Design of Experiments (DOE)
07 August 2024
**Understanding Design of Experiments (DOE)**
Design of Experiments (DOE) is an important tool that helps to reduce the amount of experiments required, and study the interaction between variables. It is an important set of tools that allows the experimenter to identify critical variables and variable interactions without performing hundreds or even thousands of experiments.
**Traditional Experimentation Methods**
In the recent past, almost all scientists and engineers performed experiments by changing one variable at a time (while holding all others constant). Depending upon what the system or experiment is, this may not a bad method, especially if one variable is causing most of the change with the system. However, there are many instances where there is not one main variable that is causing the change in the system — but it is the interaction between one or more variables. DOE will tell you the important factors or interactions to change or experiment with.
**Traditional Experimentation Methods**
The types of DOEs include:
- Full factorial designs
- Fractional factorial designs
- Taguchi design
- Response surface designs
**Steps in Creating a DOE**
There are also many other popular DOE types that are used today. The steps to creating a DOE include the following:
- The first step to creating a DOE is to select the variables that you want to have in your experiment. This can range from 2 to 15 or more variables, depending upon your system and experiment.
- The next step is to decide whether you would like to test two levels of each (usually a high and a low level), or multiple levels of each variable.
- The next critical step is to decide what the “outputs” will be. Often, the output is a measurement that will tell you if your system or experiment is performing well or poorly.
- You must then think about how many experiments that you would like to conduct, and the maximum number of experiments that you can conduct, which typically depends upon time, materials, and resources.
- At this point, you can look at various DOE designs in a textbook such as “Design and Analysis of Experiments” by Douglas C. Montgomery, or use a software such as Minitab or Matlab to create the experimental design. The design type may have to be recreated in order to suit the number of factors, levels, or experimental runs that you would like to perform.
- Once you have your experimental design, the next step is to execute your study. If possible, try to execute it in a randomized order in order to reduce variability associated with time or other covariants.
**Data Analysis in DOE**
After the experiments have been conducted, the outputs can be entered into a software, or you can calculate the variances and means manually. There are several plots that can help you to analyze the data from your DOE:
- Main effects plot
- Interaction plot
- Multivariate chart
- 3D Surface plots
There are also many other traditional plots that can be used to study the DOE dataset, such as:
- scatter plots
- box plots
- CDF plots
- probability plots
- as well as many other types.
Examples of the main effects, interaction, and multivariate plots are shown below: