Abstract

We design convolutional neural network models to simulate acoustic multiple scattering from cylindrical structures. The models are trained on data generated by multiple scattering theory to approximate total scattering cross section (TSCS) at different wavenumbers. In our first case study, the input of convolutional neural networks are binary images of positions of the cylinders, and the output is the TSCS, evaluated at eleven discrete values of wavenumber. We consider different planar configurations with five different numbers of cylinders ranging from 2 to 10. Our results show a good correspondence between predicted values and actual TSCS. However, we find that the training accuracy decreases as the number of scatterers increases. In a second case study, we develop convolutional autoencoder models, which take scatterer configuration images as input and, in the process of reproducing the same configurations at the output, reduces the dimensionality of the data to a small number of latent variables. We then use the latent variables in various inverse designs, producing scatterer configurations given values of TSCS. Despite its limitations the autoencoder proves to be a promising tool in searching for desired scatterer configuration without expensive computational cost.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call