Factorial design 1

[Karikoga]

I hope I will explain my problem clearly. I want to do pilot plant tests to first screen the significant factors out of five factors, then carry out a response surface DOE to optimise the plant settings based on a few of the significant factors.

My problem is that it is not possible to set any of the factors to their exact values according to the matrix, e.g. pressure 100.0 bar, temperature 80.0 oC. There will be a slight variation in all of these values. For some of the factors the variation can not simply be ignored, not too much, but it is significant. Is there a way of taking this into account?

I hope someone will be able to assist.

Thanks

 

[GregLocock]

There is a slightly complex theoretical answer, in which you examine noise in your experiment, or there is a rather easier if less efficient answer, which is to measure the sensitivity of your responses to small changes in each factor and then use those to 'correct' your main DOE results.

Personally I'd be inclined to make sure that your levels are set say at least 5 times the natural variability in the factor apart, and use the nominal level in your analysis.



Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[Karikoga]

Thanks Greg. This really helps. Thanks again.

 

[Karikoga]

Thanks Greg. This really helps. Thanks again.