Using an encoder-decoder convolutional neural network to predict the solid holdup patterns in a pseudo-2d fluidized bed

H. Bazai, E. Kargar, M. Mehrabi*

*Corresponding author for this work

Research output: Contribution to journalArticleScientificpeer-review

Abstract

In this paper, the capability of combined use of computational fluid dynamics (CFD) and data-based deep learning to predict fluidized beds' complex behavior without solving transport equations is being examined. A convolutional neural network (CNN) is trained to anticipate fluidized bed volume fraction contours based on the numerical simulations' results and data-based machine learning. The trained CNN receives the first ten frames from the CFD as input and predicts the next frame. This process continues until all the required frames are obtained. The results show CNN's superior spatial learning capability and how its combination with CFD can reduce the required computational power without compromising accuracy.
Original languageEnglish
Article number116886
Number of pages7
JournalChemical Engineering Science
Volume246
Early online date24 Jun 2021
DOIs
Publication statusE-pub ahead of print - 24 Jun 2021
MoE publication typeA1 Journal article-refereed

Keywords

  • CFD
  • Deep learning
  • Fluidized bed
  • convolutional neural networks

Fingerprint

Dive into the research topics of 'Using an encoder-decoder convolutional neural network to predict the solid holdup patterns in a pseudo-2d fluidized bed'. Together they form a unique fingerprint.

Cite this