A Comparative Study of Predictive Models for Nafion-117 IPMC Soft Actuators

Abstract

Ionic polymer-metal composites are electro-active soft actuators that, when stimulated by an electric field, convert electrical energy into mechanical energy. This study focuses on an ionic polymer-metal composite soft actuator that has been realized with Nafion-117 and platinum electrodes. Three black-box models, i.e., curve fitting, multi-layer perceptron, and long short-term memory recurrent neural network, are designed based on the forces exerted by the soft actuator at fixed displacements when stimulated by different voltages. The capability of the three black-box models to predict unseen forces is evaluated and compared. This study shows that the multi-layer perceptron has the best predictive capability in capturing the dynamics of unseen force data, with a root mean square error of 0.109 mN and computation time of $13 \\mu \\mathrm{s}$