Enhanced Design of soft pneumatic finger for bionic Hands Applications Using ANN and PSO

Abstract

Soft robotics is growing as a leading approach to the development of soft bionic limbs. This paper proposes a solution to the problems caused by the complex structure, limited flexibility, and heavy weight of conventional rigid robotic hands. The proposed method focuses on designing an anthropomorphic pneumatic flexible finger using the pneumatic networks (PneuNets) structure. The particular features of this actuator's motion are determined by adjustments to both the geometry of the embedded chambers and the material properties of the walls. This paper focused on manipulating the geometry factor of the soft finger structure in order to modulate the resulting bending precisely. A finite element analysis of twelve models of the soft finger with different geometric parameters yields significant results. Increasing the wall thickness from 2 to 3 mm reduces deformation by approximately 13%. However, the number of chambers per joint has a greater effect on deformation. Models with an extra chamber per joint show an approximately 41% increase in deformation at the same input pressures. Additionally, a feedforward Artificial Neural Network (ANN) has been developed to facilitate a deeper understanding forward kinematics for the soft finger. This enables the prediction of the bending angle, and end-tip coordinates from the input pressure.