Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

TA68

n TA69 West Bldg 106A Modeling, Monitoring, and Prediction in Complex Systems Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Xiaochen Xian, Madison, WI, 53705, United States Co-Chair: Kaibo Liu, UW-Madison, Madison, WI, 53706, United States 1 - Radiomics 2.0: Integration of Machine Learning and Mechanistic Models for Brain Tumor Mapping Using Multiparametric Magnetic Resonance Imaging Nathan B. Gaw, Graduate Research Associate, Arizona State University, 699 S. Mill Ave, Tempe, AZ, 85258, United States, Jing Li Glioblastoma (GBM) is the most aggressive type of brain cancer. Using multiparametric Magnetic Resonance Imaging (MRI) to quantify intra-tumor cell density distribution for each patient is critical for precision treatment. This falls in the field of “radiomicsö. Both data-driven machine learning (ML) and cell- biology-based mechanistic models have been developed, but each model has its limitations. We develop a first-of-its-kind hybrid model that integrates semi- supervised ML and a mechanistic model of GBM cell proliferation and invasion (PI), namely the ML-PI model. ML-PI achieves higher accuracy than ML or PI used alone, and produces biologically compliant cell density maps. 2 - A Generic Health Index Approach for Multisensor Degradation Modeling and Sensor Selection Minhee Kim, University of Wisconsin-Madison, Madison, WI, United States, Changyue Song, Kaibo Liu With recent development in sensor technology, multiple sensors have been widely adopted to monitor the degradation of a single unit simultaneously. This study proposes a novel data fusion method that constructs a one-dimensional health index via automatically selecting and combining multiple sensor signals to better characterize the degradation process. In particular, we develop a new latent linear model that constructs the health index and selects informative sensors in a unified manner. Simulation studies and a case study are presented to illustrate the effectiveness and evaluate the prognostic performance of the proposed method. 3 - Causation-based Monitoring and Diagnosis for Multivariate Categorical Processes with Ordinal Information Xiaochen Xian, Madison, WI, 53705, United States, Jian Li, Kaibo Liu In this paper, we leverage Bayesian networks to characterize Multivariate Categorical Processes (MCPs) with a causal structure, where the categorical variables can be either nominal, ordinal, or a combination of both. We develop one general control chart and one directional control chart, both of which fully exploit the causal relationships and the ordinal information for better process monitoring and diagnosis. Numerical simulations have demonstrated the superiority and robustness of our method in detecting and diagnosing the conditional probability shifts of nominal factors as well as the conditional latent location shifts of ordinal factors. 4 - Change Point Detection in a Multivariate Process Rong Pan, Arizona State University, School of Computing Informatics & Decison Sys, P.O. Box 878809, Tempe, AZ, 85287- 8809, United States, Steve Rigdon In this talk we present a computational Bayesian method for change point detection and compare its performance with other traditional control charts.

4 - Exploring the True Meaning Behind Words: The Lexicon Creation for Investor Sentiment

Keli Xiao, Stony Brook University, College of Business, Stony Brook, NY, 11794, United States, Liang Zhang

We develop an effective lexicon based on multi-sources for accurately differing positive and negative sentiments of investors, and help direct financial comments to their actual meanings. Specifically, we apply an efficient prediction-based neural network model to produce domain-specific lexicons for short financial texts, such as bullish and bearish comments in tweets, blogs and news headlines. We evaluate the generated investor sentiment lexicon based on the performance in unsupervised classification and supervised linear classification, as well as in a nonlinear deep learning method for sentiment classification. Our results shows that new lexicons outperforms existing lexicons.

n TA68 West Bldg 105C

Advanced Maintenance Models Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Yisha Xiang 1 - Maintaining Systems with Heterogeneous Spare Parts David Tarek Abdul-Malak, University of Pittsburgh, 1702 Jancey Street, Pittsburgh, PA, 15206-1146, United States, Jeffrey P. Kharoufeh We present a periodic maintenance model for systems maintained with spares that originate from a heterogeneous population, i.e., one with multiple system qualities that are visually indistinguishable. At inspection epochs, a failed system can be correctively repaired or replaced, and a non-failed system can be preventively repaired, replaced, or allowed to continue operating. Repairs place the same system back into service, allowing an operator to update their belief about the active system’s quality and utilize this information for decision making. A mixed observability Markov decision process model is formulated and structural results, as well as numerical illustrations, are presented. 2 - Maintenance Model for a Stochastially Degrading System with Individually Repairable Components Nooshin Yousefi, Rutgers University, Piscataway, NJ, United States, Jingyu Chen, David W. Coit An optimal maintenance model is presented for systems with individually repairable components subject to degradation and random shocks. Previous research often focuses on individual components or one component systems, or systems of components that are replaced at the same time. In practice, it is more economical to repair individual components. However, when this happens, each component within the system can have a different age at the beginning of an inspection interval, and once the system gets into steady-state behavior, the component ages become independent. To minimize system cost rate, optimal on- condition thresholds and a system inspection interval are determined. 3 - Condition-based Maintenance Optimization for Multi-component Systems: A Stochastic Programming Approach Yisha Xiang, Lamar University, 2626 Cherry Engineering Building, Beaumont, TX, United States, Zhicheng Zhu Condition-based maintenance (CBM) is an effective way to reduce system failure and operating costs with real-time information incorporated. However, most studies of CBM focus on single-component systems, which are not applicable to multi-component systems due to the various interactions between components. In this research, we propose a novel multi-stage model by using stochastic programming, which aims at providing optimal maintenance decisions for complex multi-component systems. Chance constraints are incorporated into this model to guarantee the satisfaction of predetermined budget and the requirement of system’s reliability and availability. 4 - Do Routine Breast Cancer Screenings Save Lives? A Maintenance-based Model for Evaluating Current Screening Suzan Alaswad, Zayed University, Khalifa City A, Office FF1-2- 010, Abu Dhabi, United Arab Emirates, Sinan Salman There is controversy regarding breast cancer screening - many scientists debate on the benefits of routine screenings on reducing deaths. In this paper, we evaluate the efficacy of current breast cancer screening policies using a condition-based maintenance model. Screening policies similar to inspection policies can be used to reduce cancer mortality rates. In this study, we model survival time from cancer diagnosis to breast cancer mortality using hypertabastic proportional hazards model. The progression of breast cancer is modeled using a partially observable Markov model. The combined model is used to find the reduction of breast cancer mortality under different screening policies.

268