An In-Process Reliability Measurement System for Mobile Robot
Development
A Robotics Ph.D. Thesis Proposal
by
Jack Silberman


Civil and Environmental Engineering Department
Carnegie Mellon University
October 1996
Abstract

There is a current need in the mobile robot community for a measurement
system that transforms mobile robot development into a measurable and
controllable process. Experiences from previous designs are not being
effectively recorded. Defects are sometimes quickly fixed and then
forgotten. Certain defects recur repeatedly because new designers do not
have the past experience or because a defect was not reported because of
the lack of a measurement system. If this defects are not recorded, then
experiences are lost. This lost of information has a high cost and must
be reversed.
The method for addressing this problem involves collecting information
regarding defects and their causes in the process of designing,
producing, and using a product such as a mobile robot. When extracted
and analyzed through the use of a data collection, data visualization,
and data interpretation system, this information can be used to improve
a product. That is, processes can be improved based on analyses of
defect data. =

However, there are problems using common analysis techniques (both
quantitative and qualitative). Quantitative analysis does not consider
origin, cause, or the effect of defects. Qualitative analysis does not
abstract from details, so it is difficult to quantify process- related
data. In order to improve a product, a methodology is needed that will
draw on the advantages of these two systems while minimizing the
disadvantages. The process measurement recommended here gives developers
fast feedback that takes advantage of the benefits of each method; that
is, it extracts cause-effect relations and enables reliability
predictions from quantifiable data. The method suggested here is
Orthogonal Defect Classification (ODC) which links statistical defect
models and qualitative analysis in a systematic methodology. The goal of
ODC is to generate an in-process measurement system that will extract
information from classified defects with cause-effect relationships. The
proposed work extends the ODC concept to the development of mobile
robots including the necessary tools for interactive use. The
end-product of this research is an interactive methodology for
in-process feedback for mobile robot development. =

This proposed methodology differs from IBM=92s ODC methodology in that th=
e
new methodology address issues of complex electromechanical systems and
allows extraction of components reliability information.
The validation of this methodology will be accomplished by examining the
defect rate differences when the methodology is applied vs. when not
applied. Another benefits of this research is the development of tools
that will allow monitoring of project maturity (that is, the indication
of when a product is mature enough to be released).