Compliant Mechano-Intelligent Systems

In this seminar a variety of mechanical systems will be presented that physically embody
intelligence via the deformation of constituent compliant elements. So long as their elements do
not yield, these systems, called compliant mechano-intelligent systems, achieve high
repeatability, experience no backlash, avoid friction and hysteresis, produce no wear or
particulates, and can be fabricated on the micro-scale. Consequently, they can be made to learn
behaviors or perform complex computations with sufficient precision, accuracy, and speed in a
small enough volume to enable artificial-intelligent metamaterials. Specifically, this seminar will
review the concept of mechanical neural networks (MNNs) as architected materials that
autonomously learn their behaviors and properties with increased exposure to unanticipated
ambient loading conditions. Like how artificial neural networks (ANNs) learn by tuning their
scalar weights to map know input and output data, MNNs are physical versions of ANNs in that
they learn by tuning the stiffness of their constituent beams to map known input and output
forces and displacements. This seminar will also introduce and compare both digital and analog
approaches for computing mathematical operations in harsh environments where electronics
would fail. Such environments include: radiation-rich, extreme temperature, chemically
corrosive, large magnetic field, excessively dusty, highly statically charged, or other
environments that experience large fluctuations in ambient conditions or surges in electricity.
The digital approach leverages flexure-based bi-stable memory bits and logic gates joined
together to perform calculations using discrete states of deformation, while the analog approach
uses a single one-piece structure that deforms according to continuous input displacements.