Prediction of Acoustic Fields Using a Lattice-Boltzmann Method and Deep Learning

Abstract

Using traditional computational fluid dynamics and aeroacoustics methods, the accurate simulation of aeroacoustic sources requires high compute resources to resolve all necessary physical phenomena. In contrast, once trained, artificial neural networks such as deep encoder-decoder convolutional networks allow to predict aeroacoustics at lower cost and, depending on the quality of the employed network, also at high accuracy. The architecture for such a neural network is developed to predict the sound pressure level in a 2D square domain. It is trained by numerical results from up to 20,000 GPU-based lattice-Boltzmann simulations that include randomly distributed rectangular and circular objects, and monopole sources. Types of boundary conditions, the monopole locations, and cell distances for objects and monopoles serve as input to the network. Parameters are studied to tune the predictions and to increase their accuracy. The complexity of the setup is successively increased along three cases and the impact of the number of feature maps, the type of loss function, and the number of training data on the prediction accuracy is investigated. An optimal choice of the parameters leads to network-predicted results that are in good agreement with the simulated findings. This is corroborated by negligible differences of the sound pressure level between the simulated and the network-predicted results along characteristic lines and by small mean errors.

A review: Applications of machine learning and deep learning in aerospace engineering and aero-engine engineering

“…[65] evaluated LBM using the momentum propagation problem. The results demonstrated that For compressible Navier-Stokes (N-S) equations, the fourth-order Runge-Kutta approach was 12.3 times more accurate and quicker than the centralized fourth-order scheme; LBM reduces computational effort while providing computational accuracy and easy access to data; In order to forecast the far-field acoustic field in a two-dimensional square space with monopole sources, and irregularly spaced round and rectangular items, Rüttgers et al [69] trained a deep ANN using LBM.…”

A generic deep learning framework for propagation and scattering of acoustic waves in quiescent flows

“…Some make use of built-in tools of convolutional networks, such as periodic padding to encode simple boundary conditions such as periodic conditions [7]. Other works dealing with more complex geometries employ binary inputs [8] or signed distance functions [9] to encode the geometry information. However, such works only tackle steady-state problems, such as RANS predictions [8], or Sound Pressure Level (SPL) estimation [9].…”

“…Other works dealing with more complex geometries employ binary inputs [8] or signed distance functions [9] to encode the geometry information. However, such works only tackle steady-state problems, such as RANS predictions [8], or Sound Pressure Level (SPL) estimation [9].…”

Lettuce: PyTorch-based Lattice Boltzmann Framework

“…This approach, however, has limited transferability as it is primarily data-informed and does not encode the underlying physics. Furthermore, Rüttgers et al [33] have applied deep learning methods to the lattice Boltzmann method to predict the sound pressure level caused by objects. They introduced an encoder-decoder convolutional neural network and discussed various learning parameters to accurately forecast the acoustic fields.…”