The Application of Machine Learning to Active Tensegrity Structures

Abstract

This thesis explores an application of machine learning to solve a civil engineering problem. The problem is to maintain the deck of an active tensegrity bridge flat and planar regardless of the loading the structure experiences. The problem must be solved quickly enough so that the structure can adapt instantaneously. Two machine learning algorithms, artificial neural networks with back-propagation learning and random forests are implemented and adapted to solve the problem. Each algorithm performs well, reducing the deck’s deflection by an average of 1.3%. The best performing algorithm, random forests, is then used to create a simulation where different loads are applied to the active tensegrity bridge and the structure reacts to maintain a flat and planar deck.