DOE Abstract	FAQ	Resources	Articles	Presentations

Send any additional questions to: larry@doetraining.com

1) Why might you choose DOE over a single factor experiment? 

A few of the benefits of designed experiments include ... 

a)  Interaction data between control variables.  DOE allows you to understand how each control variable impacts product function, and yields quantitative information about how these variables interact with one another.  A single factor experiment cannot provide quantitative b) Quantitative data describing each independent variable investigated.  If you want to know not just which variables you need to work with, but how they're impacting your process, what settings you need to strive for, and expected outcomes, your designed experiment is the source of your most extensive, quantitative information! 

c)  Predictive models which provide insight to input variable levels not actually tested.  Your DOE will provide information about factor levels not actually researched or measured, shortcutting your time to new target and process settings.

d)  Variability data without performing any additional experiments.  You'll have gathered enough information during the course of your experimental runs to learn how each factor contributes to variability and what you can do to minimize that variability. 

e) Efficiency in time, money and materials.  Running a designed experiment of 2 or more variables provides substantiative information about each variable, their interactions, variability, system curvature, process weaknesses, as well as location and dispersion effects.  How else can you gather so much valuable information, allowing you to address your product, process, predictive equations and shift to a functional target? 

2)  I have never run designed experiments, how  do I get started? 

I recommend that you review a DOE checklist and access a book on the subject.  Almost every book on DOE provides a checklist or decision flow chart to assist you.  Process Technologies offers a complimentary checklist, and a list of reputable books on DOE.  Check it out here on the web-site.

Running experiments is the easy part; the analysis is paramount.  I recommend a user friendly software package for the analysis portion.  You will spend more then the usual cost of about  $1000 writing programs on excel and you will worry even more about those calculations when your company starts to implement tens or hundreds of thousands of dollars of improvements based upon your recommendation and hand (or excel) calculations.  Many packages are available on the market: Stat-Ease, Minitab, JMP,

Stat-A-Matrix, and others.  Most of these organizations have websites which will allow you to use or evaluate their packages at no obligation.  (Process Technologies uses, markets, and supports Stat-Ease Inc. products.)

3)  What are the most critical aspects of performing a DOE?

Several important aspects include (but are not limited to): 

a)     Selecting a, or better yet several, response or quality characteristic(s).  You need to know what you what from the experiments before starting.  Select characteristics which reflect the function or performance you wish to target, ie. strength, torque, time, miles per gallon, etc.

b)     Performing  “good” experiments.  A good experiment requires:

1.       Keeping variables which are not under study at a constant value throughout       the experimental series.  “keep your constants, constant”.

2.       As you change control variables from one experiment to the next, maintain the control variable levels as consistently as is possible.

      These steps reduce the variability or noise within the experiment, improving your  confidence in the results.  

c)     Evaluating a sufficient sampling.  Most experimenters’ seek highly confident results, but fail to understand the relationship between the number of samples and confidence in the results.  Most any DOE or statistics book will provide equations or tables for assisting the experimenter with this calculation. 

d)     Confirmation testing.  This step will assist you in evaluating the predictive model generated from your tests.   Don’t make recommendations without verification tests to confirm your test results.  Confirmation testing builds confidence in your model.

4)  Should I perform tests with variable levels set at 2 or 3 levels?

If you are testing for system curvature, then 3 levels is a must.  However, for screening tests, one typically need run only 2 levels.  Knowing the purpose of the test will assist you with these decisions.