Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

SD71

1 - Intrinsic Decomposition of Acoustic Emission and Nonparametric State Machine Model for Real-time Monitoring of Precision Micro- machining Process Zimo Wang, Texas A&M University, Industrial & System Engineering Dept, college station, TX, 77843, United States, Satish Bukkapatnam We present an approach using intrinsic time-scale decomposition and Dirichlet processes Gaussian state machine to analyze the transient behaviors of Acoustic Emission signals. Experimental studies showed that the presented approach could timely detect changes when the precision micro-machining process undergoes variations on material structures and/or fundamental cutting mechanisms. 2 - Defect Classification using Ensemble Convolutional Neural Network in Semiconductor Manufacturing Hyung-Seok Kang, Samsung Electronics, Hwaseong-si, Gyeonggi-do, Korea, Republic of, Jaewoong Shim, Kim Kil Soo, Seung Hoon Tong In semiconductor manufacturing, visual inspection is a fundamental process for defect detection and classification. In the case of 3D-Stacked DRAM with VIA (vertical interconnect access) technology, electrical inspection can be performed after assembly, so that defects in the VIA process can be detected only by appearance. The inspection equipment shoots multiple images at different angles with high and low magnification for failure analysis. In this study, we apply the ensemble Convolutional Neural Network (CNN) which utilizes multiple images in combination, and also conduct experiments on open set recognition that recognizes untrained types considering the actual environments. 3 - Prioritize Interaction Effects on Wafer Defects for Multistage Semiconductor Fabrication Based on Applied Association Rule Mining Jinsik Kim, Samsung Electronics, Hwaseong, Gyeonggi-do, Korea, Republic of, Jaewoong Shim, Chanhwi Jung, Doh Soon Kwak, Kunhan Kim, Seung Hoon Tong The methodology of finding the single cause of defects in semiconductor manufacturing has been studied a lot. However, as the product matures, the single causes are largely resolved and necessity of the search for the interaction of two factors emerges. The number of combinations of two or more factors in a complex manufacturing process is enormous, so it is impossible to inspect all cases. We propose a methodology to identify interaction of factors that leads to Wafer Defects using association rule mining algorithm. In addition, we validate this methodology by applying it on real-world data. 4 - A Scale-invariant Method for Quality Inspection of Objects Created by Additive Manufacturing Additive Manufacturing (AM) has shown great advantages in producing objects with complex geometric features. In broader applications, the functionality of printed parts is significantly affected by not only the quality but also the relative positioning of key features. In this research, a new scale-invariant quality profile is established based on the point clouds of printed parts obtained from a laser scanner. The new quality inspection method is able to quantify the fabrication non-conformity of individual features and the key features’ relative positioning. Numerical examples are provided to demonstrate the effectiveness and efficiency of the proposed quality inspection method. n SD71 West Bldg 106C Joint Session ICS/Practice Curated: Advances in Computational Stochastic Optimization and Applications Sponsored: Computing Sponsored Session Chair: Harsha Gangammanavar, Southern Methodist University, Dallas, TX, 75275, United States 1 - Monte Carlo Tree Search with Sampled Information Relaxation Dual Bounds Lina Al-Kanj, Princeton University, Olden Street, Sherrerd Hall, Office 115, Princeton, NJ, 08542, United States, Daniel Jiang, Warren B. Powell MCTS is a well-known strategy for solving sequential decision problems, particularly in the area of game-play AI. We propose a new technique called Primal-Dual MCTS that utilizes sampled information relaxation (Brown et. al., 2010) bounds on potential actions in order to make tree expansion decisions. The approach shows promise when used to optimize the behavior of a driver navigating a graph while operating on a ride-sharing platform. Yu Jin, Ph.D Student, University of Arkansas, 4207 Bell Engineering Center, Fayetteville, AR, 72701, United States, Haitao Liao

2 - Variable Sample-size Stochastic Approximation Scheme for Two-stage Stochastic Economic Dispatch Wendian Wan, Pennsylvania State University, 351 Leonhard Building, University Park, PA, 16802, United States, Uday Shanbhag, Mort David Webster This talk introduces the development of variable sample-size stochastic approximation schemes for two-stage stochastic convex programs. Our focus lies in applying this class of schemes towards the solution of a broad class of adaptive stochastic decision-making problems arising in the operation of large-scale power systems. In this talk, we consider the stochastic economic dispatch problem. In such problems, a first-stage dispatch is made contingent on taking a recourse decision when the uncertainty reveals itself. Preliminary numerics reveal that the proposed schemes provide accurate solutions but require far less time than traditional approaches like standard stochastic approximation. 3 - Stochastic Decomposition for Two-stage Stochastic Linear Programs with Random Cost Coefficients The Stochastic Decomposition (SD) algorithm has been a computationally proficient tool to tackle real-scale stochastic optimization problems arising in practical applications. In this talk we present new enhancements to this sequential sampling-based algorithm to address two-stage stochastic linear programs with random cost coefficients. We demonstrate their performance through results from our computational experiments. 4 - Simulation-based Hybrid Stochastic Approximation using Common Random Numbers Marie Chau, Virginia Commonwealth University, 1015 Floyd Avenue, Richmond, VA, 23220, United States, Jong J. Lee, Michael Fu Common Random Numbers (CRN) is a variance reduction method, which can be used to increase the typical O(n-1/3) convergence rate of gradient-free stochastic approximation (SA) to match the optimal O(n-1/2) convergence rate of gradient- based SA. Secant-Tangents AveRaged (STAR) and adaptive Secant-Tangents AveRaged (aSTAR) simultaneous perturbation SA are hybrid methods, which combine gradient-free and gradient-based SA algorithms. By applying CRN assumptions, STAR and aSTAR can achieve the optimal asymptotic convergence rate under milder conditions. n SD72 West Bldg 211A Decision Process: Practical and Methodological Issues in MCDM/A Sponsored: Multiple Criteria Decision Making Sponsored Session Chair: Adiel Teixeira De Almeida, Universidade Federal de Pernambuco, Recife PE, 50630-970, Brazil Co-Chair: Danielle Costa Morais, Universidade Federal de Pernambuco, Recife - PE, 52020-212, Brazil 1 - Preference Elicitation for Multicriteria Group Decision Making using FITradeoff Adiel Teixeira De Almeida, Universidade Federal de Pernambuco, Cx Postal 7462, Recife PE, 50630-970, Brazil, Eduarda Asfora Frej Preference elicitation of Decision Makers (DMs) is one of the most relevant tasks in multicriteria group decision making (MCGDM). The facilitation process uses knowledge in the intersection of several topics, such as: analytical models, cognitive process of DMs and the organizational and social interaction of a DMs’ group. In this work the decision process considers a framework for building multicriteria decision models. FITradeoff (Flexible and Interactive Tradeoff) method is used for preference elicitation in this MCGDM process, with criteria aggregation by an additive model in MAVT scope, using partial information. A decision support system, available at www.fitradeoff.org, is applied. 2 - Group Decision Model for Logistics Problem using Ordered Weighted Distance José Leao Silva-Filho, Master, Universidade Federal de Pernambuco, Recife, Brazil, Danielle Morais In corporative scenario, logistics is characterized by extreme competition, requiring efficiency in its processes. Decisions in this field can be complex depending on their strategic level and the number of Decision-Makers involved. This study proposes a group decision model for strategic logistics problem, using ordered weighted vectors. A numerical simulation of a real company relocation is presented to illustrate the applicability of the model. Harsha Gangammanavar, Southern Methodist University, Department of EMIS, P.O. Box 750123, Dallas, TX, 75275, United States, Yifan Liu, Suvrajeet Sen

116