Non-Newtonian Fluid Simulation Based On Physical And Data-Driven Model

Fluid is a common phenomenon that could be widely seen in nature,daily life and various industrial applications.The simulation of fluid phenomenon,especially fluid animation,is always a research hotspot in the field of computer graphics.With the development of technology,people began to pursue more realistic and magnificent simulation scenes,real-time and efficient fluid animation effects.Therefore,physics-based fluid animation technology came into being.Since traditional fluid animation technology can already simulate the motion of Newtonian fluids realistically,I proposed to extend this series of technique to non-Newtonian fluid animation and solve some important issues.To address the deficiencies of limited model expression ability,complex numerical calculations,difficult fluid-solid interaction,and time-consuming large-scale simulating calculations,a series of simulation algorithms for non-Newtonian fluid animation are proposed,which focus on three aspects including non-Newtonian fluid algorithm acceleration,boundary processing,and large-scale simulation acceleration.The main contributions of this paper are as follows:1)To slove the problems of low simulation efficiency and poor numerical stability of non-Newtonian fluid animation,the prediction-correction SPH method is extended to deal with non-Newtonian fluids.First,the velocity of fluid particles is predicted based on the external forces,and then be corrected by setting an independent intensity coefficient for each fluid particle to satisfy the incompressible condition in the local area.The iterative scheme it applied to meet the global incompressibility.Next,the viscosity stress tensor is corrected and updated iteratively according to the predicted velocity field until the viscosity is stable.Last,the semi-implicit Euler method is used to update the position of each particle iteratively.This method can ensure that the simulation animation still has high numerical stability under a large time step,which greatly improves the simulation efficiency2)To solve the problem of unreasonable boundary conditions of the existing SPH non-Newtonian fluid simulation,a boundary treatment method suitable for non-Newtonian fluid simulation under the prediction-correction method is proposed,to alleviate the loss of physical properties when simulating fluid.In this method,solids are sampled as single-layer boundary particles,and the corresponding density calculation formula is given.Fluid particles are interacting directly with the boundary particlesunder our new interaction force calculation formula.This method can be integrated into the existing prediction-correction-based non-Newtonian fluid simulation algorithm framework.Besides,under friction boundary conditions,non-Newtonian fluids can exhibit more diverse physical characteristics by adjusting the parameters,and have a good sense of reality effect.3)Traditional physical-based fluid simulation systems are time consuming and requires large computational resources to generate large-scale fluid flows.Data-driven methods provide us a new method of data-driven to accelerate simulations.This paper presents a novel end-to-end deep learning neural network that can automatically generate a model for fluid animation based-on Lagrangian fluid simulation data.This approach synthesizes velocity fields with irregular Lagrangian data structure using neural network.Every fluid particle is treated independently and identically.We use symmetric functions to capture space structure and interactions among particles and design different network structures to learn various hierarchical features of fluid.We test this method using several data sets and applications in various scenes with different sizes.Our experiments show that the model is able to infer velocity field with realistic details such as splashes.In addition,it is also able to predict the velocity field of large simulation scene with the trained model of the small scene in the same type.Our method not only solved the problem of slow simulation speed and large memory consumption when simulation large scene using traditional method,but also solves the problem of requiring large number of large scene data sets during model learning process when using the other data-friven methods.This method shows significant speed-ups,especially on large scene simulations.