Modeling and Machine Learning Aided Analysis of a Claw-Less Magnetically Coupled Ball-Drive Design

Ball-driven mobility platforms have shown that spherical wheels can enable substantial freedom of mobility for ground vehicles. Accurate and robust actuation of spherical wheels for high acceleration maneuvers and graded terrains can, however, be challenging. In this paper, a novel design for a magnetically coupled ball drive is presented. The proposed design utilizes an internal support structure and magnetic coupling to eliminate the need for an external claw-like support structure. A model of the proposed design is developed and used to evaluate the slip/no-slip operational window. Due to the high-dimensional nature of the model, the design space is sampled using randomly generated design instances and the data is used to train a support vector classification machine. Principal component analysis and feature importance detection are used to identify critical parameters that control the slip behavior and the feasible (no-slip) design space. The classification shows an increase in the feasible design space with the addition of, and increase in, the magnetic coupling force. Based on the results of the machine learning algorithm, FEA design tools and experimental testing are used to design a spherical magnetic coupler array configuration that can realize the desired magnetic coupling force for the ball drive.