|
Comparative Study on Battery State of Health Prediction Models with Manually Selected vs CNN-extracted features |
|
Suyeon Sohn, Ha Eun Byun, Hwayeon Kum, Jay H Lee(KAIST, Korea) |
| Accurate prediction of a lithium ion battery’s state of health (SOH) is important as it can provide clues on how charging and discharging patterns affect the performance degradation and alert the user of any future performance issue. Machine learning based SOH prediction methods have made significant progress, but most of them only cover a narrow range of cycling protocols and rely on manual feature extraction which can be case-specific. In this study, two SOH prediction models for a broader datasets are proposed: one uses manually extracted features and the other uses the convolutional neural networks (CNN) for automatic feature extraction. |
|
|
Modeling and Scale-up of Multi-tube Membrane Reactor for Hydrogen Production Using Computational Fluid Dynamics |
|
Beomsu Kim, Hongbum Choi, Jay H Lee(KAIST, Korea) |
| The multi-tube membrane reactor is simulated using the computational fluid dynamics (CFD) in order to investigate how the placement of the tubes in the reactor affect its performance. The single-tube membrane reactor is first designed and then extended to the multi-tube case. Hydrogen concentration profiles and temperature profiles of multi-tube membrane reactor are investigated with respect to the number of membrane tubes are increased gradually – to 2, 4, 6, 8, 10 and 12. High levels of hydrogen recovery and methane conversion are found to be maintained until the number of membrane tube is 6 before showing significant declines as more tubes are added. |
|
|
Design of Hydrogen Isotope Separation System based on Bayesian Optimization |
|
Jae Jung Urm, Damdae Park(Seoul National University, Korea), Jae-Uk Lee, Min Ho Chang(Korea Institute of Fusion Energy, Korea), Jong Min Lee(Seoul National University, Korea) |
| This paper presents a study on the design of hydrogen isotope separation system based on Bayesian optimization. A simulation program for a hydrogen isotope separation system process was developed in an optimization-based framework. It was utilized to calculate the total amount of tritium in a process design. A feed stage and an equilibrator attached stage was determined to minimize the total tritium holdup under three product quality constraints. For the Bayesian optimization problem, the product quality constraint were imposed as penalty functions. The optimal design was found at the 37th iteration, which is about 0.8% of the total number of evaluations required for brute search. |
|
|
Model-free Design of Experiments Framework for Optimal Process Design using Process Simulators |
|
Boeun Kim(Princeton University, United States), Kyung Hwan Ryu(Sunchon National University, Korea), Seongmin Heo(Dankook University, Korea) |
| Conceptual process design is a key step to the commercialization of novel technologies, where a network of reaction-separation-recycle processes is synthesized to produce target products from given raw materials. Such design is often realized using commercial process simulators. In this work, model-free design of experiment methods are implemented to optimize process designs developed using Aspen Plus: one using response surface method with central composite design, one using response surface method with Latin hypercube sampling, and one using Bayesian optimization. Performances of different methods are compared using representative chemical process examples. |
|
|
Development of a Surrogate Model Based on 2D First Principle Equations for Open Rack Vaporizer |
|
Suk Hoon Choi(Seoul National University, Korea), Dong Hwi Jeong(University of Ulsan, Korea), Jong Min Lee(Seoul National University, Korea) |
| Open rack vaporizer (ORV) is a heat exchanger that vaporizes LNG into natural gas using seawater as a heat medium. The fluid and heat flow analysis of ORV is generally performed through computational fluid dynamics (CFD), but it has a heavy computational burden, making it impossible to be used for optimization or control of the system. In this paper, a high fidelity surrogate model for the system is developed using parameterized two-dimensional (2D) first principle equations of heat transfer. Among the model parameters, key parameters to be estimated are selected using the list of parameter ranking by sensitivity analysis. The parameters are estimated and the results are analyzed. |
| |