|Nutek, Inc. Quality Engineering Seminar, Software, and Consulting ( Since 1987) Site: DOE-DG|
Join our Taguci DOE Application Discussion Group (TDA-DG) with Googleand get answers to your application questions. Participate in discussions and benefit from answers to questions by others. Our members include individuals involved in DOE/Taguchi applications all over the globe.
Who answers the questions?
Members do. The DOE-DG group is expected to be self supporting.
A: Major attractions of the Taguchi approach are cited below.
Q: How is the Taguchi experimental design technique different from the design of experiment (DOE, classical) technique introduced by R. A. Fisher?
A: For most simpler experiments, the Taguchi experiments are same as the fractional factorial experiments in the classical DOE. Even for common experiments used in the industry for problem solving and design improvements, the main attractions of the Taguchi approach are standardized methods for experiment designs and analyses of results. To use the Taguchi approach for modest experimental studies, one does not need to be an expert in statistical science. This allows working engineers on the design and production floor to confidently apply the technique.
While there isn't much difference between the two types of DOE for simpler experiment designs, for mixed level factor designs and building robustness in products and processes, the Taguchi approach offers some revolutionary concepts that were not known even to the expert experimenters. These include standard method for array modifications, experiment designs to include noise factors in the outer array, signal-to-noise ratios for analysis of results, loss function to quantify design improvements in terms of dollars, treatment of systems with dynamic characteristics, etc. - (RKR 041115)
Q: My response is measured in terms of attributes like good or bad. Please explain what I should do for analysis method for finding optimum condition.
A: For common experiment designs and analysis of the results, you will need to quantify the results in terms of numerical values. Generally, you should attempt to define an expanded scale than just two numbers for good and bad. So, instead of assigning 1 for good and 0 for bad, assign numbers like 5 for good and 0 for bad. Of course, you will ten need to make efforts to identify the gray areas between the two extremes like not so bad as 2, not so good as 4, etc. Remember that this numerical evaluation is just for the purpose of experimental study and may not have any significance when you go to production.
In defining the range of your scale, make it as wide as you are able to define the incremental numbers between the two extremes. If you know how to tell 15 from 16 apart then go for a scale between 0 to 20. If not, keep the range between 0 to say 10. If you cannot distinguish between what's 8 and what's 9, keep the range between 0 to 5. (RKR 041103)
(Click here to review List of Nutek