A Study of Instability Mechanisms Driving Cloud Cavitation: Experimental Analysis Using New Advanced Velocimetry and Evaluation of Numerical Models

Abstract

Hydrodynamic cloud cavitation, characterized by the periodic shedding and violent collapse of large vapor structures, poses significant risks to the integrity and performance of hydraulic machinery. Despite extensive research, the precise interaction between competing shedding mechanisms, specifically re-entrant jets and condensation shocks, remains a subject of debate, largely due to the limitations of experimental diagnostics in resolving flow within opaque, dense multiphase regions. This dissertation addresses these challenges by integrating advanced deep learning-based diagnostics with rigorous experimental characterization and computational fluid dynamics (CFD) to elucidate the physics of cloud shedding in a converging-diverging Venturi geometry.

To overcome the limitations of traditional Particle Image Velocimetry (PIV) in unseeded, high-void-fraction flows, a domain-specific optical flow framework was developed. A Recurrent All-Pairs Field Transforms (RAFT) network was fine-tuned on a novel synthetic dataset derived from CFD simulations, explicitly designed to bridge the domain gap between computer vision benchmarks and multiphase flow textures. This fine-tuned model (RAFT-CloudCav) demonstrated a 37% reduction in end-point error compared to pre-trained baselines and was the only method capable of qualitatively resolving complex flow features such as the re-entrant jet and the Kelvin-Helmholtz (K-H) roll-up in high-speed experimental footage (130,000 fps). Sensitivity analysis established that a temporal resolution exceeding 32,500 fps is a prerequisite for accurate feature tracking in this regime.

Leveraging this high-fidelity velocimetry and a novel pixel-intensity "Source Term" analysis, the experimental campaign identified the K-H instability as a primary trigger for shedding, challenging the conventional view that shear instabilities are secondary in macro-scale flows. The analysis revealed a coupled mechanism where shear-induced vortex roll-up initiates local collapse events, which subsequently generate the acoustic perturbations necessary to drive upstream-propagating condensation shocks. To predict the transition between shedding topologies, a new dimensionless criterion based on the product of the Cavitation number and Reynolds number ($Gamma = sigma cdot Re$) was proposed. Validation across multiple scales confirmed that regimes where $Gamma < 10^5$ exhibit multi-cloud shedding, while regimes where $Gamma > 1.5 times 10^5$ transition to single-cloud shedding.

Finally, the experimental observations were used to validate an incompressible Homogeneous Mixture Model (HMM) simulation using OpenFOAM. The study confirmed that the standard Merkle cavitation model, when coupled with Scale-Adaptive Simulation (SAS) turbulence modeling, accurately captures the shedding frequency and the condensation front propagation speeds (3.4-5.7 m/s) observed experimentally. This work establishes a robust, data-driven framework for analyzing multiphase flows, offering new predictive criteria for the design of robust hydraulic systems.