I am designing an experiment to minimize the number of failures in wood-based materials arising from the application of wood screws. The first part is screening for significant factors and 2-way interactions. I am using a single material type and cross-section (1x1 LVL rails with the screws inserted on a specific face, with the same torque and speed) so my experimental factors are:
- Length of the rail (continuous)
- Distance of the screw from the end of the rail (x position, continuous)
- Distance of the screw from the edge of the rail (y position, continuous)
- Screw diameter (4 standard sizes, discrete)
- Screw length (3 standard lengths, discrete)
As I see it I have to basic options for response variable:
- Failed/Pass (meaning that the screw cracked the piece or not, categorical)
- Failure rate (I test the same treatment 10-20 times and input the % of failures)
I am using SAS JMP to aid with the DOE and it provides a 24 treatment table with no repetitions. My question is:
Can I simply run each individual treatment 10 times and calculate the ratio of failed components (0%-100%) and call this my response result? Or does this introduce any bias or statistical errors as the original design contemplates no repetitions and the original response would be simply Fail/Pass?
In the end I would like to provide a predictive model to keep failed parts to a minimum.