2016 INFORMS Annual Meeting Program

40 / 561

INFORMS Nashville – 2016
40
2 - Online Adaptive Sampling And Estimation For Clustered
Anomaly Detection
Hao Yan, Georgia Institute of Technology,
In point-based sampling and sensing system, adaptive exploration in complex
sampling space can dramatically reduce the sampling time. Most of the existing
techniques focus on reducing the overall fitting error for the entire sampling
space. However, in many application, such as anomaly detection, only sparse
clustered anomalous regions are important. In this paper we develop two
adaptive sampling strategies together with estimation methods to recover the
clustered region and discuss their properties to balance the space filling property
and focus sampling near the anomalous region. Finally, the proposed
methodology is validated by simulation study and real datasets in Guided Wave
Experiment.
3 - A Penalized (log)-Location-Scale Tensor Regression Model For
Residual Useful Lifetime Prediction
Xiaolei Fang, Georgia Institute of Technology,
Kamran Paynabar, Nagi Gebraeel
We develop a penalized prognostic model whose covariates are tensor-based
degradation signals. To address the ultrahigh dimensionality challenge, the
coefficient tensor is decomposed as a product of some basis matrices (CP
decomposition) or a product of a core tensor and some factor matrices (Tucker
decomposition) . Instead of estimating the coefficient tensor itself, we estimate
these basis matrices or core tensor and factor matrices, which have far much
smaller dimensionalities. Two algorithms with global convergence property are
developed for model estimation. The effectiveness of our models is validated using
a simulation study and an infrared image-based degradation signal dataset.
4 - Multivariate Profile Monitoring Based On Sparse Multichannel
Functional Principle Component Analysis
Chen Zhang, National University of Singapore,
, Hao Yan, Jianjun Shi
This paper presents a new monitoring framework for multi-channel profile data.
In particular, we first propose a sparse multichannel functional principle
component analysis (SMFPCA) to model multiple profiles, SMFPCA on one hand
can capture the auto-correlation structure of profile data well, and on the other
can allow flexible cross-correlations of multiple or even high-dimensional profiles
with different features. Then using SMFPCA scores, we further propose a
monitoring scheme that can detect sparse out-of-control changes efficiently.
Numerical studies together with a real example in the semiconductor
manufacturing demonstrate the application and effectiveness of our methods.
SA67
Mockingbird 3- Omni
Journal of Quality Technology Invited Session
Sponsored: Quality, Statistics and Reliability
Sponsored Session
Chair: Fugee Tsung, HKUST, Hong Kong,
1 - Bayesian Life Test Planning For Log-Location-Scale Family
Of Distributions
Yili Hong, Virginia Polytechnic Institute,
This paper describes Bayesian methods for life test planning with censored data
from a log-location-scale distribution. We use a Bayesian criterion based on the
estimation precision of a distribution quantile. A large-sample normal
approximation gives a simplified, easy-to-interpret, yet valid approach to this
planning problem, where in general no closed-form solutions are available. We
present numerical investigations using the Weibull distribution with type II
censoring. We also assess the effects of prior distribution. A simulation approach
of the same Bayesian problem is also presented.
2 - Multivariate Exponentially Weighted Moving-average Chart For
Monitoring Poisson Observations
Nan Chen, National University of Singapore,
In this talk, we develop a feasible multivariate monitoring procedure based on the
general multivariate exponentially weighted moving average (MEWMA) to
monitor the multivariate count data. The multivariate count data is modeled
using Poisson log-normal distribution to characterize their interrelations. We
systematically investigate the effects of different charting parameters and propose
an optimization procedure to identify the optimal charting parameters. To further
improve the efficiency, we integrate the variable sampling intervals (VSI) in the
monitoring scheme. We use simulation studies and an example to elicit the
application of the proposed scheme.
SA68
Mockingbird 4- Omni
QSR Refereed Research Session
Sponsored: Quality, Statistics and Reliability
Sponsored Session
Chair: Hui Yang, Pennsylvania State University, University Park, PA,
United States,
1 - An Optimum Design Of Laser-based Additive Manufacturing
Experiments by Leveraging Analogous Prior Data
Amir Massoud Aboutaleb, Mississippi State University, 139 A, Park
Circle, Starkville, MS, 39759, United States,
Linkan Bian
Most of Lase-Based Additive Manufacturing studies do not use a systematic
approach for optimizing process parameters for desired part properties. Existing
design-of-experiment methods require two stages of experiments: a large batch of
initial experiments and multiple smaller batches of sequential experiments. Our
method directly utilizes experimental data from previous studies to guide the
sequential optimization experiments of the current study.
2 - Model Transfer via Equivalent Effects Of Lurking Variables
Arman Sabbaghi, Purdue University, West Lafayette, IN, 47907,
United States,
Qiang Huang
The transfer of a model across different settings of lurking variables is addressed
with a novel framework that fuses the Rubin causal model with the effect
equivalence concept. A Bayesian methodology for model transfer is developed
and applied to transfer deformation models across additive manufacturing
environments
3 - Residual Useful Lifetime Prediction Using a Degradation
Image Stream
Xiaolei Fang, Georgia Institute of Technology,
1546 Woodlake Dr NE, Apt F, Atlanta, GA, 30329, United States,
Kamran Paynabar, Nagi Gebraeel
This paper proposes a new methodology for RUL prediction of a system using a
sequence of degradation images. The methodology integrates tensor linear algebra
with traditional location-scale regression widely used in reliability and prognosis.
Two optimization algorithms with a global convergence property are developed
for model estimation.
4 - Statistical Modeling For Spatio-Temporal Degradation Data
Xiao Liu, IBM, 1101 Kitchwan Road, Room 29-252, Yorktown
Heights, NY, 10598, United States,
Kyongmin Yeo, Jayant Kalagnanam
This paper investigates the modeling of an important class of degradation data,
which are collected not only over time but also from a spatial domain. Like many
traditional degradation models which rely on stochastic processes, a space-time
random field is constructed, through a novel approach, for modeling the spatio-
temporal degradation process.
SA69
Old Hickory- Omni
2016 Edelman Finalists Reprise – I
Sponsored: CPMS, The Practice Section
Sponsored Session
Chair: Michael A Trick, Carnegie Mellon University, Pittsburgh, PA,
United States,
1 - Operations Research Transforms The Scheduling Of Chilean
Soccer Leagues And South American World Cup Qualifiers
Andres P Weintraub, Universidad de Chile, Dept De Ingenieria
Industrial, Republica 701 Casilla 86-D, Santiago, Chile,
, Fernando Alarcon, Guillermo Duran,
Luis Ramirez, Hugo Munoz, Mario Ramirez, Denis R. Saure,
Sebastian Souyris, Rodrigo Wolf-Yadlin, Gonzalo Zamorano,
Matias Siebert, Jaime Miranda, Mario Guajardo
For the past 12 years, we have applied OR techniques to schedule soccer
leagues in Chile. Using integer programming-based methods, it is decided
which matches are played in each round, taking into account various
objectives. We have scheduled more than 50 tournaments using this
approach, resulting in an estimated economic impact of about $59 million.
Because of the high portability of these techniques, we have used them
successfully to schedule sports leagues in other countries and the South
American qualifiers for the 2018 Soccer World Cup. Furthermore, the
methods used in this application have been disseminated widely, helping to
promote OR as an effective tool for addressing practical problems.
SA67