Application of neural network fitting for modeling the pneumatic networks bending soft actuator behavior

Abstract

Soft actuators have recently gained a lot of interests as an emerging topic, although complete methodologies for modeling soft actuators are still missing. Identifying and forecasting the behaviour of soft actuators is difficult due to the nonlinear behaviour of the materials used, the complicated geometries they form, and the wide range of motions they produce. In this paper, we demonstrated how to use neural network technology to describe the motion and produced force that the pneumatic network bending soft actuator can create at various input pressures. To confirm the results, three separate neural network models for three different modeling modes were constructed and evaluated with different input data sets. First, the dimension model, which deals with changes in the form and geometry of the soft actuator and their influence on its response at various pressure inputs. Second, the free force model, which simulates the motion of a soft actuator in free space without any external disturbances. Finally, the blocked force model, which may simulate a real-world soft actuator that is subjected to an external force. The input data sets were created with ABAQUS/CAE software, which replicates the behavior of the soft actuator and uses this data to train the neural network models.