DOE and Cp (capability analysis)
Six Sigma – iSixSigma › Forums › Old Forums › General › DOE and Cp (capability analysis)
 This topic has 10 replies, 5 voices, and was last updated 13 years, 8 months ago by Jose Ramon.

AuthorPosts

January 31, 2008 at 1:39 am #49234
Jose RamonParticipant@JoseRamon Include @JoseRamon in your post and this person will
be notified via email.Thanks a lor always for your comments.
I have this question: In the Measure step of the DMAIC I`ve trying to calculate the Ppk in order to calculate Six Sigma value. (continuous data). The data that I have is not normal so I found what distribution it. The distribution is 3 Parameter Loglogistic. When I use it Ppk and all the capabilities cannot be estimated. This is because the data is nonnormal? or because the data have been transformed ?
If I do not calculate Ppk this is not 100% a need ?
A Master Black Belt 4 month ago told me this: Waiting time follow a Poison distribution ?. I am not sure about this since I can find the right distribution for a data. Is this correct ?
DOE
If I have 100 data and I will perform an experiment 2k4 full factorial. This means = 16 datas I will use.
How can I select which one to use…between the 100. I did it at randome but is this correct ?. What do you recomend ?
Thanks so much
Jose Ramon0January 31, 2008 at 5:03 am #167999
Chris SeiderParticipant@cseider Include @cseider in your post and this person will
be notified via email.It sounds like you do not understand how to use the designed experiment tool. You do no select past conditions to fit your 2k design and plug those numbers into the array and analyze the results.
You need to set your factors and levels and then carefully execute your experiment being sure to minimize any change in other potential inputs to your process. After getting your results, then you can analyze the results. FYI, it is very possible to do a DOE with the output of interest being the standard deviation. Just be sure to run replicates or repeats but do NOT make the mistake of doing multiple measurements on the same item and call that your process variation–it would be only your measurement system variation. When you get your repeats and replicates, then calculate your standard deviation at each corner of your experimental design and place that in one column (assuming you are using Minitab) and then analyze with the output being that one column.
Good luck…
0January 31, 2008 at 6:27 am #168006Ramon,The wait time does not follow a Poisson distribution, it follows the Exponential distribution. The two are close but different.
Poisson distribution applies to situations such as the probability for finding a defect on a machine, not the wait time because the time to wait is more of an interval on a continuum.
If you use Minitab, you should be able to find the distribution that your data follow.
Why would you want to find the Ppk, if you can find the PPM?
If you run a Full Factorial design with two levels and 4 factors, you get 16 runs, I do not understand what you mean by 16 Data. If you can be a little more explicit on what you are trying to do, maybe we will be able to help you.0January 31, 2008 at 11:30 am #168015
Jose RamonParticipant@JoseRamon Include @JoseRamon in your post and this person will
be notified via email.Thanks a lot for your comment but now I am just a little bit confused.
You are saying that I set all my factors right ? and the response too ?? the response is what I what to test in the experiment right…
Let analyze this example: If I have 2 department ( A and B ). I measure the waiting time when customers come to get an item. First customers come to A and the B. If the measure are:
A (Min) B(Min)
5 4
4.5 3
6 2
8 7
This is the data that I take from the process. So the run an experiment about waiting time in A and B. What I need to do ?. what I understand now (thanks so much for correct my mistake) is that I set the waiting time that is good for me and see what the output is right ?. If I made a mistake please let me know.
And then in the case I have 100 datas of A and B…Can I design my experiment base on this data ?.
Please let me know and thanks so much for your comments
Jose Ramon0January 31, 2008 at 11:40 am #168016
Jose RamonParticipant@JoseRamon Include @JoseRamon in your post and this person will
be notified via email.Thanks Joe for your comments..Thanks so much.
I am a little bit confused with the DOE. When I said 16 this mean 16 runs but the question was How to set this data. As I sent a post to Seider . The questions are : If you have 2 department and you measure waiting time in both departments. If you have 100 datas from A and B. and in the experiment you have 4 factor to be test (Xs)….the response is waiting time…but this response how do you selected ?. what i understand now is that I do it no base on that data but base on what I want ?…..maybe I made I mistake …….
Thanks a lot for your comments about the Ppk….you`re right.
Thanks and waiting for your response.
Jose Ramon0January 31, 2008 at 11:53 am #168018
Jose RamonParticipant@JoseRamon Include @JoseRamon in your post and this person will
be notified via email.to collect experimental data for A and B waiting time must be taken from the process right ?. Thanks so much for you help. I think that after your next response I will undestand better about this.
Please could you explain just a little more about “run the experiment at each corner”. Does this mean limits of the experiment ?
0January 31, 2008 at 12:40 pm #168019Why in the world are you talking DOE in the Measure phase of a
project?Waiting time does not require DOE, try learning about queueing and
the simple modeling of a queue.0January 31, 2008 at 1:25 pm #168021
6Sigma Below The BeltParticipant@6SigmaBelowTheBelt Include @6SigmaBelowTheBelt in your post and this person will
be notified via email.For Ppk
If the data is not normal and have not been transformed, compare the AD value among different distributions and see as to what kind of distribution (e.g. Weibull) is appropriate.
If the data have been transformed, be sure to transform the spec limits as well to get the right Ppk value.
For DOE
No need for you to worry about the 100 data, if your combination has 16 data to collect then randomly experiment then whatever the results are, get 16 data.0January 31, 2008 at 2:43 pm #168024
Chris SeiderParticipant@cseider Include @cseider in your post and this person will
be notified via email.Yes Jose.
Set your two factors to the levels from your experimental design and then observe the result.
Take note of Stan’s comments….if you are in the measure phase (I did not know if the two questions were related), then don’t bother doing a DOE yet. Gather data, watch the process, and analyze even BEFORE you attempt DOE’s or else you don’t know for sure which inputs and what levels to set the inputs at.0January 31, 2008 at 3:04 pm #168026This is an answer that can come only from the “Original” Stan,the fake
one can’t introduce such an answer….0February 1, 2008 at 4:55 pm #168080
Jose RamonParticipant@JoseRamon Include @JoseRamon in your post and this person will
be notified via email.Stan thanks a lot for your answers. I have a couple of question .
1. I know that the DOE do not help you to reduce waiting time. (please let me know if I am wrong) If I have 2 departments A and B and in A (2,3) person in B (1,2) (this mean 2 in the morning and 3 in the afternoon) the output for instance Y1 = 60 sec (A.M) and Y2 = 65 Sec (afternoon) so I can run an experiment to know according to the response that I need (for instance 40sec) and see what happen with Y1 and Y2. If Y1 is > that the last one then this mean that 2 persons is not enough to accomplish what I am investigating , on the other hand If Y1 < that the last one this mean they can. My analysis is correct ?
I think Six Sigma Project are a combination of tools and depend on how we use and of course knowing the tools we can get a good result.
2. When I get the Ys equation in the DOE and this equation has a lot of interaction ( for instance), let us say this equation:
Y = 34+4*A2*B+5*C7*E – 4.87A*B + 6*A*C + 7*A*B*C – 8*A*B*D + 54*A*B*C*D +A*B*C*E + 87*A*B*C*D*E
with excellent fit and p value < 0.005. Something could be wrong ?. What this suggest ? a Good equation ?.
3. I have two Book of Six Sigma ( very Good Books) the first one use the DOE in the Improvement state steps in the 3 step of the DMAIC model. The other states that the DOE is use in the analysis phase. One more point: the first one (Title of the Book : Six Sigma for Green Belt and Champions) and the Second ( The Six Sigma Black Belt Handbook). Which one is correct ?. My point of view is that DOE can be use in both and this depend on how we perform the analysis….Am I right ?
Thanks a lot for you comments
Jose Ramon0 
AuthorPosts
The forum ‘General’ is closed to new topics and replies.