DOEattribute
Six Sigma – iSixSigma › Forums › Old Forums › General › DOEattribute
 This topic has 12 replies, 10 voices, and was last updated 11 years, 11 months ago by Bower Chiel.

AuthorPosts

November 18, 2009 at 4:29 pm #52944
TinocoParticipant@anthony Include @anthony in your post and this person will
be notified via email.I want to find out how to analyze a DOE when you have attribute (pass/fail) response.
For example, Im doing a DOE on laser marking the parts. First phase was a screening just to get an acceptable laser mark. Some pieces burned so they were fail, some pieces didnt mark at all and those were also fail. So I have a table that has all these passes and fails but it doesnt tell me how to adjust my parameters to get where I want to go.0November 18, 2009 at 6:44 pm #186917
…thanapostParticipant@...thanapost Include @...thanapost in your post and this person will
be notified via email.So you want to build a reliable optimization model using experimental data which consists of a binary Dependent Variable with some combo of factors/covariates for your Independent Variables?
I am no expert, but in your shoes I would consider going back to ‘design’ and finding a way to use a continuous response…you’ll get much greater precision in your model and the analysis will be straightforward…..otherwise, all I can think offer is a binary logistical regression approach…good luck…and verfiy.0November 18, 2009 at 7:23 pm #186921
Robert ButlerParticipant@rbutler Include @rbutler in your post and this person will
be notified via email.It sounds to me like you have a continuous variable that you are forcing into an attribute response. You said, “Some pieces burned so they were fail, some pieces didnt mark at all and those were also fail In other words things are going from burning to not burning and everything in between. Somewhere in there is the correct amount for marking – how is this correct amount measured – depth of mark, measure of char, or????
0November 18, 2009 at 7:40 pm #186922Compute the percentage pass (or fail) for each cell, combination of 1 and +1 factors, in your design. Use that result as the dependent variable in the analysis. The factor level combinations are your set of independent variables in the matrix.
Several points, some of which were covered with the previous responses. If you have too few units in each treatment combination, you may not see any discrimination in the results, or if the results are too close to zero or 100%, you may well need a transformation applied in order to analyze the data appropriately.
If the type of failure is also a concern, no mark versus burned, than you may need multiple response variables to address that concern, also. A Pareto of the failure types by treatment combinations might also be of use before going too much further.0December 22, 2009 at 8:15 am #187687Where R U going ?
For DOE you need atleast two parameters for which the result is measured. In your case result is Pass/Fail.
But where are parameters ?
Temperature, Intensity of Laser, Time of Exposure etc.
0December 22, 2009 at 9:13 am #187688Hai Anthony,
Ask your experts to give a score on the quality of the mark on a scale of 1 to 10: a “5” is considered perfect marking; a “1” is for ‘burned’ and a “10” is for ‘no mark’. In this way you have a ranking of the laser marking from ‘too much’ (1) through ‘good’ (5) until ‘too little’ (10).Use this score as the Y of your DoE analysis and optimize to a Target of “5” (you don’t want the maximum or minimum Y). Don’t forget to check the residuals.
For finding the perfect optimal setting this method probably is too crude but for a screening it will work all right.Maybe you even can finetune with decimal points if all products are between “4” and “6”.
Good luck, Remi0December 22, 2009 at 11:26 am #187690Hi
Your “Y” is following a binomial distribution.
First, for each factorlevel combination find our the no. of success and failures. Then caluclate “p” which is success/failure. Then take squareroot of p. For this squareroot of p, take the trignometric function “Sine inverse”. Now use this Sine inverse root p value as your “Y” and carry out DOE (more like a oneway ANOVA).
Do let me know your results.
All the best!!!!0December 22, 2009 at 11:26 am #187691Hi
Your “Y” is following a binomial distribution.
First, for each factorlevel combination find our the no. of success and failures. Then caluclate “p” which is success/failure. Then take squareroot of p. For this squareroot of p, take the trignometric function “Sine inverse”. Now use this Sine inverse root p value as your “Y” and carry out DOE (more like a oneway ANOVA).
Do let me know your results.
All the best!!!!0December 22, 2009 at 11:37 am #187692Bizarre!
0December 22, 2009 at 6:09 pm #187707
roartyParticipant@cognition Include @cognition in your post and this person will
be notified via email.Hi Thejasvi,
Can you explain why you recommend this series of steps.0December 23, 2009 at 1:50 am #187721Hi
These steps are carried out to meet the properties of normal distribution (remember, we are transforming a binomial distribution into a normal distribution)0December 23, 2009 at 3:27 am #187722He who is good with the hammer , thinks everything is a nail !
0December 23, 2009 at 12:46 pm #187733
Bower ChielParticipant@BowerChiel Include @BowerChiel in your post and this person will
be notified via email.Hi
A few comments: –
1. Information on the arcsine transformation may be found at http://demonstrations.wolfram.com/TheArcsineTransformationOfABinomialRandomVariable/
2. An example of the use of the transformation in a DoE scenario may be found at http://www.asq.org/data/subscriptions/qp/2004/0304/qp0304million.html
3. The book by Ellis R Ott and others entitled Process Quality Control: Troubleshooting and Interpretation of Data has a chapter entitled Troubleshooting with attribute data which has lots of sound advice on factorial experimentation where the responses are based on counts.
Of course, prior to any formal analysis or transformation of the data, one can often gain valuable insights by plotting the data from a factorial experiment in main effects and interaction plots.
Best Wishes
Bower Chiel
0 
AuthorPosts
The forum ‘General’ is closed to new topics and replies.