Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

MB76

3 - Experiments with Two Relaxations for Multiobjective Integer Linear Programs Serpil Sayin, Koc University, College of Admin Sciences and Economics, Rumeli Feneri Yolu Sariyer, Istanbul, 34450, Turkey We study two relaxations of multiobjective integer linear programs. The first one is the result of an LP relaxation of the feasible set. This leads to a formulation that can be solved as a multiobjective linear programming (MOLP) problem. The second one is a convex hull relaxation of the problem in the outcome space. This relaxation makes it possible to relate extreme supported nondominated solutions of the original problem and the nondominated set of the relaxed problem. We then conduct some preliminary experiments to evaluate the information delivered by the relaxations for some different problem types. 4 - MSEA 2.0: A Multi-stage Exact Algorithm for Multi-objective Pure Integer Linear Programming in Julia Aritra Pal, Sr Operations Research Specialist, BNSF Railway Company, 6301 Sabbatical Street, Apt 938, Fort Worth, TX, 76131, United States, Hadi Charkhgard We present a new exact method for multi-objective pure integer linear programming, the so-called Multi-Stage Exact Algorithm (MSEA). The method combines several existing exact and approximate algorithms in the literature, either to compute the entire efficient set or to compute the minimum of a linear function over the entire efficient set of any multi-objective pure integer linear program. The proposed method supports execution on multiple processors and is available as an open source Julia package (MSEA.jl), in GitHub. Another desirable feature of the package is that users can easily customize the package to develop their own custom-built exact solvers for their specific problems. n MB75 West Bldg 212B Quantitative Methods Supporting Operational Decisions Sponsored: Military and Security Sponsored Session Chair: William Caballero, Air Force Institute of Technology, WPAFB, OH, 45424, United States 1 - Drone-aided Border Surveillance with an Electrification Line Battery Charging System Seon Jin Kim, University of Houston, 422 Cypress Vista, Houston, TX, 77094, United States, Gino J. Lim Mobile and fixed border surveillance systems are often used to enhance the comprehensive situational awareness along the U.S. border lines. Drawbacks of such systems include limited operating capability, blind spots, and physical fatigue of field agents. The use of drones is an ideal way to overcome these issues in border patrol activities. Therefore, we present a drone-aided border surveillance system with E-line battery charging systems to wirelessly charge drones during the flight to extend flight duration. For operational purposes, an optimization model and solution algorithm to schedule drone flights for the proposed system. 2 - A Multi-objective Trilevel Optimization Model for Integrated Air Defense System Penetration Brian J. Lunday, Air Force Institute of Technology, 2950 Hobson Way, Department of Operational Sciences, WPAFB, OH, 45433, United States, Aaron M. Lessin, Raymond R. Hill We present a trilevel math programming formulation in which an intruder identifies subsets of a defender’s air defense batteries to respectively destroy and degrade, subject to budget constraints; a defender subsequently adjusts their array of defensive assets subject to certain time and movement limitations; and the intruder selects a penetration path. Within this framework, the intruder and defender respectively seek to minimize and maximize the expected exposure time of the intruding aircraft to engagements. We present a reformulation to a bilevel program, customized heuristics, and the results of heuristic performance on synthetic-but-representative scenarios. 3 - Approximate Dynamic Programming for Military Medical Evacuation Dispatching Policies Phillip Rolland Jenkins, Air Force Institute of Technology, WPAFB, OH, 45424, United States, Matthew J.D. Robbins, Brian J. Lunday Military medical planners must consider how aeromedical evacuation (MEDEVAC) assets will be dispatched prior to engaging in combat operations. We formulate a Markov decision process model to examine the MEDEVAC dispatching problem. We develop and test two distinct approximate dynamic programming (ADP) solution techniques. The first technique utilizes least-squares temporal differences (LSTD) learning, whereas the second technique leverages neural network (NN) learning. A notional planning scenario is examined to determine the efficacy of our ADP solution techniques. Results indicate that the NN policies substantially outperform both the LSTD and currently practiced policies.

4 - Solving the Heterogeneous Multi-stage Weapon Target Assignment Problem with Adaptive Dynamic Programming Darryl K. Ahner, United States Army, 135 Eastwick Court, Dayton, OH, 45440-3647, United States, Alexander Gill Kline, Carl R. Parson The weapon target assignment (WTA) problem seeks, within an air defense context, to assign interceptors (weapons) to incoming missiles (targets) to maximize the probability of destroying the missiles. In the Dynamic WTA (DWTA), there is knowledge of targets that pose an immediate threat and it is known, to a probability distribution, how many and what type of targets will follow in a subsequent stage. This paper develops a real-time, near optimal solution technique for the heterogeneous DWTA which utilizes the CAVE Algorithm. Further, it compares the results to an optimal Markov Decision Process for smaller problem instances and a baseline policy for larger problem instances. n MB76 West Bldg 212C Data-driven Approaches for Smart Manufacturing System Analysis, Monitoring, and Control Emerging Topic: Design and Control of Manufacturing Systems Emerging Topic Session Chair: Xiaoning Jin, Northeastern University, Boston, MA, 02115, United States Co-Chair: Weihong Guo, Rutgers, The State University of New Jersey, Rutgers, Piscataway, NJ, 08854, United States 1 - Process Variation Modeling and Monitoring for Interconnected Additive Manufacturing using Cloud Data Hui Wang, PhD, Florida State University, FL, United States, Jie Ren, Arriana Nwodu, Tarik Dickens Additive manufacturing (AM) has its flexibility of creating a high variety of products with complex structures. However, due to frequent changes in production demands, AM processes usually do not have sufficient data to establish a baseline for effective quality monitoring and process control. This talk envisions a solution by using an inter-connected AM environment based on cloud platforms, by which production data from multiple AM processes can be shared and exchanged with each other to jointly learn AM process variations. A case study demonstrates that the proposed algorithm. 2 - NARNET-based Degradation Analysis under Time-Varying Operating Conditions Xiaoning Jin, PhD, Northeastern University, Boston, MA, United States, Anqi He We present a new prognostic modeling method based on nonlinear autoregressive neural network (NARNET) for computing remaining useful life (RUL) of degraded systems under dynamic operating conditions. Our approach consists of two processes: (1) an offline training process for modeling the degradation and failure zones using run-to-failure sensor measurements; (2) an online remaining useful life (RUL)prediction algorithm. The operating conditions are forecasted by a NARNET model based on the unit’s operating history. We show that the prognostic model provides more accurate RUL prediction, demonstrated with an aircraft turbine engine degradation dataset. 3 - Image Analytics for Prognostics and Health Monitoring in Advanced Manufacturing System Shenghan Guo, Rutgers, The State University of New Jersey, Piscataway, NJ, United States, Weihong Guo Automatic sensing devices and computer systems have been widely adopted by the automotive manufacturing industry, which is able to record machine status and process parameters nonstop. The objective of this study is to explore and exploit the large amounts of recorded data to facilitate system prognostics and health monitoring. The proposed method represents machine status sequences in moving-window images and then extracts features using texture analysis. Classification results show that the proposed method can effectively distinguish between normal and abnormal processes. A case study on a critical asset from an automotive manufacturing plant is provided.

175