COMPARISON OF APPROACHES:
Just to recap, let's look at the major characteristics of our two methods side by side:
Process Knowledge:
- DOE assumes no understanding of the fundamental mechanisms governing the process
we are investigating.
- Taguchi assumes we have a certain understanding of the process and the interactions
that are likely to exist between inputs.
Combinations of Inputs Tested:
- DOE tests all combinations of input levels, or some symmetrical subset (such as one
half or one fourth).
- Taguchi tests a small fraction of all possible combinations, but in a manner that
allows us to calculate the affects of all inputs on the output.
Noise Factors:
- DOE traditionally ignores Noise Factors, although they could be added to the
experimental plan if desired.
- Taguchi makes use of Noise Factors to test robustness of the system and
find optimal inputs.
Understanding of Variability:
- DOE ignores variability in the process; it assumes a deterministic nature to
the system, and finds combinations of input variables that maximize or minimize
output, as the case may be.
- Taguchi assumes a stochastic nature to the system; it looks at both the levels
of output and the variability of the output; it lets us select levels of input variables
to maximize or minimize output or to minimize variability of output (i.e., maximize
robustness).
Confirming Experiment:
- DOE requires no confirming experiment, since all combinations of inputs were
tested (in a full factorial). In effect, the confirming experiment was taken care
of in the original experimental plan.
- Taguchi recommends a confirming experiment just to make sure, since the
winning set of inputs was probably not part of the original experimental plan.
WHICH APPROACH SHOULD YOU USE?
Hopefully, this article will have given you enough insights into the similarities and differences between these two powerful techniques for you to decide on a case by case basis which should be used, and when. In general, however, think DOE when you have no idea about the fundamental mechanisms governing your process, when you have no idea about interactions between your inputs, and when experiments take so long that you absolutely must get it right the first time. Think Taguchi when you have a fairly firm grasp of the underlying processes and are just trying to optimize an additional notch, when robustness or consistency of output is just as important as maximizing (or minimizing) your output, when you cannot afford to test all possible combinations of inputs, and when you have the luxury of going back and doing a final confirming test later. Finally, remember that these two techniques are really only two ends of a spectrum of possible optimization methods, and you are free to adapt elements of each to your own specific needs and situations as circumstances warrant. As always, try to understand the principles, and then do what makes sense.
ADDITIONAL READING:
Barrentine, Larry B., An Introduction to Design of Experiments: A Simplified Approach, ASQ, 1999.
Bendell, Disney, and Pridmore, Taguchi Methods: Applications in World Industry, IFS Publications, Bedford UK, 1989.
Lochner and Matar, Designing for Quality, Quality Resources and ASQ Press, 1990.
Montgomery, Douglas C., Design and Analysis of Experiments, 5th Edition, Wiley & Sons, 2000.
Ross, Phillip J., Taguchi Techniques for Quality Engineering: Loss Function, Orthogonal Experiments, Parameter and Tolerance Design, 2nd edition, McGraw-Hill, 1995.
Roy, Ranjit K., Design of Experiments Using the Taguchi Approach: 16 Steps to Product and Process Improvement, Wiley & Sons, 2001.
Schmidt and Launsby, Understanding Industrial Designed Experiments, 3rd edition, Air Academy Press, 1992.
Taguchi, Genichi, Taguchi on Robust Technology Development, ASME Press, 1993.
Taguchi, Chowdhury, and Taguchi, Robust Engineering: Learn How to Boost Quality While Reducing Costs & Time to Market, McGraw-Hill, 1999.
Wilson, Millar, and Bendell, Taguchi Methodology Within Total Quality, IFS Publications, Bedford, UK, 1990.
