Computational Homogenization of Yarn-Level Cloth

We report on the state of the art in efficient simulation of yarn-level frabric from the computer graphics (SIGGRAPH) community. We first present our method for reproducing the qualitative effects of knitted and woven fabrics effects using a thin-shell continuum mechanics simulation. We accomplish this through numerical homogenization: we first use a large number of simulations of individual knotted and colliding yarns to build a model of the potential energy density of the cloth, and then use this energy density function to compute stresses in a continuum mechanics simulator. We model several yarn-based materials, including both woven and knitted fabrics. Our model faithfully reproduces expected effects like the relative stiffness of woven fabrics, the high compliance and anisotropy of knitted fabrics, and the coupling of stretching and curling behaviors in stockinette knit patterns. This method also has computationally convenient side effects, like the ability to re-use pre-computed yarn geometry to efficiently display millions of deforming threads for real-time animations.

We then applied this methodology to the inverse-modeling of yarn-level mechanics of cloth, based on the mechanical response of fabrics in the real world. We first compiled a database from physical tests of several different knitted fabrics used in the textile industry. These data span different types of complex knit patterns, yarn compositions, and fabric finishes, and the results demonstrate diverse physical properties like stiffness, nonlinearity, and anisotropy. We then developed a system for approximating these mechanical responses with yarn-level cloth simulation.

This talk summarizes three years of publications in the computer graphics and simulation community.