Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SC69

2 - Degradation in Common Dynamic Environments Zhi-Sheng Ye, National University of Singapore, E1-07-26 Computing lab, 10 Kent Ridge Road, Singapore, 119260, Singapore, Qingqing Zhai Degradation studies are useful for reliability evaluation. Because of limited test resources, several test subjects may have to share a test rig. The common environments experienced by subjects in the same group introduce significant inter-individual correlations in their degradation. In the study, the Wiener process is used to model product degradation, and the group-specific random environments are captured using a stochastic time scale. Both semiparametric and parametric estimation procedures are developed for the model. Performance of the maximum likelihood estimators is validated theoretically and by simulation. The proposed models are illustrated by an application to LED data. 3 - A Nonparametric Adaptive Sampling Strategy for Online Monitoring of Big Data Streams Xiaochen Xian, Madison, WI, 53705, United States, Andi Wang, Kaibo Liu With the rapid advancement of sensor technology, a huge amount of data is generated in various applications, which poses new and unique challenges for Statistical Process Control (SPC). In this paper, we propose a Nonparametric Adaptive Sampling (NAS) strategy to online monitor non-normal big data streams in the context of limited resources, where only a subset of observations are available at each acquisition time. In particular, the proposed method integrates a rank-based CUSUM scheme and an innovative idea that corrects the anti-rank statistics with partial observations, which can effectively detect a wide range of possible mean shifts when data streams follow arbitrary distributions. n SC69 West Bldg 106A Joint Session QSR/Practice Curated: QSR Student Interaction Session Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Dongping Du, Texas Tech University, Lubbock, TX, 79409, United States Co-Chair: Raed Al Kontar 1 - QSR Student Interaction Session The Student Interaction Session is designed for QSR student to build their professional network, show up their talents, and learn from invited guests. The session consists of students’ introduction, student and guest interactions, and Best Student Poster Competition. Each participant will have 2 minutes to deliver an elevator speech about his/her research interests and accomplishments; Senior QSR members and guests will be invited to interact with attendees and share experience; A panel of judges will select a poster competition winner, which will be announced at the QSR business meeting. Dongping Du, Texas Tech University, 2500 Broadway, P.O. Box 43061, Lubbock, TX, 79409, United States Joint Session QSR/DM/Public Sector: Functional, Image and Shape Data Analysis; Methodology and Applications Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Hao Yan, Tempe, AZ, 85281, United States Co-Chair: Kamran Paynabar, Georgia Institute of Technology, Atlanta, GA, 30332-0205, United States 1 - Online Monitoring for Metal Based Additive Manufacturing Processes using a Dual Control Charting System Based Thermal Image Tensor Decomposition Mojtaba Khanzadeh, Mississippi State University, Starkville, MS, 39759, United States, Linkan Bian, Wenmeng Tian Real-time thermal image streams captured from AM processes are regarded as most informative signatures of the process stability. The objective of this paper is to develop a statistical process control to detect process changes based on predefined distribution of the monitoring statistics. There are two major challenges: Complex spatial interdependence exists in the thermal images; the thermal images suffer from a large data volume and low signal-to-noise ratio. Multilinear principal component analysis approach is used to extract low dimensional features and residuals. Subsequently, an online dual control charting system is proposed by leveraging multivariate T^2 and Q control charts. n SC70 West Bldg 106B

2 - Lifetime Prediction of Condition Monitoring Signals using Monotonic B-splines with Infinite Support Salman Jahani, University of Wisconsin-Madison, Department of Industrial/Systems Engineering, 1513 University Avenue, Madison, WI, 53706, United States, Raed Kontar, Shiyu Zhou, Dharmaraj Veeramani Most sensor-based condition monitoring signals in practice are not observed after the failure threshold which results in truncated signals and are subject to high level of noise. This study claims a mixed effects model based on B-splines of infinite support procedure to account for truncation and noise using extrapolation feature and monotonicity conditions of B-splines with infinite support. The performance of the proposed procedure is further investigated through analysis based on the numerical studies and case study with real world data from automotive lead acid batteries. 3 - Time-warped Sparse Representation for Nonnegative Functional Data Analysis Chen Zhang, Tsinghua University, Beijing, 100084, China This paper proposes a novel framework for learning time-warped sparse non- negative factorization of functional data. The proposed method not only can guarantee the extracted bases and their coefficients to be positive and interpretable, but also can handle weakly correlated functions with different features. Furthermore, the proposed method incorporates time warping into feature learning, and allows the bases of different samples to have temporal deformations. To solve the learning task, we propose an efficient framework of estimation algorithms based on a greedy variable selection approach. 4 - Machine Annotation of Post-hurricane Satellite Imagery for Identifying Damages Quoc Dung Cao, University of Washington, Seattle, WA, United States After a hurricane makes a landfall, emergency managers and first responders commonly assess the damage by driving around the impacted area (also known as a windshield survey). Recently, post-event images captured by satellites started to facilitate the damage assessment, but the process still largely relies on human visual inspection. To improve the efficiency and accuracy of damage assessment, we propose to automatically label damages using image classification techniques. The method is applied to the case study of 2017 Hurricane Harvey. n SC71 West Bldg 106C Joint Session ICS/Practice Curated: Decomposition Algorithms for Power Grid Optimization Sponsored: Computing Sponsored Session Chair: Deepak Rajan, Lawrence Livermore National Laboratory, Livermore, CA, 94551, United States 1 - Optimizing Power System Restoration using Mixed Integer Linear Programming Ignacio Aravena, Universite Catholique de Louvain, Voie du Roman Pays 34, Center for Operations Research and Econometri, We present a novel framework for optimizing power system restoration and black-start allocation as MILPs. First, we build piece-wise linear approximations of power flow equations that account for the regime of excess of reactive power typical during restoration. Then, we propose a specialized integer L-shaped algorithm that decouples the power flow equations from the combinatorial dynamics of the restoration process. The method allows sharing cuts between time periods and formulating feasibility cuts over electrical islands. We present numerical results for modified IEEE test systems and models of the WECC and the Chilean power grids demonstrating the effectiveness of the proposed approach. 2 - Scalable Decomposition for Stochastic Unit Commitment Jean-Paul Watson, Sandia National Laboratories, 7305 Blue Louvain-la-Neuve, B-1348, Belgium, Deepak Rajan, Georgios Patsakis, Jennifer Rios, Shmuel S. Oren We discuss recent advances in scenario-based decomposition algorithms for solving realistic, large-scale stochastic unit commitment problems. We focus on mitigating the difficulty of instances with diverse renewables production patterns, which can cause serious degeneration in practical convergence behaviors. Mitigation proceeds through the use of cross-scenario cuts. We will discuss experimental results and newly created openly available benchmarks. Cypress Avenue NE, Albuquerque, NM, 87113-2065, United States, David L. Woodruff, Bernard Knueven

86