A method for removal of reflection artifact in computational fluid dynamic simulation of supersonic jet noise

Taeyoung Park, Hyun‐Shik Joo, Inman Jang, Seung-Hoon Kang, Won‐Suk Ohm, SangJoon Shin, Jeong‐Won ParkThe Journal of the Acoustical Society of Korea 2020 · DOI

Abstract

Simulating the acoustic environment of supersonic jets is essential for protecting aerospace payloads, yet these simulations are often hampered by artificial wave reflections at the boundaries of finite computational domains. Traditionally, "sponge layers" are used to damp these reflections, but they significantly increase computational costs. This paper introduces a post-processing technique that adapts the "two-microphone method" to separate outgoing noise from boundary reflections mathematically. Applied to a Mach 1.8 supersonic jet simulation, the method successfully isolated the physical acoustic field, offering a software-agnostic solution that can reduce total computation time by approximately 14% by eliminating the need for extensive buffer zones.

Introduction

The intense acoustic loading generated by rocket exhaust during liftoff poses a significant risk to sensitive satellite payloads. Predicting this noise typically involves a two-step process: high-fidelity Computational Fluid Dynamics (CFD) to model the turbulent flow near the nozzle, followed by an acoustic transport model to predict noise propagation.

A major challenge in the CFD step is the management of domain boundaries. Because computational grids are finite, sound waves hitting the edge of the grid reflect back into the domain, creating "reflection artifacts" that corrupt the acoustic data. The standard mitigation strategy involves implementing a "sponge layer"—an artificial damping zone. However, sponge layers require larger grids and meticulous parameter tuning, adding substantial overhead to already massive simulations.

Methodology

The proposed method shifts the focus from preventing reflections during the simulation to removing them during post-processing.

CFD Simulation and Data Extraction

The researchers performed a Detached Eddy Simulation (DES) of a Mach 1.8 supersonic jet (nozzle pressure ratio 5.4) using a grid of approximately 40 million cells. Notably, the simulation was conducted without a traditional sponge layer at the boundaries to intentionally allow reflections.

Schematic of the computational domain showing the jet plume and the Kirchhoff surface used for data extraction.

Wave Separation Logic

The core of the technique is an adaptation of the "two-microphone method" used in physical acoustics. Instead of physical sensors, the method utilizes the pressure ($p$) and its radial derivative ($\partial p/\partial r$) extracted from a cylindrical "Kirchhoff surface" within the CFD domain.

By transforming this data into the frequency-wavenumber (k-space) domain via Fourier transforms, the complex wave field is decomposed into two components:

  1. Diverging waves: The physical noise generated by the jet moving outward.
  2. Converging waves: The unwanted artificial reflections moving inward from the boundaries.

The "clean" acoustic field is reconstructed by setting the converging wave coefficients to zero and performing an inverse Fourier transform.

Results and Discussion

The method was first verified using a simplified 1D cylindrical wave problem, which confirmed the mathematical accuracy of the separation logic. When applied to the high-fidelity supersonic jet case, the technique effectively isolated "Mach wave radiation"—the primary noise source—from the cluttered reflection artifacts.

Comparison of Sound Pressure Level (SPL) and Power Spectral Density (PSD) with and without the reflection removal processing.

As shown in the spectral analysis, the Overall Sound Pressure Level (OASPL) and Power Spectral Density (PSD) were consistently lower after processing. This confirms that the boundary reflections were artificially inflating the noise predictions.

Efficiency Gains

The computational benefits are significant. For the Mach 1.8 case, implementing a proper sponge layer would have increased the grid count by 14% and the computational volume by 35%. In a simulation requiring 120 days of compute time, this post-processing method could save approximately 17 days. Furthermore, while the CFD simulation takes months, the artifact removal script runs in minutes on standard hardware.

Conclusion

This research demonstrates that boundary reflections in CFD are mathematically deterministic and can be removed without the need for computationally expensive damping zones. The method is software-agnostic, making it easily applicable to data from commercial solvers like ANSYS FLUENT. Future work will focus on validating these results against large-scale simulations using traditional sponge layers and investigating whether reflected waves significantly alter noise generation at the source.

Summary updated

BibTeX
@article{park2020method,
  author = {Taeyoung Park and Hyun‐Shik Joo and Inman Jang and Seung-Hoon Kang and Won‐Suk Ohm and SangJoon Shin and Jeong‐Won Park},
  title = {{A method for removal of reflection artifact in computational fluid dynamic simulation of supersonic jet noise}},
  journal = {The Journal of the Acoustical Society of Korea},
  year = {2020},
  doi = {10.7776/ask.2020.39.4.364},
}
Figure