DOE Design of Experiments: Plan Randomized Experiments and Analyze Effects
What is Design of Experiments (DOE)?
Design of experiments systematically arranges factor-level combinations to extract the most information from the fewest runs, identifying which factors significantly affect a response and how they interact. Unlike one-factor-at-a-time (OFAT) experiments, DOE estimates main effects and interactions simultaneously, so synergistic factor effects are not missed. It is the core tool for parameter optimization and robustness improvement, built on the three principles of randomization, replication and blocking.
When to Use It
Use full factorial designs when there are few factors, about 4 or fewer, and experiment cost is acceptable, to estimate all main effects and interactions. Use fractional factorial designs, 2^(k-p), when there are more than about 5 factors and you are screening for the key ones. After the key factors are found, response surface designs (CCD, Box-Behnken) fit a second-order model to locate the optimum, and Taguchi designs add robustness against noise factors.
How to Use It (Step by Step)
Define the response, the factors and their level ranges, then choose the design type with center points and replicates. The tool generates the experiment plan table with coded levels (-1/0/+1), the run order randomized to reduce systematic effects, and block assignments if needed. Execute the runs in that order, fill in the response data, and review the main-effects plot, interaction plot, Pareto effect plot and ANOVA table to identify key factors and the best level combination.
Key Formulas / Example
A full factorial with k two-level factors requires 2^k runs: 2 factors give 4 runs, 3 give 8 and 4 give 16; three-level designs need 3^k runs. A fractional factorial 2^(k-p) runs a fraction 1/2^p of the full design, for example 2^(5-2) uses 8 of 32 runs, at the cost of confounded effects described by the resolution. ANOVA decomposes the sum of squares into main effects and interactions, and the Pareto plot ranks effect sizes for significance screening.