Session: 7.7.2 - Numerical Methods for Multiphase Flows II

Paper Number: 156513

156513 - Automating Computational Fluid Dynamics Simulations for Bubble Velocity Detection in Glass Melters Using Machine Learning 

Abstract: 

Analyzing computational fluid dynamics (CFD) simulations to detect the velocity of rising bubbles in a glass melter can be both time-consuming and laborious. To streamline this process, we developed an automated workflow using Maestro Workflow Conductor, integrated with the Ultralytics YOLOv11 machine learning algorithm and the STAR-CCM+ commercial CFD software. This workflow automates the setup, execution, and analysis phases of the CFD simulations, significantly reducing the manual effort and time required. By ensuring seamless job submission and comprehensive result verification, the workflow leverages YOLOv11 to accurately compute bubble velocities and categorize results into bins. This categorization allows for a detailed analysis of how average velocities change as bubbles rise through molten glass. Our innovative approach not only enhances efficiency but also ensures consistent and reliable results, making it a valuable tool for researchers in fluid dynamics and related fields. For our specific application, the automated workflow will be used to facilitate the development of momentum source terms in the CFD model, eliminating the need to resolve the bubbles using a volume of fluid formulation. The melters used for waste glass vitrification at the Hanford site employ bubblers near the base to inject air into the highly viscous melt pool, enhancing convection. This increased convection boosts the melt rate of the waste slurry fed to the melter, thereby increasing the plant's production rate. Different bubbler configurations and various waste glass formulations are being evaluated, necessitating numerous CFD simulations. To improve the turnaround time of these simulations, momentum source terms can be used to replicate the convective flow patterns created by the bubblers. However, these momentum source terms need to be tuned to the specific conditions within the melter, including melt viscosity, bubbling rate, and other parameters, requiring many computer simulations. The automated workflow provides CFD analysts with an easier way to develop these momentum source terms.

Presenting Author: Donna Guillen Idaho National Laboratory

Presenting Author Biography: Donna Post Guillen is the Group Lead for Materials Performance and Modeling in the Materials Science and Manufacturing Department at Idaho National Laboratory. Dr. Guillen has over 40 years of research engineering experience and has served as principal investigator/technical lead for numerous multidisciplinary projects encompassing waste heat recovery, combustion, heat exchangers, power conversion systems, nuclear reactor fuels and materials experiments, waste vitrification, and advanced manufacturing. Her core area of expertise is computational modeling of energy systems, materials, and thermal fluid systems. She is experienced with X-ray and neutron beamline experiments, computational methods, tools and software for data analysis, visualization, application development, machine learning and informatics, numerical simulation, and design optimization. As Principal Investigator/Technical Lead for the DOE Nuclear Science User Facility Program, she has engaged in irradiation testing of new materials and performed thermal analysis for nuclear reactor experiments. She actively mentors students, serves in a leadership capacity as well as routinely chairs and organizes technical meetings for professional societies (ANS, ASME, TMS), provides subject matter reviews for proposals and technical manuscripts, has published over 70 peer-reviewed journal articles, 150+ conference papers, received three Best Paper awards, authored numerous technical reports, and has written/edited several books and proceedings.