I have got 98.16% of R2 value for the linear model generated for my experimental design of L9 orthogonal array using Taguchi method. Further I also performed Response Surface Methodology (central composite design) and got 99.69% of R2 value for the quadratic model generated with no significant interactions but only one parameter out of three has significant quadratic effect.
1) do a Lack of FIT F test.
2) calculate Adjusted R2 and Prediction R2 as well and compare the two models based on this properties, in general, a model with higher prediction R2 is more preferable in case you are interested in using your derived model for prediction of response at different experimental settings in factor space.
3) as a side analysis, analyse the Quadratic surface you have generated, retaining all the terms, and find out what kind of surface it is, through eigen-analysis of the associated Hessian matrix.
i believe, you can go with the quadratic model.
Yes, it's good to do that. The linear model shows that there is a relationship, but the response surface test shows the relationship between explanatory and response variables :)
Yes, but u canot use taguchi method directly. U can use full factorial DOE and select significant parameter their ranges and after u can use Taguchi
If u hv problem call me on 9665696022
- Badges
- Science method
- Similar topics
- Methodology