2016 INFORMS Annual Meeting Program

WA67

INFORMS Nashville – 2016

3 - Data-driven Diagnosis For Asthma Control Status In Smart Asthma Management System Based On Correlated Gamma-based Hidden Markov Model Junbo Son, Assistant Professor, University of Delaware, Newark, DE, United States, sonjunbo@gmail.com, Shiyu Zhou, Patricia Brennan Driven by the IoT, a smart asthma management (SAM) system has been implemented in practice. The SAM system includes rescue inhalers with a wireless connection and the system records the inhaler usage and transmits the data to a centralized server. To effectively manage the asthma, a statistical model based on the patient monitoring data from the SAM system is crucial. In this research, we propose a data-driven diagnostic tool for assessing underlying asthma control status of a patient based on hidden Markov model (HMM). The proposed correlated gamma-based HMM can visualize the asthma progression to aid therapeutic decision making and its promising features are shown in both simulation and case study. 4 - Reliability Analysis Considering Dynamic Material Local Deformation Wujun Si, Wayne State University, fk9456@wayne.edu, Qingyu Yang, Xin Wu We conduct reliability analysis utilizing dynamic material local deformation information. A novel multivariate general path model with a new variance-based failure criterion is proposed. A two-stage parameter estimation method is developed to overcome the computational complexity. Both simulation studies and physical experiments are conducted for verification and illustration. WA67 Mockingbird 3- Omni Data Analytics for System Improvement II Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Xi Zhang, Peking University, Beijing, China, xi.zhang@pku.edu.cn Co-Chair: Kaibo Liu, Universityof Wisconsin, Madison, 1513 University Avenue, Madison, WI, 53706, United States, kliu8@wisc.edu 1 - Statistical Process Control Of Stochastic Textured Surfaces Anh T Bui, Northwestern University, Evanston, IL, United States, atbui@u.northwestern.edu, Daniel Apley We develop a defect monitoring and diagnostic approach for manufactured products that have stochastic textured surfaces (e.g., textiles or material microstructures). We first use generic supervised learning methods to characterize the stochastic behavior of “normal” in-control samples of the textured surfaces. Based on the residuals of the supervised learning model applied to new samples in a statistical process control context, we propose two spatial moving statistics for detecting local aberrations in the textured surfaces. We illustrate the approach using simulated and real examples. 2 - Causation-based Process Monitoring And Diagnosis For Multivariate Categorical Processes Xiaochen Xian, the University of Wisconsin, Madison, WI, xxian@wisc.edu Statistical surveillance for multivariate categorical processes have attracted more and more attentions. In many applications, causal relationships may exist among categorical variables, where the shifts at upstream variables will propagate to their downstream variables. We employ Bayesian network to characterize such causal relationships and integrate it with the statistical process control technique. We propose two control charts for detecting shifts in the conditional probabilities of the multiple categorical variables that are embedded in the Bayesian network. Both simulation and real case studies are used to demonstrate the effectiveness of the proposed schemes. 3 - A Thermal Field Estimation Method Based On Spatial-temporal Dynamics Using Multi-channel Sensor Data Xi Zhang, Peking University, Beijing, China, xi.zhang@pku.edu.cn, Di Wang, Kaibo Liu Thermal field profile is one of the critical issues for the quality assurance of the grain warehouse. However, only limited sensors are afforded to characterize the dynamics in the grainhouse, leading to an inappropriate decision for grain maintenance. This article presents a field estimation approach to model spatio- temporal dynamics of warehouse temperature through integrating thermodynamics model and spatiotemporal stochastic processes. Specifically, we integrate a 3-D unsteady heat transfer model into a Gaussian Markov random field to achieve a parsimonious representation of spatial patterns. Simulation and real case are conducted to show the effectiveness of the developed method.

4 - Multivariate Ordinal Categorical Process Control Based On Log-linear Modeling

Jian Li, Xi’an Jiaotong University, Xi’an, China, jianli@mail.xjtu.edu.cn, Junjie Wang, Qin Su

The quality of products or services is sometimes measured by multiple categorical characteristics, each of which is classified into attribute levels such as good, marginal, and bad. There is usually natural order among these attribute levels. By assuming that each ordinal categorical quality characteristic is determined by a latent continuous variable, this work incorporates the ordinal information into an extended log-linear model and proposes a multivariate ordinal categorical control chart. Simulations show that the proposed chart is efficient in detecting location shifts and dependence shifts in the corresponding latent continuous variables of ordinal categorical characteristics. WA68 Mockingbird 4- Omni Statistical Models for Computer Experiments Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Qiong Zhang, Richmond, VA, United States, qzhang4@vcu.edu 1 - Efficient Gaussian Process Modeling For Computer Experiments Yibo Zhao, Rutgers State University of New Jersey, Piscataway, NJ, United States, yz346@scarletmail.rutgers.edu We study the problem of simultaneous variable selection and parameter estimation in Gaussian process models. Conventional penalized likelihood approaches are attractive but the computational cost of the penalized likelihood estimation (PMLE) or the corresponding one-step sparse estimation (OSE) can be prohibitively high as the sample size becomes large. This is because the likelihood function heavily involves operations of a covariance matrix of the same size as the number of observations. To address this issue, this article proposes an efficient subsample aggregating (subagging) approach with an experimental design-based subsampling scheme. The proposed method is computationally cheaper, yet it can be shown that the resulting subagging estimators achieve the same efficiency as the original PMLE and OSE asymptotically. The finite-sample performance is examined through simulation studies. Application of the proposed methodology to a data center thermal study reveals some interesting information, including identifying an efficient cooling mechanism. 2 - Change-point Detection For Spatial-temporal Organ Image Data Shuyu Chu, Virginia Tech, cshuyu@vt.edu, Xinwei Deng, Ran Jin The demand for organ transplantation increases rapidly, but only a limited number of viable organs is available. Poor preservation and evaluation cause many organs to be discarded. Current evaluation methods are often inaccurate or result in organ damage. There is a great need for accurate non-invasive evaluation methods. In this work, we focus on detecting quality changes in organs under preservation by only using biomedical thermal image data. Scalable Gaussian processes with expressive spectral mixture kernels is applied on the large multidimensional image data to conduct model fitting and inference. A real case study will be used to elaborate the performance of the proposed method.

3 - Asymmetric Process For Stochastic Simulation Qiong Zhang, Virginia Commonwealth University, qzhang4@vcu.edu

Quantiles serve as important measurements in stochastic simulation. In simulation practice, we need statistical methods to model these quantiles for optimization or calibration. However, the traditional Gaussian process model often fails to capture the behavior of quantiles if the sample path is not long enough. To resolve this issue, we will introduce the asymmetric process for modeling the quantiles in stochastic simulation. Numerical results will be provided to show the effectiveness of this new approach.

386