A Composite Neural Gas-Elman Network that Captures Real-World Elastic Behavior of 3D Objects

This paper employs a neural gas network to obtain a compressed model of 3D geometry of objects, which accounts for elastic behavior as well. Based on the output of the network, we are able to cluster the object into areas of similar geometry and elasticity and then represent the elastic behavior of each cluster by an Elman neural network that models force-displacement behavior without explicit computation of elastic parameters. This approach allows us not only to recover the elastic parameters in the sampled points (those points for which we have measurements) but also provides us with an estimate on the elastic behavior in points that are not part of the sampling point set. The comparison of the Elman network with the three-element viscoelastic model indicates that the neural approach estimates better nonlinear elastic behaviors than its counterpart