ML-Enhanced Multiphysics Modeling for Packed-Bed Thermal Energy Storage Optimization
PI: Olumide Ogunmodimu (EMS)
Additional Team Member: Arash Dahi Taleghani (Energy & Mineral Engineering)
EME Teaching Assistance and support from the department will support the students.
As the need for reliable, affordable energy storage increases, packed-bed thermal energy storage (TES) systems have become essential for Concentrating Solar Power (CSP) plants. Packed beds utilizing granular media provide substantial thermal capacity, cost efficiency, and scalability1. However, there is a notable research gap in accurately forecasting and optimizing the complex interrelated phenomena that affect the performance of these systems. Figure 1 illustrates how TES operates within a modular CSP framework and computational simulation of the TES operation, along with machine learning (ML) capability that leverages computational data for predictive purposes. Heated particles are transported to the top of the tower using a Particle Lift and enter the Falling Particle Receiver (FPR), where they harness concentrated solar energy. These particles are then weighed in a Weigh Hopper before being placed into a Hot Storage Bin. Heat from these stored particles is transferred to a supercritical CO₂ (sCO₂) loop via a heat exchanger, facilitating power generation. After cooling, the particles are gathered in a Cold Storage Bin for recirculation. As depicted in Figure 1, this arrangement enables high-temperature, efficient, and recyclable thermal energy storage for advanced concentrated solar power systems. Current modeling techniques often use overly simplified assumptions, which do not adequately capture the complex interactions among granular flow dynamics 5, heat and mass transfer, flow-induced segregation, and uneven heating 2. This shortcoming has resulted in suboptimal designs and discrepancies between modeled predictions and actual performance. In particular, the impact of particle size distribution and variable flow conditions on thermal efficiency is still not well understood and inaccurately represented in existing models.
Preliminary numerical studies have shown that particle size significantly affects heat transfer rates, storage efficiency, and energy retention. Specifically, smaller particles demonstrate lower conduction resistance, quicker heating, and greater energy storage efficiency in both pure conduction and convection-enhanced scenarios. However, these simplified models assume uniform temperature distributions and overlook detailed particle dynamics and local fluid flow patterns, emphasizing the need for more advanced multiphysics approaches.
To overcome these limitations, this research introduces a comprehensive multiphysics modeling framework that integrates machine learning and digital learning tools for enhanced simulation and analysis. The approach combines DEM to capture granular flow dynamics, CFD for simulating fluid flow and heat transfer, and leverages MATLAB for advanced data processing, visualization, and integration with machine learning algorithms. Incorporating machine learning into the workflow facilitates the rapid prediction of complex interrelated behaviors, such as granular segregation and heat transfer efficiency, by learning from large simulation datasets generated by DEM and CFD. This not only accelerates simulations but also enables the creation of surrogate models for predictive analysis. Additionally, the use of MATLAB provides robust tools for preprocessing, analyzing high-volume multidimensional data, and training machine learning models to extract actionable insights, ultimately supporting digital learning and improved decision-making in system optimization. This integrated computational approach addresses the critical need for more accurate prediction tools that can guide the development of next-generation TES systems with improved efficiency, reliability, and cost-effectiveness for renewable energy applications.
Objectives
This research aims to develop and rigorously validate an advanced multiphysics simulation framework that couples the DEM and CFD for detailed modeling of granular flow and heat transfer in packed-bed TES systems. To enhance the framework’s capability, machine learning (ML) approaches, specifically supervised regression models such as Random Forest Regressors and Gradient Boosting Machines, will be integrated to systematically optimize particle properties, flow geometries, and operational conditions for smarter and more efficient TES designs. The three main objectives of this research are:
1. To create a fully coupled DEM-CFD simulation platform that accurately captures heat transfer, particle-fluid interactions, and granular flow dynamics during both charging and discharging cycles, enabling a detailed investigation of all critical physical phenomena in packed-bed TES operations.
2. To employ Random Forest and Gradient Boosting regression techniques, trained on simulation-generated datasets, for predicting key TES performance indicators such as temperature profile evolution, pressure drop, and storage efficiency, thus enabling targeted optimization of particle size distribution, flow geometry, and operating parameters for enhanced system performance.
3. To validate the predictive accuracy and reliability of the ML models by benchmarking their outputs against lumped MATLAB simulation results and available experimental data, ensuring robust, generalizable guidance for data-driven TES system design optimization.
By integrating an advanced physics-based simulation with data-driven machine learning optimization, this research addresses current challenges in TES system design and operation, including particle dynamics, the enhancement of uniform heat distribution, and the impact of size on thermal storage during charging and discharging.
Methodology
Multiphysics DEM-CFD Modeling: We would employ the DEM to simulate the movement, interaction, and heat exchange between individual particles within a packed bed, capturing critical phenomena such as collisions, segregation, and particle-scale thermal interactions. Simultaneously, we will apply CFD to model the fluid phase, accurately resolving the flow behavior and convective heat transfer as the fluid moves through and interacts with the granular bed. We plan to implement a coupling approach using advanced software frameworks such as Corheos and ANSYS Fluent-EDEM to achieve a comprehensive and realistic system simulation. This will allow us to synchronize the particle and fluid phases, enabling detailed analysis of multiphase flow dynamics and energy transfer processes within the packed bed.
Simulation Design and Parameters: The simulation design will incorporate a comprehensive range of variables to capture the complexities of the system. Particle distribution will be explored by simulating both uniform (monodisperse) and varied (polydisperse) particle sizes to evaluate their influence on system behavior, including flow dynamics and heat transfer efficiency. The operating regimes will encompass static bed conditions as well as dynamic phases involving heat input (charging) and heat extraction (discharging), representing realistic cycle variations encountered in practical applications. Various geometric configurations will be assessed, focusing on different hopper and channel layouts to determine their impact on particle flow and thermal performance. Additionally, boundary conditions will be systematically varied, including parameters such as heat flux, flow rate, and inlet temperature, to ensure the robustness and broad applicability of the simulation models across different operating scenarios.
Machine Learning Integration: ML techniques will be integrated into the simulation framework to enhance predictive capabilities and optimize system performance. Data preparation will involve extracting and organizing simulation results to construct a comprehensive training dataset, incorporating features such as particle size distribution, geometric configuration, flow rate, and heat flux. During model development, supervised learning algorithms, including Random Forest, Gradient Boosting, and Neural Networks, will be employed to create models capable of delivering rapid predictions of TES performance and identifying potential operational risks. The optimization application will utilize these ML models to pinpoint optimal parameter sets aimed at achieving improved heating uniformity and minimizing particle segregation. The ML-driven recommendations will be validated against both simulation outputs and available experimental data to ensure reliability and practical relevance. Professor Arash’s lab at the Energy Institute will supply experimental data to validate the results of the DEM simulations, specifically focusing on how particle size affects thermal storage efficiency.
Innovation and Scientific Advancement
This project introduces an integrated multiphysics and machine learning (ML) workflow that provides a comprehensive platform combining advanced simulation techniques with data-driven optimization tailored for TES systems. The approach facilitates design exploration by investigating the effects of different particle distributions and geometric configurations, enabling the customization of system designs to achieve improved heat transfer and consistent flow behavior. Additionally, the incorporation of rapid, data-driven optimization significantly accelerates the identification of optimal system parameters, reducing reliance on repeated, time-consuming computational experiments and enhancing the efficiency of the design process.
Anticipated Deliverables
The project is expected to deliver validated simulation tools, including a robust DEM-CFD model capable of accurately capturing the key physical behaviors observed in packed-bed TES systems. In addition, predictive machine learning solutions will be developed to forecast system performance and recommend design improvements with high reliability. These deliverables and preliminary results will serve as the foundation for comprehensive follow-on proposals to the Department of Energy (DOE) and National Science Foundation (NSF), targeting their solar energy and energy storage initiatives. The team has established relationships with manufacturers of particulate materials (proppants and slags) who can serve as potential industrial sponsors in the future. These partners may also provide valuable industry perspective and access to material properties for model validation, strengthening future funding applications through demonstrated industrial relevance. The findings and methodologies will be disseminated through peer-reviewed publications, and opportunities for engagement with industry stakeholders will be explored to promote the application of the research outcomes.
Significance and Impact
This research will advance the state-of-the-art in TES system design for CSP applications, addressing practical challenges such as non-uniform heating, flow instabilities, and particle segregation. The integration of machine learning will provide a powerful tool for rapid design iteration and optimization, enabling cost-effective and efficient thermal energy storage solutions. The methodologies developed will also have broader applications in other granular flow and heat transfer systems, such as chemical reactors and energy-efficient industrial processes.
Budget
The requested funding amounts to $36,972. This budget allocation will cover stipends (2 x $16,656) and fringe benefits (2 x $1,829.75) for two graduate assistants in grade 14 over two semesters. The balance of their fee will be provided through Teaching Assistance at the EME. Also included in the budget are 80 cores of high-speed memory at $2,000 and 60 hours of ICDS RISE Team at $4,800. The entire budget is designated from July 2025 to June 2026.