INFORMS 2021 Program Book

INFORMS Anaheim 2021

WE30

3 - Design Variable-Sampling Control Charts using Covariate Information Kai Yang, University of Florida, Gainesville, FL, 32610-3010, United States Statistical process control charts are widely used in the manufacturing industry for monitoring quality variables. In practice, the quality variables are often affected by certain covariates, and it should improve the performance of a control chart if the covariate information can be used properly. However, because of the complex relationship between the quality variables and the covariates, it is quite challenging to properly use covariate information in process monitoring. To address this problem, we propose a general framework to design a variable- sampling control chart by using covariate information. Our proposed chart is self-starting and can well accommodate stationary short-range serial data correlation. It should be the first variable-sampling control chart in the literature that the sampling intervals are determined by the covariate information. WE30 CC Room 207D In Person: Big Data Analytics in System Monitoring and Anomaly Detection General Session Chair: Yuanxiang Wang, University of Southern California, Los Angeles, CA, 90007, United States 1 - Penalized Linked Component Analysis for Spatial-Temporal Burst Detection in Water Distribution Systems Shenghao Xia, University of Arizona, Tucson, AZ, United States, Jian Liu, Kevin Lansey Detecting bursts from spatial-temporal (ST) hydraulic data is critical for water distribution system management. Traditional anomaly detection methods based on basis expansion are inefficient and inaccurate in detecting and localizing bursts from data continuously collected from multiple potential locations. This research proposes a new method based on Penalized Linked Component Analysis, which extracts ST anomaly features by differentiating the commonly shared normal features and individual anomaly features. Penalization is adapted in the algorithm to increase the sensitivity of location detection and reduce the rate of false alarm. The effectiveness of the proposed method is demonstrated with a simulated case study. 2 - Wavelet Gaussian Process for Monitoring Image Profiles Runsang Liu, Pennsylvania State University, State College, PA, 16803, United States, Hui Yang, Bryan D. Vogt Advanced sensing is increasingly invested in the manufacturing industry to cope with the complexity of new additive manufacturing (AM) technologies. However, imaging data from manufacturing processes have complex correlations and often contain hidden information that cannot be revealed on a single scale, which jeopardizes the effectiveness of image-based process monitoring. In this paper, we investigate the wavelet Gaussian Process monitoring of layerwise image data from multiscale spatial-frequency perspectives. The proposed methodology is evaluated and validated using both simulated data and real-world layerwise imaging data from an additive manufacturing process. 3 - Learning and Predicting Shape Deformation Through an Extended Convolution Learning Framework for Additive Manufacturing Yuanxiang Wang, University of Southern California, Los Angeles, CA, 90007-3952, United States, Cesar Alexander Ruiz, Qiang Huang Geometric accuracy control is critical for precision additive manufacturing (AM). To learn and predict the shape deformation of 3D printed products from a limited number of training samples, we extend the previously developed fabrication- aware convolution learning framework to a broader class of geometries by constructively incorporating spherical and polyhedral shapes into a unified model. The modeling approach generalizes the 2D cookie-cutter functions to 3D shapes and considers the spatial correlation among neighboring regions and across different shapes through a novel distance metric. The framework approximates 3D freeform shapes by using spherical and polyhedral patches. Case studies show the promise of the framework for shape accuracy control of complex geometries.

WE32 CC Room 208B In Person: Topics in RM General Session Chair: John G. Turner, University of California Irvine, University of California Irvine, Irvine, CA, 92697, United States 1 - Managing Retail Inventory and Pricing in the Presence of Stochastic Purchase Returns Alys Liang, Michigan Ross, Ann Arbor, MI, 48104, United States In US alone, returns cost retailers a total of hundreds of billions of dollars annually in the last few years. It is generally accepted in the industry that returns are inevitable and often considered as the necessary cost of doing business. In this paper, we consider a single warehouse/store joint inventory and pricing problem in the presence of stochastic purchase returns. A purchase returned could be re- stocked for resale after inspection. A key feature of our model is that we allow a general class of return time distributions. This problem is very challenging to solve optimally since we need to keep track the return status of all purchases made in the past. We propose an easy-to-implement joint inventory and pricing policy and show that it is near optimal in the setting with a large annual market size, which is a practically relevant setting for many product categories. 2 - Assortment Optimization with Multi-item Basket Purchase under The Multivariate MNL Model Chengyi Lyu, University of Colorado Boulder, Boulder, CO, United States, Stefanus Jasin, Sajjad Najafi, Huanan Zhang We incorporate customers’ multi-item purchase behavior into the assortment optimization problem which we consider under the Multivariate MNL (MVMNL) model. Under MVMNL, products are clustered into different groups, and a customer can simultaneously purchase from as many groups as possible, where at most one product gets selected from each group. We first show that the revenue- ordered assortment may not be optimal. Nonetheless, we show that under some mild conditions, a certain variant of this property holds (in the uncapacitated assortment problem) under the MVMNL model—the optimal assortment consists of revenue-ordered local assortments in each group. We develop FPTAS for capacitated and uncapacitated assortment problems. Our analysis reveals that disregarding multi-item purchase behavior can have a significant negative impact on a retailer’s profitability. Chair: Guanzhou Wei, Fayetteville, AR, 72701, United States 1 - Fine-Grained Spatio-temporal Pollution Forecasting and Hotspot Identification Using Low-cost Sensors in Delhi: A Three-Year Study Shiva R. Iyer, PhD Student, New York University, New York, NY, United States The city of Delhi has 32 air quality monitors over an area of about 900 sq km, but we do not have information on fine-grained variations in air quality in the city in order to reason about citizen exposure and identify hotspots. We have installed 28 low-cost sensors, many of them concentrated in the south Delhi region. We have developed a generic definition of “hotspots” in terms of spatio-temporal variations, using which we validate some known hotspots and discover new ones. We have also designed a novel model combining geostatistics and deep learning that is able to make spatio-temporal pollution predictions by the hour with an MAPE of about 10% across all locations. 2 - The Economic and Environmental Impact of Sharing Economy Business Model Fahimeh Rahmanniya, PhD Candidate, University of Tennessee, Knoxville, Knoxville, TN, United States, Paolo Letizia, Paolo Roma Sharing economy models involve owners renting out their poorly utilized assets to renters, generally through an online platform. In this research, we analytically study the impact of sharing economy business models on the firms’ profits and the environment. The analysis shows that sharing economy can be a win-win from these two perspectives, and it can also outperform other consumption models such as pure sales and servicizing. WE36 CC Room 210B In Person: Sustainability/Big Data Contributed Session

178