K. DONDA: Low-frequency Absorbing Acoustic Metasurfaces: Deep-learning Approach and Experimental Demonstration

Abstract:
The advent and development of acoustic metamaterials/metasurfaces in recent years has overturned conventional means in all aspects of acoustic waves propagation and manipulation. In the context of sound absorption, it has offered an unprecedented expansion of our ability to attenuate low-frequency sound beyond the classical physical limits. The main aim of this dissertation is to conceive and design acoustic metasurfaces for extreme low-frequency absorption (<100Hz). First, the concept of multicoiled metasurface absorber (MCM) is proposed. The effectiveness of its physical mechanism is theoretically, numerically, and experientially demonstrated. The presented MCM is capable of fully absorbing acoustic energy at an extremely low frequency of 50Hz with a deep subwavelength thickness (λ/527). To circumvent the conventional physics- and rule-based approaches and accelerate the design process, a novel deep learning-based framework is introduced in this dissertation. Specifically, the convolutional neural network (CNN) and conditional generative adversarial networks (CGAN) are implemented to simulate and optimize complex metasurface absorber structures. The developed deep learning-based framework for the acoustic metasurface absorber can be potentially extended to the design and optimization of other acoustic devices. This dissertation provides a new way for deep learning-enabled acoustic metasurface designs that will allow the physical acoustics community to focus more on truly creative ideas. This will be led to solving complex design problems that have yet to be explored by the machine, rather than on tedious trial and error processes.

Keywords:
acoustic metasurfaces and metamaterials, low-frequency absorption, deep learning, convolutional neural network, conditional generative adversarial network, inverse design.

Composition of the jury:
> Reporters:
- Abdelkrim KHELIF, CNRS senior researcher, Université de Franche-Comté
- Yan PENNEC, Professor, Université de Lille
> Examiners:
- Agnès MAUREL, CNRS senior researcher, ESPCI, Paris
- Nico DECLERCQ, Professor, Georgia Tech Lorraine
- Jean-François GANGAUFFER, Professor, Université de Lorraine
> Director of thesis:
- Badreddine ASSOUAR, CNRS senior resarcher, Université de Lorraine