Informs Annual Meeting 2017

SD70

INFORMS Houston – 2017

2 - IIE Transactions Jianjun Shi, Georgia Institute of Technology, H. Milton Stewart School of, Industrial and Systems Eng, Atlanta, GA, 30332-0205, United States, jianjun.shi@isye.gatech.edu 3 - Journal of Quality Technology Fugee Tsung, HKUST, Clearwater Bay Road, Hong Kong, Hong Kong, season@ust.hk 4 - Technometrics Daniel Apley, Northwestern University, Dept of IEMS, Room C150, Tech Institute, 2145 Sheridan Road, Evanston, IL, 60208-3119, United States, apley@northwestern.edu 5 - Quality Engineering Murat Caner Testik, Hacettepe University, Muhendislik Fakultesi, Endustri Muhendisligi Bolumu, Ankara, 06800, Turkey, mtestik@hacettepe.edu.tr 6 - Quality Technology and Quantitative Management Jing Li, Arizona State University, School of Computing Informatics & Decision Sy, P.O. Box 878809, Tempe, AZ, 85287-8809, United States, jing.li.8@asu.edu 371C High-dimension Spatio-Temporal Analysis I Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Jian Liu, University of Arizona, Tucson, AZ, 85721, United States, jianliu@email.arizona.edu Co-Chair: Shyam Ranganathan, shyam81@vt.edu 1 - A Flexible Method for Building Degradation Index from Multivariate Degradation Signals Yili Hong, Virginia Tech, 213 Hutcheson Hall, Department of Statistics, Blacksburg, VA, 24061, United States, yilihong@vt.edu, Zhongnan Jin, Zhiyang Zhang, Xinwei Deng, Ran Jin, Sam Davanloo Tajbakhsh Most of the existing research on degradation modeling assumes that the degradation index is provided. Modern sensor technology allows one to collect multi-channel sensor data that are related to the underlying degradation process, which may not be sufficiently represented by any single channel. Thus, constructing a degradation index is a fundamental step in degradation modeling. In this paper, we develop a general approach for degradation index building based on an additive-nonlinear model with variable selection. The approach is more flexible than a linear combination of sensor signals, and it can automatically select the most informative variables to be used in the degradation index. 2 - A Predictive Analytic Methodology via Modeling High-dimensional Signals with Missing Observations Xiaolei Fang, Georgia Institute of Technology, 1546 Woodlake Dr NE, Apt F, Atlanta, GA, 30329, United States, xfang33@gatech.edu, Kamran Paynabar, Nagi Gebraeel Many complex engineering systems operate in harsh environments that often have a significant impact on the quality of the raw sensor data due to errors in data acquisition, communication, read/write operations, etc. Consequently, the resulting degradation signals often contain significant levels of missing and corrupt observations. In this talk, we present a scalable prognostic methodology capable of modeling poor quality (ultra)high-dimensional degradation signals to predict residual lifetime. Numeric studies are used to evaluate the performance of the proposed model. 3 - Detecting and Locating Pipe Burst with Spatial and Temporal Analysis Yifei Yuan, University of Arizona, Tucson, AZ, United States, yifeiyuan@email.arizona.edu, Jian Liu, Kevin Lansey A pipe burst is a major water distribution system failure and will result in lower water pressures at customer taps. A large number of meters are usually installed in the pipe systems to continuously monitor the water pressure, from spatially distributed locations. The data collected from these meters forms a spatial and temporal information stream. In this research, a spatio-temporal data analysis method is proposed to effectively and efficiently detect and locate the pipe burst. A simulation based case study was conducted to validate the method. SD68

4 - Neighborhood Vector Autoregressive Model for High-dimensional Time Series Shyam Ranganathan, Department of Statistics, Virginia Tech, Blacksburg, VA, United States, shyam81@vt.edu, Xinwei Deng Vector autoregression (VAR) models are popular for studying multiple dependent timeseries, with broad applications in analyzing sensor data, imaging data, and network data. One key challenge for VAR models in high dimensions is on how to insightfully explore the inherent dependency in the data to enable meaningful interpretation, and accurate prediction. To tackle this challenge, we present a neighborhood VAR model to explore meaningful sparse structures induced by underlying interactions among multiple time series, thus achieving accurate prediction with low computational cost. Numerical examples are shown to evaluate the performance of the proposed method. 371E Big Data Contributed Session Chair: Iman Vasheghani Farahani, North Carolina State University, Raleigh, NC, United States, ivasheg@ncsu.edu 1 - Application of a Novel Nonlinear Support Vector Machine Based Feature Selection Algorithm for Fault Detection & Diagnosis Melis Onel, Texas A&M.Energy Institute, College Station, TX, We present a new data-driven framework for fault detection and diagnosis, a critical task in industry processes to attain a safe operability and minimize loss of productivity and profit. Central to the framework are novel theoretical and algorithmic developments in nonlinear support vector machine (SVM)-based feature selection which encapsulates highly nonlinear relationships between features, thus improving fault detection model accuracy and guiding fault diagnosis. We present results of a recent extensive benchmark dataset consisting 90,400 batches with numerous and diverse fault types. 2 - Energy Optimization in Water and Wastewater Cycle Ali Reza Asadi, PhD Candidate, Wayne State University, 4815 Fourth Street, Room 2033, Detroit, MI, 48201, United States, ali.asadi@wayne.edu Implementation of big data techniques in the water/wastewater treatment cycle provide grounds for minimization of overall energy consumption as well as improvements in food and water safety. This methodology allows consideration of new threats such as water reservoir pollution, microorganisms and antibiotics resistance in to the optimization model. 3 - Estimating an Inverse Mean Subspace Jiaying Weng, Graduate Student, University of Kentucky, 300 Alumni Drive, Apt 179, Lexington, KY, 40503, United States, jiaying.weng@uky.edu, Yin Xiangrong We develop a new sufficient dimension reduction method called Fourier transformation inverse regression to capture the directions in the central subspace. Under linearity condition, the space spanned by Fourier transformation inverse regression would be in the central subspace. By forming a kernel matrix using Fourier transformation, we construct asymptotic dimensional test via weighted chi-squared distribution. Our new method performs better than existing methods in the simulation study and real data analysis. In addition, we further develop our new method for variable selection and for large p and small n problems. 4 - An Investigation into Online Video Sharing as a Source of Consumer Safety Hazard Reports United States, melis@tamu.edu, Chris A.Kieslich, Yannis A. Guzman, Christodoulos A. Floudas, Efstratios N. Pistikopoulos SD70

Leila Nasri, Virginia Tech, Blacksburg, VA, United States, leilan@vt.edu, Alan S. Abrahams, Johnathon P. Ehsani

This study focuses on online video-based product reviews as possible sources to detect safety hazard issues. Reviewing over 15,000 possible product reviews from YouTube, we show that the set of smoke words are a more accurate predictor of safety hazards in video-based product reviews than sentiment words.

125