Mechanical neural networks: Architected materials that learn behaviors

Area of Science:

• Materials Science
• Robotics
• Artificial Intelligence

Background:

• Few synthetic materials possess autonomous learning capabilities like living tissues.
• Existing materials struggle to adapt to changing conditions while learning new behaviors.

Purpose of the Study:

• To introduce mechanical neural networks (MNNs), architected materials capable of autonomous learning and adaptation.
• To investigate the factors influencing MNN learning capabilities.

Main Methods:

• Developed mechanical neural networks (MNNs) by tuning the stiffness of constituent beams.
• Fabricated an example lattice to demonstrate simultaneous learning of multiple mechanical behaviors.
• Studied the impact of lattice size, configuration, algorithm, and stiffness tunability on MNN learning.

Main Results:

• Demonstrated MNNs' ability to learn multiple mechanical behaviors concurrently.
• Identified key parameters affecting MNN learning efficiency and adaptability.
• Showcased the potential for AI-driven materials with adaptive properties.

Conclusions:

• Mechanical neural networks represent a significant advancement towards intelligent, adaptive materials.
• This work provides a foundation for developing artificial-intelligent (AI) materials that learn and evolve.
• MNNs offer a novel approach to designing materials with unprecedented responsiveness.