Session: 20-17-01: Rising Stars of Mechanical Engineering

Paper Number: 173027

Embodied Structural Computing for Real-Time Stiffness Adaptation 

Mechanical adaptation of stiffness and shape seeks to emulate the efficiency of biological structures. We strive to make aircraft as maneuverable as birds through morphing, package space structures tens of meters in size into rocket launch fairings using patterns emerging during leaf growth, and are inspired by adaptive damping in animals to create legged-locomotion with rapid response in robotic systems. While current approaches pre-program adaptation to known stimuli, significant promise lies in embodied physical intelligence where structures autonomously sense, compute changes, and adapt. Adaptive architected materials with these functionalities would offer robust mechanical adaptation to unknown environments. However, computing the necessary mechanical changes requires global optimization, taking hours and thousands of iterations. This research will develop architected materials with controllable distributed stiffness and embedded information processing to enable real-time (~seconds) stiffness adaptation. The non-linear mechanics of the architected materials will offload much of the computational burden, with only simple algebraic operations falling to the software to control the structure. 

Given an input and desired mechanical response, an adaptive architected material in our approach learns the required mechanical changes through iterative physical interaction with the environment. In each iteration, the structure is loaded, a distributed sensor network measures a physical quantity (e.g., strain), and a model-free algorithm adjusts stiffness of metamaterial elements based on sensor measurements. Preliminary results demonstrate an architected material beam with adaptive stiffness maintaining a desired tip displacement, independent of the applied loading. Convergence in 7 - 15 cycles is achieved in simulation, for loads varying by up to 67%. This research quantifies the the speed and accuracy of physical learning in quasi-static problems, develops distributed sensing and computing approaches, and integrates mechanical sensing. The work will demonstrate learning in a cantilever beam architected material under a variety of loading scenarios. The physical learning algorithms will be developed in simulation with the structural mechanics captured by reduced order finite element models and then validated using physical prototypes of the stiffness-changing beam.

This research overcomes the computation bottleneck in mechanically adaptive structures by leveraging the mechanics of architected materials for computing, enabling real-time adaptation in response to environmental unknowns. We gain a fundamental understanding of how the mechanics of a structure can be used to "learn" a mapping between input and output without software-based optimization. The reliance on physical computations is especially relevant for extreme environments with limited electronics as well as for cybesecure systems. Applications include disturbance rejection in aircraft wings through stiffness adaptation, adaptive vehicle damping in rough terrain, and rapid shape morphing for electromagnetic devices.

Presenting Author: Maria Sakovsky Stanford University

Presenting Author Biography: Maria Sakovsky's work focuses on the use of shape adaptation to realize space structures with reconfigurable geometry, stiffness, and even non-mechanical performance . She received her PhD in Space Engineering from Caltech in 2018, where she developed a deployable satellite antenna based on origami concepts utilizing elastomer composites. She concurrently worked with NASA’s Jet Propulsion Laboratory on developing cryogenically rated thin-​ply composite antennas for deep space missions. She was awarded the ETH Zurich Postdoctoral Fellowship in 2018 as well as the DARPA Young Faculty Award in 2024.

Authors:

Maria Sakovsky Stanford University