This page is being retired. Please go to the Design of Experiments Study Guide page.
A designed experiment can help you choose between alternatives and select key factors affecting a response even despite uncontrollable noise factors. You can also use Response Surface modeling to hit a certain target, reduce variability in a process, maximize or minimize a response, make a process more robust despite uncontrollable noise and even pursue multiple goals.
Basic steps of DOE
- Define objective(s)
- Gather knowledge about the process
- Develop a list and select your variables
- Assign levels to variables
- Conduct experiments
- Data analysis and conclusions
Items to avoid when conducting the experiment:
- Unwarranted assumptions of the process.
- Undesirable combinations of the factors.
- Violation of known laws of physics.
- Too large or small design sizes.
- Inappropriate confounding.
- Imprecise measurement.
- Unacceptable prediction error (Type 1 & type 2 errors).
- Undesirable run order.
reference amazon Systematic Approach to Planning for a Designed Industrial Experiment – technometrics.
Planning and Organizing Designed Experiments
- Define interactions
- Understand how to analyze interactions
Balanced experiment: each factor appears the same number of times.
Full factorial experiment: At least one trial for all possible combinations of factors and levels
One Factor at a Time (OFAT)
Explain the one factor at a time (OFAT) approach
Randomized block plans
Latin square designs
Experiments can be abbreviated numerically.
2^5 means that there are 5 factors at 2 levels.
Fractional Design Ex.
2^5-2 means that there are 5 factors at 2 levels and 2 generators. The generator determines what effects are confounded or combined with one another.
We would call this example an 1/L^g fractional factorial.
Thus, 2^5-2 is a 2 level, 5-factor, 1/4th fractional design.
Design of Experiments Examples
Design of Experiments Videos
Good SlideShare presentation: