Mechanical Neural Networks: Materials with Variable Physical Properties

Materials made of mechanical neural networks can learn to adapt their physical properties

According to my colleagues and myself, a new material type can improve its ability in dealing with unexpected forces by learning from a unique lattice with connections that are variable stiffness.

The new material is an architected material that gets its properties more from its geometry and design than the material it’s made of. Consider Velcro-like fabric closures, such as hook-and loop. No matter if it’s made of cotton, plastic, or another substance. The material will be sticky as long as the fabric has stiff hooks on one side and fluffy loops on the other.

We modeled the architecture of our new material after an artificial neural net–layers and nodes interconnected that can be trained to perform tasks by adjusting how much weight they give each connection. We hypothesized a mechanical lattice made up of physical nodes can be trained to acquire certain mechanical properties through adjusting the rigidity of each connection.


Materials Made of Mechanical Neural Networks Can Learn to Adapt Their Physical Properties