Informs Annual Meeting Phoenix 2018

INFORMS Phoenix – 2018

WC68

3 - Driver Retention in Ride-sharing Services: An Empirical Study Kyungsun (Melissa) Rhee, University of Washington, Seattle, WA, 98105, United States, Jinyang Zheng, Yong Tan, Fei Ren With the growing popularity of global sharing economy market, on-demand ride- sharing services have become an important transportation channel. However, there is a rising concern on retaining drivers in ride-sharing platforms; they do not necessarily earn a significant amount of profit due to an increasing competition between the drivers and frequent order cancellations. By leveraging data from Chinese leading on-demand ride-sharing platform, we examine how spatial and temporal characteristics of passengers’ trips affect drivers’ behaviors of using ride-sharing apps. We extend the investigation to find out how drivers such behaviors affect their decision on staying/leaving the platform. n WC68 West Bldg 105C Joint Session QSR/Practice Curated: Data Analytics for Advanced Manufacturing Systems Sponsored: Quality, Statistics and Reliability Sponsored Session Chair: Wenmeng Tian, Mississippi State University, MS, 39762, United States Co-Chair: Hongyue Sun, University at Buffalo, Buffalo, NY, 14260, United States 1 - Spatial-temporal Modeling for the Additive Manufacturing Process of Ti-6Al-4V Weihong Guo, Rutgers, The State University of New Jersey, 96 Frelinghuysen Rd, CoRE Rm 220, Piscataway, NJ, 08854, United States, Shenghan Guo, Linkan Bian This study focuses on modeling the data indicating temperature and heat transfer within the melt pool and heat affected zone atop a thin-walled structure of Ti- 6Al-4V during its additive manufacture. A spatial-temporal model is developed to quantify the temperature, heat transfer, and laser-induced, powder-fed melt pool characteristics. The proposed method is integrated with Bayesian inferences for online adaptive modeling and monitoring. A novel control chart is developed to monitor the model parameters to support decision making. The performances of several spatial-temporal models are compared using experimental data. 2 - Customized Modeling for Multi-stage Additive Manufacturing Systems Hongyue Sun, University at Buffalo, 319 Bell Hall, Industrial and Systems Engineering, Buffalo, NY, 14260, United States, Ramanarayanan Purnanandam, Chi Zhou Most existing works on additive manufacturing (AM) focus on the single-stage printing at a local 3D printer, and fail to analytically quantify the effect of material preparation and post-processing stages on the printed part quality. To fully investigate AM, multiple stages (both AM and its up-stream and down-stream processes) need to work synergistically. However, the complex functional variable relationships among multiple stages pose challenges for conventional data analytics models. In this work, we propose customized models to address the problem. Both simulation and a constrained-surface stereolithography (SLA) process are used to demonstrate the proposed method. 3 - Online Monitoring for Additive Manufacturing Processes Based on Image Sequence Analysis Wenmeng Tian, Mississippi State University, P.O. Box 9542, Mississippi State, MS, 39762, United States, Mehrnaz Esfahani, Linkan Bian The melt pools in thermal images are regarded as the most informative process signature in metal-based additive manufacturing (AM) processes, and thus can be used for real-time process monitoring. However, how to effectively extract features from a series of melt pools for anomaly detection in each fabricated layer is still an open question. In this work, we propose a novel feature extraction approach by formulating the melt pool contours of one entire layer as a 3D space- time object. Low dimensional features can be extracted to characterize the shape variability in the 3D object. A case study based on a thin wall fabrication process is used to validate the proposed approach.

n WC69 West Bldg 106A Statistical Methods for Modern Reliability Data Analysis Sponsored: Quality, Statistics and Reliability Sponsored Session

Chair: Wujun Si, Wayne State University, Detroit, MI, United States Co-Chair: Qingyu Yang, Wayne State University, Detroit, MI, 48202, United States 1 - A General Repair Model with Dynamic Covariate Information and Application in Optimal Maintenance Planning Wujun Si, Wichita State University, Wichita, KS, 67206, United States, Qingyu Yang We propose a novel general repair model when dynamic covariate information is presented. Based on the repair model, we develop an optimal maintenance planning strategy subject to dynamic covariates. A maximum likelihood estimation method is developed to estimate the model parameters. A simulation study is implemented to illustrate the developedmethods, and a real world case study is conducted to demonstrate the proposed model. 2 - Event Prediction for Individual Unit Based on Time-to-event Data Collected in Teleservice Systems Akash Deep, University of Wisconsin Madison, Madison, WI, United States, Dharmaraj Veeramani, Shiyu Zhou We present a semi-parametric method to predict the event occurrence for an individual unit in real-time using time-to-event data. The units we consider are prone to experience the event multiple times during their usage lifetime, and these events are critical to overall performance of the system. The hazard of event-occurrence is modeled using Cox PH model and the distinction of units is achieved using a frailty parameter. The method features an on-line updating scheme, therefore, it can provide real-time prediction of the occurrence of next event. We demonstrate the efficacy of frailty and the updating scheme through comprehensive numerical experiments and a case study based on real-world data. 3 - Analysis of Large Repairable System Reliability Data with Static System Attributes and Dynamic Sensor Measurement Xiao Liu, University of Arkansas Leveraging modern statistical learning and conventional repairable system reliability analysis methods, this work investigates a statistical analysis approach which integrates the Random Forests algorithm and the classical statistical reliability data analysis methods. 4 - Optimal Maintenance Policies for Multi-level Preventive Maintenance with Complex Effects Yisha Xiang, Lamar University, Yue Shi Existing maintenance literature mainly focus on a single type of preventive maintenance action and often make simple assumptions regarding the effects of maintenance activities. Many important maintenance effects are overlooked in these models, e.g., deterioration rate reduction and random duration of a maintenance effective period. In this research, we consider joint planning for multiple levels of preventive maintenance with complex effects, and formulate the problem as a Markov decision process. Structural properties of the optimal policies are investigated by minimizing the total expected discounted costs. Numerical examples are conducted to validate our proposed model. n WC70 West Bldg 106B Practice- Quality & Reliability Engineering I Contributed Session Chair: Mejdal Alqahtni, Rutgers Univ., Bound Brook, NJ, 08805, United States 1 - Power Systems Reliability Assessment with Variance Reduction and Markov Chain Monte Carlo Daniela B. Almeida, PSR, Centro Empresarial Rio, Praia de Botafogo 228 / 1701-A Rio de Janeiro, RJ, Rio de Janeiro, 22250- 145, Brazil, Carmen L. Borges, Gerson C. Oliveira The methodology combining Variance Reduction and Markov Chain Monte Carlo significantly reduces the computation time and number of samples required for reliability studies compared to the standard Monte Carlo simulation. Some results from a study with real power systems are presented to highlight the effectiveness of the proposed method.

492