I am new and self taught in the area of DOE and after having read a lot of information on experimental designs I believe I have more doubts than right ideas.
My goal is to find the optimum parameters of a machine for which I obtain the highest mechanical properties of a part in the lowest time and cost. For this I have a 3-level 5 parameter design, i.e. too much testing...!
My idea is to first perform a L27 Taguchi fractional factorial in order to determine the most influential factors and then perform a 3 factor, 2 level, full factorial so that I can also identify factor interactions.
My question is wether I will be able to find from this data an optimum compromise for these 3 responses or do I need to as well do Box-Behnken for this (I clearly dont understand very well what the results will represent). Also, is there an easier path?! (most likely)
Thank you very much!
Dear Alberto
In the realm of DOE, mostly three types of studies are possible :
I) System Identification
II) Process modeling
III) Process optimization
Some of the mentioned options could be considered by proper design, but not necessarily all of them. So it is impotent to chose your suitable method of DOE. If your factors weren't in a right range for optimization, operating an RSM model may be not such efficient that should be. Many researchers - especially outside of mechanical engineering area- believes Taguchi Latin square designs are outdated and imperfect, I am not one of them, but I also suggest using another model. Such as a D-Optimal design. I hope my answer, be helpful to you.
Dear Mohammad,
First of all thank you for taking the time to answer.
Why is it that you recommend a D-optimal design over Taguchi method?
Also, do you agree with the following steps of a full-factorial with the most important parameters to find the relations between them and later a box-behnken design to find the optimum choice of parameters?
Once again thank you!
Dear Alberto
Everything in science and engineering has an exclusive history. Taguchi designs belonged to "age of big and slow computers." In this era, mostly everything was calculated by the bare hand, and of course, the priority of DOEs were simplicity and no need for computers. After this short introduction, What is the main problem of Taguchi designs? Interactions! The most of Taguchi designs cannot be used for determination of interaction effects, and in any system, this is a very fundamental drawback.
So why I recommend using D-Optimal designs? Smaller runs in comparison to factorials or fractional factorial designs in addition to flexibility in doing what you exactly want by applying some constraints. They called optimal because they are optimal in determining effects. But your design choice is wholly related to your statistical package which you access to it and your experiences.
Finally, I agree with you about doing preliminary designs (such as you mentioned maybe factorial designs), but be aware of tracing effects in the initial design. If you see that in your experiments the optimal value for a factor is placed in the boundaries (like minimum or maximum of it), it is a big issue. So you must redesign your factor range to a bigger one in these cases for your Box-Behnken design . If you do that, I think you can optimize your system quickly.
Please see:
https://www.itl.nist.gov/div898/handbook/pri/section5/pri521.htm
and
https://www.science.gov/topicpages/t/taguchi+design+method
Thank you again for your consideration
Central composite rotatable design (CCRD) module of response surface methodology (RSM) is preferred technique for DOE. In this technique, researcher can explore input parameters range upto five levels (lowest, low, middle, high and highest).
In addition, desirability function optimization under RSM can be used for simultaneously optimization of multiple objectives under defined constraints.
Best wishes, Er. Talwinder Singh
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA