Session: 17-01-01: Research Posters

Paper Number: 144933

144933 - Deep Learning Assisted Design Optimization of Porous Structure for Enhanced Multiphase Heat Transfer With Oscillating Flow 

Multiphase flow and heat transfer in porous media occur in a variety of applications, including oil recovery, geothermal systems, and fuel cells. Furthermore, heat transfer can be enhanced in the presence of oscillating flows and engineered porous structures due to the increase of surface area and the improvement of enhanced convection, such as Stirling engines, pulsating heat pipes (PHP), and oscillating water columns (OWC). There is a diverse range of porous media in nature, and they are playing a vital role in terms of fluid and heat transfer in their ecosystem. The role of coral rock in the ocean is multifaceted, from ecosystems to regulating global climate. Coral reefs slow down ocean currents and waves, and the complex porous structure encourages water to move slowly between the reefs, which helps deposit sediment and keeps the water clear. Additionally, the oscillating flow of seawater through coral structures helps distribute heat throughout the ocean, which is critical for maintaining temperature balance in coral reef ecosystems.

However, due to the complex internal structures of the porous media, there are some limitations when exploring the multiphase heat transfer and oscillating flow in porous media, including computational cost and manufacturing constraints. Recently, Artificial Intelligence (AI) has been playing a transformative role in revolutionizing thermal engineering for design optimization and performance improvement. Researchers have been applying AI-assisted methods to generate and analyze engineered porous structures, such as topology optimization, Graph Neural Networks, and Variational Autoencoders (VAEs).

Inspired by the coral rocks and their living environment in terms of oscillating flow, in this work we will employ a coral rock for enhanced multiphase heat transfer. The coral rock will be scanned using Microcomputed tomography (micro-CT), and then image reconstruction will be performed to obtain the internal structure of the scanned coral rock. The typical pore size of the coral rock we applied is ranging from 5 μm to 5 mm. We will apply a trained Generative Adversarial Networks (GAN) model to learn and regenerate the porous structure based on computational fluid dynamics (CFD) simulations. The GAN will be trained based on the CT-scanned coral rock image stacks. The reconstructed 3D internal structures will be used for multiphase oscillating flow simulation using Palabos, a Lattice Boltzmann method (LBM) based open-source software for mesoscopic scale simulations. To accelerate and improve the efficiency of structure evaluation, we will apply a trained Convolutional Neural Networks (CNN) model to evaluate the multiphase oscillating flow and thermal performance of the given porous structure. The GAN model will learn from the database and then generate a predicted porous design. The predicted porous design will then be evaluated by the CNN model for its multiphase heat transfer performance, and then the predicted performance will be sent to the loss function. Therefore, the loss function will be treated as the input of the optimizer, where the optimizer will give GAN feedback and go to the next loop until the epoch number achieves the expected setting. Eventually, we will have an optimized porous structure design based on our database.

The implementation of our hybrid deep learning model facilitates an understanding of the flow and heat transfer within porous media, enabling the prediction and optimization of designs for enhancing multiphase heat transfer with oscillating flows that are inspired by natural coral structures.

Presenting Author: Lichang Zhu University of Houston

Presenting Author Biography: Lichang Zhu is currently a PhD student under the supervision of Dr. Ben Xu at the Department of Mechanical Engineering in University of Houston.

Authors:

Lichang Zhu University of Houston
Ben Xu University of Houston