Taguchi methods in internal combustion engine optimisation

[摘要] ENGLISH ABSTRACT:Statistical experimental design techniques are powerful tools that are often approached with suspicionand apprehension by experimenters. The trend is to avoid any statistically structured and designedexperimentation program, and to rather use the traditional method of following ones gut feel. Thisapproach, more often than not, will supply a satisfactory solution, but there is so much moreinformation availablefor the same amount of effort.This thesis strives to outline the method and application of the Taguchi methodology of experimentaldesign. The Taguchi method is a practical, statistical experimental design technique that does not rely onthe designer's knowledge of the complex statistics typicallyneeded to design experimental programs, afact that tends to exclude design of experiments from the averageengineers' toolbox. The essence of thestatistical design of experiments is this: The traditional method of varying one variable at a time andinvestigating its effect on an output is no longer sufficient. Instead all the input variables are varied atthe same time in a structured manner. The output trends resulting from each input variable are thenstatisticallyextracted from the data in the midst of the variation.Taguchi method achieves this by designing experiments where every level of every input variable occursan equal number of times with every level of every other input variable. The experimental designs arerepresented in orthogonal arrays that are chosen and populated by the experimenter by following asimple procedure.Four case studies are worked through in this text and, where possible, compared to the traditionalapproach to the same problem. The case studies show the additional information and time savingsavailablewith the Taguchi method, as well as clearlyindicating the importance of using a stable systemon which to do the experiments. The Taguchi method generated more information in fewerexperiments than the traditional approaches as well as allowing analysis of problems too complex toanalysewithout a statisticaldesign of the experimentation procedure.

[发布机构] Stellenbosch University