This Novel "Mechanical Neural Network" Architected Material Learns to Respond to Its Environment

Built using voice coils, strain gauges, and flexures, this mechanical marvel acts as a physical neural network — "learning" as it moves.

Researchers from the University of California at Los Angeles and the University of Twente have come up with an adaptive material capable of actively responding to changing conditions — in their own words, a "mechanical neural network" that can learn.

"This research introduces and demonstrates an artificial intelligent material that can learn to exhibit the desired behaviors and properties upon increased exposure to ambient conditions," claims Jonathan Hopkins, mechanical and aerospace engineering professor at the UCLA Samueli School of Engineering and research lead on the project. "The same foundational principles that are used in machine learning are used to give this material its smart and adaptive properties."

This mechanical device is a functioning neural network, and could one day boost aircraft efficiencies or provide reactive armor. (📹: Flexible Research Group, UCLA)

The material itself somewhat bulky, at least in prototype form: The mechanical neural network the team has created takes the form of beams arranged in a triangular lattice pattern, each being made of a voice coil, typically used in speakers to convert magnetic fields into the movement required to make sound, strain gauges, and flexures that mean the beams can alter their length as required.

An optimization algorithm takes input from the strain gauges then calculates how the material should react to the forces being applied at any given moment — and while early prototypes had issues with lag between input and reaction, the team spent five years ironing out the kinks to the point where the mechanical neural network material can "learn" and react in real-time.

"An example lattice was fabricated to demonstrate its ability to learn multiple mechanical behaviors simultaneously," the team explains of its prototype, "and a study was conducted to determine the effect of lattice size, packing configuration, algorithm type, behavior number, and linear-versus-nonlinear stiffness tunability on MNN learning as proposed. Thus, this work lays the foundation for artificial-intelligent (AI) materials that can learn behaviors and properties."

The current prototype is roughly the size of a microwave oven, but the team is keen to push forwards with miniaturization and extension into the third dimension — and envisions applications including use in aircraft wings to morph the shape in response to wind patterns to boost efficiency, adding reactive rigidity to buildings to better withstand earthquakes and other disasters, shockwave-deflecting reactive armor, or even the creation of surfaces able to perform acoustic imaging.

The team's work has been published in the journal Science Robotics under open-access terms.

Gareth Halfacree
Freelance journalist, technical author, hacker, tinkerer, erstwhile sysadmin. For hire:
Latest articles
Sponsored articles
Related articles
Latest articles
Read more
Related articles