Informs Annual Meeting 2017

SA67

INFORMS Houston – 2017

2 - Large Neighborhood Heuristics for the Two-echelon Multi-depot Vehicle Routing Problem Arising in City Logistics

3 - Diagnosis of Multi-scale Spatial Point Interaction Qiang Zhou, University of Arizona, Rm314, 1127 E James E. Rogers Way, Tucson, AZ, 85721, United States, q.zhou@arizona.edu We propose a novel diagnosis method for spatial point data in manufacturing, based on decomposition of a K function-based T2 statistic. The method can independently analyze point interactions at multiple spatial scales, which is particularly useful for fault diagnosis when the process is out-of-control.

Jiang Da-Pei, Tongji University,School of Economics and Management, Tongji Building A, Siping Road 1500, Shanghai, China, 56jdp@tongji.edu.cn

Two-echelon distribution systems are quite common in city logistics under electronic commerce. First-level vehicles located in the depots carry the demands from depots to the satellites, then demands are transfer to the second-level vehicles, then delivery to customers and fulfillment the time windows. In this paper, we introduce a two-echelon multi-depot vehicle routing problem with tasks synchronization problem (2E-MDVRP-TS). We propose an adaptive large neighborhood search to solve this problem. Our method used existing operators that in the literature and developed some new operators designed specifically for the problem considered, later evaluated on some benchmarks. 3 - An Exact Algorithm for the Pickup and Delivery Problem with Time Windows and its Variants Ali Alyasiry, The University of Queensland, St. Lucia Campus, Brisbane, 4072, Australia, a.alyasiry@uq.edu.au, Michael Forbes Research on exact methods to solve the pickup and delivery problem with time windows (PDPTW) and its variants has mainly focused on branch and price and cut algorithms. We propose a novel exact approach based on fragments - a series of pickup and delivery requests starting and ending with an empty vehicle. Using fragments, we formulate an optimistic network flow model with side constraints and use lazy constraints to cut off any illegal solutions generated while solving the resultant integer program. Results confirm that this approach is competitive with existing methods. 4 - Multi-hop Driver-parcel Matching Problem with Time Windows Marco Schutten, University of Twente, Enschede, Netherlands, m.schutten@utwente.nl, Wenyi Chen, Martijn Mes Crowdsourced shipping can result in significant economic and social benefits (e.g., faster deliveries and congestion reduction). With the aim of using the spare capacities along the existing transportation flows of the crowd to deliver small-to- medium freight volumes, we introduce the multi-driver multi-parcel matching problem and propose a general ILP formulation, which incorporates drivers’ maximum detour, capacity limits, and the option of transferring parcels between drivers. The numerical study shows that crowdsourced shipping can be an economic viable and sustainable option. Chair: Zhou Qiang, University of Arizona, zhouq@email.arizona.edu Co-Chair: Jianguo Wu, University of Texas-El Paso, El Paso, TX, 79968, United States, jwu2@utep.edu 1 - Multivariate Profile Monitoring using Gaussian Process Models Salman Jahani, Research Assistant, University of Wisconsin- Madison, Madison, WI, 53706, United States, jahani@wisc.edu Recently, considerable attention has been paid to profile monitoring. While most of the research is focused on monitoring univariate profiles, this study focuses on multivariate profile monitoring using a multivariate Gaussian process (MGP) model. In this regards, first, using MGP the structure of profiles is determined; then based upon this baseline structure, a distance-based statistic is proposed to monitor the stability of profile over time. 2 - Size Distribution Estimation of 3D Particle Clusters in Metal Matrix Nanocomposites Considering Sampling Bias Jianguo Wu, University of Texas-El Paso, 500 W University Ave, Engineering Building, Room A-244, El Paso, TX, 79968, United States, jwu2@utep.edu Nanoparticle clustering phenomenon is a critical quality issue in metal matrix nanocomposites manufacturing. Accurate estimation of the 3D cluster size distribution based on the 2D cross-section images is essential for quality assessment, quality control, and process optimization. The existing studies often draw conclusions with observable samples, which are inherently biased because large clusters more likely to be intersected by SEM images compared with small ones. This paper takes into account this sampling bias and proposes two statistical approaches, namely, the maximum likelihood estimation (MLE) and the method of moments (MM), to estimate the distribution parameters accurately. SA67 371B Analysis of Complex Manufacturing Data Sponsored: Quality, Statistics and Reliability Sponsored Session

SA68

371C QSR Best Student Paper Award Competition Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Xinwei Deng, Department of Statistics, Virginia Tech, Blacksburg, VA, 24061, United States, xdeng@vt.edu 1 - Nonparametric Modeling and Prognosis of Condition Monitoring Signals: A Transfer Learning Approach Based on Multivariate Gaussian Convolution Processes Raed Al Kontar, University of Wisconsin-Madison, 1513 University Avenue, Room 3255, Madison, WI, 53706, United States, alkontar@wisc.edu In this paper, an alternative view on modeling condition monitoring signals is proposed. This view draws its roots from transfer learning and is based on sharing convolved latent functions between training and testing units. The advantageous features of our method are highlighted with a case study for automotive lead-acid batteries. 2 - S3T: An Efficient Score-statistic for Spatio-Temporal Surveillance Junzhou Chen, Georgia Institute of Technology, Atlanta, GA, United States, na@tbd.org We present an efficient score statistic to detect the emergence of a spatially and temporally correlated signal, which is called the S3T statistic. The signal may cause a mean shift, and (or) a change in the covariance structure. The score statistic can capture both spatial and temporal change and hence is particularly powerful in detecting weak signals. Our score method is computationally efficient and statistically powerful. The main theoretical contribution is an accurate analytical approximation on the false alarm rate of the detection procedure, which can be used to calibrate a threshold analytically. Simulated and real-data examples demonstrate the good performance of our procedure. 3 - Adaptive Importance Sampling for Determining Resistance Levels in Reliability-based System Design Qiyun Pan, University of Michigan, 1205 Beal Avenue, Department of IOE, Ann Arbor, MI, 48109-2117, United States, qiyun@umich.edu In order to design reliable systems, resistance level determination becomes crucial in many applications. At the design stage, resistance level can be estimated using stochastic simulations. However, estimation results using the crude Monte Carlo sampling are highly uncertain. We present a new adaptive algorithm to improve the estimation accuracy. 4 - Bayesian Sequential Calibration using Detailed Sample Paths Bo Wang, Rensselaer Polytechnic Institute, Troy, NY, United States, wangb13@rpi.edu Simplified simulation models are often used to guidethe decision-making for complex stochastic systems. To faithfullyassess the mean performance of the real system, we developa new calibration approach incorporating the detailed output- sample paths in a sequential manner, that can efficiently use thesimulation resources and achieve better accuracy.

26