A machine learning application for form-finding of tensegrity structures

Abstract

Tensegrity (tensional integrity) is a structural principle where rigid elements (struts) under compression are held together by a network of elastic elements (cables) under tension. Tensegrity structures have many applications in modelling the natural world. Tensegrity research has been applied to fields including robotics, art, architecture, and biology. In recent years, computer simulation has been introduced as a tool to allow researchers to design, build, and simulate tensegrity structures. Structures designed both as physical models and in simulation software can require several iterations of fine adjustments. In this thesis, we develop a form-finding application to reduce the iterative adjustments required when designing a tensegrity structure. Form-finding is the process of finding a structural configuration capable of a state of self-stressed equilibrium – when tension and compression stabilize the structure. Our form-finding application uses a tensegrity structure that is not necessarily in an equilibrium state represented as a graph as input, and produces either (a) failure when no equilibrium state is possible, or (b) a fully attributed labeled graph of a tensegrity structure in an equilibrium state. In this thesis, we use an efficient fitness function and genetic algorithms to find the stable state of a tensegrity structure. Through our form-finding application, we aim to promote the use of computer simulation, and collaboration between tensegrity researchers.