This document discusses optimal designs as an alternative to standard response surface designs when: (1) The experimental region is irregular shaped, (2) The model is nonstandard with categorical or different level factors, or (3) There are unusual sample size requirements to reduce runs. It provides examples and discusses D-optimal designs, which minimize variance of regression coefficients. Software like R, Minitab, and Design Expert can generate D-optimal designs and are discussed in the context of irregular regions, reducing runs, and handling different factor levels and model types.
This document discusses optimal designs as an alternative to standard response surface designs when: (1) The experimental region is irregular shaped, (2) The model is nonstandard with categorical or different level factors, or (3) There are unusual sample size requirements to reduce runs. It provides examples and discusses D-optimal designs, which minimize variance of regression coefficients. Software like R, Minitab, and Design Expert can generate D-optimal designs and are discussed in the context of irregular regions, reducing runs, and handling different factor levels and model types.
This document discusses optimal designs as an alternative to standard response surface designs when: (1) The experimental region is irregular shaped, (2) The model is nonstandard with categorical or different level factors, or (3) There are unusual sample size requirements to reduce runs. It provides examples and discusses D-optimal designs, which minimize variance of regression coefficients. Software like R, Minitab, and Design Expert can generate D-optimal designs and are discussed in the context of irregular regions, reducing runs, and handling different factor levels and model types.
This document discusses optimal designs as an alternative to standard response surface designs when: (1) The experimental region is irregular shaped, (2) The model is nonstandard with categorical or different level factors, or (3) There are unusual sample size requirements to reduce runs. It provides examples and discusses D-optimal designs, which minimize variance of regression coefficients. Software like R, Minitab, and Design Expert can generate D-optimal designs and are discussed in the context of irregular regions, reducing runs, and handling different factor levels and model types.
Optimal designs are an alternative when a standard
response surface design cannot be applied : i. An irregular experimental region ii. A nonstandard model iii. Unusual sample size requirements
i) An irregular experimental region
The region of interest is not a cube or a sphere Example : Factor1 = adhesive applied Factor 2=cure temperature Problems: If low adhesive, low cure temperature (The part will not bond) If high adhesive, high cure temperature (damage occur)
ii) A nonstandard model
Specific model Categorical variables Factor 1 = Yes, No Factor 2 = Low, Medium, High
Different level factor
iii) Unusual sample size requirements
Reduce the number of runs required by a
standard response surface design Problems: i) Time-consuming ii) Too expensive
What is D-optimal ? D-optimal design is one of widely used optimality criteria. D-optimal designs minimize the overall variance of the estimated regression coefficients by : i.
maximizing the determinant of XX
ii.
or minimizing the determinant of inversed XX
A measure of the relative efficiency of design 1 to design 2
by
CCD compared with D-optimal design
CCD is only 47.6 percent efficient as the D-optimal
design
CCD is almost as efficient as the D-optimal
design
Softwares 1) R 2) Minitab 3) Design Expert
R (Quantitative, Different level )
Minitab(Reduce No of Runs)
Design expert (Irregular region)
Click on the Response Surface folder tab, select D-optimal for the design and enter the lower and upper limits as shown below
Press the Edit Constraints button at the bottom of the screen and, as shown below, enter the multifactor equation derived at the outset of this tutorial:1550 2A + 70B .
Then, we choose the Point Exhange and press the Edit Model button and upgrade the default model of quadratic to Cubic.
Then choose Create candidate points,
Right click the header id and choose point type.
Press Results to see how well it designed the experiment.
