Rai Chowdhary, MS, CQE, CQM,
Six Sigma Black Belt, Business Coach
NOTE: These will be of benefit if you have some familiarity with DOEs or have taken our workshop before.
DOEs generally require the following steps:
Excellent training is available on
DOE from TEAM 2000
Call today: 1-877-HOWNWHY, or 1-877-469-6949
There are some books and training that will get you started quickly, the books I found most helpful are:
1 Understanding Industrial Designed Experiments, by Stepehen Schmidt, and Robert Launsby
2 Statistics for Experimenters, by Box, Hunter, and Hunter
3 Understanding Industrial Experimentation, by Donald Wheeler
Excellent consulting / training workshops are offered in DOE by TEAM 2000 with hands
on exercises customized to fit your needs
Please call 1-877-HOWNWHY, or 1-877-469-6949.
Website: www.hownwhy.com
In the pursuit of understanding cause and effect we can take the long road of studying one variable at a time, and miss out on interactions between variables, or we can take the DOE approach. If you have 12 variables to study, and each needs to be studied at 2 settings (a high and a low setting), the number of experiments you would have to do will be 212 or 4096 experiments to cover all permutations and combinations. Now, we know based on Pareto's principle that hardly 3 to 4 of these variables are going to be really critical - compared to the rest; so 4000+ experiments seem to be a huge waste of energy and precious resources. Moreover, most managers will balk at such proposals, and it is unlikely you will get the support needed to proceed.
Using a properly designed experiment, with 32 (using 25 partial factorial designs, or bigger screening design) runs or little more than that - a large amount of information can be quickly gathered. A few more confirmation runs and we can filter out the variables that are really critical.
The approach will involve making judicious choices on what variables to include in a given matrix, and how to go about aliasing the variables. It is always preferable to alias with 4 way or higher interactions. If you know much about the nature of the variables, aliasing with 3 way interactions can work as well.
Make your DOEs work for you - learn how to conduct sequential DOEs for complex processes
Get the right training from TEAM 2000
Call: 1-877-HOWNWHY, or 1-877-469-6949
Small DOEs may be performed by individuals, larger ones may need to be a team effort. Also, one individual may not have enough know how about the process to identify the possible variables that affect a process - this can lead to getting blind sided, and recovery after a large experiment is underway can be very difficult.
Keeping the team size to a reasonable number is recommended; large teams can get bogged down.
You can attempt to learn DOE by
trial and error, or get the right training from those who have actually done
it,
lived it, and created may products / processes with it. Why waste time and
precious resources when
you can get a running start. Call Rai Chowdhary at TEAM 2000, you will be
glad you did
Phone: 1-877-HOWNWHY, or 1-877-469-6949, or
e mail: rai_chowdhary@yahoo.com
DOE is a structured approach to conducting experiments with the purpose of understanding what variables have what kind of effects on a given product or process. The traditional way of running experiments is to identify variables, and then study the effect of each variable by holding the others fixed while each one is changed in turn. This is an excruciatingly slow process, and often leads to sub-optimal results because the opportunity to study interactions is lost.
With DOEs you can establish on a quantitative model relating cause and effect, leading to a prediction equation. This can be used to determine the settings for variables to achieve a desired response value for the product or process.
Suppose you are interested in increasing the gas mileage of your car. What are the factors that affect it? Tire pressure, ignition timing, number of passengers, tire characteristics, road surface, wind velocity, etc. Now using DOE you can quantify the effect of each of these variables - once this is done, you can use the model equation to decide what should be the speed, say, to get x mpg given that the wind is blowing against you at 15 mph, the tire pressure is 32 psi, the number of passengers is 4, and ignition timing is set at 12 degrees.
Further, once you determine the prediction equation, you can use Statistical methods such as ANOVA to decide which variables have meaningful coefficients; do not confuse this with a high correlation coefficient.
Ideally speaking the DOE ought to be performed before the final configuration of a product or process is set in place. Performing it later can mean the difference between having to "live" with sub-optimal products / processes and eventually losing out in the market place vs introducing powerful products that are hard to beat and continue to perform in the market place.
Of the many DOEs I got involved with, almost invariably the processes and products that were created from the same were so robust that the competition could not imitate the same, and we had the market to ourselves for that particular technology. An important consideration here is whether you can patent the product / process - this can give you the ability to enjoy the gains over the life of the patent. Actually the parameters identified through DOE can be used to define the bounds for patents thereby giving your company a powerful edge in the market place.
Fact: Rai Chowdhary was told
by experts in the medical field that the process of bonding
Titanium to Cobalt Chrome would not work, and he should give up on the
same.
With DOEs he found the right parameters to make it work! So much
for the "Experts".
I will mention 5 of the common mistakes I have seen occur: