Complex DOE setup - guidance neededHi
this is probably something for consulting you may say, however as it is of my personal interest too I ask this question nevertheless. I believe this might be not that uncommon in upstreaming situations, therefore it might be good to know for the future.
General situation
I have a special situation for a DOE setup where I am not sure how to easily setup the design. Here is the challenge:
- I have 6 AMBR available, each with 20 reactors (of 24), so 20 experimental runs can be run at the same time for one AMBR
- I have 3 DS (Dilution Themes, level L0,L1,L2), but level L0 must be run in AMBR 1 and 2 only
This constraint I could set up with a covariate design easily. Another approach would be a blocking factor but then we got into issues with constraining DS level L0 on AMBR 1 and 2 only. In addition with the other constraints following:
Constraints for the Design conduction
1.
- One continuous factor, lets say FX will have lower, upper and center level (as quadratic effects are of interest too)
- In one AMBR there can only be up to two different FX levels be run
- so within the 20 runs in one AMBR, only level combinations 0/-1, 0/1, -1/1 are allowed
- e.g. 10 runs FX level 0 and 10 runs FX level -1 --> all three levels appearing in those 20 runs is permitted
2.
- One continuous factor, lets name it Cell Exp. Duration (CED) will have 3 levels (-1, 0, 1) (as quadratic effects are of interest too)
- CED levels can be combined with DS levels only in max 6 combinations within the 20 runs in one AMBR
- e.g. 20 runs in one AMBR but only combinations L0/-1, L0/1, L2/1, L3/-1, L3/0, L3/1, and in the next AMBR another up to 6 combinations, and so on
Additional information
3. besides these factors I have another 12 continuous factors with 2 levels
- Note:
- they came up with 120 runs (6 AMBR times 20 reactors)
- they have model terms with interaction and quadratic effects
- BUT they removed several effects they do not think are of interest (see Design-01, if you want to see what effects they are interested in
- For the sake of this design creation the effects are secondary important, the structure needs to be fulfilled.
Attached is
- a covariate table for allowed AMBR/DS combinations.
- a model effects table for effects of interest in the generated blocking design (does not reflect the constraints)
- a initial blocking design (blocking factor mimics AMBR) Design-01.jmp which does not fulfill the constraints on CED-DS combinations number nor DS-AMBR constraing for L0
- a factor table for the initial design
@Ryan_Lekivetz and team
Re: Complex DOE setup - guidance neededFor completeness:
I tried an approach to build the structure of the DOE Setup and after the design has been generated I used a design for all the other parameters and used them as covariates (however, to fulfill the requirements, this was a manual approach to have it somehow random but keeping the requirements. Nothing to really recommend with that many parameters 
The customer reduced the requirements (after we discussed the situation) to be able to find a design.
Reviewing the design restrictions the most difficult one was to have certain dilution themes only in the first two ambers. Because this will also have a huge impact on the information you will gain due to few variation and no information from the other ambers on these themes.
Re: Complex DOE setup - guidance neededOk, so how about doing a DSD for the 14 factors for each level of the dilution theme.
Do all the runs for each DS on the appropriate AMBR.
The runs will need to be split into two blocks.
Re: Complex DOE setup - guidance neededThanks Byron for the suggestion. Going back to the original design requirements this would work if there wouldn't have been the constraints for CED/dilution theme combination restrictions to have in each AMBR just a max of 6 combinations (out of 9). One could argue, only 2 combination per AMBR would fulfill this, and finally one death one needs to die
At the end, even with lots of power nowadays in design generators, restricting a design with too many requirements has its limit, so you may need to accept to release some and follow a more iterative path with less requirements.
