exclude factors after DOE
Six Sigma – iSixSigma › Forums › Old Forums › General › exclude factors after DOE
 This topic has 7 replies, 6 voices, and was last updated 12 years ago by Robert Butler.

AuthorPosts

November 25, 2009 at 12:00 pm #52971
Hi,
I set up a full factorial DOE for some welding tests: 3 factors, 2 levels and 3 replicates, resulting in 24 runs. I had to use destructive tensile testing to get the results, but with the high level of one of the factors the weld did not break. The tensile test therefore returned some high load values, but they don’t really represent an actual breaking load of the weld.
Since using this factor (high level) the welds did not break, I am convinced this is a significant factor. When analyzing the DOE using Minitab, this factor is the only one with a pvalue below 0.05. Another one has a pvalue of 0.052.
Since the returned values for this factor aren’t accurate (no break), can I use these results in the DOE and conclude that’s the only siginificant factor? Or do I have to exclude this factor and recalculate the DOE using only 2 factors to check if the factor with pvalue 0.052 becomes siginificant?
Your input would be greatly appreciated. Thanks!
K0November 25, 2009 at 2:15 pm #187046Hai K,
Remark: since you didn’t include centerpoints you can not check if the linear model is correct (maybe a nonlinear model would be more correct).
My assumption is that for each run you performed a break test and measured the load value at the point of breaking.
Due to physics we can (=I) assume that for the nobreaks the load value of the moment of breaking would have been higher than the measured value of the DoE. So the Y value in the DoE is for these points an under estimate. As a consequence: a significant X would stay significant if we would do the correct DoE. You can simulate this by replacing the nobreaks with a load value Y that is 10% higher than the one measured. This simulates that you just missed the breaking point. You can then see what happens with the DoE results. WARNING: when you do this you are inventing results. So only do this to get a feeling of the behaviour of the X’s on the Y and not to get real results.
There is also the possibility that an X that you found to be insignificant happens to be significant in reality but due to the problems during the DoE you can not see this anymore.
Remi0November 25, 2009 at 4:23 pm #187056
AllattarParticipant@Allattar Include @Allattar in your post and this person will
be notified via email.I think perhaps this data may better fit a reliability study.
The regression with life data may be of more use. As we do not know the breaking point of some figures create a new column with values Failed and Censored. Failed to indicate breaks, censored to indicate a non break. Use the data as right censored, and in the censoring options tell it to use the column identifying the censoring.
Put the distribution to use as smallest extreme value, as I would expect break tests to follow this shape, rather than normal.
(I am not talking about the shape of the results when suggesting set the distribution to smallest extreme value. Before someone gets smart.)
You will have to set up the dialogue box in the same way as the general linear model in the model section. You must put categorical factors in the Factors box, as well as in the model selection.
That is just my first consideration about these results, becuase of unbroken parts, and it being a break test.0November 26, 2009 at 5:59 am #187082Thanks for your input Remi (same to Allattar)!
Remi, you said “There is also the possibility that an X that you found to be insignificant happens to be significant in reality but due to the problems during the DoE you can not see this anymore.” This is what I feared.
Can’t I create a new full factorial DOE, but this time with 2 factors, again 2 levels and 3 replicates, resulting in 12 runs. So without the factor I had the trouble with? Than I can just fill in the values I already measured and analyze it with Minitab. Would this give me correct results about the first two factors and tell me wheter they’re significant? As I already know the third one is significant.
Thanks!
K0November 26, 2009 at 6:23 am #187083
Bower ChielParticipant@BowerChiel Include @BowerChiel in your post and this person will
be notified via email.Hi K
Would it be possible for you to post what sounds like an interesting data set? You could call the factors A,B and C and simply indicate low or high and the response y for each test. You could even code the responses by subtracting the same fixed number from each result and dividing the answers by another fixed number.
Best Wishes
Bower Chiel0November 26, 2009 at 8:43 am #187084Hai K,
yes and No. You could do it but the 3d factor may influence the result as a lurking factor.
Example.
Suppose your ‘real (unknown) model’ is: Y= X1 + X2 + 20*X3 + 100*X1*X3If you leave out the significant X3 two things happen (both bad for conclusions):Your SSE (error=noise) gets large due to 20*X3 so your Rsq will drop to a low value
Your X1 influence may look large due to the X1X3 interaction so the X1 will have a low pvalue as a result
I recommend that you do the Analysis on 2X (costs only PC time) and check how the Residues look and how large the Rsq has become. If everything is OK the pvalue of X1 can be used otherwise it is only a weak indication (and you already have that now).
Remi0November 26, 2009 at 10:10 am #187086K,I agree with your approach of running the experiment again with only 2 factors. The very fact that the weld did not break at High of one of the factors (in all the replications I suppose) is good enough to make this factor a significant one. Once we have the final significant factors, the optimisation can be carried out. – Sharry
0November 26, 2009 at 2:28 pm #187089
Robert ButlerParticipant@rbutler Include @rbutler in your post and this person will
be notified via email.I too would like to see the deidentified data set as Bower proposed, however, your thought concerning the variables, particularly the notion of “recalculate the DOE using only 2 factors to check if the factor with pvalue .052 becomes significant” indicates you don’t understand what the design is doing for you.
Since you said the design was 3 factors at two levels this would suggest you ran a basic 2**3 design and replicated it 3 times. The point of a design is that the variables in the design are varied in such a way that their effects on the outcome are independent of one another. Consequently, there’s no point in “excluding the factor” the design already did that for you and what you have is what you will get whether you include that factor or not.
What would be worth doing is a simple graphical plot of the data where the X axis corresponds to the experiments ranked in Yates order and the Y axis represents the response. Plot the results for the three replicates using different plot symbols for each of the three and visually examine this graph to see if there is some kind of offset between one of the replicates and the other two or to see if there is some kind of trending in the responses as you move from the first to the second to the third replicate.
If you should see one offset from the other two it would be worth rerunning the analysis with the two that are in “visual” agreeement and if there is some kind of trend as you move from the first to the third replicate then it would be worth adding the variable of time to your analysis and again rerunning the analysis to see what you get.
As for the issue of failure to break the simplest is that offered by Remi. The problem as he noted is that you are pretending that the nonbreak value is in fact a measure at break. As also noted you can run this as a failure analysis but unless you have software that will permit weibull analysis with time and independent factors this will be difficult0 
AuthorPosts
The forum ‘General’ is closed to new topics and replies.