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

The study targets low-frequency traffic noise below 500 Hz, which is hard to mitigate with conventional resonators because the wavelengths are long. The authors use an inverse-design workflow for phononic-crystal-based acoustic metamaterial resonators. They combine Gaussian-process surrogate modeling with a genetic algorithm, and iteratively retrain the surrogate using FEM-validated designs to reduce the number of expensive FEM evaluations. After scaling, a 2.5D prototype achieved about 20× pressure amplification near 490 Hz, and a revolved 3D cavity showed amplification above 30× with about 14 dB transmission loss near the target frequency. When integrated with a mass-loaded five-PZT stack, the device produced 5.5 Vpp and 0.25 mW under 100 dB SPL, indicating simultaneous noise mitigation and localized energy harvesting.