ROBUST DESIGN FOR SOFTWARE DEVELOPMENT - Software Design for Six-Sigma: A Roadmap for Excellence - page 468

robust design is aimed at reducing the loss caused by a variation of performance from
the target value based on a portfolio of concepts and measures such as quality loss
function (QLF), signal-to-noise (SN) ratio , optimization, and experimental design.
Quality loss is the loss experienced by customers and society and is a function
of how far performance deviates from the target. The QLF relates quality to cost
and is considered a better evaluation system than the traditional binary treatment of
quality (i.e., within/outside specifications). The QLF of a functional requirement, a
design parameter, or a process variable (generically denoted as response y) has two
components: mean ( µ y ) deviation from targeted performance value ( T y ) and variance

y ). It can be approximated by a quadratic polynomial of the response of interest.

( σ

18.2 ROBUST DESIGN OVERVIEW

In Taguchi's philosophy, robust design consists of three phases (Figure 18.2). It begins
with the concept design phase followed by the parameter design and tolerance design
phases. It is unfortunate to note that the concept phase did not receive the attention it
deserves in the quality engineering community, hence, the focus on it in this topic.
The goal of parameter design is to minimize the expected quality loss by select-
ing design parameters settings. The tools used are quality loss function, design of
experiment, statistics, and optimization. Parameter design optimization is carried out
in two sequential steps: variability minimization of

y and mean (

µ y ) adjustment
to target T y . The first step is conducted using the mapping parameters or variables
(x's) (in the context of Figure 13.1) that affect variability, whereas the second step
is accomplished via the design parameters that affect the mean but do not adversely
influence variability. The objective is to carry out both steps at a low cost by exploring
the opportunities in the design space.

σ

18.2.1 The Relationship of Robust Design to DFSS

Let us discuss this relationship through an example. Consider a digital camera ap-
plication in which the central processing unit (CPU) unit uses illumination levels to
produce images of a specified quality. A measurement system measures illumination
levels and feeds it into the CPU. The system measures the performance of a camera
in four dimensions: luminance, black level, signal-to-noise level, and resolution.
For each dimension, an illumination level is defined at which the camera fails. The
highest of these (i.e., the worst performance dimension) is defined as the minimum

Concept Design Parameter Design

Tolerance Design

FIGURE 18.2 Taguchi's robust design.