Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.
Showing results for
Search instead for
Did you mean:
Our World Statistics Day conversations have been a great reminder of how much statistics can inform our lives. Do you have an example of how statistics has made a difference in your life? Share your story with the Community!
Questions about the factor value range in Custom DOE
May 23, 2019 1:43 AM(661 views)
Hi~ I'm engineer to develope a antenna.
I want to prediect the antenna performance by manufacturing tolerance using the DOE.
Manufacturing tolerance vary by vendor. (ex. A vendor tol. : +/-0.004, B vendor tol. : +/-0.016) So I defined the range of factor values widely so that they can contain all the vendor's tolerance value. I selected the range of each vendor's tolerance using "Nomal truncated" in pofiler of Fit model and checked the response(VSWR)'s SD value. When the range of factor's value varies as in above case, is it correct to simulate with "normal truncated"? I made two designs to verify it and compared to standard deviation(SD) values of responese(VSWR). In case1, range of factor(ppg_tol) is +/-0.004 and it of case2 is +/-0.016. In profiler of fit model, SD of case1 is 0.09 and SD of case2 is 0.05.
I want to know why the two SD values are different. (I used custom design and both designs have the same condirions except ppg_tol)
And How is the best DOE design method to compare various factor's ranges?
The profiles shown in the two cases are different. It seems that the terms in the model and the parameter estimates for case 1 are not the same as for case 2. Such differences could easily account for the difference in the simulation results.
Also, case 2 appears to have fewer terms, so the RMSE is higher. You can see that the confidence interval on the profiles is different for the two cases. That difference would also change the simulation results.
I think that most of the reason for the difference is the change in the model between case 1 and case 2, though.
I still do not know much about what you are studying but let me suggest an approach for you to consider.
I assume that your response is continuous. I further assume that you have specifications for this response. So I would design your experiment using factor ranges as wide as possible. Your analysis (after confirming that the model has no bias) with the simulation should provide you with the failure probability.
The simulation depends on the model and the model depends on the data. A good model should accurately predict good and bad responses. The factor levels (ranges) should never be limited by expectation of a good response level.