DOE with repeated responses

[tgeorge]

Most DOEs I am familiar with involve only looking at 1 response for each test run. Is there any way to incorporate some sort of repeatability test inside each test run's y response?

 

[PSE]

Are your test items stable? If so, you could use a follow up "classical cause and effect" test once your DOE identifies your major factors. If your test items or evaluation equipment is not very stable, you might not get any useful results from any runs.

Cost of running the tests should be another consideration. If inexpensive, you could run multiple samples using the same factor combinations and see how the results compare. If it is expensive, I would recommend using a classical test to verify the DOE results.

Regards

 

[tgeorge]

Thanks for the tip. They are pretty stable, in my opinion, although, the more I think about it, the more it seems that doing it like you said is the way to go. Identify the major factors then look for repeatablity. Ironically, the major cost for me, in this case, is any initial setup costs.

 

[stel8]

One way of analyzing the repetitions for each run (if you can afford them) is to calculate signal/noise ratios. S/N is combining the mean and std.dev. in one statistical measure (see Taguchi DOE). best rgds

 

[tgeorge]

After some investigation into S/N ratios, I have this question:

Why is are they based on logarithmic functions?

I can understand wanting to see which drowns out which (signal vs noise). But I fail to see why that sort of function is used.