Session: 03-09-03: Design of Engineering Materials

Paper Number: 100241

100241 - Acoustic Metamaterial Design Using Deep Reinforcement Learning 

The last decade has witnessed a surge of scientific publications in which deep learning,  reinforcement learning, and generative modeling were applied in different areas of science and engineering. Recent advances in the field of machine learning have enabled a novel data-driven approach with a great promise to solve problems such as the inverse design in the acoustic metamaterial. In this talk, we present our findings and methodologies for applying reinforcement learning to the design of acoustic metamaterials [1]. As the application of the method, here we design an acoustic cloak. Our reinforcement learning agents are capable of discovering configurations of cylindrical scatterers situated in water and interacting with an acoustic plane wave which minimizes their total scattering cross-section (TSCS). These agents are successful in the optimization of two different parametric designs, i.e. radii and positions of each scatterer. Our motivation is to eventually create an acoustic cloaking device. Once built, this device will render objects invisible to incoming waves in the bandwidth for which it is optimized. The design parameters for our cloaking device are the positioning of multiple cylindrical scatterers on a two-dimensional (2D) grid (position adjustment) or the radius of each cylindrical scatterer on a 2D grid (radius adjustment). Our aim is to compare the performance of two different RL algorithms for each of these parameters.

The resultant designs produced by reinforcement learning algorithms such as double deep Q-learning network and deep deterministic policy gradient algorithms are comparable and in some cases superior to those produced by the gradient-based optimization solver such as fmincon. Through evaluating the gradients of TSCS and other information about the state of the configuration, the reinforcement learning agent perturbatively adjusts design parameters, considering multiple scattering between the scatterers. As each adjustment is made, the reinforcement learning agent receives a reward negatively proportional to the root mean square of the TSCS across a range of wavenumbers. By maximizing its reward per episode, the agent discovers designs with low scattering. However, significant computational resources are required to train reinforcement learning models to completion [1]. This poses a significant challenge when attempting to increase the complexity of a design; as training times increase to unfeasible levels. Methods to counteract this challenge are discussed in this talk. These methods include the utilization of the Julia programming language as well as multiprocessing and multithreaded programming. Together they create a synergy that reduces training times by orders of magnitude.

 

References:
[1] T. Shah, L. Zhuo, P. Lai, A. Rosa-Moreno, F. Amirkulova, and P. Gerstoft. "Reinforcement learning applied to metamaterial design," J. Acoust. Soc. Am. 150(1), 321-338 (2021).

 

Presenting Author: Feruza Amirkulova San Jose State University

Presenting Author Biography: Dr. Feruza Amirkulova is an Assistant Professor of Mechanical Engineering at San Jose State University. She has served as a faculty member in the Mechanical Engineering Department at SJSU since 2018. She graduated from Rutgers University with a Ph.D. and MSc in Mechanical and Aerospace Engineering. Previously, she completed her Ph.D. in Technique, specializing in Civil Engineering from Samarkand State University. She received her BSc and MSc in Mathematics from Samarkand State University, graduating with honors.<br/><br/>Dr. Amirkulova has served as the Chair of Special Session at the Acoustical Society of America, the Co-chair of Special Session SV WiE 2019, and a Grant review panelist at the National Science Foundation. She has also served as the invited reviewer at Acoustica, Journal of Acoustical Society of America, Wave Motion and Mechanics Research. She enjoys cooking, running, traveling, and spending time with her children and grandchildren in her free time.<br/><br/>Dr. Amirkulova is the recent recipient of the Guidry Family Faculty Teaching Fellow appointment. With her new appointment, she plans to engage students’ learning in the areas of conducting sound and vibration measurements, and hands-on experiences using simulations.

Authors:

Feruza Amirkulova San Jose State University
Tristan Shah San Jose State University