$Account.OrganizationName
Design of Experiments
Finding the Power Factors in Your Process
February 2007 - Vol 4, Issue 1
In This Issue
Sign Up
Quick Links
Greetings!

Winter has finally arrived in Vermont. Right now it is snowing and and today we are having a relative heatwave with temperatures around 10°F. Last night it was -15°F! After a brisk morning snowshoe trek, I spent sometime today looking at seed catalogs and planning out my summer garden. That got me thinking about Design of Experiments (DOE) of all things! That's because DOE techniques were first developed by Sir Ronald Fisher in his efforts to increase agricultural yields in the early 1900s. Thanks to Fisher and those who built on his work, today we have techniques that can be used in manufacturing and other fields to help determine the key factors in processes. Knowing those key factors will enable us to optimize the process for improved quality and productivity.

Design of Experiments is one the most powerful, yet least understood and used, of the improvement tools available to manufacturing organizations. The financial payback period achieved from using DOE, especially screening experiments, is often measured in months and weeks, not years. What other investment in time and resources can generate that level of return over and over again?

Some believe DOE is an advanced SPC technique. It is not. In discussions regarding quality, we have a tendency to use “control” and “improvement” interchangeably, but the implied meanings of the two terms are fundamentally different. By nature, control mechanisms prevent change. But improvement is change –specifically change for the better. The statistical tools of control, SPC, are designed to prevent change, not cause it. SPC helps us maintain the status quo. The statistical tools of improvement are the families of designed experiments. DOE is the antithesis of control.

However, SPC and DOE are complementary. It is best to ensure a process is stable before introducing deliberate changes with a designed experiment. Without process stability (SPC), the experimental results may be confounded with special causes of variation. DOE analysis techniques are based on the ability to change variables systematically and then determine if one output is statistically different from another. If a process is unstable and full of special causes of variation, we will have difficulty in determining if differences in experimental outputs are due to the changes introduced through the experiment or due to the instability of the process. Application of SPC gives us the solid foundation needed to effectively use DOE.

Ray Mikulak is president of Resource Engineering, Inc. You can contact him at 800-810-8326 or 802- 496-5888 or by e-mail at raym@reseng.com.
The major types of Designed Experiments are Full Factorials, Fractional Factorials, Screening Experiments, Response Surface Analysis (RSA), EVOP and Mixture Experiments. Our DOE course focuses on Screening Experiments. Screening experiments are the ultimate fractional factorial experiments. They literally screen the factors, or variables, in the process and determine which are the critical variables that affect the process output and from a manufacturing standpoint, usually give the “biggest bang for the buck.”

There are two major families of screening experiments: Drs. Plackett and Burman developed the original family of screening experiments matrices in the 1940s. Dr. Taguchi adapted the Plackett– Burman screening designs. He modified the Plackett– Burman design approach so that the experimenter could assume that interactions are not significant, yet could test for some two-way interactions at the same time.

Our DOE: Screening Experiments course consists of three units: Background for DOE, Plackett-Burman Experiments, and Taguchi Techniques. Each unit contains lessons that divide the content into manageable learning segments. At the end of each unit, learners have access to a Challenge to test their comprehension of the body of knowledge covered in the unit.

Carolyn Burke is Sr. Customer Service Representative. She can be reached toll free directly at 866-791-1003 or by e-mail at carolynb@reseng.com.
A DOE effort will fail if not properly planned. The team or individual responsible for the experiment needs to take the time to think through the entire activity. Without good planning, the designed experiment might yield poor results or, even worse, lead to misleading conclusions. Working through these 8 considerations will help ensure a successful experimental outcome.



  1. Design and Communicate the Objective

The objective will generally be one of three forms: The “Biggest” (to maximize the response), the “Smallest” (to minimize the response) or the “Closest-to-Target” (to hit a target)

  1. Define the Process

Define the boundaries of the process to be experimented upon.   This could be just internal processes or it could include the full extended process in which the processes of suppliers and/or customers are studied along with internal processes.

  1. Select a Response and Measurement System

Responses are the outputs, or the dependent variables, of the process. In analyzing a designed experiment, you can use as many responses as you are willing to measure.  A good measurement system is one that is accurate, repeatable, reproducible, stable, and linear.  Taking good samples is a critical aspect of the measurement system.  The samples from each experimental run must be representative of the response during that run.

  1. Ensure that the Measurement System is Adequate

Make sure the measurement system has been calibrated.  If the measurement system is not repeatable and reproducible, the results of the designed experiment will not be valid.  It is prudent to conduct a GR&R before investing in the time, effort and funds for conducting a designed experiment.

  1. Select Factors to be Studied

Factors are the independent variables that will affect the response; select those factors that should have the greatest impact on the response.  Ensure that it is practical, feasible, and cost effective to select a factor to be studied and to change its level.

  1. Select the Experimental Design

The type of design is highly dependent on the number of factors to be studied.  Screening experiments are usually the best design choice early in an experimental sequence when many factors are to be explored.

  1. Set Factor Levels

Be bold and set the levels at the edges of the operating window for the process when conducting screening experiments.

  1. Final Design Considerations

Final considerations include:  Selecting the experimental matrix to use; deciding how to estimate the experimental error and planning the experiment so that any external sources of variation are minimized.
Here are some links that will help broaden your knowledge of DOE.

  1. From the NIST Sematech Engineering Statistics Handbook (Choosing an experimental design), a exceptional resource on DOE techniques: 

http://www.itl.nist.gov/div898/hand book/pri/section3/ pri3.htm

  1. From StatSoft, Experimental Design (Industrial DOE), an excellent online reference “textbook” on experimental designs: 

http://www.statsoft.com/textbook/stexdes. html

  1. An article from Quality Digest (written by Mark J. Anderson and Shari L. Kraber of Stat-Ease, Inc.) entitled: Eights Keys to Successful DOE:

http://www.qualitydigest.com/july99/html/doe.ht ml

  1. From Quality America, Inc., a short article contrasting an 8-run Fractional Factorial with a Taguchi L8 design and an 8-run Plackett-Burman design:

http://www.qualityam erica.com/knowledgecente/articles/CJKpickadesign.ht m

  1. From Stat-Ease, Inc., a fun article reporting on a DOE conducted to determine how to select the important factors and determine factor settings to make microwave popcorn: 

http://www.statease.com/pubs/popcorn.pdf  

  1. Who is Sir Ronald Fisher from Wikipedia: 

    http://en.wikipedia.org/wiki/Ronald_Fisher

We hope that this issue of our newsletter has gotten you thinking about how you might be able to use design of experiments in your job and on problem- solving teams. While the heavy duty statistics behind DOEs might be a little daunting, there are lots of software programs out there today to help you do the job. What the software can't tell you is where to use a DOE or how to set one up. We hope that we have gotten you thinking about that with our newsletter. Next month our newsletter will focus on Measurement Systems Analysis.

Stay warm,
Robin McDermott
Resource Engineering, Inc.

phone: 802-496-5888 or 800-810-8326