Page 614 - Design for Six Sigma a Roadmap for Product Development

Design Optimization: Advanced Taguchi Robust Parameter Design 567

design parameters in a robust parameter design, we will actually kill
several birds with one stone. This is because

1. Higher dynamic S/N means that the system will follow the ideal
function with the least amount of variation. This means that the
main transformation process behind the system high-level function
can consistently perform at different transformation levels and has
excellent controllability and thus will guarantee that the high-level
function of the system will perform well and consistently.
2. The goal of the robust parameter design is to make the system
perform its main function consistently under the influence of noise
factors. If robustness is achieved, it will eliminate many potential
symptoms in the downstream stages. Because focusing on symptom(s)
is equivalent to working on one noise factor at a time, it is highly
inefficient and time-consuming and we could run into cycles.
3. In summary, focusing functional quality and using dynamic S/N will
ensure real quality and save product development time.

15.4.2 Why do we use linear ideal function
and dynamic S/N?
Dynamic S/N is derived from linear ideal function. From Fig. 15.10,
using dynamic S/N as the robustness measure clearly will penalize
■ Variation
■ Nonlinearity
■ Low sensitivity

It is easy to understand why variation is undesirable. We have already
discussed at great length the need to reduce the variation. But why is
nonlinearity undesirable? Nonlinearity is a form of complexity, from
an axiomatic design viewpoint. The second axiomatic design principle
stated that if there are several design concepts and all of them can
deliver required functions, the design with the least complexity will be
preferred. Robust parameter design is equivalent to selecting a good
design by using dynamic S/N as a benchmark, and penalizing nonlin-
earity will help select a good design. Linearity is a proportionality
property; for a signal, it is a desirable property.
Also, if we conduct a robust parameter design at an early stage of
product development, we may conduct it on a small laboratory scale or
on a computer. If many nonlinearities already exist in the signal-
response relationship, this relationship may become even more com-
plicated in large-scale production; the whole signal-response relation
may become uncontrollable or unpredictable.