A machine learning approach to volume tracking in multiphase flow simulations: a thesis in Mechanical Engineering
Abstract
Multiphase flow is referred to flow of two or more immiscible phases (liquid, gas and solid). It is encountered in many essential natural phenomena and industrial processes. Computational simulations have emerged as a powerful and reliable tool for multiphase flow research to further current understanding and uncover new insights. They complement and are often strong alternatives to experimental methods, especially in studies where experiments are unfeasible or prohibitively expensive, due to, for example, short length or time scales or complex geometries. An essential component of multiphase flow simulations is capturing the dynamics of the interface separating the immiscible phases and tracking the phase volumes. Various methods have been proposed to achieve this, including the front tracking, level set, and volume-of-fluid (VOF) methods. The VOF method has become one of the most commonly used approaches to volume tracking and is the focus of this thesis. In VOF, the most common solutions are performed in two steps: interface reconstruction followed by flux calculation for volume advection. They represent a significant computational cost in VOF based multiphase flow simulations. In this work, a new approach using machine learning(ML) is used to generate a general advection function in a two-dimensional VOF scheme, which bypasses interface reconstruction and flux calculation. Although ML functions require a larger upfront cost to train, the resulting functions may be less computationally expensive to use when compared to traditional VOF methods. The data set in this work was generated from translation and rotation of a circle under various spatial and temporal resolutions. The ML training was performed using MATLAB’s Deep Learning Toolbox. To find an optimal neural network configuration, a grid search method based on the validation performance was used. Additionally, a rating system was developed to assess the overall performance of each function, as a potential alternative to solely relying on validation performance. This thesis presents results from commonly used advection tests to evaluate performance of volume tracking methods. In ideal conditions, the computation speed up is four times compared to the conventional VOF method, although determining ideal conditions will require further testing. In terms of accuracy, however, the VOF method remains superior.
Keywords
Not available
Specialisations
Similar theses
- Hybrid Machine Learning and Time-Series Modelling to Forecast Product Demand: A CRISP-DM Approach Niklas Virtanen · University of Vaasa · 2026
- Distributed Intelligence for 3D Object Tracking in Autonomous Vehicles, A Multi-Sensor Approach to Object Tracking and Planning for Self-Driving Cars, Real-Time 3D Perception and Path Planning in Autonomous Systems,,Vision-Based Localization and Distributed Architecture for Driverless Cars,A Hybrid Deep Learning Framework for Object Detection and Trajectory Planning in Autonomous Vehicles Manas Madine · University of Massachusetts Amherst · 2023
- Machine Learning with Conformal Prediction for Predictive Maintenance tasks in Industry 4.0 : Data-driven Approach Liu, Shuzhou ; Mulahuko, Mpova · Jönköping University · 2023
- Time Series Algorithms in Machine Learning - A Graph Approach to Multivariate Forecasting Ryan Zhou · Trent University · 2020
- DEVELOPMENT OF ACTIVE MACHINE LEARNING METHODS FOR PREDICTIVE MAINTENANCE PROCESSES AND REMAINING OF USEFUL LIFE (RUL) PREDICTION IN INDUSTRY 4.0 Esmaail Albarazi · 2026
- Predictive Maintenance in Modern Radar Systems using Machine Learning Klas Thorgren · Lund University · 2026
