Six Sigma strives to prevent process variation because variation hinders a process’s ability to reliably and consistently deliver high-quality products or services. The Design of Experiments (DOE) method allows quality teams to simultaneously investigate multiple potential causes of process variation.
DOE is also is also known as Designed Experiments or Experimental Design and begins by identifying the major factors that could cause process variance. The Designed Experiments tool contains three elements. For example, if the DOE were used on the process of making a pizza the elements would include the following:
- Factors – These are inputs to the process. Factors are considered as either controllable or uncontrollable variables. Factors in the pizza example include the oven, dough, sauce, and toppings.
- Levels – These are the potential settings of each factor. The levels in the pizza making process are the temperature of the oven, the cooking time and the amount of sauce and toppings used.
- Response – This is the output of the experiment. DOE strives for a measurable output that is influenced by the factors and their differing levels. The response or output from the example is how the pizza tastes.
Select the Factors
There can be a number of inputs in a process that can affect the output. The factors that are most relevant to the end result are the ones most important to DOE. These factors can be selected by the project team in a brainstorming session. In ordinary circumstances where time and budget are finite, the team should limit the experiment to six or seven key factors. These factors are controlled by setting them at different levels for each run.
Set the Levels
Once the factors have been selected, the team must determine the settings at which these factors will be run for the experiment. The example of cooking a pizza demonstrates that some factors are measured in numbers, such as oven temperature and cooking time. Some factors are qualitative such as which toppings are used; they are measured in categories and are converted into coded units for linear regression analysis.
The more levels that are identified for each factor the more trials will be required to test these levels. To ensure that an optimal number of levels are selected, focus on a range of interest. This range includes settings used in the normal course of operations and also may include settings of more extreme scenarios. The greater the difference in factor levels the easier it becomes to measure variance.
Evaluate the Response
The response is the outcome of the experiment. Outcomes are most helpful in improving the process when they can be measured in quantitative terms rather than in qualitative attributes. A response that is quantifiable makes the experiment well suited to the additional scrutiny of statistical regression techniques.
Design of Experiments allows inputs to be changed to determine how they affect responses. Instead of testing one factor at a time while holding others constant, DOE reveals how interconnected factors respond over a wide range of values, without requiring the testing of all possible values directly. This helps the project team understand the process much more rapidly.