2016 INFORMS Annual Meeting Program

WD66

INFORMS Nashville – 2016

3 - How Understanding The Sensitivity And Stability Of Preferences Among Colorectal Cancer Screening Alternatives Could Lead To “better” Medical Decisions M. Gabriela Sava, Assistant Professor, Clemson University, Clemson, SC, United States, msava@clemson.edu, Luis G Vargas, Jerrold H May, James G Dolan Patients are faced with multiple alternatives when selecting the preferred method for colorectal cancer screening, and there are multiple criteria to be considered in the decision process. We model patients’ choices using a multi-criteria decision model and propose a new approach for characterizing the idiosyncratic preference regions for individual and group of patients. We show how insights derived from the sensitivity and stability of patients’ preferences could be used within the medical decision making process. 4 - Quick Anp - A New Approach To Anp Sensitivity Analysis Elena Rokou, Chief Research Officer, Creative Decisions Foundation, Pittsburgh, PA, 15213, United States, erokou@creativedecisions.net, Bill Adams The proposed approach consists of two phases, in the first one each decision maker fills out a very short version of the ANP questionnaire. This way the initial point of views are collected by the negotiator. The initial questionnaires give the needed input to define what are the points of greater conflict and which judgments have a primary role in the final decision outcome. In the second phase the team focuses only on those conflicting points that have great impact on the outcome. The focal point of this work is to present a new type of sensitivity analysis for single level ANP models. WD66 Mockingbird 2- Omni Data Analytics For System Improvement III Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Abdallah A Chehade, University of Wisconsin-Madison, Madison, WI, United States, chehade@wisc.edu Co-Chair: Kaibo Liu, University of Wisconsin-Madison, 1513 University Ave, Madison, WI, 53706, United States, kliu8@wisc.edu 1 - Sensory-updated Failure Threshold Estimation For Remaining Useful Life Prediction Abdallah A Chehade, University of Wisconsin-Madison, chehade@wisc.edu, Kaibo Liu The rapid development of sensor technology led to significant research efforts in remaining useful life (RUL) prediction. Such efforts often consider that a unit fails when it crosses a failure threshold, which is estimated offline. Unfortunately, such failure threshold estimation may not be valid due to the stochastic nature of the underlying degradation mechanism. In this talk, we propose a novel data fusion model that combines the information from the degradation profiles of historical units and the in-situ sensory data from an operating unit to online estimate and update the failure threshold distribution of this unit in the field. This approach is expected to help better online predict the RUL. 2 - Hard Failure Prediction Based on Joint Models With Extended Hazard Jianing Man, City University of Hong Kong, Kowloon, Hong Kong, jianinman2-c@my.cityu.edu.hk, Qiang Zhou Remaining useful life (RUL) prediction is essential for the prognostics and health management (PHM) to guarantee the system performance. We use joint models for the individual units (or systems) which subject to hard failure, including the random effects model for degradation signals and the extended hazard (EH) model for time-to-event data. The EH model is a general model that includes the proportional hazards (PH) model and accelerated failure time (AFT) model as special cases. A two-stage method and a Bayesian updating scheme are used in the offline estimation and online prediction separately. 3 - To Integrate Or Not? Covariance Selection In Gaussian Process Modeling Ran Yang, Northwestern University, RanYang2011@u.northwestern.edu, Daniel Apley Power exponential, Gaussian, and Matern are the most commonly used covariances for Gaussian process modeling of simulation response surfaces. A recently proposed class of fundamentally different, integrated covariance functions has been shown to work remarkably well for simulation models of many real physical systems. We demonstrate that likelihood and leave-one-out cross validation can both reliably select the best covariance model for a given response surface and data set and, in particular, determine whether an integrated covariance function should be used.

4 - Diagnostic Monitoring And Fault Diagnosis In Large Scale Multivariate Process Via Compressive Sensing And Optimization Screening Yan Jin, University of Washington, yanjin@uw.edu, Shuai Huang Smart manufacturing has been an emerging concept in many industries that highlights unprecedented connectivity of manufacturing infrastructure and abundance of sensors for real-time monitoring of many system entities. While it provides a data-rich environment, how to effectively model the variations of system entities and synthesize decentralized information into global situational awareness have been challenging issues. To tackle these challenges, we propose an integrated framework that unifies multivariate statistical monitoring, compressive sensing, and convex optimization. The advantages of proposed method are demonstrated through both simulations and real world application. WD67 Mockingbird 3- Omni Dynamic Maintenance/Reliability Planning Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Anahita Khojandi, UTK, Address, City, TN, 00000, United States, khojandi@utk.edu Co-Chair: Murat Kurt, Merck & Co., Inc., Addrress, Pittsburgh, PA, United States, murat.kurt7@gmail.com 1 - Combined Condition-based Maintenance And Repairman Routing Optimization Bram de Jonge, University of Groningen, Groningen, Netherlands, b.de.jonge@rug.nl, Lisa M Maillart, Oleg A Prokopyev Existing studies on maintenance optimization for multiple machines generally ignore the required travel times to move from one machine to another. We consider the problem of a single repairman who is responsible for the maintenance activities of a set of geographically distributed machines with condition monitoring. The problem is formulated as a Markov decision process and insights are obtained on when to relocate and when to carry out preventive and corrective maintenance activities. 2 - Maintenance Of Degrading Servers Stored In A Stack Mahboubeh Madadi, Louisiana Tech, madadi@latech.edu, Lisa M Maillart, Charles Richard Cassady, Shengfan Zhang Inspired by queueing systems in which the servers are stored in a stack and arriving customers are served by the server on the “top” of the stack, we consider an M/M/n/n queue under a Most Recently Busy (MRB) service discipline in which the operating cost of each server increases in its cumulative time-in-use. More specifically, we formulate a continuous time Markov model to characterize the transient utilization of each server and to determine optimal maintenance policies of various forms. 3 - Condition-based Repair Prioritization In Repairable Inventory Supply Chains Chiel van Oosterom, Eindhoven University of Technology, Eindhoven, Netherlands, c.d.v.oosterom@tue.nl, Joachim Jacob Arts, Geert-Jan Van Houtum We propose a model for exploiting condition information to dynamically prioritize repairs in a capacitated repair shop. The repair shop supports a system with a number of different repairable components. The system is down whenever a component fails and no ready-for-use spare part is available for that component. The objective in prioritizing repairs is to maximize the long-run availability of the system. 4 - Joint Optimization Of Replacement And Inspection Decisions For Two-unit Standby Redundant Systems With Non-silent System Failures Anahita Khojandi, University of Tennessee, Knoxville, TN, United States, khojandi@utk.edu, Murat Kurt We consider a two-unit standby redundant system in which individual unit failures are silent, but simultaneous unit failures cause system shutdown. We propose a Markov decision process to jointly determine inspection frequency and preventive repair decisions to minimize the total expected operational cost, including inspection, repair and failure costs. We analytically establish properties of the value function and the optimal policy, derive insights from a wide range of numerical examples, perform extensive sensitivity analysis, and discuss algorithmic enhancements.

476