Informs Annual Meeting 2017

TC16

INFORMS Houston – 2017

TC16

TC17

332F Operations Management Contributed Session Chair: Yen-Ju Chiang, National Cheng Kung University, Tainan, Taiwan, r48031016@mail.ncku.edu.tw 1 - Recurrent Neural Networks with Long Term Dependencies in Machine Tool Wear Diagnosis and Prognosis Jianlei Zhang, Research Assistant, North Carolina State University, Raleigh, NC, United States, jzhang45@ncsu.edu, Binil Starly Automated real-time tool condition monitoring is critical to improving productivity and part quality in manufacturing processes. This paper proposes a Recurrent Neural Network (RNN) approach for online diagnosis and prognosis of the tool wear. It avoids an analytic model for specific tool wear model, and aims to capture the long term dependencies. Without increasing the complexity of the Neural Networks, our approach can realize multi-step ahead tool wear prediction and Remaining Useful Life forecast. This is realized by applying real time indirect measurements obtained from vibration signals and power signals during the machining process. 2 - Comparing Multiple Linear Profiles for Determining Alternative Processes Chen-ju Lin, Associate Professor, Yuan Ze University, No. 135 Yuan Tung Rd., Taoyuan, 320, Taiwan, chenju.lin@saturn.yzu.edu.tw, Yuan-Chung Chuang A product can be manufactured by using alternative processes to reach a predefined product specification. Having several alternative processes can have flexible production scheduling, increase capacity utilization, or reduce lead time. To maintain consistent product performance, this research aims at identifying the alternative processes having similar quality performance to the current production process. Linear profile data for quality characteristic is investigated. A step-up procedure is developed to distinguish profiles from a controlled profile. The simulation results show that the proposed method is more powerful in differentiating profiles than using Bonferroni adjustment. 3 - A Comprehensive Learning Model for Determining Optimal Investments in a Dyadic Chain Didun Peng, Purdue University, Krannert Graduate School of Management, west lafayette, IN, 47906, United States, peng67@purdue.edu, Robert Plante, Jen Tang We explore the optimal investment in quality improvement efforts within a centralized dyadic supply chain, in which a manufacture outsources product components to several suppliers. The end product’s quality depends on several performance measures, each of which is affected by the interaction of the various supplier sourced components. Manufacturer and suppliers both incur quality- related cost, encouraging both parties to invest in quality-improvement efforts for each component. We attribute quality improvement and concurrent degradation to suppliers’ comprehensive learning curve, which considers autonomous and induced learning, and their respective knowledge depreciation. 4 - Product Branding in the International Context a Case of Norwegian Fashion Apparel Manufacturing Industry Rhythm Wadhwa, NTNU, Teknologiveien 22, Gjoevik, 2821, Norway, rhythmwadhwa@hotmail.com The paper investigates the product branding and supply chain issues of Norwegian fashion brands and their positioning in the international apparel retail industry. The research questions explored include current state challenges, knowledge and perception of brands, regional differences regarding fashion and consumer behavior. Qualitative semi structured interviews and survey research methodology was used for investigation. 5 - The Myths and Clarification for the Theory of Constraints Kan Wu, Nanyang Technological University, School of MAE, 50 Nanyang Ave, Singapore, 639798, Singapore, kan626@gmail.com The theory of constraints was proposed in the mid-1980s and has had a significant impact on the productivity improvements in many manufacturing systems. While it is intuitive and easy to understand, its conclusions are mainly derived from deterministic settings or based on the first moment results. Since production systems are stochastic in general, some of its conclusions are not rigorous and have to be modified. In this study, we show that the process of ongoing improvement may lead to unfavorable outcomes and a throughput bottleneck should be planned for certain types of machines. Specifically, if a system has no bottleneck, every station will be a bottleneck. 6 - Market Anomalies of Open Innovation Yen-Ju Chiang, Student, National Cheng Kung University, No.1, University Road, Tainan, 701, Taiwan, r48031016@mail.ncku.edu.tw, Shuang-Shi Chuang Past research focus on close innovation, bus we emphasize open innovation which means cooperate with other organizaions by their knowledge. And we adopt the

340A Scheduling & Queueing Sponsored: Applied Probability Sponsored Session Chair: Mor Harchol-Balter, Carnegie Mellon University, Pittsburgh, PA, 15213, United States, harchol@cs.cmu.edu 1 - Scheduling on Queues under Imperfect Information: Stability and Robustness Zheng-Yuan Zhou, Stanford University, Stanford, CA, 94305, United States, zyzhou@stanford.edu, Mehdian Saied, Nicholas Bambos Scaling up schedulers in computing systems (e.g., data centers) and communication networks may require sampling queue depths at random times and/or communicating the information with substantial random delays over relatively large distances to the schedulers. Imperfect state information (noisy, delayed, partial) can severely affect system performance in terms of backlog congestion and even queue stability. We discuss canonical scheduling examples and examine their stability and performance under increasingly “looser” state information in the interest of evaluating their robustness. 2 - Optimally Scheduling Jobs with Multiple Tasks of Unknown Duration Ziv Scully, Carnegie Mellon University, Pittsburgh, PA, 15213, United States, zscully@cs.cmu.edu, Guy Blelloch, Mor Harchol-Balter, Alan Scheller-Wolf We consider optimal job scheduling where each job consists of multiple tasks, each of unknown duration, with precedence constraints between tasks. A job is not considered complete until all of its tasks are complete. Traditional heuristics, such as favoring the job of shortest expected remaining processing time, are suboptimal in this setting. Furthermore, even if we know which job to run, it is not obvious which task within that job to serve. In this talk, we characterize the optimal policy for a class of such scheduling problems. We show the policy is simple to compute in many practical cases. 3 - Near Optimal Performance via Fixed-width Parallelization Benjamin Berg, Carnegie Mellon University, Pittsburgh, PA, 15213, United States, bsberg@cs.cmu.edu Running jobs in parallel is an excellent way to reduce their mean response time. In applications such as jobs running on multiple cores, the user chooses the level of parallelization for her jobs. We propose instead that the system should choose the level of parallelization so as to minimize mean response time across jobs. Interestingly, we find that a fixed level of parallelization suffices. While one might imagine that a system should dynamically change a job’s level of parallelization based on the current system state, our work shows that the right static level of parallelization yields near-optimal performance. Joint work with: Jan-Pieter Dorsman and Mor Harchol-Balter 340B Special Session: Machine Learning Sponsored: Applied Probability Sponsored Session 1 - Machine Learning and Operations Research Gah-Yi Ban, London Business School, Regent’s Park, London, NW1 4SA, United Kingdom, gban@london.edu Machine Learning for Applied Probability Society. TC18

342