I would really appreciate help regarding a project I've been given.
First off, I'm an undergrad engineering student with no experience or training in statistics.
I've been given a task to improve the accuracy of a rubber wheel abrasive wear test machine (example: http://www.kondex.com/weartestmachine.html). The test results seem to vary randomly even when the material tested stays the same.
It was decided that using robust parameter design to optimize the machine would be a good idea. My boss suggested using Taguchi methods and gave me Minitab 16 to use. I've identified 8 parameters to test, 3 controls and 5 noise.
However, there seems to be a lot of controversy about Taguchi methods in the literature. I've found many sources saying they shouldn't be used, so I'm not sure anymore. The alternatives offered are presented very technically, so I don't know how I could do them with just Minitab.
First, response surface methods. These are suggested often, but I don't understand how they are used in minimizing output variance, they seem to be just for optimizing the output mean.
This study concludes that a Taguchi control/noise factor array design should be used, but not analyzed using signal-to-noise ratio, but something called response model analysis. I have no idea how that could be done using Minitab.
Then there's randomized block designs, split-plot designs and the modern Definitive Screening Designs, which seem pretty neat, but I wouldn't know how to apply these.
Also, would it be better to first do some kind of a screening experiment and then follow-up with something more resource-heavy?
If you have any insights, I'd be much obliged.
