Efficient modelling and simulation of soft tissue deformation using mass-spring systems

Abstract

In this paper, we use a mass-spring system to simulate facial soft tissue deformation resulting from the bone realignment at the lower jaw area. Since the materials concerned often exhibit significant nonlinearity, correct simulation parameters are needed to capture the nonlinear characteristics in order to achieve satisfying simulation accuracy. We propose a neural network identification method that takes mass-spring structure into account and uses only two neural networks to identify these parameters, which are usually nonlinear functions. An adaptive learning rate formula is also introduced to improve the simulation accuracy and convergence speed.