TY - JOUR
T1 - Inverse design and experimental verification of an acoustic sink based on machine learning
AU - Gao, Nansha
AU - Wang, Mou
AU - Cheng, Baozhu
AU - Hou, Hong
N1 - Publisher Copyright:
© 2021 Elsevier Ltd
PY - 2021/9
Y1 - 2021/9
N2 - Recently, advances in artificial intelligence research, in particular “machine learning”, made it much easier to predict the structural parameters for a specific desired acoustic performance. In this paper, we build a acoustic sink model by using transfer matrix method, and establish a database of 79,730 lines in the frequency domain of 0–5 kHz. The proposed CNN model, which consists of two building blocks (encoder and decoder), is expected to predict the corresponding geometric parameters of the selected (expected) absorption curve. Geometric parameters include diameter a of the neck structure, diameter b of the cavity, neck length c, and thickness e of porous material. Through the comparison of four groups of prediction curves, target curves and test curves, it can be concluded that accuracy of prediction is high in the sound absorption area. This study verifies the feasibility of machine learning method in the inversion design of acoustic functional devices, which is suitable for performance structure inversion prediction of complex acoustic structures, and has potential application in shortening the design cycle of acoustic products.
AB - Recently, advances in artificial intelligence research, in particular “machine learning”, made it much easier to predict the structural parameters for a specific desired acoustic performance. In this paper, we build a acoustic sink model by using transfer matrix method, and establish a database of 79,730 lines in the frequency domain of 0–5 kHz. The proposed CNN model, which consists of two building blocks (encoder and decoder), is expected to predict the corresponding geometric parameters of the selected (expected) absorption curve. Geometric parameters include diameter a of the neck structure, diameter b of the cavity, neck length c, and thickness e of porous material. Through the comparison of four groups of prediction curves, target curves and test curves, it can be concluded that accuracy of prediction is high in the sound absorption area. This study verifies the feasibility of machine learning method in the inversion design of acoustic functional devices, which is suitable for performance structure inversion prediction of complex acoustic structures, and has potential application in shortening the design cycle of acoustic products.
KW - Acoustic sink
KW - Inverse design
KW - Machine learning
KW - Porous material
KW - Resonant sound absorption
UR - http://www.scopus.com/inward/record.url?scp=85105327846&partnerID=8YFLogxK
U2 - 10.1016/j.apacoust.2021.108153
DO - 10.1016/j.apacoust.2021.108153
M3 - 文章
AN - SCOPUS:85105327846
SN - 0003-682X
VL - 180
JO - Applied Acoustics
JF - Applied Acoustics
M1 - 108153
ER -