Informs Annual Meeting 2017

WC68

INFORMS Houston – 2017

WC69

5 - Spectral Graph Theoretic Sensor-based Monitoring in Laser Powder Bed Fusion Process Prahalada Rao, rao@unl.edu, Mohammad Montazeri, Brandon Lane, Jarred Heigel The goal of this work is to monitor the laser powder bed fusion (LPBF) additive manufacturing (AM) process using heterogeneous sensors incorporated inside the machine. This will allow opportune preventive action to avert propagation of build defects across multiple layers. In pursuit of this goal, a commercial LPBF machine was integrated with three types of sensors, namely, a photodetector, high-speed video camera, and SWIR thermal camera with the following objective: to develop and apply a spectral graph theoretic approach to monitor the LPBF build status from the data acquired by the three sensors. 371C Quality and Reliability Session Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Young Myoung Ko, Pohang University of Science and Technology, Pohang, 790-784, Korea, Republic of, youngko@postech.ac.kr 1 - A New Multi-response Optimization Approach using Bayesian Seemingly Unrelated Regression Model Jianjun Wang, Associate Professor, Nanjing University of Science and Technology, Nanjing, 210094, China, jjwang@njust.edu.cn This paper presents a new multi-response optimization (MRO) approach integrating a quality loss function with a posterior predictive function in a single framework of Bayesian seemingly unrelated regressions (SUR) modeling and optimization. The proposed approach not only measures the conformance probability through the posterior predictive function, but also takes into account the expected quality loss with the quality loss function. The results show that the proposed approach may give more reasonable results than the existing approaches when both conformance probability and quality loss are important issues. 2 - Hard Failure Prediction with Stochastic Degradation Signals using Wiener Process Jianing Man, City University of Hong Kong, Dept of SEEM, General office, Room P6600,6/F, AC1,City U. of Hong Kong, Kowloon Tong, Kowloon, Hong Kong, jianinman2-c@my.cityu.edu.hk, Qiang Zhou Remaining useful life is one of the most important part for the conditional based maintenance (CBM) and prognostics and healthy management (PHM). In this paper, we consider the stochastic process for the degradation path using wiener process to predict the RUL for hard failure, with degradation signals and time-to- event data. The time-to-event data is modeled by the proportional hazards (PH), by considering the stochastic feature of the degradation signals. The simulation studies and a case study are conducted to show the performance of the method. 3 - Introducing Randomness into Control Chart Sampling Strategies to Better Protect Against Cyber-physical Attacks on Production Systems Lee Wells, Western Michigan University, 2827 Daventry Ave., Portage, MI, 49024, United States, lee.wells@wmich.edu, Ahmed Elhabashy, Bill Woodall, Jaime Camelio With recent technological advancements, production systems have become more susceptible to cyber-attacks. Such attacks can be designed to both alter product dimensions and avoid detection by current Statistical Process Control tools. This work discusses a novel approach to better detect attacks against manufacturing, which is based upon introducing randomness into control chart sampling strategies. Several different random sampling strategies will be introduced and their performances under varying attack scenarios will be analyzed. In addition, this work will propose new control chart performance metrics for analyzing situations where cyber-physical attacks are possible. 4 - Condition-based Selective Maintenance Optimization for a Large-scale Non-Markovian System Euihwan Lee, Pohang University of Science and Technology, 77, Cheongam-Ro, Nam-Gu, Pohang, 790-784, Korea, Republic of, euihwan.lee@postech.ac.kr, Young Myoung Ko, Eunshin Byon We consider selective maintenance that repairs severely degraded units in the system consisting of massive units. Under the assumption that units degrade independently in a finite number of states, we derive a fluid model that approximates the mean behavior of the system’s health condition. Our simulation study indicates that even if only a subset of units gets repaired, the system would asymptotically become a regenerative process as the maintenance operations are repeated over time. Based on this observation, we optimize the maintenance scheduling that triggers the maintenance operations based on the fraction of units at each degradation state in order to minimize long-run maintenance costs. WC68

371D Optimization, Convex Contributed Session Chair: Matthew Saltzman, Clemson University, Clemson, SC, United States, mjs@clemson.edu 1 - Optimization Based Heuristics for the Graph Partitioning Problem The graph partitioning problem is concerned with partitioning the vertices of a graph into a pre-specified number of clusters, so as to minimize the total weight of the edges cut. We use optimization formulations to design two approximation algorithms that can be used in large scale graph partitioning problem. Using valid inequalities and lower bounds play an important role in strengthening the solution of the Linear Programming relaxation of the problem which has been used in these two proposed methods. 2 - Optimal Number of Inspection and Rework Servers for Feed-back Queues in Steel Pipe Manufacturing Shahriar J. Hossain, Graduate Research Assistant, Louisiana State University, 4141 Burbank Dr. Apt. 3, Baton Rouge, LA, 70808, United States, msjhossain1@gmail.com, Bhaba R. Sarker This research aims to find the optimum number of inspection and rework servers in a steel pipe manufacturing system. If a repairable defect is detected at the inspection station, the pipe is sent to the rework station, while the reworked pipes are fed back to the inspection station for re-inspection. Thus, two intertwined queues of pipes are generated. The size of queues and number of servers are constrained by the availability of space. The total cost function is formulated as a nonlinear programming problem. The model is solved optimally to minimize the total cost. 3 - Irregular Shape Nesting via Branch and Bound using Custom Relaxations Akang Wang, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, PA, 15213, United States, akangw@andrew.cmu.edu, Christopher Hanselman, Chrysanthos Gounaris We present a custom-built linear relaxation and an associated branch-and-bound (BB) approach for solving irregular shape nesting problems to global optimality. We utilize a novel branching scheme based on circle-packing constraints, and we integrate a dynamic circle-adding procedure into our BB implementation, in order to enforce shape non-overlap. We also utilize a domain reduction scheme to speed up our algorithm. Extensive computational studies are performed on benchmark instances. 4 - Lagrangian Relaxation of the Defender Attacker Defender Constrained Shortest Path Problem Gary Lazzaro, Permanent Military Professor, United States Naval Academy, 1903 Baltimore Annapolis Blvd., Annapolis, MD, 21409, United States, lazzaro@usna.edu, Matt Carlyle The first application of tri-level Defender Attacker-Defender (DAD) optimization models to the constrained shortest path (CSP) problem. The side constraint in DAD CSP breaks the network structure of a shortest path problem into a challenging tri-level integer program. We create new solution procedures by merging the attacker model with Lagrangian relaxation of the operator model into a single formulation that can obtain fast heuristic solutions. We combine our heuristic algorithm with traditional methods to obtain provably optimal or near- optimal solutions. Our results on medium and large networks show that our innovations can significantly outperform traditional nested decomposition. 5 - Open-source Solvers for Quadratic Assignment and Related Problems Matthew J. Saltzman, Associate Professor, Clemson University, Dept of Mathematical Sciences, Martin Hall Box 340975, Clemson, SC, 29634-0975, United States, mjs@clemson.edu, Peter M. Hahn, Amir Roth, Monique Guignard A recent COIN-OR poject makes available exact solvers for the QAP and related problems of increasing interest, with wide application, and for which exact solution is very challenging. These problems address economic issues in the location and design of factories, hospitals, depots, transportation hubs and military bases.Our exact solvers produce strong bounds using Adams-Sherali RLT and include techniques to speed up branching and fathoming of search trees. We describe these codes and suggest a taxonomy and framework for combining existing methods and computer tools that can be modified and extended to make inroads in solving this growing class of optimization problems. Ali Hassanzadeh Kalshani, Graduate Student, University of California-Irvine, 1723, Verona Place, Irvine, CA, 92617, United States, ahassan2@uci.edu, Emre Alper Yildirim

526