Design of Experiments
A Design of Experiment (DOE) is the process of determine the interaction of KPIVs (Key Process Input Variables) on the output of a process. It attempts to quantify the relationship of the variables in order to optimize the settings for that process. A key point of the design of experiment process is that it changes several variables at once. That allows the statistics behind the process to identify interactions between the KPIVs. The design of experiments methodology is closely associated with Six Sigma.
To conduct a DOE, different levels for each variable are identified, and a testing plan is developed. Obviously, the more variables you are investigating, the more combinations you require.
For example, an oven may have the KPIVs of cooking time (Long/Short), temperature (High/Low), and humidity (High/Low). The possible combinations might look like this table:
|
KPIV |
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
|
Time |
L |
L |
L |
L |
S |
S |
S |
S |
|
Temp |
H |
H |
L |
L |
H |
H |
L |
L |
|
Humidity |
H |
L |
H |
L |
H |
L |
H |
L |
The systematic approach of a DOE prevents missing important interactions and helps process owners optimize the variables that affect output.
The Design of Experiments process has two main drawbacks. The first is its complexity. It relies heavily on statistics, and thus is not a tool for the faint hearted. The other main drawback is the number of runs of the experiment you need to do to get statistical significance. Because you need several data points for each combination of variables, the testing requirement explodes in a hurry. A DOE evaluation of a process can take considerable investment.
Other problems include an inability to control certain variables, problems identifying all the variables that affect the process, and a lack of linearity in variables that makes it hard to identify optimum settings. These drawbacks tend to limit the frequency with which DOE is used to solve problems.
People often, though, attempt informal experimentation, which is not DOE. In most cases, this is done as a trial and error, with little statistical scrutiny. In many straightforward cases, simple experimentation can shed light on a process. In complex situations, though, observation of the results without any quantitative data may let you go down the wrong path.
Despite its limited use in most companies, DOE is a powerful tool. A well-run experiment can provide a great understanding of the variables impacting a process, how they interact, and how to set them to get the optimal results.
2 Comments
John Hunter · October 18, 2010 at 7:54 am
Design of Experiments is a very powerful tool. And it is not used nearly enough. I think one of the best things about six sigma is the much higher use of doe in six sigma than other improvement efforts. Still even in six sigma it is a vastly underutilized tool.
As you mention, just plain experimenting (pilots, pdsa…) is great too. And is also used far far far too little. Experimenting must be done sensibly to avoid tampering and confusing the standard processes but this is easy to do.
Jeff Hajek · October 18, 2010 at 8:16 am
John,
I think most problem solving tools are under used, not just the sophisticated ones. I suspect that we’ve become such an action oriented society that we think any time spent in contemplation is wasted. We even coin terms like ‘analysis paralysis’ that show how we feel about diving into the details.
Like all things, though, these tools require practice, and that means investing time up front to get good at them. That creates an even greater feeling of waste, and more inertia about using the tools.
Thanks for commenting.
Jeff