UCLA

This dissertation builds the foundational knowledge required for creating a general material capable of adapting and learning to meet changing requirements. The novel material demonstrates its ability to learn mechanical behaviors through changes in its structure in an experimental setting which represents a broad leap in the capabilities of architected materials. Architected materials are systems which derive their apparent bulk properties from their structure rather than their chemical composition. In this dissertation, we propose that a mechanical structure with many similarities to the compositional framework used in artificial intelligence (AI) that can enable a material to learn, adapt, and relearn behaviors. The presented architected material is inspired by the artificial neural network (ANN) and uses mechanically analogous elements called a Mechanical Neural Network (MNN). To demonstrate the MNN’s potential, we constructed a simulation tool as well as a physical apparatus to show an MNN’s ability to learn and then uses these tools to explore several factors for increasing an MNNs utility. Specifically, by improving the accuracy of the learned behaviors, increasing an MNN’s potential to learn more properties simultaneously, and decreasing the amount of time required to obtain new behaviors. The findings from these studies are then used to explore avenues for future microscale MNNs by creating designs that can be scaled arbitrarily. The work shown in this dissertation has the potential to have dramatic effects on the material selection process and the life cycle of future products by granting flexibility in design requirements and giving the freedom to reteach behaviors to compensate for aging and damage.