Session: 7.7.1 - Numerical Methods for Multiphase Flows I

Paper Number: 157813

157813 - A Hardware Accelerated Euler Lagrange Algorithm for Simulating Acoustic Cavitation of Microbubbles 

Abstract: 

Acoustic cavitation refers to the phenomenon where rapid oscillations in pressure, caused by high-frequency sound waves (typically ultrasound), create, grow, and collapse small gas-filled bubbles in a liquid. These oscillations can occur due to the alternating compression and rarefaction phases of the sound waves. Acoustic cavitation has several applications in both biomedicine and engineering areas. One such application is the microbubble assisted high intensity focused ultrasound (mb-HIFU) therapy. This involves using the acoustic energy of ultrasound and thermal effects of bubble oscillations to increase the temperature of a target tissue and cause cell necrosis. While experiments have indicated preliminary success of this method, the exact mechanism behind how microbubbles contribute to the thermal effects is still not clear. Due to the complexity of experimentally isolating the ‘acoustic-to-thermal energy conversion’ mechanisms at the microbubble scale, we use numerical simulations, validated by experimental data, to explore these mechanisms in detail. In our simulations, the ultrasound field is modeled using the compressible Navier-Stokes equations on a fixed grid, while microbubbles are represented in a Lagrangian framework. The dynamics of the bubbles are captured using the Keller-Miksis model, which accounts for the compressibility of the surrounding medium. We apply a two-way coupling approach between the Eulerian and Lagrangian frameworks via local volume averaging. Time-averaged heat deposition terms are derived from the bubble-acoustic field and then used to solve a bio-heat transfer equation, which models the resulting temperature field over extended time scales. Using this numerical methodology in a traditional CPU-based algorithm utilizing Message Passing Interface (MPI) communication protocols is computationally expensive especially for large number of microbubbles. Further, the presence of microbubbles in only a few of the processor domains can cause load imbalance resulting in computational inefficiency. To address this challenge, we develop a GPU-based hardware acceleration of our multiscale CFD model. We use OpenACC to offload all computationally intensive tasks to GPUs. After applying the initial conditions, the state variables are copied to the GPUs which is in charge of the massive computations. While transferring results back to the CPU can be time-consuming, this operation is unusual since it typically occurs during periodic data saves every several hundred simulation steps. To ensure compatibility and portability, we use standard OpenACC directives, which support a variety of compilers, including NVHPC, GNU, and Cray (CCE) for both NVIDIA and AMD GPUs. This hardware acceleration of our multiscale CFD model enables us to efficiently perform parametric studies of key variables across a wide range of therapeutically relevant conditions to elucidate the mechanisms through which microbubbles enhance acoustic-to-thermal energy conversion in HIFU therapy. We will present the computational method, show validation results for several cases and demonstrate acceleration obtained compared to a traditional CPU based algorithm. 

Presenting Author: Diego Vaca Revelo Worcester Polytechnic Institute

Presenting Author Biography: