期刊
CHEMICAL ENGINEERING JOURNAL
卷 409, 期 -, 页码 -出版社
ELSEVIER SCIENCE SA
DOI: 10.1016/j.cej.2020.128163
关键词
Reactor; A computational fluid dynamics (CFD); Deep Neural Network (DNN); Butadiene; Genetic Algorithm (GA)
资金
- National Research Foundation of Korea (NRF) - Korea government (MSIT) [62163-01]
- BK 21 FOUR program
- Korea Institute for Advancement of Technology (KIAT) - Korea Government (MOTIE) [P0008475]
This study proposes a unified framework for mathematical modeling, simulation, prediction, and optimization of chemical processes, utilizing Deep Neural Network and Genetic Algorithm for reaction condition optimization. The results indicate that the framework can achieve higher reactor performance indices compared to conventional optimization methods.
The fight against climate change and environmental pollution have set the agenda for efficient reaction systems design. It has necessitated now more than ever, the careful design and optimization of chemical processes that result in carbon production. In this study, a unified framework for the mathematical modelling and simulation, prediction, and optimization of reaction systems is proposed. A computational fluid dynamics (CFD) simulation of butadiene synthesis over a ferrite catalyst in a 3D shell and multi-tubular reactor was executed. A rigorous mathematical formulation of the process kinetics and multiscale heat transfer was incorporated into an Open-FOAM CFD model using a porous media. The CFD model was validated against experimental data using gas concentrations, temperature, and CO2 partial pressure with a maximum error of 3.2% (97% accuracy). The developed model was then used to generate data sets for both prediction and optimization of the reaction conditions in terms of temperature, flow rate, and feed compositions using Deep Neural Network (DNN) and Genetic Algorithm (GA). Another surrogate model was developed for temperature control (cooling side) optimization by utilizing the coolant flow rate, flow directions (co-current and countercurrent), coolant types (water and solar salt), and velocity. Dynamic evolution and steady-state contours of the species concentration distributions were predicted with our DNN model with only an error of 1.26% (98.74% accuracy). Our results clearly indicated that higher reactor performance indices that exceed those of the conventional optimization approach in terms of conversion (99.98%), yield (93.28%), and selectivity (92.3%) are obtainable with the proposed framework. The mathematical models, DNN surrogate models, and the genetic algorithms that were integrated as a unified framework can be adopted for designing other reactor systems.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据