Speaker
Description
This study presents Nalu-Acoustics, a GPU-accelerated high-order finite-difference solver for acoustic propagation analysis over large domains. The solver employs dispersion-relation-preserving (DRP) spatial discretization and low-dissipation low-dispersion Runge–Kutta (LDDRK) time integration to minimize numerical dispersion and dissipation. Perfectly matched layers (PML) are applied at domain boundaries to suppress spurious reflections and reduce the computational domain size. The solver is implemented in CUDA with multi-GPU support via domain decomposition, currently utilizing up to eight GPUs. A single-GPU configuration achieved an 11-fold speedup over the equivalent CPU implementation. To validate the solver, the authors conducted transmission loss measurements in shallow waters near the Southwest Offshore Wind Farm demonstration site. Over the 100–1000 Hz one-third octave bands, the predicted transmission loss agreed with the measurements within a mean deviation of 1.0 dB, comparable to the measurement standard deviation. The validated solver was applied to noise propagation from the 20-turbine demonstration site, using source data from an offshore wind turbine at rated power. The simulation resolved multi-source interference and reflection at the sea surface and seabed. The results indicate that Nalu-Acoustics can be used to estimate noise impact zones around offshore wind farms and to support environmental assessment of underwater noise exposure to marine life.