Hardware-accelerated Phase-Averaging Simulates Cavitating Bubbly Flows with 16-fold Performance Gains and 8% Accuracy

Understanding the complex behaviour of bubbly flows presents a significant challenge in many engineering applications, from sonar systems to medical ultrasound. Diego Vaca-Revelo, Benjamin Wilfong, Spencer Bryngleson, and Aswin Gnanaskandan, from Worcester Polytechnic Institute and Georgia Institute of Technology, have developed a new computational method that dramatically accelerates simulations of these flows. Their work introduces a hardware-accelerated approach to phase-averaging, allowing researchers to model the dynamics of individual bubbles or statistically represent entire bubble populations with unprecedented efficiency. Validated against both analytical solutions and experimental data, the method achieves up to a sixteen-fold speedup compared to conventional CPU-based simulations, opening new possibilities for detailed investigation and optimisation of bubbly flow phenomena.

Microbubble Enhanced Ultrasound Flow Simulation Validated

Scientists have developed and thoroughly tested a sophisticated computational tool for simulating microbubble-enhanced high-intensity focused ultrasound (MB-HIFU). This research focuses on accurately modeling the complex interplay between ultrasound waves, microbubbles, and surrounding fluids, essential for optimizing HIFU therapies. The new solver, called MFC, captures the intricate physics of this process, paving the way for more effective and precise treatments. HIFU is a non-invasive therapeutic technique used for cancer treatment and targeted drug delivery. Introducing microbubbles enhances HIFU’s effectiveness by increasing ultrasound absorption and creating localized cavitation.

MFC combines Eulerian and Lagrangian methods, tracking both the fluid and individual microbubbles, and utilizes phase-averaged equations to efficiently model large bubble populations. It employs advanced numerical techniques and leverages GPUs for significant performance acceleration, allowing for simulations of unprecedented complexity and resolution. The researchers validated MFC by comparing simulation results with experimental data and analytical solutions, demonstrating its accuracy and reliability. The solver successfully captures key phenomena like bubble oscillation, cavitation, and acoustic wave propagation, promising to improve HIFU treatment parameters and provide a deeper understanding of MB-HIFU. Future work will focus on incorporating more complex bubble models and simulating tissue heterogeneity.

Acoustic Bubble Suspension Multiscale Simulation Method

Researchers have developed a high-performance solver for simulating acoustically driven bubbly suspensions, addressing the challenge of modeling phenomena across vastly different scales. The solver utilizes the compressible Navier-Stokes equations to model the fluid, representing dispersed bubbles using two approaches: a volume-averaged model tracking individual bubbles and an ensemble-averaged model statistically representing the bubble population. This allows for a flexible and efficient approach to simulating complex bubbly flows. The volume-averaged model explicitly tracks bubble motion, while the ensemble-averaged model represents the collective behavior of the bubble population through a discretized distribution of bubble sizes.

Both models utilize the Keller-Miksis equation to accurately capture bubble dynamics under pressure fluctuations. The team significantly accelerated the solver by utilizing OpenACC directives to offload computations to NVIDIA A100 GPUs, achieving a sixteen-fold speedup compared to a conventional CPU. Rigorous tests demonstrate the robustness and scalability of the method across both CPU and GPU platforms, confirming its efficiency for simulating complex bubbly flows. These results establish the proposed method as a robust, accurate, and efficient tool for multiscale simulation of acoustically driven dilute bubbly suspensions.

GPU Accelerated Simulation of Bubbly Suspensions Validated

This work presents a comprehensive validation and performance analysis of a hardware-accelerated solver designed to simulate acoustically driven bubbly suspensions. The solver employs both volume-averaged and ensemble-averaged subgrid bubble models, offering flexibility and efficiency in simulating complex bubbly flows. The team modeled the fluid using the compressible Navier-Stokes equations and validated the volume-averaged model against both analytical solutions and experimental data, achieving root-mean-squared errors of less than eight percent for single-bubble oscillation and collapse scenarios. Experiments reveal a sixteen-fold speedup on a multi-GPU system compared to a conventional CPU, demonstrating the significant performance benefits of hardware acceleration.

The ensemble-averaged model further reduces computational cost by solving a single set of averaged equations, offering increased efficiency for large bubble populations. Weak and strong scaling tests demonstrate good scalability across both CPU and GPU platforms, confirming the robustness and efficiency of the method for large-scale simulations. The team’s work delivers a significant advancement in multiscale simulation techniques for dilute bubbly suspensions, enabling more comprehensive and computationally feasible studies of these complex systems.

GPU Acceleration Validates Bubbly Flow Simulations

This research presents a validated, hardware-accelerated solver for simulating acoustically driven bubbly flows, employing both volume-averaged and ensemble-averaged models to represent the dispersed bubble phase within a liquid. The team successfully demonstrated the solver’s accuracy through rigorous testing, achieving excellent agreement with both analytical solutions and experimental data for single bubble dynamics and spherical bubble collapse, with root-mean-squared errors consistently below eight percent. The study highlights a significant performance advantage for the GPU-accelerated solver, achieving up to a sixteen-fold speedup compared to CPU-based simulations on a multi-GPU system. The ensemble-averaged model further reduces computational cost by representing the bubble population statistically, offering substantial savings over simulating individual bubbles. Scalability tests confirm the solver’s efficiency across increasing computational resources. The current solver provides a robust and efficient tool for investigating complex bubbly flows and their applications.