DOE with more than one response

Sarang-
We use a software package called DOE Pro which allows analysis of multiple responses putting the affects of the factors on the mean and standard deviation side by side so that you can compare the individual responses.
The analysis of DOE can be overwelming for those who do not have a lot of experience doing them. Most of the work of a DOE though is done before running the actual parts. Without learning more about what your doing, my only suggestion is this:
If your not using something that gives you that multiple response capability, one place to start would be to pareto the effects of each response, then sum those ranks to see what factors affect the responses the most. I’m sure there are other ways though.
If you are using DOE Pro, let me know and I’ll try to walk you through further analysis.

Doug,  that’s the way it’s done.  You build independent models for each Y.  After you have the models you assemble them in some kind of spreadsheet and use their simultaneous predictions to identify optimum operating regions.  Some computer packages have efficient spreadsheets that allow you to set up optimum criteria – minimum acceptable values for the Y’s, bounds on the X’s etc. These packages typically have some kind of search algorithm which will then search all of the predicted combinations to identify those that are closest to what you want.
  If your package doesn’t offer this then you can assemble all of the equations in an Excel spreadsheet and run them side by side with the columns containing your X responses.  The only problem is that you will have to do all of the searching manually and if you have a lot of X’s or a lot of Y’s this can be very time consuming.

There is another way to analyze multiple Y’s in a single DOE analysis.  In a Taguchi design, they call it the Overall Evaluation Criterion (OEC).  All that needs to be done is the assign a weighting factor to each Output (Y) that you are measuring for the DOE.  By assigning a weighting value to each Y, you can now calculate a single value to run through your DOE analysis.
The power of this approach is that you can take into account all the important Outputs in a DOE in 1 step.  This ensures that you really are selecting the design that has the best results with respect to all of your Outputs (Y’s) not just 1 Output at a time.
Using OEC is redundant, of course, if you have already shown that the “Best Design” is the same for each individual Output (Y) when you perform your DOE Analysis on each Output 1 at a time.

Sarang,
1- Analysis corrlation betwin Ys.If Ys are independent you can analysis responses independently.
2-refer to book “analysis and design of certain quantitative multiresponse experiments”by R.Gnanadesikan etal 1971.

I and my Professor invented a new Tool to solve the multi response optimization problem named as CDFT approach. This methodology avoids the uncertainity in the previous methods with lesser number of trials

KT,
Most effective way would be to compare the surface plots from both the responses after equalizing the scales and looking for areas that are favorable to both the responses. This would give you an eyeball fit idea of how do these responses interact at given values of input variables.
The second thing ofcourse would be take out the prediction equations on an Excel sheets and look at the values generated when the inputs are altered. You could then draw a line graph of the values generated and look at where the two responses could be optimized.
Hope this helps.
Regards,
Prateek

(1) If  responses  are more than one, in Response surface Methodology,  Desirability Function can be used. It converts multi variate problem in to single variate problem by giving common scale of values to each response. Lot of literatures has come to prove this approach. But the only disadvantage is it requires a full factorial experimental trials.
(2) The other method is Taguchi’s Multi response Optimization. While solving any problem according to this method, engineers has to take their own decission in selecting the optimal levels of the parameters. So, this method often bring some degree of uncertainity.
To overcome above drawbacks researchers are finding new methods. CDFT approach is a new method to solve the multi response optimization problems.  It is the Combination of Desirability function and Taguchi method. If the problem is solved according to this method avoids uncertainity and the results are produced in less number of trails.

If you are an expert…Don’t keep overposting your messages. Literate people on DoE can anyway go through single posting and move across related threads.
Another question, can you enlighten with practical examples where you have used this, not just theory. I have good books on theory.
Cheers!
 

Intersting to hear about new methods for multiresponse optimization. Who knows where information about the CDFT appropach can be found?
Thanks
Ermanno Oberrauch
 

sarang:
what kind of you Doe which you did? full factorial or fractional factorial or others? First you should collet the 2 responses in your experiment, then you can analyze them with Minitab’s Doe function,(Optimization..)