Seminar on Industrial Mathematics and Statistics
Friday, November 19, 1999
8:00 A.M. - 12:00 Noon
Department of Mathematics and Statistics
College of Arts and Sciences
372 Science and Engineering Building
Presentation I: Regression Trees as a Tool for Understanding Quality
and Warranty Data
Abstract: In working to improve the quality and reliability of a
complex
product or system, there is a big difference between knowing that a
problem exists and understanding which parts of the product or system
may
be contributing to the problem. Recursive binary partitioning in the
form
of regression trees can be used on warranty or manufacturing inspection
data to develop understanding of which factors split the better
performing
systems from the ones which are not performing as well. These factors
are
investigated to find the root cause of the engineering issues. This
presentation will summarize recursive binary partitioning and the
regression tree algorithm. Then, several case studies will be used to
illustrate the strengths and weaknesses of recursive binary partitioning
on quality data. Examples will include both warranty data and in plant
manufacturing inspection data.
Speaker: Ellen Barnes, Ford Motor Company.
Ellen is an internal consultant at Ford Motor Company. She guides teams
through the engineering problem solving process. She is a registered
professional engineer, and has a MS in Applied Statistics from Oakland
University, a BS in Mechanical Engineering from Columbia University in
New
York, and a BA in Mathematics and Physics from Grinnell College in
Grinnell, Iowa. Her current area of research is applying and developing
data mining techniques on warranty data so that engineers have a better
definition of the problems causing the warranty.
Presentation II: Statistical Analysis of System Life with Masked
Cause-of-Failure
Abstract: Consider a system with K components which can be viewed as
either multiple failure modes or different risk factors acting on the
system. A system failure occurs at the earliest onset of any one of
these
risk factors. Under this competing risks framework, when the exact
cause
of the system failure can be identified, a detailed Failure Mode Effect
Analysis (FMEA) can be carried out in a routine manner. In reliability
applications, one however frequently encounters system life-data, where
the cause of failure cannot be exactly identified, but can only be
narrowed down to a subset of the K potential failure modes. In
statistical literature , such data is termed as masked failure data.
Masking is often the manifestation of an attempt to expedite the process
of repair by replacing the entire subset of components responsible for
failure instead of carrying out a second-stage resolution or autopsy
that
can be prohibitively expensive and time-consuming. In this talk, I
shall
present a brief overview of the statistical methods used in analyzing
such
data. A special emphasis will be given to the Bayesian methodologies
used
in this context, that have proved to be extremely useful and promising
thus far. Finally, I shall present some biomedical applications of the
scenario under discussion arising from certain clinical trials.
Speaker: Ananda Sen, Oakland University.
Ananda received his Ph.D. from the Department of Statistics at the
University of Wisconsin, Madison in 1993. He is currently an Associate
Professor of Statistics in the Department of Mathematics and Statistics
at
Oakland University, Michigan. His major research interest lies in
life-testing and reliability. Along-side pursuing a steady teaching and
research career, he is a consultant in projects for the Ford Motor
Company
Presentation III: Modular Vehicle Architectures Using Integration
Analysis Techniques
Abstract: The trend within the automotive industry is towards modular
systems. The automotive manufacturers separate the vehicle into modular
systems (chunks), which may be built and tested off line before
assembled
for vehicle installation. Modular systems provide the ability to
achieve
product variety through the combination and standardization of
components.
In this paper, a methodology that combines the system modeling,
integration analysis, and optimization techniques for development of
modular electrical/electronic systems is presented. The approach
optimizes integration and interactions of the electrical/electronic
system
elements and creates functional and physical modules for the system.
The
approach proposed in this paper is systematic and can be used to support
product development and decision making in engineering design. The
application of the approach is illustrated with an industrial example
from
the automotive industry, i.e., design of a cockpit system. The
Hatley/Pirbhai methodology is used for modeling functional requirements
of
the cockpit of a vehicle. The Hatley/Pirbhai requirements model defines
the interfaces (interactions) to support the functions of a cockpit
system. Once the interfaces among the functions are identified, an
incidence matrix of the interfaces is developed. A clustering algorithm
is used to identify clusters in the incidence matrix, group the
functions
in the cockpit of a vehicle, and create electronic modules. A
Hatley/Pirbhai architecture model is developed to represent the system
design. A detailed discussion on the importance of system modeling in
design of modular systems and on the constraints that limit the
development of modular vehicle systems is also presented.
Speaker: Gary Rushton, Visteon Automotive Systems.
Gary has an MS in Automotive Systems Engineering from the University of
Michigan. He is currently working as a systems engineering technical
specialist with Visteon Automotive Systems, an enterprise of Ford Motor
Company. At Visteon he has worked on audio software, subsystems product
development/design, diagnostics, vehicle system architectures and
design,
and cockpit systems design.
Presentation IV: Cluster Analysis Techniques: Models and Algorithms
Abstract: Cluster analysis is concerned with grouping of objects into
homogeneous clusters (groups) based on the object features. In this
presentation, two basic formulations of the clustering model, i.e.,
matrix
formulation and integer programming formulation, are discussed.
Algorithms useful for clustering binary incidence matrices into
mutually
separable or partially separable clusters are also presented. The
presentation concludes with a discussion of several relevant research
issues. The significance of the cluster analysis techniques arises from
the fact that they are widely used to solve various design optimization
problems, i.e., design of modular products and design of manufacturing
systems.
Speaker: Armen Zakarian, University of Michigan-Dearborn.
Armen received his Ph.D. in Industrial Engineering from The University
of
Iowa, Iowa City in 1997. He is an Assistant Professor of Industrial and
Manufacturing Systems Engineering at the University of Michigan -
Dearborn. His research interests include development of products and
systems, reliability and risk analysis of process models, and modeling
and
analysis of manufacturing systems.