A fast parametric deformation mechanism for virtual reality applications

Virtual reality technologies have been adopted in a wide variety of applications for its interactive ability and realistic senses. Despite early implementations regard VR only as a medium for lively animation; a practical VR work must deliver precise deformation on virtual objects based on real-time interactions. The exact ability is especially important for users who utilize VR to do collaborative design, for it will greatly reduce the amount of on-line computations on operating substance-based interactions, and consequently facilitates the collaboration. Therefore, this research will employ neural networks to memorize the deformation behavior of solid objects, and then perform instant and accurate deformations in the virtual environment. The proposed method also allows design variations for parametric features, and uses feature parameters as variable switches to adjust the deformation mechanism. There are three steps in the method: (1) For a sample object, generate force-induced deformations using the finite-element method; (2) memorize the surface displacements with artificial neural networks; and (3) convert the parametric deformation matrices into Behavioral Modules for the virtual reality engine. In the implementations, ANSYS is used to generate model deformations, and MATLAB is used to perform neural training. Finally, a virtual environment is built using Virtools where customized Building Blocks are created to present interactive deformation behavior. Experiments were carried out on an Intel XEON workstation with nVIDIA Quadro4 750GL display device. Sample workparts are tested to examine the ability of the method. The results show that both training accuracy and real-time capability are more than satisfactory.