The designs recommended immediately below the objectives are the minimum recommended design. A more capable design for more complex objectives that are farther down the list can also be used. A larger, more complex design will produce better estimates and lead to stronger conclusions.
When the objective is to optimize the process and there are more than 6 factors, it is common to start the experiment at the screening objective and take small steps, learning more about the process at each step, and building upon that knowledge to make the next step more efficient. Taking small steps in the learn as you go approach often results fewer runs and better overall understanding of the process than one large design.
If you need to quickly trouble-shoot a process and get results fast.
Test a product, process, or method for ruggedness. The real world objectives are to verify that a test method works for all operators or that a process is stable when used over intended ranges of the factors.
For both of the above, the mission is to run a bare minimum number of runs, but, manage to test all variables simultaneously. If no factors are significant, then changes made to the factors do not significantly affect the product, process, or method. If something is significant, we can conclude the process is not stable. But to understand the cause of the instability, follow-up experiments are necessary.
For the above, use a two-level factorial design that is color-coded red (resolution III).
Screening experiments are used to narrow down a large list of potential factors to a smaller list of factors that have significant effects on the process.
For the above use a two-level factorial design that is color-coded yellow (resolution IV) or a Min-Run Screening design. Another option is a response surface, supersaturated Definitive Screening Designs.
Follow-up experiments are necessary to characterize interaction and understand why certain factor combinations work or fail.
Characterize main effects, interactions and even curvature to gain better understanding of the process being studied.
For the above use a two-level factorial design that is color-coded white (full factorial) or green (resolution V or higher). Other acceptable two-level designs are the Min-Run Characterize, and Irregular Res V designs. Add center points to any two-level factorial design to estimate curvature effects.
Optimize the process by finding the best factor settings to achieve goals such as maximizing, minimizing or hitting a target value separately for each response.
For any of the above, use a response surface design that is appropriate to the problem. Standard designs such as the central composite or Box-Behnken are good for up to a quadratic model. Optimal designs have more flexibility and allow for higher-order models.
Formulation work. If the process is a mixture, such as a drug formulation, a chemical composition, or even how to allocate a budget, then use a Mixture Design. These designs produce runs to model the responses in terms of the relative proportions of the components. Because proportions are being used, as one component is increased the sum of the other components must decrease. The sum of the component proportions is always 1. Mixture optimal designs are most commonly used because they allow the most flexibility in the component ranges.
Hard-to-change factors are best handled with split-plot designs. Look at the topic on Randomized vs. Split-plot designs to make the decision. Split-plot designs can be used to screen and characterize
Next step: Identify the responses to be measured.