Speaker
Description
The growing use of design methods based on machine learning and optimization has increased the need for methods capable of generating large, high-quality acoustics datasets. To address this, we present a stochastic 3D geometry generation algorithm for irregular, ventilated acoustic metamaterials that support resonant behavior in sound transmission loss applications. The method operates on a voxel grid, using a flood-growth rule with lateral expansion and inter-layer propagation to synthesize fully connected air duct networks. Finite element simulation model used for the analysis of the generated geometries is validated experimentally sing ten cube-structured geometries fabricated as laser-cut birch plywood specimens. Validation employs four-microphone transmission loss measurements in an impedance tube over 500–6400 Hz. Across all designs, the numerical model reproduces experiments with an average absolute discrepancy of 2.3 dB, standard deviation of 1.4 dB, and maximum of 11.8 dB. Although no specific optimization was targeted, some samples already show a promising broadband transmission loss, reaching up to 58 dB at 2650 Hz. Overall, the proposed method provides a validated foundation for data-driven design of ventilated metamaterials.