Explainable machine learning determines effects on the sound absorption coefficient measured in the impedance tube

Area of Science:

• Acoustics
• Materials Science
• Data Science

Background:

• Impedance tube measurements of sound absorbing materials exhibit poor reproducibility.
• Standardized methods lack precise definitions for measurement setup and specimen preparation, leading to uncertainty.

Purpose of the Study:

• To identify and quantify factors influencing sound absorption coefficient using machine learning.
• To determine frequency ranges most affected by setup variations in acoustic measurements.

Main Methods:

• Utilized machine learning models on a dataset of over 3000 absorption spectra.
• Analyzed variations in cutting technologies, operators, specimen dimensions, and mounting methods.
• Applied explainable AI techniques to interpret model findings.

Main Results:

• Specimen thickness and operator were identified as highly influential factors affecting sound absorption.
• A directional, non-random relationship was found between operator and absorption coefficient.
• Specific frequency ranges were pinpointed as being most sensitive to setup choices.

Conclusions:

• Explainable machine learning offers a promising approach for discovering knowledge from acoustic measurement data.
• Controlling operator influence is crucial for improving the reproducibility of sound absorption measurements.
• The study highlights the need for more precise definitions in acoustic testing protocols.