ROBUST DESIGN FOR SOFTWARE DEVELOPMENT - Software Design for Six-Sigma: A Roadmap for Excellence - page 491

4. Location of bugs
a. Current process: Because they need to change only a single parameter for
each test, they can easily notice whether changed items or parameters involve
bugs.
b. Orthogonal array: Locations of bugs are identified by looking at the numbers
after the analysis.
5. Judgment of bugs or normal outputs
a. Current process: They easily can judge whether a certain output is normal
or abnormal only by looking at one factor changed for the test.
b. Orthogonal array: Because they need to check the validity for all signal
factors for each output, it is considered cumbersome in some cases.
6. When there are combinational interactions among signal factors
a. Current process: Nothing in particular.
b. Orthogonal array: they cannot perform an experiment following combina-
tions determined in an orthogonal array.

Although several problems remain before they can conduct actual tests, they believe
that through the use of our method, the debugging process can be streamlined. In
addition, because this method can be employed relatively easily by users, they can
assess newly developed software in terms of bugs. In fact, as a result of applying
this method to software developed by outside companies, they have found a certain
number of bugs.

18.10 SUMMARY

To briefly summarize, when using robustness methods, the DFSS team first needs to
determine the design or control factors that can be controlled. These are the factors
in the DFSS team for which the team will try different levels. Next, they decide on
an appropriate orthogonal array for the experiment. Then, they need to decide how
to measure the design requirement of interest. Most SN ratios require that multiple
measurements be taken in each run of the experiment; so that the variability around
the nominal value otherwise cannot be assessed. Finally, they conduct the experiment
and identify the factors that most strongly affect the chosen SN ratio, and they reset
the process parameters accordingly.

APPENDIX 18.A

Analysis of Variance (ANOVA)
Analysis of variance (ANOVA) 16 is used to investigate and model the relationship
between a response variable ( y ) and one or more independent factors. In effect,

16 ANOVA differs from regression in two ways; the independent variables are qualitative (categorical), and
no assumption is made about the nature of the relationship (i.e., the model does not include coefficients
for variables).