Harnessing Physical Reservoir Computing in Nonlinear Mechanical Metastructures

Mechanical metamaterials and metastructures have gained increasing interests due to their unique properties and promising engineering applications. With the recent advances in new technologies, researchers are attempting to embed more intelligence directly within the mechanical domain of functional metastructures, enabling them to perform intelligent tasks such as sensing the environment, changing properties and geometries, and making decisions in an even more proactive and autonomous manner, as compared to traditional systems with add-on electronics-based intelligence. However, the existing efforts mainly rely on ad-hoc designs to achieve limited tasks within pre-defined configurations, and there is lack of a systematic foundation for constructing and integrating the different aspects of mechano-intelligence, such as observation, learning, decision-making and execution. As a first step toward the systematic design of mechanical metastructures with mechano-intelligence, this research proposes to advance the state of art by exploiting the versatile computing capability embedded in nonlinear metastructures through an efficient machine learning framework called Physical Reservoir Computing. We show that the nonlinear and high-dimensional dynamics of the mechanical metastructure can be rich computational resources to perform various benchmark tasks without re-designing the structure, indicating the versatile computing potential of the metastructure. Further, we investigate the relationship between the parameters and the computational performance of the proposed system, revealing how the structural dynamics may affect the computing power and how to maximize the performance of the metastructure reservoir. Overall, this research examines a novel and effective approach to achieve mechano-computing capability in mechanical metastructures, which could form the basis towards creating future systems with more integrated mechano-intelligence.