Structural Parameter Design Of Micro-perforated Plate Based On Deep Learning

The micro-perforated plate is a new type of resonant sound-absorbing material,which can perform directional sound absorption according to the noise spectrum,and has good application value in the engineering field.The sound absorption coefficient of the micro-perforated plate mainly depends on the structural parameters such as the micro-hole diameter,plate thickness,perforation rate,and air cavity depth.Traditional structural parameter design needs to be adjusted continuously through formula calculation and combined with experience,until a good effect is obtained.Such a process often faces many problems,such as cumbersome calculations and difficulty in directionally meeting target requirements.In order to realize the process from goal to result more directly,this article adopts deep learning method to carry out the reverse design of the structure.The thesis mainly includes the following main contents:(1)In order to directly obtain the appropriate parameter variables according to the target characteristics,this paper adopts a neural network structure of a kind of auto-encoder,and designs the structural parameters through the connection of a forward prediction network and a reverse design network.This approach overcomes the "non-uniqueness" problem faced by reverse design,and at the same time avoids the error amplification problem caused by a single reverse network.Then,the trained neural network model was used to design the structural parameters of the single-frequency high-absorption micro-perforated plate at different frequencies,and the finite element analysis was used to verify the rationality of the design results.(2)This article expands the demand for single-frequency sound absorption to broadband sound absorption,and realizes it by increasing the number of layers.Then the deep learning model is used to obtain the design parameters of the double-layer micro-perforated plate according to the sound absorption curve of the frequency band,and the finite element simulation results further explore the specific situation of the broadband sound absorption of the micro-perforated plate designed by deep learning.The results show that the trained neural network model has learned the non-linear relationship between the target sound absorption coefficient curve and the structural parameters,and can carry out reasonable parameter design according to the characteristics of the target curve,and design the results from the demand through the deep learning method.Has a good sound absorption performance.(3)For the full-frequency sound absorption requirements in practical applications,this paper uses a neural network model to design the four-layer micro-perforated plate structure parameters for ultra-wideband sound absorption in the 2500 Hz frequency range.Then combined with the finite element simulation,it is proved that the deep learning method still has a good design ability for the four-layer micro-perforated plate under multi-parameters.According to the design results,the general rules of the influence of the micro-perforated plate structure parameters on the sound absorption performance are summarized.In the context of vibration and noise problems,the deep learning-based micro-perforated plate structure design makes full use of the powerful calculation and fitting capabilities of neural networks,and provides new ideas for solving these problems.At the same time,this method also explores a technical route from demand capture,directional design to product realization in terms of acoustic material application in the engineering field.