I have never used DOE. I want to perform optimization of three parameters that affect my experiment. The parameters are pH (9.0-11), concentration of Lm (5-100uM) and concentration of POV (5-200 ug/mL). I will gratefully appreciate it if someone can lend a helping hand. Thank you
For optimization, you need at least three levels for each independent variable (i.e., your parameters) because it allows you to fit a quadratic model, which is often sufficient for optimization purposes. So, you can choose Central Composite Design (CCD), Box-Behnken Design (BBD), or a 3^3 factorial design.
CCD requires more runs and is usually chosen to optimize processes with 2 independent variables. However, having 5 points (2 factorial, 2 axial, and 1 center) for each level makes this design safer. Rotability is another feature of CCD.
BBD requires the least number of runs in comparison to CCD and full factorial design. From my experience, it is also the most often used design in papers. However, while it lacks design points in the corners of the design space, there are areas of poor prediction capability.
The 3^3 full factorial design emerges as the safest option. It requires more runs, but the prediction capability is also very good. Moreover, if you are not sure if your parameters are truly significant, you can initially conduct a 2^3 factorial design to determine the effect of the parameter and then add a "0 level" run to determine the curvature and optimize.
There are more designs like D-optimal, but since it is your first time, I would suggest the 3^3 factorial design. It is more tedious, but you will be less annoyed by unclear interpretation.
We have chosen a design, so you have a direction to follow. However, I will not explain it in detail here, so I recommend "Design and Analysis of Experiments" by Douglas C. Montgomery.
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Materials Science
- Materials
- Inorganic Materials
- Metals
- Alloys
- Steel