.Kevin,I would like to take a step back and ask do you know what is causing the defects in the first place? The defects you are tracking appear to be lagging indicators and I wonder if you are tracking the leading indicators as well? Are you wanting to graph the defects at this point to establish a baseline and/or to see if it is getting worse or better? or to see where to go find out what is causing the defects? I know this sounds basic but if you look at the discussions above the argument started because people have different opinions for chart uses based on what you are going to do with the information. Develop your metrics on what your process is doing and on what you intend to do with the numbers you collect. Then have the above discussion.
0 -
AuthorPosts
The forum ‘General’ is closed to new topics and replies.