Date of Award
Fall 12-20-2024
Level of Access Assigned by Author
Open-Access Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Mechanical Engineering
Advisor
Justin Lapp
Second Committee Member
Zhihe Jin
Third Committee Member
Andrew Goupee
Additional Committee Members
Yifeng Zhu
Johannes Grobbel
Abstract
Numerical simulation of high-temperature granular media for predicting particle movements and heat transfer has recently been in high demand. An increasingly popular method for calculating particle movement is the Discrete Element Method (DEM), which includes rotational and translational degrees of freedom. By modeling individual particles and their interactions in detail, DEM can capture complex behaviors that continuum models might not represent as effectively. In DEM, the model calculates elastic forces at the contact points between particles using springs and accounts for energy dissipation—such as friction and damping effects—through dampers in both normal and tangential directions. Employing heat transfer equations, i.e., conduction, radiation, and convection for the particles and boundaries, is needed for a comprehensive model. The scope of the thesis is developing an in-house solver for DEM, including conductive and radiative heat transfer for particle-particle and particle-wall pairs, as well as convective heat transfer between particles and interstitial fluid. The code is compatible with Computational Fluid Dynamics (CFD) software for the computation of the force interactions and the convective heat transfer and/or mass transfer between phases due to the presence of the particles in the flow field.
The DEM solver is developed on Graphical Processing Units (GPUs) to improve computational speed. The GPU-based DEM solver improves stability, performance and seamlessly integrates into commercial or open-source CFD software. A key innovation is eliminating a prerequisite for network communication between solvers, previously required for cross-platform coupling. This is accomplished by a direct coupling method that employs Dynamic-Linked Libraries (DLLs). The solver directly imports cell geometry information from the CFD software for particle-wall and particle-fluid interactions.
Furthermore, the solver optimizes memory usage by streamlining the particle-cell search algorithm, eliminating the cells' searching grid. This ensures the solver is compatible with a wide range of CFD cell types, providing high geometric flexibility. The approach simplifies the simulation process by incorporating CFD cell information directly into the GPU-based DEM solver. Additionally, a new method for computing void fraction is introduced to enhance the compatibility for various types of CFD cells compared to existing void fraction methods. Performance analysis shows a significant boost in computational speed compared to CPU-based and GPU-based solvers. The solver's compatibility with polyhedral meshes—a vital advantage for complex geometries—is demonstrated through testing against a referenced study simulating an immersed-tube fluidized bed.
The last section aims to enhance the computational efficiency of simulating radiative heat transfer in granular media by employing machine learning. Three novel methods for accelerating the computational simulations of high-temperature radiative heat transfer in granular media are presented. The study introduces a novel algorithm for neighbor searching that locates and sorts multi-sized particles based on their distance with minimal computation. By adapting the traditional use of regression recurrent neural networks for time-series data to a regression function for distance-ordered data, LSTM and GRU cells are utilized in a deep recurrent neural network to predict radiation view factors of particle and face neighbors through distance-ordered sequential data. Two new transformation methods are proposed to preprocess this sequential data. Through these transformation methods, particle and face neighbors are treated as general objects, allowing a single regression neural network to predict the view factor of any number of particles and faces sequentially. This approach enables the prediction of view factors for different-sized objects (particles and faces), offering a novel contribution to the field. The models demonstrate high precision for objects located within the short-range radiation region. Although objects situated at a long distance from the emitting particle tend to have an over-prediction of the view factor, their negligible contribution to overall radiation heat transfer results in a high coefficient of determination for all neighbors (both particles and faces). This method offers orders-of-magnitude computational acceleration compared to conventional Monte Carlo ray-tracing methods.
Recommended Citation
Kianimoqadam, Alireza, "Computational Acceleration of a CFD-DEM Solver for Dynamic Granular Media with Consideration of Energy Equations" (2024). Electronic Theses and Dissertations. 4093.
https://digitalcommons.library.umaine.edu/etd/4093
Files over 10MB may be slow to open. For best results, right-click and select "save as..."
