Summary:
UCLA researchers in the Department of Mechanical and Aerospace Engineering have developed a new type of meta-material that can actively adapt its mechanical properties from exposure to external stimuli.
Background:
As technologies have evolved, humans developed new methods to produce unique materials, such as plastics, steels and composites, whose mechanical properties can be customized during production to meet a particular demand. These methods have allowed for the tailored synthesis and production of materials with novel and desirable mechanical properties. However, once the material is chosen and manufactured into a device or product, its mechanical properties are usually fixed. There are limited examples of materials, such as shape memory polymers and piezoelectric ceramics, that exhibit varying mechanical properties after production. However, these materials are subject to their physical limits, difficult to control, and do not possess the inherent capability to learn. They therefore cannot achieve a specified mechanical property in response to varying external stimuli. There remains an unmet need for smart materials that are tunable and responsive to their environment.
Innovation:
Professor Hopkins and his research team have invented a new kind of architected meta-material referred to as mechanical neural networks (MNNs). These consist of an array of interconnected tunable beams which are driven by forces or displacement inputs. It is a physical embodiment of an artificial neural network (ANN) that can autonomously adapt its mechanical properties via prolonged exposure to external forces. The MNN is able to learn and control its bulk mechanical properties (e.g., shape morphing, acoustic wave propagation, and mechanical computation) and bulk properties (e.g., Poisson’s ratio, modulus, and density) by actively tuning the local properties/stiffness of each individual beam. This invention is the first mechanical meta-material that shows learning capabilities analogous to how the ANNs map inputs to outputs. Additionally, the tunable beams used in this invention can be replaced by a variety of materials that continuously exhibit the desired mechanical properties previously learned from an external influence. This new type of meta-material introduces a novel class of smart, artificially intelligent materials. The patent application for this technology was published on January 25, 2024: Mechanical neural networks: architected materials that learn behaviors.
Demonstration Video:
Compliant Mechanisms that LEARN! - Mechanical Neural Network Architected Materials - YouTube
Potential Applications:
• Aircraft wings
• Robotic or vehicular skin
• Construction materials
• Acoustic imaging
• Sports equipment (protective gear; training aids)
• Controllable mechanical components and supporting structure
Advantages:
• Autonomously adaptable mechanical properties
• Mechanical properties can be actively tuned after material selection
• Learning abilities to adapt to unforeseen conditions
Development to Date:
Invention was reduced to practice in a laboratory setting and successfully demonstrated.
Related Papers and Patents (from the inventors only):
Mechanical neural networks: Architected materials that learn behaviors
Lee, R. H., Sainaghi, P., and Hopkins, J. B. (May 18, 2023). "Comparing Mechanical Neural-Network Learning Algorithms." ASME. J. Mech. Des. July 2023; 145(7): 071704. https://doi.org/10.1115/1.4062313
Press Release:
UCLA Engineers Design AI Material That Learns Behaviors and Adapts to Changing Conditions
Related Patent:
US 10065322 B2: “Actively Controlled Microarchitectures with Programmable Bulk Material Properties,” Hopkins, J.B., Song, Y., assigned to The Regents of the University of California, September 2018
Reference: UCLA Case No. 2022-317