Modeling of Soft Object Deformation using Finite Element Differential Neural Networks Academic Article in Scopus uri icon

abstract

  • © 2018 This paper presents a nonparametric modeling based on Finite Element Differential Neural Networks (FEDNN) of soft object deformation dynamics. The construction of the adaptive model is based on the Finite Element Method (FEM) and the Differential Neural Network (DNN) methodology. The input is taken by the displacement at each node collected from experimental data obtained from a motion capture system. A soft object sample is characterized using an equipment for stress tests and the nodes are collocated in its surface. The nodes information is used only to train the FEDNN. To verify the qualitative behavior of the suggested methodology, here the estimated trajectories are compared with the Motion Capture spatial position vector of the surface of the sample soft object. The adaptive laws for weights ensure the closeness of FEDNN trajectories to the tissue dynamics.

publication date

  • January 1, 2018