Parameter prediction and inverse design of metamaterial acoustic absorber based on artificial neural network

Metamaterial sound absorbers have received widespread attention due to their deep subwavelength characteristics, while the design and optimization of metamaterial acoustic absorbers mainly relies on parameterized scanning method at present. However, this method largely relies on artificial physical intuition and design experience, which poses challenges in the process of multi-parameter optimization, such as consumption of massive computing resources and much time, and difficulties in obtaining global optimal design. For this reason, this article proposes a convenient and efficient design and optimization scheme for metamaterial sound absorbers based on artificial neural network algorithms, including forward prediction of the overall acoustic characteristics for the sound absorber based on primitive geometric parameters, as well as inverse design of the required primitive geometric structure based on the target acoustic spectra. Moreover, a deep sub-wavelength broadband sound absorber with an average sound absorption at 92.6% in the range of 300−400 Hz is also achieved with this method, which is described a compact thickness at 36.2 mm. The optimization scheme proposed in this article displays the advantages of multi-parameter synchronous optimization, fast calculation speed and wide applicability.