Taguchi Methods and Optimization for Robust Software (Digital Short Cut)

The software industry stands on the brink of an era of dramatic change. We expect the industry to continue the restructuring process already begun, emerging as a much smaller number of horizontally structured firms mostly doing business with each other. As software becomes highly "componentized," the industry will begin to resemble the automotive industry, with many small firms making parts, but only a few large ones assembling them into finished products. Software automation in the form of application generation technology will become the norm as system analysts and other domain specialists become the new application programmers, writing in specification languages. Meanwhile, the more talented of today's application programmers will become system programmers, writing the meta-compilers that will transform specification language codes into Java and C application programs. It is still true that new technologies do not replace old technologies, at least not at first; in their infancy, they merely supplement them. Chapters 16, 17, 18 and 19 of the book Design for Trustworthy Software address the transition period during which robust, trustworthy software is still created by current technology and processes as the new technology and its streamlined processes emerge. This short cut is a reproduction of Chapter 17 of Design for Trustworthy Software. It illustrates how Taguchi's quality loss function provides a measure of the overall loss to society when a product fails to meet its target functionality and reliability. It describes howsignal-to-noise ratio measures the positive quality contribution from controllable or design factors versus the negative quality contribution from uncontrollable or noise factors. It presents Taguchi Methods involving seven steps, beginning with a clear statement of thedesign problem and ending with a confirming statisticalexperiment showing how parameter choices will enhance robustness. An example from electrical circuit design is presented, because it is much more similar to software design than mechanical design, where Taguchi Methods have found their largest applications. A more detailed example from software design or product improvement builds on the previous example. Lastly, this short cut describes Taguchi's development and application of an earlier technique involving Latin squares or orthogonal matrices to allow the evaluation on multiple parameters simultaneously. It illustrates how his use of orthogonal matrices permits