Reinforcement learning applied to metamaterial design

Hybrid supervised and reinforcement learning for the design and optimization of nanophotonic structures

From higher computational efficiency to enabling the discovery of novel and complex structures, deep learning has emerged as a powerful framework for the design and optimization of nanophotonic circuits and components. However, both data-driven and exploration-based machine learning strategies have limitations in their effectiveness for nanophotonic inverse design. Supervised machine learning approaches require large quantities of training data to produce high-performance models and have difficulty generalizing beyond training data given the complexity of the design space. Unsupervised and reinforcement learning-based approaches on the other hand can have very lengthy training or optimization times associated with them. Here we demonstrate a hybrid supervised learning and reinforcement learning approach to the inverse design of nanophotonic structures and show this approach can reduce training data dependence, improve the generalizability of model predictions, and significantly shorten exploratory training times. The presented strategy thus addresses several contemporary deep learning-based challenges, while opening the door for new design methodologies that leverage multiple classes of machine learning algorithms to produce more effective and practical solutions for photonic design.

An acoustic cloaking design based on topology optimization

In this work, we explain how to utilize the topology optimization method for the design of acoustic cloaks based on the principle of scattering cancellation. To take account of the challenging fabrication restriction, we impose boundary control inside the optimization objective function and enforce hyperbolic tangent projection to minimize the gray transition regions of the optimized design. In addition, a filter based on the Helmholtz differential equation is used to remove any tiny structures due to the effect of discretized grids. Then, we fabricate the designed cloaks and conduct the experiments in a couple of representative set-ups to validate the proposed design approach. The experiments are conducted inside both air and water. We found that the current cloaking design performs much better in air than in water and reveal the associated reason. Overall, this work paves the way for the acoustic cloaking design, fabrication, and experiments for future practical applications.

Design of resonant elastodynamic metasurfaces to control S0 Lamb waves using topology optimization

Control of guided waves has applications across length scales ranging from surface acoustic wave devices to seismic barriers. Resonant elastodynamic metasurfaces present attractive means of guided wave control by generating frequency stop-bandgaps using local resonators. This work addresses the systematic design of these resonators using a density-based topology optimization formulated as an eigenfrequency matching problem that tailors antiresonance eigenfrequencies. The effectiveness of our systematic design methodology is presented in a case study, where topologically optimized resonators are shown to prevent the propagation of the S₀ wave mode in an aluminum plate.

Intelligent design of the chiral metasurfaces for flexible targets: combining a deep neural network with a policy proximal optimization algorithm

Recently, deep reinforcement learning (DRL) for metasurface design has received increased attention for its excellent decision-making ability in complex problems. However, time-consuming numerical simulation has hindered the adoption of DRL-based design method. Here we apply the Deep learning-based virtual Environment Proximal Policy Optimization (DE-PPO) method to design the 3D chiral plasmonic metasurfaces for flexible targets and model the metasurface design process as a Markov decision process to help the training. A well trained DRL agent designs chiral metasurfaces that exhibit the optimal absolute circular dichroism value (typically, ∼ 0.4) at various target wavelengths such as 930 nm, 1000 nm, 1035 nm, and 1100 nm with great time efficiency. Besides, the training process of the PPO agent is exceptionally fast with the help of the deep neural network (DNN) auxiliary virtual environment. Also, this method changes all variable parameters of nanostructures simultaneously, reducing the size of the action vector and thus the output size of the DNN. Our proposed approach could find applications in efficient and intelligent design of nanophotonic devices.

Development and Optimization of Broadband Acoustic Metamaterial Absorber Based on Parallel-Connection Square Helmholtz Resonators

An acoustic metamaterial absorber of parallel-connection square Helmholtz resonators is proposed in this study, and its sound absorption coefficients are optimized to reduce the noise for the given conditions in the factory. A two-dimensional equivalent simulation model is built to obtain the initial value of parameters and a three-dimensional finite element model is constructed to simulate the sound absorption performance of the metamaterial cell, which aims to improve the research efficiency. The optimal parameters of metamaterial cells are obtained through the particle swarm optimization algorithm, and its effectiveness and accuracy are validated through preparing the experimental sample using 3D printing and measuring the sound absorption coefficient by the standing wave tube detection. The consistency between the experimental data and simulation data verifies feasibility of the proposed optimization method and usefulness of the developed acoustic metamaterial absorber, and the desired sound absorption performances for given conditions are achieved. The experimental results prove that parallel-connection square Helmholtz resonators can achieve an adjustable frequency spectrum for the low frequency noise control by parameter optimization, which is propitious to promote its application in reducing the noise in the factory.

Broadband acoustic lens design by reciprocity and optimization

A broadband acoustic lens is designed based on the principle of reciprocity and gradient-based optimization. Acoustic reciprocity is used to define the pressure at the focal point due to a source located in a far-field and to relate the response by a configuration of scatterers for an incident plane wave. The pressure at the focal point is maximized by rearranging the scatterers and supplying the gradients of absolute pressure at the focal point with respect to scatterer positions. Numerical examples are given for clusters of cylindrical voids and sets of elastic thin shells in water.

Gaussian-Based Machine Learning Algorithm for the Design and Characterization of a Porous Meta-Material for Acoustic Applications

The scope of this work is to consolidate research dealing with the vibroacoustics of periodic media. This investigation aims at developing and validating tools for the design and characterization of global vibroacoustic treatments based on foam cores with embedded periodic patterns, which allow passive control of acoustic paths in layered concepts. Firstly, a numerical test campaign is carried out by considering some perfectly rigid inclusions in a 3D-modeled porous structure; this causes the excitation of additional acoustic modes due to the periodic nature of the meta-core itself. Then, through the use of the Delany–Bazley–Miki equivalent fluid model, some design guidelines are provided in order to predict several possible sets of characteristic parameters (that is unit cell dimension and foam airflow resistivity) that, constrained by the imposition of the total thickness of the acoustic package, may satisfy the target functions (namely, the frequency at which the first Transmission Loss (TL) peak appears, together with its amplitude). Furthermore, when the Johnson–Champoux–Allard model is considered, a characterization task is performed, since the meta-material description is used in order to determine its response in terms of resonance frequency and the TL increase at such a frequency. Results are obtained through the implementation of machine learning algorithms, which may constitute a good basis in order to perform preliminary design considerations that could be interesting for further generalizations.

Conditional Wasserstein generative adversarial networks applied to acoustic metamaterial design

This work presents a method for the reduction of the total scattering cross section (TSCS) for a planar configuration of cylinders by means of generative modeling and deep learning. Currently, the minimization of TSCS requires repeated forward modelling at considerable computer resources, whereas deep learning can do this more efficiently. The conditional Wasserstein generative adversarial networks (cWGANs) model is proposed for minimization of TSCS in two dimensions by combining Wasserstein generative adversarial networks with convolutional neural networks to simulate TSCS of configuration of rigid scatterers. The proposed cWGAN model is enhanced by adding to it a coordinate convolution (CoordConv) layer. For a given number of cylinders, the cWGAN model generates images of 2D configurations of cylinders that minimize the TSCS. The proposed generative model is illustrated with examples for planar uniform configurations of rigid cylinders.

Introduction to the special issue on machine learning in acoustics

The use of machine learning (ML) in acoustics has received much attention in the last decade. ML is unique in that it can be applied to all areas of acoustics. ML has transformative potentials as it can extract statistically based new information about events observed in acoustic data. Acoustic data provide scientific and engineering insight ranging from biology and communications to ocean and Earth science. This special issue included 61 papers, illustrating the very diverse applications of ML in acoustics.