Mechanical neural networks: Architected materials that learn behaviors

Abstract

Aside from some living tissues, few materials can autonomously learn to exhibit desired behaviors as a consequence of prolonged exposure to unanticipated ambient loading scenarios. Still fewer materials can continue to exhibit previously learned behaviors in the midst of changing conditions (e.g., rising levels of internal damage, varying fixturing scenarios, and fluctuating external loads) while also acquiring new behaviors best suited for the situation at hand. Here, we describe a class of architected materials, called mechanical neural networks (MNNs), that achieve such learning capabilities by tuning the stiffness of their constituent beams similar to how artificial neural networks (ANNs) tune their weights. An example lattice was fabricated to demonstrate its ability to learn multiple mechanical behaviors simultaneously, 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.

Deep Learning in Mechanical Metamaterials: From Prediction and Generation to Inverse Design

“…Reproduced with permission. [ 246 ] Copyright 2022, American Association for the Advancement of Science. c) Deep‐sea creature, hexactinellid sponge Euplectella aspergillum , with hierarchical assembly of the siliceous skeletal lattice.…”

“…For instance, a mechanical mechanical logic gate can only perform certain types of logic operations, [245,250,253] and a mechanical neural network can only respond to specific types of force and displacement. [246] In contrast, Western Pacific hexactinellid sponges, such as Euplectella aspergillum, which lack true tissues and organs, can survive extreme deep-sea environments owing to their cylindrical latticelike structure with at least six hierarchical levels (Figure 19c). Moreover, biological neural networks learn to respond to a variety of external stimuli from the environment by adjusting both topology and weights (Figure 19d).…”

Signal Propagation in Resettable Mechanical Logic

“…We identify rst law limits to signal propagation through bistable elements as well as AND gates and link the scale-independent parameters to simulated and measured performance limitations for logical operations. Future work building on these deterministic methods could enable mechanical computation with inputs keyed to speci c, multi-domain environmental temperature, biological, chemical, magnetic, vibration, or acceleration signals with applications such as powerless rare-event detection on space missions 28,29 , extreme temperature data storage 30 , authentication tagging of high value items 31,32 , disposable health sensors 33 , and smart materials 34 .…”

Embodying Multifunctional Mechano‐Intelligence in and Through Phononic Metastructures Harnessing Physical Reservoir Computing

“…Previous studies have attempted to address certain aspects of intelligence into the design of various systems. [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15][16][17][18][19] While promising, these mechano-intelligence (MI) efforts have been primarily limited to specific cases with narrowly defined functions. Overall, a systematic foundation remains lacking for effective synthesizing and integrating the various intelligent elements, namely information perception, decisionmaking, and commanding, in multifunctional structures.…”