On thedynamics of a magnetostriction-based soft robotic manipulator: Closed form and Machine Learning approaches

The impetus of this study is to investigate the dynamics of a flexible manipulator. The model consists of two beams, each of which is surrounded by an electric coil. Each beam consists of two layers, including Metglas 2605SC as the magnetostrictive material and steel. Once an electric current is applied to the coils surrounding the beams due to generation of magnetic field the Metglas on the steel undergoes mechanical deformation which is accordingly conveyed to the whole structure. The end-effector control parameters are the two currents applied to the electric coils. The governing motion equations based on Euler-Bernoulli beam assumptions and large deformation conditions are derived and discretized to reduced order model based on Galerkin technique. The resulted equation is then numerically solved by means of Runge-Kutta method in Matlab software. To control the end-effector position of the manipulator in the defined work space, the relation between the applied electric currents and the end effector position and orientation were determined by means of the solution of the governing equations. However, in order to achieve the inverse relation between the essential currents needed for manipulator to reach a desired position, deep machine learning methods were used. Two learning algorithms including a four-layered artificial neural network and a neurofuzzy network were applied to capture the inverse dynamics relation. The results revealed that the network based on fuzzy logic exhibited less error than the neural network.