Fracture Simulation of Elastic Objects Using the Tresca Yield Criterion

초록

This paper presents two approaches to enhance realism in virtual environments by accurately simulating the destruction of elastic objects: fracture generation using deep learning (DL) and residue emission based on the yield criterion. These approaches apply a particle-based material point method (MPM) to simulate the non-linearity during the destruction process. Fracture generation using DL utilizes high-quality MPM-based simulation data. The method learns based on training data from various environments and inference models, enabling high-quality fracture generation processes even on low-performance hardware. Residue emission based on the yield criterion obtains an approximation formula using the Tresca yield criterion to determine the destruction of an elastic object. The physical phenomenon of the elastic object resisting external forces is the principal stress. The difference between the maximum and minimum principal stresses caused by external forces is used to approximate the destruction criterion. At the moment of destruction of the elastic object, the accumulated force causes the residues to be emitted. Experiments were conducted in the same environment to compare the traditional MPM and the proposed method. Traditional MPM requires a high computational cost to simulate detailed compression fractures, which results in simple tearing. In contrast, the proposed method simulated the emission caused by stress at the moment of fracture with 20% less computational cost. Consequently, it allows for realistic compression fracture simulations of various types of elastic objects, such as the juice from compressed grapes or the yolk oozing from a soft-boiled egg.

키워드