Informs Annual Meeting 2017

TD66

INFORMS Houston – 2017

4 - Routing and Scheduling Optimization in Road Transport Operations

3 - Practical Heteroskedastic Gaussian Process Modeling and Design for Large Simulation Experiments Mickaël Binois, University of Chicago, Chicago, IL, United States, mbinois@uchicago.edu, Robert B. Gramacy, Michael Ludkovski We present a unified view of likelihood based Gaussian progress regression for simulation experiments exhibiting input-dependent noise. Replication plays a key role in that context as it allows to perform inference for all parameters, bypassing full-data sized calculations. We then borrow a latent-variable idea from machine learning to address heteroskedasticity, leveraging both the computational and statistical efficiency of designs with replication. We further propose to create and sequentially enrich designs with a tunable degree of replication. 371C Quality Engineering Invited Session Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Murat Caner Testik, Hacettepe University, Ankara, 06800, Turkey, mtestik@hacettepe.edu.tr 1 - An Economic Off-line Quality Control Approach for Unstable Production Processes Rong Pan, Arizona State University, School of Computing Informatics & Decison Sys, P.O. Box 878809, Tempe, AZ, 85287- 8809, United States, rong.pan@asu.edu, Zhenlu Chen, Lirong Cui An economic off-line inspection/disposition approach is proposed, which incorporates manufacturing variation. This approach includes a new inspection algorithm for inspections based on cost minimization and utilizes a specified confidence level for identifying in-control items. 2 - A Discrete Spatial Model for Wafer Yield Prediction Yield analysis is one of the key concerns in the fabrication of semiconductor wafers. In this paper, we propose a novel discrete spatial model based on defect data on wafer maps for analyzing and predicting wafer yields at different chip locations. More specifically, based on a Bayesian framework, we propose a hierarchical generalized linear mixed model, which incorporates both global trends and spatially correlated effects to characterize wafer yields with clustered defects. Both real and simulated data are used to validate the performance of the proposed model. 3 - Charting Methods to Address Unknown Weights and Autocorrelation with Cyber Security Applications Theodore T. Allen, Ohio State University, 210 Baker Systems, 1971 Neil Ave, Columbus, OH, 43210-1271, United States, allen.515@osu.edu Literature relating to demerit and other multinomial charting methods is reviewed. A novel Bayesian charting method generalizing with convergence properties is introduced. Its applications to two problems in cyber security monitoring and addressing false positives are explored. 4 - ERLDAT2: X-bar Analysis Enhancements to the EWMA Run Length Distribution Analysis Tool TD68 Kaibo Wang, Tsinghua University, Department of Industrial Engineering, Tsinghua University, Beijing, 100084, China, kbwang@tsinghua.edu.cn ERLDAT provided quality practitioners an easy-to-use web-based tool to evaluate the on-target and off-target run length performance of the EWMA control chart. Performance metrics implement in ERLDAT included the average run length, variance of the run length and user-specified quantiles of the run length distribution. Seven options were provided to users to help design and evaluate EWMA control charts. ERLDAT2 extends the original ERLDAT application by providing the run length performance information for a comparable X-bar control chart versus EWMA control charts. Hiram Moya, University of Texas, McAllen, TX, 78501, United States, hiram.moya@utrgv.edu, Douglas Timmer

Onur Can Saka, Optimization & Analytics Unit Manager, Borusan R&D, Istanbul, Turkey, osaka@borusan.com Onur Can Saka, Optimization & Analytics Unit Manager, Koc University, Istanbul, Turkey, osaka@borusan.com, Sibel Salman The aim of this study is to develop a methodology to optimize the daily planning of road transport operations of a third-party logistics provider. The problem is represented on a time-space network and time windows are imposed on both pick-ups and deliveries, while synchronization requirements arise at cross-docks. We propose a construction algorithm that generates routes considering predefined rules for distinct vehicle types. The resulting set of routes is then inserted into a comprehensive mixed-integer linear programming model to find the least-cost assignment of orders to routes and vehicles satisfying problem constraints. We test and validate our model using real-life data of the company. 371A Best Dissertation Sponsored: Transportation Science & Logistics Sponsored Session Chair: Lavanya Marla, U. of Illinois at Urbana-Champaign, 104 S Mathews Avenue, Room 216E, Urbana, IL, 61801, United States, lavanyamarla@cmu.edu 1 - Best Dissertation Finalists Halit Uster, Southern Methodist University, Dallas, TX, United States, uster@smu.edu Finalists and winner(s) of the best dissertation award, announced in the TSL Business Meeting on Monday, will present their dissertation work. 371B Design and Analysis of Experiments with Heterogeneous and Functional Responses Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Lulu Kang, Illinois Institute of Technology, Chicago, IL, 60616, United States, lkang2@iit.edu 1 - A Sequential Split-conquer-combine Approach for Gaussian Process Modeling in Computer Experiments Ying Hung, Rutgers State University of New Jersey, 110 Frelinghuysen Road, 501 Hill Center Busch Campus, Piscataway, NJ, 08854-8019, United States, yhung@stat.rutgers.edu Gaussian process (GP) models are widely used in the analysis of computer experiments. However, two critical issues remain unresolved. One is the computational issue in GP estimation and prediction where intensive manipulations of an n-by-n correlation matrix are required and become infeasible for large sample size n. The other is how to improve the naive plug-in predictive distribution which is known to underestimate the uncertainty. In this talk, we introduce an unified framework that can tackle both issues simultaneously. The proposed framework is demonstrated by a data center example based on tens of thousands of computer experiments generated from a computational fluid dynamic simulator. 2 - An Empirical Adjustment of the Uncertainty Quantification in Gaussian Process Modeling Ran Yang, Northwestern University, Dept. of Industrial Engineering, Rm C230, 21, Evanston, IL, 60208, United States, ranyang2011@u.northwestern.edu, Daniel Apley Gaussian process (GP) models is the standard surrogate models for deterministic computer response surfaces, due in part to their built-in mechanism for providing uncertainty quantification (UQ) in the form of prediction intervals on the response. However, their parameter estimation methods tend to favor good prediction at the expense of poor UQ. We develop a post-processing method that takes a fitted GP model and its training data, and empirically adjusts the built-in GP UQ so that it is in better agreement with the actual uncertainty in the predictions. We show that it substantially improves the accuracy of the UQ, especially for nonstationary response surfaces that have more complex behaviors. TD66 TD67

400