Presentation Information

[3C03]Reduced-order Modeling of Heat Transfer in Gas–Solid Flows with Deep Learning

*Michael Castro1, Shuo Li1, Kai-en Yang1, Toshiki Imatani1, Mikio Sakai1 (1. UTokyo)

Keywords:

discrete element method,computational fluid dynamics,heat transfer,machine learning,surrogate modeling,proper orthogonal decomposition,long-short term memory,neural network

Reduced-order models (ROM) are essential to Industry 4.0 since they can generate real-time data about equipment. The ROMs have been developed for modeling gas-solid flows in solids handling equipment. Heat transfer must also be modeled in some equipment, but this has yet to be studied extensively. In this study, an ROM of a fluidized bed accounting for gas-solid flows and heat transfer is developed. To this end, a full-order model (FM) of the fluidized bed was formulated using coupled computational fluid dynamics and discrete element method (CFD-DEM) with heat transfer. Next, a Lanczos-based proper orthogonal decomposition (LPOD) is applied to decompose the FM simulation results into its modes and coefficients. Afterwards, the ROMs are developed by employing long-short term memory (LSTM) neural networks to predict the system dynamics within a reduced space spanned by a set of POD modes. It was found that LSTM-based ROMs can accurately predict the flow fields involved in gas-solid flows and heat transfer in bubbling fluidized beds. ROMs for other fluidization regimes will be developed in future work.

Comment

To browse or post comments, you must log in.Log in