Identifying sources of variability in impedance tube sound absorption measurements with multivariate explainable machine learning

Characterising the acoustic properties of porous materials is essential for predicting the effectiveness of acoustic treatments in rooms, vehicles and aircraft interiors. Impedance tube measurements are commonly used to determine acoustic properties such as the sound absorption coefficient (SAC); however, their reproducibility is often affected by non-standardised specimen preparation procedures. In particular, deviations from the nominal tube diameter and other geometric inconsistencies may alter the measured acoustic response. This study investigates the influence of specimen thickness, diameter, rotational angle and material on the SAC curves. A data-driven classification framework is employed to assess whether these factors leave distinguishable signatures in the SAC, using both univariate and multivariate formulations. Model interpretability tools are used to identify the frequency ranges most sensitive to variations of each factor, enabling a physically interpretation of the results. The analysis demonstrates that variations in diameter, material and thickness produce distinct and measurable modifications of the SAC, whereas the rotational angle has a negligible influence. While the effects of thickness and material are well established in the literature, diameter deviations of 2% are shown to significantly alter the acoustic response, highlighting the role of edge-related phenomena. The higher multivariate performance further suggests the presence of interaction among factors that are not captured by single-factor analyses. The SHAP-based analysis, identifies the low-frequency band ( < 600 Hz) as the most sensitive to variation in SAC for the factors under investigation.