Active-Learning-Guided Acoustic Metamaterial Resonators for Low-Frequency Noise Suppression and Piezoelectric Energy Harvesting

초록

Low-frequency traffic noise below 500 Hz is difficult to mitigate because its long wavelengths require impractically large conventional resonators. Here, we report an active-learning-guided inverse-design approach for scalable phononic-crystal-based acoustic metamaterial resonators that simultaneously suppress low-frequency noise transmission and harvest acoustic energy. The approach combines Gaussian process regression surrogate modeling with genetic algorithm optimization to efficiently explore high-dimensional cavity geometries. By iteratively retraining the surrogate with FEM-validated designs, the active-learning process guides the search toward high-performance structures while reducing costly FEM evaluations compared with conventional GA optimization. After geometric scaling, the 2.5D prototype derived from the nine-point optimized cavity achieved a pressure amplification factor of approximately 20 near 490 Hz, while the revolved 3D cavity exhibited amplification exceeding 30 and a transmission loss of approximately 14 dB near the target frequency. Integrated with a mass-loaded five-PZT stack, the device generated 5.5 Vpp and 0.25 mW under 100 dB SPL, corresponding to a normalized power density of 0.58 mu W Pa-2 cm-3. These results demonstrate a route toward multifunctional piezoelectric acoustic devices for noise mitigation, localized energy harvesting, and self-powered sensing.

키워드