2015 Informs Annual Meeting

WB72

INFORMS Philadelphia – 2015

WB73 73-Room 203B, CC Bayesian Data Analytics for Quality and Reliability Assurance Sponsor: Quality, Statistics and Reliability Sponsored Session Chair: Mingyang Li, Assistant Professor, University of South Florida, 4202 East Fowler Avenue, Tampa, United States of America, mingyangli@usf.edu 1 - Density Estimation with Indirect Data Park Chiwoo, Assistant Professor, FAMU-FSU College of Engineering, 2525 Pottsdamer Street, Tallahassee, FL, 32310, United States of America, cpark5@fsu.edu, Xin Li We present a general problem of estimating a probability density of a random variable when there are only observations of other random variables that are correlated to the random of interest. The application of the approach to estimating particle size distribution with dynamic light scattering and/or small X-ray scattering data will be also presented. 2 - Bayesian Hazard Modeling of Heterogeneous Lifetime Data with an Unknown Number of Sub-populations Mingyang Li, Assistant Professor, University of South Florida, 4202 East Fowler Avenue, Tampa, FL, United States of America, mingyangli@usf.edu, Jian Liu Lifetime data collected from reliability tests or field operations often exhibit heterogeneity. To quantify such heterogeneous data with an unknown number of sub-populations, a Bayesian hazard modeling approach is proposed. It features in jointly estimating model parameters and determining the number of sub- populations. Effective and efficient sampling schemes are further developed to comprehensively address the model estimation difficulty when non-conjugate priors involve. 3 - A Semi-markov Random Field Gaussian Process Models for Forecasting in Complex Systems We introduce a Semi-Markov random field approach to extend Gaussian process models for multi-step forecasting in complex systems with transient behaviors. Experimental studies indicate that forecasting approach can predict the onsets of epilepsy episodes 1 min earlier compared to other contemporary methods tested. 4 - Real-time Monitoring for Advanced Manufacturing Processes using a Novel Greedy Bayesian Method Kaveh Bastani, Research Assistant, Virginia Tech University, 106 Durham Hall (MC 0118) 1145 Perry Str, Blacksburg, VA, United States of America, kaveh@vt.edu, Zhenyu Kong The objective of this work is to realize real-time monitoring of process conditions in advanced manufacturing processes. To achieve this objective we propose an approach invoking the concept of sparse representation for multiple sensor signals, and subsequently, develop a novel greedy Bayesian method (GBM) to approximate the sparse solution. We validate the effectiveness of the proposed approach in real-time monitoring of a fused filament fabrication additive manufacturing process. WB74 74-Room 204A, CC Reliability I Contributed Session Chair: Shufeng LI, University of Houston, Room NT0403, 4401 Wheeler, Houston, TX, 77004, United States of America, sli33@uh.edu 1 - Maintenance Policies for a Deteriorating System Subject to Non-self-announcing Failures Onur Bakir, Associate Professor, Istanbul kemerburgaz University, Mahmutbey Dilmenler Caddesi, No:26, Bagcilar, Istanbul, 34217, Turkey, onur.bakir@kemerburgaz.edu.tr We evaluate various maintenance policies for systems subject to continuous time Markovian deterioration which may result in non-selfannouncing failures. The decision maker inspects the system periodically at the decision epochs, identifies the current state; and chooses an available action. The objective is to minimize the expected long-run cost rate. We provide a numerical example to analyze the effect of various cost parameters on the optimum inspection period and policy. Zimo Wang, Texas A&M University, 3131 TAMU, College Station, TX, United States of America, zimowang@tamu.edu

3 - Modeling Location Diffusion, Resource Allocation and Rebalance in Car Sharing Industry-a Zipcar Exam Wei Chen, Assistant Professor, York College of Pennsylvania, 441 Country Club Rd, York, PA, 17403, United States of America, wchen@ycp.edu In this study, we build two novel models to tackle three car-sharing management questions: 1). Car Station Location Selection; 2), Car Station Size problem; and 3).Car Relocation Numbers. Our models require less inputs and offer a quick analytic result to answer three operational issues. Particularly, we use Zipcar as an example to illustrate how our models work, it turns out that our models can perform well and achieve expectations. 4 - Routing Decision Strategy and Resource Allocation Planning for a Two-Echelon Rescue Delivery System Yi Liao, Southwestern University of Finance and Economics, Liutai Ave 555, School of Business Administration, Chengdu, China, yiliao@swufe.edu.cn, Hanpeng Zhang One of the most important post-catastrophe goals is the effective and efficient allocation of rescue resources. We consider the combined problem of allocating rescue resources between two main warehouses and planning their delivery strategies to several local distribution centers. We suggest three routing strategies- simple, mixed and dynamic mixedóand analyze the effects that different routing strategies have on rescue resource allocation and relief performance. WB72 72-Room 203A, CC Omni-channel Commerce and Analytics Sponsor: Quality, Statistics and Reliability Sponsored Session Chair: Dahai Xing, Research Staff Member, IBM Research, 1101 Route 134 Kitchawan Rd, Yorktown Heights, NY, 10598, United States of America, dxing@us.ibm.com 1 - Demand Modeling in the Presence of Unobserved Lost Sales Shivaram Subramanian, IBM Research, 1101 Kitchawan Road, Yorktown Heights, NY, 10598, United States of America, subshiva@us.ibm.com, Pavithra Harsha We present an integrated optimization approach to parameter estimation and missing data imputation for calibrating discrete choice demand models where one or more choice alternatives are censored. We jointly determine the prediction parameters associated with the customer arrival rate, as well as their preferences in an assortment. We share experimental results for instances arising in a variety of industrial settings. The results achieved indicate the efficacy of the proposed methods. 2 - Big Data Solutions for Omni-channel Fulfillment Planning and Optimization Ajay Deshpande, Research Staff Member, IBM Research, 1101 Route 134 Kitchawan Rd, Yorktown Heights, NY, 10598, United States of America, ajayd@us.ibm.com, Yingjie Li, Dahai Xing, Brian Quanz, Arun Hampapur, Ali Koc, Xuan Liu Retailers look to leverage their store networks to fulfill omni-channel demand. Our research efforts focus on developing two Big Data solutions. The Network Planner provides rapid what-if analysis to find the most effective fulfillment plan. The Optimizer dynamically optimizes sourcing of online orders while balancing conflicting business goals. 3 - How Can Manufacturers Help Brick-and-mortar Stores Fight with “Showrooming”? Dahai Xing, Research Staff Member, IBM Research, 1101 Route 134 Kitchawan Rd, Yorktown Heights, NY, 10598, United States of America, dxing@us.ibm.com, Tieming Liu We study a supply chain in which an online retailer free-rides a brick-and-mortar retailer’s sales effort. The free riding effect reduces brick-and-mortar retailer’s desired effort level, thus hurts the overall supply chain performance. We examine the selective rebate contract with price match in two scenarios: the online channel is owned by or independent of the manufacturer. We show that the contract can allocate the supply chain system profits arbitrarily between the players.

424