Data Analysis and Modeling Applications
Monday 2 June, 10:20--12:10
- Moderator: Michael SINCLAIR
- Response Analysis
- Leslie M. MOORE, Joanne R. WENDELBERGER
- Los Alamos National Laboratory, Los Alamos, New Mexico,
USA
Dynamic systems may be modeled using systems of
differential equations. In some cases, an analytical solution
to a system may be obtained, and a model may be fit using
standard fitting techniques such as nonlinear estimation.
In other cases, an analytical solution is not available,
and the solution to a system must be obtained numerically.
A statistical approach to the modeling of dynamic systems
using experimental data is proposed and applied to an application
involving the degradation of the estane binder in a high explosive
compound. Optimized model fits are obtained by minimizing the error
between the model predictions and the experimental data.
Simple empirical models as well as theoretical differential
equation systems are considered. Uncertainty bounds for the
resulting predictions are considered. Approaches for obtaining
uncertainty bounds including parametric techniques and resampling
methods are discussed.
[Joanne R. Wendelberger, Statistics Group, MS-F600, Los Alamos National Laboratory,
Los Alamos, NM 87545, joanne@lanl.gov]
- James M. LANDWEHR, Scott A. VANDER WIEL
- Bell Labs, Lucent Technologies, Murray Hill, NJ. USA
Multi-mode optical fiber preforms have a refractive
index profile that decreases from the center of a core region
to its edge. The precise shape of the profile is critical to
a fiber's information carrying capacity as measured by bandwidth.
While the overall profile shape is approximately quadratic, there are
general departures and local ripples, corresponding to the layer
structure of the glass, whose effects are not adequately predicted
by current theory. Our work involves graphical and numerical
methods for relating the preform refractive index profile shapes
to bandwidth, so that the lathes which manufacture preforms can
be fine-tuned to increase bandwidth. The statistical problem can
be thought of as modeling a response based on an extremely high
dimensional explanatory variable space -- essentially continuous
functions of certain special types -- plus separate discrete covariates
representing other important aspects of the manufacturing process.
We describe methods we have developed for processing, smoothing,
and comparing the measured refractive index profiles and their
associated bandwidths, and some initial experimental results in
which manufacturing regimes intended to produce different profile
shapes are compared.
[James M. Landwehr, Room 2C-257, Bell Laboratories, 700 Mountain Ave., Murray Hill, NJ 07974, jml@bell-labs.com]
- W. J. OWEN, W. J. PADGETT
- Department of Statistics, University of South Carolina, Columbia, USA
Cumulative damage models for tensile strength of composite materials (complex
systems) are developed based on a multiplicative damage process and ``initial
strength'' distributions derived from inherent material flaw concepts. The
resulting material strength distributions are three-parameter generalizations
of the Birnbaum-Saunders distribution and include the effect of ``size'' of the
material specimen on its tensile strength. Estimation and asymptotic
distribution theory are investigated, and the model fits to tensile strength
data for carbon composite materials are presented. Comparisons of the new models
with corresponding additive damage models and with the power-law Weibull model
are also given based on the data.
[W. J. Padgett, Department of Statistics, University of South Carolina, Columbia, SC 29208, padgett@stat.sc.edu]
- Shyamal D. PEDDADA
- Division of Statistics, University of Virginia, Charlottesville, VA, USA
- Robert J. MCDEVITT
- Naval Surface Warfare Center, Dahlgren Division, Dahlgren, VA, USA
Given a sequence of images ${\cal I}$ and ${\cal J}$, of
a collection of ice floes, taken at times $t_0$ and $t_1$
respectively, in this talk we address the problem of tracking ice floes between
the two images. We shall describe a new tracking algorithm called
the Least Average Residual Algorithm (LARA) which was introduced in
Peddada and McDevitt (1996, IEEE Transactions on Geoscience and Remote Sensing).
LARA is based on the {\it nonhomologous nonlinear regression} that was
introduced in Peddada and Chang (1996, {\it JASA}) and
Peddada and McDevitt (1996, {\it IEEE Transactions on Geoscience and Remote Sensing}).
LARA performs a constrained search for ice floes by determining an
appropriate search space for each floe to be tracked.
It takes a stepwise approach with suitable
decision rules at various stages by taking into account various
distinguishing characteristics such as (i) the geometry
of the ice field, (ii) size of an ice floe, (iii) geometry of the floe.
LARA also attempts to detect broken and amalgamated floes.
[Shyamal D. Peddada, Division of Statistics, University of Virginia, Charlottesville, VA 22903, sdp7u@virginia.edu]
- Kin LAM
- Hong Kong Baptist University, Hong Kong
- H. C. YAM
- University of Hong Kong, Hong Kong
It is discovered that the CUSUM techniques used by people in
the manufacturing industry can be adapted to yield a trading strategy
in the financial market. The filter trading strategy familiar to
people in finance is found to be a particular case of CUSUM
procedures. A more general form of the CUSUM techniques will yield
new trading strategies which have intuitive appeals. Trading
characteristics of such strategies will be investigated using CUSUM
techniques.
[Kin Lam, Department of Finance and Decision Sciences, Kowloon Tong, Hong Kong, lamkin@hkbu.edu.hk]
Back to JRC program
Back to JRC Home Page