Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

MD33

3 - Modeling a Battery Swapping Station for Electric Vehicles with Time-varying Demand Kyle Hovey, North Carolina State University, Raleigh, NC, United States, Yunan Liu, Xiang Li We model a battery swapping station (BSS) for electric vehicles that has a fixed supply of available battery charging slots. When customers arrive, the BSS will replace their drained batteries with those supply batteries having the highest charge that also meet an acceptable quality level (e.g., charge greater than 80% of capacity). Our model captures three realistic features of this process: i) time- varying arrivals, ii) non-linear battery charging, and iii) random residual charges. We take a queueing theory approach by keeping track of the real time ages of the batteries that are being charged. We give asymptotic performance analysis as the scale of the system increases. 4 - Relationship Between Uncertainties in Supply Chain and Its Risk Perception Yuji Sato, Professor, Chukyo University, 101 Yagotohonmach, Showa, Nagoya, Aichi, 466-8666, Japan This paper clarifies the relationship between uncertainties in supply chain and its risk perception. Rapid spread of globalization pushes firms to face the higher level of uncertainty, where they must formulate necessary and sufficient strategy for supply chain management. Such a strategy, however, involves a broad range of factors, including some that are subjective. This paper addresses this issue by refining existing structural model of supply disruption. A case study was conducted that demonstrates the applicability of the approach based on the proposed structural model to the real markets in both developed and developing countries. 5 - The Shortest Path Approaches with Time Window in Multi-graph Networks Nasibeh Zanjirani Farahani, PhD Student, University of Missouri, E3443 Thomas and Nell Lafferre Hall, Columbia, MO, 65211, United States, Moein Enayati, James S. Noble, Ronald G. McGarvey One of the common problems of Service Network Design is finding the shortest path. Networks as the infrastructure of the logistics are the graphs which are used to show the communications of different nodes by connection edges. Those types of networks that consider multiple edges in between two nodes are multigraphs and shortest path problems are rarely discussed on them. In this research different methods of solving shortest path problems for multigraph networks with time window are explained and new approaches to solving such problems are classified and explained. In the end, a new mathematical model and a metaheuristics algorithm to find the best path in such networks are proposed. n MD32 North Bldg 222B Research with Autonomous Vehicles and Platooning Sponsored: TSL/Freight Transportation & Logistics Sponsored Session Chair: Ann Melissa Campbell, University of Iowa, Iowa City, IA, 52242- 1994, United States 1 - Dynamic Operation of Autonomous Vehicle Fleets for Urban Mobility Applications Hani S. Mahmassani, Northwestern University, Transportation Center, 600 Foster Avenue, Evanston, IL, 60208-4055, United States, Michael Hyland We present dynamic fleet assignment strategies for autonomous vehicle fleet operations under different business models for urban goods delivery. Recent results from shared passenger mobility applications are also discussed. 2 - Dynamic Synchronization of Truck Platoons Anirudh Kishore Bhoopalam, Erasmus University, Rotterdam, Netherlandsl, Niels Agatz, Rob A. Zuidwijk Automated vehicle technology enables the formation of platoons in which virtually linked trucks drive closely behind one another. Platooning helps to reduce fuel consumption and emissions. In this study, we look at the dynamic planning of platoons of two trucks. To determine the platoons and the associated synchronized truck routes, we present an exact algorithm and several quick heuristics. We perform numerical experiments on different instances to study the impact of the maximum detour length and the waiting time on the total system- wide travel costs. 3 - Use of Autonomous Robots for Single Package Delivery Iurii Bakach, University of Iowa, Iowa City, IA, 52240, United States, Ann Melissa Campbell, Jan Ehmke Autonomous robots can help making last-mile deliveries more efficient and customer friendly. We examine the use of robots to deliver individual packages in urban last-mile delivery. We will look at the number of robots required and time of completion as compared with traditional truck delivery models. We will examine different assumptions on the speed of travel, geography, and working hours to understand where autonomous robots are particularly valuable.

4 - The Impact of Parking on the Use of Autonomous Vehicles in Urban Delivery Sara Reed, University of Iowa, Ann Melissa Campbell, Barrett Thomas In the dense road network of urban areas, the time spent to find parking can impact the overall time of parcel delivery. With the use of autonomous vehicles, this time can be minimized allowing the vehicle to be in continuous travel while being used for the loading of parcels at determined locations. With capacity constraints on a delivery person between loading times, we look to analyze the relationship between these constraints and estimated time of parking. This work aims to examine the competitiveness of using autonomous vehicles for delivery in urban areas. 5 - Autonomous Truckload Delivery Kamal Lamsal, Barrett Thomas, Ann Melissa Campbell The impact of autonomous vehicles will be huge. If the vehicle no longer requires a driver, the drive time across country can be reduced. One conservative estimate is that all driving across interstate highways is done with trucks on autonomous settings and handled by a regular driver from the exit of the highway to the final destination. We study the impact of autonomous drivers on truckload shipping. We assume that if a driver is used to make the delivery, the FMCSA regulations on drivers applies. If an autonomous vehicle is used, the FMCSA regulations do not apply on interstate portion . We experiment with different distributions of deliveries to understand the impact of delivery location on final costs. Joint Session ORAM/QSR/Practice Curated: Data Analytics Methods for Smart Manufacturing Systems Monitoring and Control Emerging Topic: OR and Advanced Manufacturing Emerging Topic Session Chair: Mohammed Shafae, University of Arizona, Tucson, AZ, 85743, United States Co-Chair: Dazhong Wu, University of Central Florida, FL, United States 1 - A Feature-Based Data-Level Fusion Model for Degradation Modeling and Prognostics Yupeng Wei, Pennsylvania State University, PA, United States, Dazhong Wu, Janis Terpenny The rapid development of sensor technologies has enabled multiple in-situ sensors to monitor the degradation status of operation units. To achieve an accurate prediction of remaining useful life (RUL), multiple sensor signals should be fused. This work presents a new data-level fusion methodology for degradation signals based on statistical features, which is designed to provide much more accurate features to better support the prediction of RUL. In addition, 4 machine learning algorithms are facilitated to predict the RUL based on features extracted from the fused signal. Our methodology was evaluated through a degradation dataset of an aircraft gas turbine engine that was generated by C-MAPSS. 2 - Parallel Computing and Network Analytics for the Monitoring of Industrial Internet-of-Things (IIoT) Machines Chen Kan, TX, United States, Hui Yang, Soundar Kumara This paper presents a new method for IIoT machine condition monitoring. First, dissimilarities among machine signatures were characterized. Then, we proposed a stochastic learning algorithm to construct a large-scale dynamic network of IIoT machines. When machine condition varies, the network structure is changed accordingly. A parallel computing scheme is further developed to significantly improve the computational efficiency. Results show the developed algorithm effectively and efficiently captures cycle-to-cycle dynamics of a machine and machine-to-machine variations in large-scale IIoT. 3 - Cloud-Based Parallel Machine Learning for Tool Wear Prediction Dazhong Wu, University of Central Florida, FL, United States Cloud computing and machine learning have the potential to advance smart manufacturing. One of the limitations of current machine learning methods is that large volumes of training data are required to train predictive models. Consequently, computational efficiency remains a primary challenge. In this presentation, we will introduce a parallel random forests algorithm. This algorithm is implemented on the cloud with varying combinations of processors and memories. This algorithm is demonstrated using condition monitoring data collected from milling experiments. n MD33 North Bldg 222C

225