Designing a Cross Trainer Using an Artificial Neural Network

Abstract

Cross-trainers are machines that use link mechanisms to mimic walking or running as part of workout sessions or rehabilitation systems. The simplest cross-trainer incorporates a crank-rocker or a crank-slider mechanism and provides a nearly elliptical path for foot motion. However, the natural human foot trajectories are far from being elliptical. Therefore, existing designs require modifications. Artificial neural networks are used for this purpose. Instead of trying to match the foot trajectory directly, here we tried to match different geometric properties of the area enclosed by the foot trajectory. Neural networks are trained to predict these geometric properties as outputs with the dimensions of the linkage as inputs. With the help of the same trained network, the “best-fit dimensions” were predicted for the desired trajectories.