2015 Informs Annual Meeting

WE72

INFORMS Philadelphia – 2015

3 - Use of Analytics to Investigate Crash Related Risk Factors Dursun Delen, Oklahoma State University, 408 Business Building, Stillwater, OK, United States of America, Dursun.Delen@okstate.edu Investigation of the risk factors that contribute to the injury severity in motor vehicle accidents has proved to be an interesting and challenging problem. In this study, employing a data-driven predictive analytics methodology along with information fusion-based sensitivity analyses, we identified the relative importance of the crash related risk factors as they relate to varying levels of injury severity. 4 - Real-time Schedule Recovery in Liner Shipping with Regular Uncertainties and Disruption Event Chen Li, Dr, Hong Kong University of Science and Technology, Dept of IELM, Clear Water Bay, Hong Kong, Hong Kong - PRC, cliad@connect.ust.hk, Dongping Song, Xiangtong Qi We study real-time schedule recovery policies for liner shipping under regular uncertainties and the emerging disruption. One important contribution is to distinguish two types of uncertainties, and propose different strategies to handle them. For regular uncertainties, we address the problem as a stochastic control problem, and develop the structural results; then we show how an emerging disruption changes the control policies. Numerical studies demonstrate the advantages of control policies. 5 - Cyber Physical Allocation to Make Efficient Bike Sharing Programs Subasish Das, Research Associate, University of Louisiana at Real-time allocation helps making the bike sharing programs efficient and productive. Cyber physical network of any bike sharing program will provide real-time status of the bike kiosks and user location. The bike sharing program can turn these information into real-time data product app. Users can use the app for the real-time info and make plan accordingly. This paper develops simulation tool to verify the research findings. WE72 72-Room 203A, CC Physical and Computer Experiments Sponsor: Quality, Statistics and Reliability Sponsored Session Chair: Matthew Plumlee, University of Michigan, 1 205 Beal Avenue, Ann Arbor, MI, 48109, United States of America, mplumlee@umich.edu 1 - Partial Aliasing Relations in Mixed Two- and Three-Level Designs Arman Sabbaghi, Assistant Professor Of Statistics, Purdue University, Department of Statistics, 150 N. University Street, West Lafayette, IN, 47907, United States of America, sabbaghi@purdue.edu Indicator functions are constructed under the orthogonal polynomial parameterization of contrasts, and applied to the study of partial aliasing, for mixed two- and three-level designs. The algebra behind calculation of indicator function coefficients is proven to be a product of individual algebraic operations for the different types of factors. Conditions for estimable interactions in mixed- level designs are established by means of this equivalence. 2 - Local Calibration of Computer Models Arash Pourhabib, Assistant Professor, Oklahoma State University, 322 Engineering North, Stillwater, OK, 74078, United States of America, arash.pourhabib@okstate.edu, Rui Tuo, Jianhua Huang, Yu Ding We propose a framework for the local calibration of parameters when a computer model is used to approximate a physical process. The proposed framework, non- parametric local calibration, acknowledges the functional dependency of parameters on control variables. We present the model in terms of a regularized optimization problem and solve it using a representer’s theorem. We also prove the consistency of the estimator obtained via this approach. 3 - Maximum Projection Designs for Computer Experiments Evren Gul, PhD Student, Georgia Institute of Technology, 251 10th Street NW, THB 504, Atlanta, GA, 30318, United States of America, egul3@gatech.edu Space-filling properties are important in computer experiments. Maximin and minimax distance designs consider only space-filling in the full-dimensional space; this can result in poor projections onto lower-dimensional spaces, which is undesirable when only a few factors are active. Latin hypercubes can improve one-dimensional projections but cannot guarantee good space-filling in larger subspaces. We propose maximum projection designs that maximize space-filling properties in all subspaces. Lafayette, P.O. Box- 44886, Lafayette, LA, 70504, United States of America, subasishsn@gmail.com

4 - Smoothing The Bumps: Sigmoidal Versus Localized Basis Functions in Gaussian Process Modeling Daniel Apley, Professor, Northwestern University, 2145 Sheridan Road, Evanston, IL, 60208, United States of America, apley@northwestern.edu In Gaussian process (GP) modeling of computer simulation data, common covariance models have localized basis functions, which can result in a bumpy fitted response surface. We propose a new class of covariance models that can be viewed as incorporating an integrator into any stationary GP (akin to the integrator in an ARIMA model), thereby resulting in sigmoidally-shaped basis functions. We contrast local versus sigmoidal basis functions and argue the advantages of the latter in GP modeling.

WE73 73-Room 203B, CC Reliability Test Design Sponsor: Quality, Statistics and Reliability Sponsored Session

Chair: Edward Pohl, University of Arkansas, Department of Industrial Engineering, Fayetteville, United States of America, epohl@uark.edu 1 - Algorithms for Optimal Allocation of Resources in Reliability Growth Testing Kelly Sullivan, Assistant Professor, University of Arkansas, Fayetteville, AR, 72701, ksulliv@uark.edu, Mohammadhossein Heydari Reliability growth testing seeks to identify and remove failure modes in order to improve the reliability of a product entering the market. We develop and test a suite of exact and heuristic algorithms for allocating limited testing resources within a series-parallel system in order to maximize the resulting system’s reliability. 2 - Application of Markov Decision Processes for Optimization of Reliability Growth Tom Talafuse, Graduate Student, University of Arkansas, Fayetteville, AR, 72701, United States of America, tom.talafuse@gmail.com, Shengfan Zhang, Edward Pohl Reliability growth occurs when failure modes are identified and corrective actions taken to improve system reliability. Planning methods allow construction of idealized growth curves to estimate the time and resources needed to reach a desired level of reliability. Since developmental testing results often deviate from this idealized curve, we propose a Markov Decision Processes approach to optimally allocate resources to improvement efforts to minimize deviation from idealized growth. Chair: Minjae Park, Hongik University, 72-1 Sangsu-Dong, Mapo-Gu, Business School, Seoul, 121-791, Korea, Republic of, mjpark@hongik.ac.kr 1 - Optimal Condition-based Imperfect Maintenance Policy for Systems Subject to Multiple Competing Risks Sara Ghorbani, American Express, 33 Hudson Street, Jersey City, NJ, 07302, United States of America, saraghorbani21@gmail.com, Elsayed A. Elsayed, Hoang Pham We develop a generalized threshold-type condition-based maintenance (CBM) policy for a system subject to multiple competing risks including degradation process and sudden failure. This model extends the existing research by considering imperfect maintenance. Furthermore, a special case of a system subject to two independent competing risks, degradation and sudden failure is studied and the numerical optimization analyses are presented. 2 - Simulation-based Reliability Evaluation of Multi-stage Multi-state Manufacturing Systems Seyed Niknam, Western New England University, 1215 Wilbraham, Springfield, MA, 01119, United States of America, seyed.niknam@wne.edu, Rogerio Peruchi This research investigates the reliability analysis of a multi-stage multi-state manufacturing system. The proposed model provides a sensible measure to assess the system situation against the best-case scenario of a production line. The proposed model incorporates not only failures that stop production but also deals with partial failures where the system continues to operate at reduced performance rates. A simulation model is developed to define the possible states in the system. WE74 74-Room 204A, CC Reliability IV Contributed Session

502