In situ estimation of the acoustic surface impedance using simulation-based inferencea).

Area of Science:

• Acoustics
• Computational Physics
• Data Science

Background:

• Accurate acoustic simulations depend on precise surface impedance boundary conditions.
• Conventional measurement methods have limitations for real-world scenarios due to simplifying assumptions.
• In situ estimation of acoustic properties is crucial for complex environments.

Purpose of the Study:

• To develop a Bayesian framework for in situ estimation of frequency-dependent surface impedances.
• To utilize simulation-based inference with neural networks for efficient parameter inference.
• To overcome limitations of conventional impedance measurement techniques.

Main Methods:

• Employed a Bayesian framework with simulation-based inference and neural networks.
• Modeled surface impedance using a damped oscillator model with a fractional calculus term.
• Validated the framework on a finite element model of a room and with impedance tube measurements.

Main Results:

• Achieved robust and accurate estimation of six individual surface impedances from 63 to 500 Hz.
• Demonstrated reliable uncertainty quantification and high predictive accuracy for complex geometries like car cabins.
• Verified well-calibrated inference through posterior predictive checks and coverage diagnostics.

Conclusions:

• The proposed Bayesian framework offers a generalizable and physically consistent method for characterizing acoustic boundary conditions.
• The approach enhances the accuracy of acoustic simulations in real-world interior environments.
• Simulation-based inference provides advantages for high-dimensional acoustic parameter estimation.