| 
BEWARE: Effective experimentation does not happen by accident. However, many ineffective approaches to DOE succeed only by accident, and not very often at that. By reviewing volumes of textbooks and sitting in on their related DOE courses, QPI has found that successes from some of these approaches are cumbersome, costly, and marginal.
What to look for? . . . beware of those who teach the following strategic flaws;
|
The "Minimal Effect"
"Of the many variables and interactions contained within an experiment, only a few will prove to have significant effects." Why would anyone spend a lot of money and time for limited return? This statement is true only of Poor DOE Strategies.
Back to top |
The Discovery Channel
Entertaining in your spare time, don't model experiments this way. Watch out for this one. "The variables within the scope of the DOE will be those that we do not understand. After all, if you knew the variable was important, why would you have to experiment with it?" This is the thinking that leads to the Minimal Effect above. Think about it, wouldn't you use variables that you believe would actually have an effect on the output? Wouldn't you like all of your variables to be significant - giving you more ways to control the product or process performance? Heck Yes! We want to acheive Maximum Effect - And Fast.
Back to top |
Three Level Phobia
Some believe "there is not much to be learned from high, low, and middle settings of your variables." They do this because it is easier to explain two-level testing of variables. Some of the most impressive knowledge that leads to robust product or process performance comes from understanding the mathematical relationship between the variable and the result. Non-Linear relationships ARE A GOLD MINE, and there are a lot of them. Testing at only two levels, you would be assuming all relationships to be Linear - Too Bad.
Back to top |
The Double T Approach
Alias "Tweak-Testing" approach. One of the most common success inhibitors. "Since the problem is already noticeable, there is no need to vary the factors beyond what the current specification or environment suggests. Just vary the factors a little bit above and below the norm." Vary the factor as much as you can, without causing death or destruction. Don't break anything and don't hurt anyone. By exploring the full real range of a variable and its outputs, you will not have to guess about what would have happened if you changed the variable just a little bit more. Couple this strategy with a High, Low, and Mid settings and you have the input to output relationship defined, enabling you to find the absolute optimum setting for the factor. In fact, if the experiment involves processing equipment, you could easily establish whether a desirable output could be achieved with the existing process. Small changes, small results.
Back to top |
Resolution Polution
Many DOE wizards favor the higher resolution designs (resolutions range from 5 down to 1). The higher the resolution, the less assumptions were made, more test trials were executed, more data was analyzed, and Not Very Many Variables were Tested. This is commonly a Full Factorial approach. An experiment is not to be chosen by its resolution. Your Choice of Experimental Design Depends on How Much You Know about your product or process. If you know quite a bit, choose a method that does not demand so much data collection. If you don't know so much, choose an Experimental Design that yields more data and analysis.
Back to top |
The Never Ending Story
This is the notion that "each experiment will answer a few, not all, questions while introducing new questions that need to be answered in further experimentation." OMG! This is great job security for the DOE guru. And guess what, if you were to abide by notions 1 through 5 above, you would be on a continuous journey seeking answers. Meanwhile, your competitors are bringing product to market.
Back to top |
|
