A Data-Assimilation Framework for Bubbly Flow Reconstruction Based on AOT-Nudging and Image Velocimetry
Loading...
Files
Date
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Waterloo
Abstract
Bubbly flows are common and important in nature and industry. For example, in lakes and oceans, bubbles affect methane emissions and oxygen exchange. In industry, bubbles in reactors affect reaction efficiency and operational safety. Many studies have attempted to reconstruct bubbly flows from experiments and investigate the bubble mechanism. However, bubbly-flow reconstruction remains challenging for several reasons. For example, bubbly flows are inherently multi-scale in time and space. Small and rapid interface dynamics (e.g., coalescence and breakup) are coupled with large-scale liquid motion, such as wakes and vortical structures. In many studies, it is also necessary to simultaneously resolve the bubble interface and the surrounding liquid-phase motion. Although both experimental and numerical methods are widely used to study bubbly flows, each approach has limitations. Experiments provide direct observations of real flows, but they are limited by spatial and temporal resolution, measurement noise, and optical accessibility. Numerical simulations can provide complete flow fields, but their accuracy depends on modeling assumptions, numerical discretization, initial and boundary conditions, and they require experimental validation.
In this thesis, a Data-Assimilation (DA) framework is proposed to synchronize numerical simulations with experimental observations for bubbly-flow reconstruction, simultaneously resolving bubble interfaces and liquid-phase velocities at the bubble scale. The experimental observations are obtained from image-based measurements, including image velocimetry for the liquid phase and bubble silhouette sequences for the gas–liquid interface. The simulation is performed using the Volume-of-Fluid (VoF) method. DA is achieved using the Azouani–Olson–Titi (AOT) nudging algorithm. In the AOT-nudging algorithm, penalty terms are embedded in the governing equations of the two-phase flow as source terms. These terms penalize discrepancies between the experimental observations and the simulations, thereby driving the simulation toward the true flow state. As a result, the reconstruction can be viewed as a simulation continuously corrected by the observations.
The proposed framework is validated using synthetic and experimental data. The validation considers bubbles rising in two dimensions (2D). The validation results show that the framework synchronizes the reconstructed bubble and liquid-phase kinematics with those from the ground truth, whereas the baseline simulation without DA deviates from the ground truth at an early stage. The framework can also reconstruct temporally and spatially resolved full-field flow states, which are difficult to obtain directly from experiments alone. A parameter study shows that the reconstruction accuracy depends on several DA parameters, providing practical guidance for parameter selection.
The thesis also develops an accompanying Uncertainty Quantification (UQ) method for the bubble interface and liquid-phase velocity. The uncertainty is estimated from the local observation–reconstruction mismatch and the measurement uncertainty. Synthetic noisy observations are used to show that the UQ is effective in bubbly-flow reconstruction: it can predict the spatial distribution and magnitudes of reconstruction errors.
Lastly, the AOT-nudging framework is extended and applied to a three-phase flow, enabling the use of inertial tracer particles to reconstruct bubbly flows. In the extended framework, the measured particle kinematics, such as velocities and accelerations, are used to infer a preliminary liquid-phase velocity based on a particle-inertia correction model. The inferred velocity nudges and updates the liquid-phase momentum equation. Next, the updated flow advects the reconstructed inertial particles. The reconstructed and observed particle locations are compared to nudge the reconstructed trajectories toward the observed ones. Using the latest reconstructed particle trajectories, the liquid-phase velocity is updated again. As a result, the DA algorithm treats inertial-particle trajectories, liquid-phase velocities, and bubble silhouettes as both observations and assimilated states. This allows the reconstructed three-phase flow to satisfy the governing equations while remaining consistent with the observations. Validations using synthetic data show that the solver can achieve accurate and physically consistent bubbly-flow reconstruction even when inertial tracer particles with high Stokes numbers are used.
In summary, this thesis proposes an AOT-nudging DA framework that provides a practical and extensible approach for simultaneous bubble-interface and liquid-phase flow reconstruction at the bubble scale, with integrated uncertainty quantification and inertial particle tracking capabilities. The proposed framework can reconstruct bubbly flows from sparse observations with accuracy, robustness, completeness, and physical consistency. The framework can be readily extended to a wide range of multiphase-flow problems for which experiments or numerical simulations alone are difficult to apply.