Well-designed experiments according to the methodology of Genichi Taguchi provide enough reliable data for interpretation. Mostly, the goals are achieved when certain targets are hit, and the main intent seems to be accomplished. Nevertheless, it can be worthwhile to turn back to the original set of data with more experience and the demand for a detailed understanding. First, a chemical case study is reconsidered. The initial task was to design an appropriate catalyst out of a combination of 27 elements（metals）to reduce the concentration of toxic nitrogen oxides in a fuel-burning process. In 54 experiments the concentration of all chemical compounds was measured and evaluated with the SNR for chemical reactions. Several elements got identified with the highest catalytic performance. Later, the goal was defined much more strictly to avoid -as far as possible- the emission of any toxic component in the exhaust gas. Chemically this translates into reactions in opposite directions, which means to develop a catalyst with contradicting properties. This ambitious task was carried out with the identical set of data. As a second example, a medical case study is rethought. The area is open-heart surgery. Surveillance data of each individual patient exist during and after the surgery. The challenge is to investigate the earliest possible indication of serious risk for the patient. Thus, appropriate therapy can start immediately to lower the mortality rate even further. The total group consists of 780 patients monitored with 29 parameters. Application of the Mahalanobis-Taguchi algorithm yielded reasonably good discrimination power to separate patients with an inconspicuous clinical picture from patients at risk. The question of the remaining error in both groups could not be answered at that point. For a deeper understanding, the same data set will be analyzed more intensely with medical expertise to further improve the survival rate regarding such severe surgeries.

View full abstract

Recently, engine development has tended to require the evaluation of increasing numbers of external components such as auxiliary emission control devices and modularized structural components, which often have to be tested repeatedly. This must be done promptly if development schedules are to be maintained. For this purpose, design for six sigma（DFSS）technology can be applied. This involves four checks:（1）Does everyone agree on the main themes of the project?（2）Are the target values reasonable?（3）Has anything been left undone, or is there anything more to do?（4）When the IDDOV steps（identify, define, develop, optimize, validate）are applied, have the issues been isolated and dealt with?

View full abstract

Among the S/N ratios that indicate the robustness of a system undergoing functional evaluation, here the S/N ratios of dynamic characteristics are considered with particular attention to those cases in which the ideal function is a zeropoint ratio. The S/N ratios under consideration are the widely used zero-point proportional S/N ratio, the energy-ratio type of S/N ratio described by Odamura or by Tetsumi et al., and the S/N ratio based on the mean square log loss and asymmetric exponential loss function described by Kawamura et al. Features derived from the formulas for each of these are shown, and a new formula for calculating the S/N ratio is proposed. The proposed S/N ratio takes the form of the ratio of the output of the function to its variance, making it compatible with the formula for the S/N ratio of a nominal-is-best characteristic. The variance in the denominator of the proposed S/N ratio indicates what is due to noise factors and to other factors, simplifying the determination of the appropriateness of a functional evaluation given by, for example, the variance ratio of the two. It is also flexible in that the S/N ratio can be calculated from any input values.

View full abstract