Moving Particle Semi-implicit Method: A Meshfree Particle Method for Fluid Dynamics

Moving Particle Semi-implicit Method

A Meshfree Particle Method for Fluid Dynamics

Seiichi Koshizuka

The University of Tokyo, Tokyo, Japan

Kazuya Shibata

The University of Tokyo, Tokyo, Japan

Masahiro Kondo

The University of Tokyo, Tokyo, Japan

Takuya Matsunaga

The University of Tokyo, Tokyo, Japan

Table of Contents

Cover image

Title page

Copyright

Preface

Chapter 1. Introduction

Abstract

1.1 Concept of Particle Methods

1.2 MPS Method

1.3 Research History of Particle Methods

References

Chapter 2. Fundamentals of Fluid Simulation by the MPS Method

Abstract

2.1 The Elements of the MPS Method

2.2 Basic Theory of the MPS Method

2.3 Outline of Simulation Programs

2.4 Exercise of Simulation

2.5 Hints for Exercises

2.6 Frequently Asked Questions

References

Chapter 3. Extended Algorithms

Abstract

3.1 Compressible-Incompressible Unified Algorithm

3.2 Explicit Algorithm Using Pseudo-Compressibility

3.3 Symplectic Scheme

3.4 Arbitrary Lagrangian-Eulerian

3.5 Rigid Body Model

3.6 Structural Analysis

References

Further Reading

Chapter 4. Boundary Conditions

Abstract

4.1 Introduction

4.2 Solid Wall

4.3 Free Surface

4.4 Inlet and Outlet Boundary Modeling

References

Chapter 5. Surface Tension Models in Particle Methods

Abstract

5.1 Surface Tension Calculation Using CSF Equation

5.2 Surface Tension Calculation Based on a Pairwise Potential

5.3 Applications of the Surface Tension Models Using the MPS Method

References

Chapter 6. Advanced Techniques

Abstract

6.1 Liquid–Solid Phase Change Model

6.2 Gas–Liquid Two-Phase Flow and Phase Change Model

6.3 Turbulence

6.4 Suppression of Pressure Fluctuations

6.5 Higher-Order Schemes

6.6 Parallel Computing

6.7 Multiresolutions

6.8 V&V and Applications

References

Index

Copyright

Academic Press is an imprint of Elsevier

125 London Wall, London EC2Y 5AS, United Kingdom

525 B Street, Suite 1650, San Diego, CA 92101, USA

50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States

The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom

Copyright © 2018 Elsevier Inc. All rights reserved.

No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Details on how to seek permission, further information about the Publisher’s permissions policies and our arrangements with organizations such as the Copyright Clearance Center and the Copyright Licensing Agency, can be found at our website: www.elsevier.com/permissions.

This book and the individual contributions contained in it are protected under copyright by the Publisher (other than as may be noted herein).

Notices

Knowledge and best practice in this field are constantly changing. As new research and experience broaden our understanding, changes in research methods, professional practices, or medical treatment may become necessary.

Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using any information, methods, compounds, or experiments described herein. In using such information or methods they should be mindful of their own safety and the safety of others, including parties for whom they have a professional responsibility.

To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any liability for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions, or ideas contained in the material herein.

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library

Library of Congress Cataloging-in-Publication Data

A catalog record for this book is available from the Library of Congress

ISBN: 978-0-12-812779-7

For Information on all Academic Press publications visit our website at https://www.elsevier.com/books-and-journals

Publisher: Matthew Deans

Acquisition Editor: Brain Guerin

Editorial Project Manager: Thomas Van Der Ploeg

Production Project Manager: Kamesh Ramajogi

Cover Designer: Victoria Pearson

Typeset by MPS Limited, Chennai, India

Preface

Seiichi Koshizuka, Kazuya Shibata, Masahiro Kondo and Takuya Matsunaga

The Finite Volume Method (FVM) has been established in 1980s for simulating fluid dynamics problems (e.g., Rhie-Chow, 1983). Unstructured grids of arbitrary shapes can be used with the techniques of coordinate transformation that is based on differential geometry. Another typical aspect of the FVM is variable arrangement that is based on sophisticated topology. The algorithm for incompressible flow employs implicit pressure correction procedure using the pressure Poisson equation. The FVM is markedly successful to apply to engineering problems. However, it still has limitations derived from the grid. Numerical errors and instabilities take place where the grid is highly distorted. Besides, grid generation needs heavy effort and various knowledge to obtain reasonable results for complicated three-dimensional geometries.

Gridless/particle methods are expected to solve the troubles concerning the grid in the FVM. Particularly, particle methods using Lagrangian description do not need to calculate the convection terms. It is revolutionary in the computational fluid dynamics because discretization of the convection term is likely to cause terrible numerical diffusion and instability, and massive studies have been dedicated to this topic for many years. The particle methods are essentially free from the troubles concerning the convection terms.

Smoothed particle hydrodynamics (SPH) developed by Lucy (1977) and Gingold and Monaghan (1977) has been used for fluid dynamics problems in astrophysics. The SPH method was limited to nonviscous and compressible flow in the early 1990s. The idea of the Moving Particle Semi-implicit (MPS) method was conceived as the extension of the FVM to a particle method; spatial discretization was based on differences and pressure Poisson equation was employed for incompressible viscous flow (Koshizuka et al., 1996; Koshizuka and Oka, 1997). Engineering problems have been solved by the MPS method. Nowadays, many techniques in the MPS and SPH methods are common: semiimplicit algorithm, explicit algorithm using pseudo-compressibility, free surface boundary condition, wall boundary condition, surface tension model, etc. However, the spatial discretizations are basically different and the resultant discretized equations are clearly distinctive.

This book provides the comprehensive knowledge of the MPS method that has been developed to date. Chapter 1, Introduction, is an introduction to the basic concept of the MPS method. Differences between the MPS and SPH methods are explained. The research history of the particle methods for continuum mechanics is summarized. Chapter 2, Fundamental of Fluid Simulation by the MPS Method, provided a detailed explanation of the basic formulation of the MPS method. This chapter is instructive for the beginners of the MPS method. Chapter 3, Extended Algorithms, explains various algorithms except for the basic semiimplicit one. They are two compressible–incompressible unified algorithms considering slight and strong compressibility, an explicit algorithm using pseudo-compressibility, symplectic schemes based on Hamiltonian system, an Arbitrary Lagrangian-Eulerian (ALE) approach, a rigid body model using a quaternion, and an elastic solid model. Chapter 4, Boundary Conditions, shows the treatment of boundaries: such as solid walls, free surfaces and inlet/outlet boundaries. Chapter 5, Surface Tension Models in Particle Methods, is used for surface motion models used in the MPS method: the model using continuum surface force and pair-wise potential force. Various advanced techniques are summarized in Chapter 6, Advanced Techniques: liquid–solid phase change, gas–liquid two-phase flow, a subparticle-scale turbulence model, numerical techniques for suppressing pressure fluctuations, high-order schemes, parallel computing techniques, and multiresolution techniques. Industrial applications are presented as verification and validation (V&V) examples.

Chapter 1, 3, and 6 were mainly written by Koshizuka, Chapter 2 and Section 6.7 were written by Shibata, Matsunaga wrote Chapter 4, and Kondo authored Chapter 5 and Section 3.6.

The development of the MPS method has been stimulated by the studies in the field of physics-based computer graphics. A calculation method of rigid bodies using quaternions is imported from the physics-based computer graphics. Furthermore, high-quality visualization makes the calculation results of the particle method very attractive, particularly complex free surface motion accompanied by splashing. Parallel computing on graphics processing unit has been realized in the MPS method in a relatively early stage due to the interaction with the physics-based computer graphics.

Another important movement influencing on the MPS method is V&V. V&V has been implemented as the technical guidelines to the computer simulation procedure to assure the credibility of the results. This is important to use the computer simulation in the industries. Application of the MPS method to the industrial problems needs to follow the V&V framework. In this book, application examples in the industries are provided in the V&V section in Chapter 6, Advanced Techniques.

In 2011, the Great East Japan Earthquake occurred suddenly. The tsunami killed many people. One of the lessons of this tsunami was that many floating objects were accompanied to destroy the houses and the structures on the coast. The particle methods should be used more for the analysis of tsunami run-up with floating objects. In addition, unbelievable reactor core melting and devastating hydrogen explosions took place in Fukushima Dai-ichi Nuclear Power Plant after the tsunami. The first application of the MPS method was the molten fuel fragmentation process in the postulated severe accident of the nuclear reactor, and the second paper of the MPS method was published in Nuclear Science and Engineering in 1996. It should be said that the particle methods would be used more for nuclear reactor safety. Visualization and V&V are more and more important to transfer the simulation results of the disasters to the public and decision makers.

The authors would like to appreciate all the members who studied or helped studying the MPS method in our laboratories. The authors also would like to express thanks to the researchers and engineers who developed and used the software using the MPS method in the companies. The contents of the book are accumulation of their huge studies and tough experiences. The kind proposal, proper guidance, and encouragement to writing manuscripts by Mr. Brian Guerin, Mr. Thomas van der Ploeg, and Ms. Swapna Srinivasan in Elsevier are highly acknowledged.

January 2018

Chapter 1

Introduction

Abstract

Particle methods are based on Lagrangian description and meshless discretization to simulate continuum mechanics. Advantages and disadvantages of the particle methods are discussed from these viewpoints. Moving particle semi-implicit (MPS) method is one of the particle methods. The fundamental idea of the MPS method is a weighted difference without the mesh. Particle interaction models are prepared for differential operators and substituted into the governing equations. Pressure Poisson equation is constructed to solve the pressure field implicitly, while the other terms are explicitly calculated. Differences between the MPS method and smoothed particle hydrodynamics (SPH) method are explained. The research history of the particle methods for continuum mechanics is summarized.

Keywords

Lagrangian description; meshless discretization; moving particle semi-implicit (MPS) method; weight function; particle interaction model; semi-implicit algorithm; incompressible flow; pressure Poisson equation

Contents

1.1 Concept of Particle Methods 1

1.1.1 Lagrangian Description 2

1.1.2 Meshless Discretization 3

1.1.3 Continuum Mechanics 5

1.2 MPS Method 11

1.2.1 Weighted Difference 11

1.2.2 Particle Interaction Models 12

1.2.3 Semi-implicit Algorithm 16

1.2.4 MPS and SPH 18

1.3 Research History of Particle Methods 20

References 23

1.1 Concept of Particle Methods

The concept of the particle methods is explained. The most important viewpoint is the comparison with the mesh methods (Fig. 1.1). The mesh methods are, for example, the finite volume method and the finite element method, which have been mainly used for the computational fluid dynamics and the computational solid mechanics, respectively. The particle methods are said as new and advanced against the conventional mesh methods. In this section, the differences between particle and mesh methods are summarized to three aspects and both advantages and disadvantages are discussed.

Figure 1.1 (A) Mesh method and (B) particle method.

1.1.1 Lagrangian Description

Particles are used in the particle methods. They are the calculation points where the variables are located. The variables are, for instance, velocity vector components, pressure, temperature, etc. The particles are equivalent to the nodes, where the variables are located, in the mesh method.

The particles move with their velocities, while the nodes do not. We can say that the particles are fixed on the moving material. This is expressed as Lagrangian description. On the other hand, the nodes of the mesh methods are fixed in space, which is expressed as Eulerian description.

In the particle methods, it is necessary to hold coordinate vector components as variables as well as the velocity vector components. This means that the additional calculation time is required to update the coordinate vector components in each time step. Actually, it is not very large in the total calculation time.

A moving particle can be regarded as a substance of the material keeping a constant mass. Image of the moving particles is that the divided materials travel with their own velocities. The velocity and coordinate vectors of a particle are considered as those of the center of gravity of the divided material. The distribution of the mass is not considered in the particle. Therefore, overlapping of the particles is not concerned. Each particle is simply expressed by a velocity vector and a coordinate vector and has a mass without considering the mass distribution.

Mass conservation is essentially satisfied in the particle methods when the number of the initial particles is kept. Addition of new particles means mass increase and deletion of existing particles means mass decrease. This characteristic of the particle method is advantageous in the computational fluid dynamics, particularly, for incompressible fluids. It has been considered that rigorous mass conservation is important for the analysis of incompressible fluids because the pressure field is highly sensitive to the mass conservation. The mesh methods are based on space and the mass conservation is derived by satisfying the continuity equation. The mesh methods rigorously keep the space by explicitly drawing the mesh. Mass is indirectly kept by using the continuity equation which represents the relation between mass and space. The continuity equation expresses change of density, which is mass divided by space. In the particle methods, mass is conserved directly without the continuity equation.

The space that is occupied by a particle cannot be rigorously determined in the particle methods. Mass distribution inside a particle is not considered. A particle simply represents its velocity and coordinate vectors of its center of gravity. Therefore, there are no typical directions or coordinates of the particle. It can be said that the particle is spherical. In the mesh methods, the mesh near the wall is often generated to have a large aspect ratio; the mesh length parallel to the wall is longer than that vertical to the wall. Such thin mesh is efficient and widely used in the computational fluid dynamics since the variable distributions change rapidly near the wall. However, the aspect ratio that is different from 1.0 is difficult in the particle methods. This is one of the disadvantages of the particle methods.

We can say that the concept of moving particles, Lagrangian description, has an advantage of essential mass conservation and a disadvantage of difficulty using the technique of the arbitrary aspect ratio.

1.1.2 Meshless Discretization

The particle methods need no mesh for discretization of the governing equations, while the mesh methods are based on the mesh. First, the function of mesh is discussed in the mesh method. As shown in Fig. 1.1, the mesh consists of segments, each of which connects two nodes. Connection means that two nodes are adjacent. Discretized equations are given as the relationships among the adjacent nodes. For example, in the two-dimensional space (Fig. 1.1), one node has four segments which reach four adjacent nodes. A difference scheme is to be constructed using the variables located at five nodes: one node and its four adjacent nodes. We can say that the mesh is used to identify the adjacent nodes which are to be used for discretization of the governing equations. The mesh explicitly shows the adjacency.

There are no such segments in the particle methods. Adjacent particles are not explicitly shown. The adjacent particles are identified in the simulation by evaluating the distance between two particles (Fig. 1.2). We assume that there are two particles i and j and their position vectors are ri and rj, respectively. When the distance r ) is smaller than the effective radius re, they are considered as adjacent. The adjacent particle j is called a neighboring particle. The distances of all combinations of two particles are evaluated so that the list of neighboring particles is completed. Discretization equations are to be constructed using the variables at the neighboring particles.

Figure 1.2 Particle interaction with neighboring particles.

Mesh generation is not necessary in the particle methods. Mesh generation is often very complicated and time consuming in three-dimensional complex domains which are solved in the industrial applications. Accuracy of the simulation result is largely affected by the mesh quality. A more accurate result is expected by using a finer mesh though it requires more computation time. Compromise is necessary in mesh generation by considering the accuracy and the computation time. A lot of know-hows are alleged to be utilized, which leads to the mesh generation complicated. In the particle methods, initial arrangement of the particles is necessary but it is much easier than the mesh generation. This is a substantial advantage of the particle methods.

In the particle methods, the neighboring particles are memorized as the neighbor list and it is updated in each time step. This is an additional procedure in each time step in contrast to the mesh generation which is carried out once in the initial process. Besides, this procedure is time consuming because the calculation of the distances between all combinations of two particles requires the order of N², where N is the total number of particles. The order of calculation is usually N¹ for the substitution of variables into the discretized equations and N¹.⁵ for typical solvers of simultaneous linear equations. The order of N² has the same order of global interaction physics, e.g., electromagnetic force and gravity, where the interaction force reaches infinity. For the search of the neighboring particles, an algorithm of the order of N¹ has already been proposed and widely used in the particle methods. This algorithm is explained in the later chapters. Thus, at this moment, the additional computation time for making a neighbor list is not a disadvantage of the particle methods any more.

The mesh has a role of dividing space as well as explicit description of adjacency. Consistent division of space (=no overlapping) is necessary for consistent integration in the finite element method and for rigorous conservation of physical quantities in the finite volume method. On the other hand, division of space is ambiguous in the particle method. However, distinct troubles or disadvantages are not found due to this at this moment.

1.1.3 Continuum Mechanics

The governing equations for continuum mechanics to be solved are mass and momentum conservation equations. In fluid dynamics, they are also called continuity and Navier–Stokes equations, respectively. The governing equations for continuum mechanics should be discretized without the help of mesh in the particle methods. As explained in Section 1.1.2, mesh has information of strict space division, which cannot be utilized in the particle methods. We need other approaches for the spatial discretization.

For example, in the moving particle semi-implicit (MPS) method (Koshizuka and Oka, 1996), weighted average of differences is the basic concept of the discretization. In the smoothed particle hydrodynamics (SPH) method (Lucy, 1977; Gingold and Monaghan, 1977), superposition of kernels is basically assumed to obtain the spatial distribution of the variables. In both of the particle methods, the mesh is not used for the discretization at all. The detailed formulation of the discretization in the MPS method is explained in Chapter 2, Fundamental of Fluid Simulation by the MPS Method.

The coordinate and velocity vectors of the particles are updated in each time step by using the discretized governing equations. The discretized governing equations involve the variables of the neighboring particles. This can be said that the particle motion is determined through the interaction with the neighboring particles. The discretized governing equations are regarded as the motion equations of the particles. Here, it should be noted that the particles are artificial for discretization in the particle methods. The fluid is represented by a finite number of artificial particles. They are not real particles, such as powder and molecules.

An example of the real particle dynamics is bead motion (Fig. 1.3). We can see the motion of beads when they are poured into a glass as shown in Fig. 1.3A. The bead dynamics can be described by Newton’s second law with friction terms. On the other hand, we can see motion of water when it is poured into the same glass as shown in Fig. 1.3B. The governing equations of water are the Navier–Stokes equations. The water behavior is different from that of the beads because the governing equations are different. However, as shown in Fig. 1.3C, the particle motion can be evaluated based on the discretized Navier–Stokes equations (=particle motion equations) in a computer. In this case, the particle motion in the computer is like water motion. This is the concept of the particle methods using artificial particles for continuum mechanics.

Figure 1.3 Particle dynamics: (A) physical bead motion, (B) physical water motion, and (C) numerical water motion using particle method.

Some other particle methods, molecular dynamics and discrete element method, are the numerical methods for the real particles. In molecular dynamics, motion of molecules is simulated. In discrete element method, motion of powder particles is simulated. These particle methods are essentially different from the present particle methods, such as MPS and SPH, for the continuum mechanics.

Here, a one-dimensional pure convection problem is discussed to understand the difference between the mesh methods and the particle methods. Pure convection of a passive scalar variable ϕ in one-dimensional space x with a constant velocity u (>0) is governed by the following particle differential equation:

(1.1)

The passive scalar variable can be temperature, chemical spices concentration, etc. The pure convection governed in Eq. (1.1) is simple movement of the variable without any change of the profile. An analytical solution is illustrated in Fig. 1.4.

Figure 1.4 One-dimensional pure convection.

First, we consider the mesh method (finite difference method) employing the Eulerian description. The one-dimensional space is discretized to a finite number of mesh points. Each point i at time step k.

First, discretization in time is carried out. The left-hand side of Eq. (1.1) is then discretized to

(1.2)

is the time interval between time steps k and k+1. Unknown variables at new time step k+1 are calculated by using known variables at old time step k:

(1.3)

The scalar variables at time step k+1 are calculated using Eq. (1.3) which is the discretized equation of Eq. (1.1). The coordinate of each mesh point is not changed because of the Eulerian description. This can be written as:

(1.4)

in Eq. (1.3) in space. Various spatial difference schemes have been proposed until now. Here, an upwind scheme, one of the basic schemes, is applied:

(1.5)

Applying Eq. (1.5) to Eq. (1.3) results in Fig. 1.5A. We can see diffusive profiles of ϕ. The analytical solution of Eq. (1.1) is the simple translational movement of the initial profile with velocity u. The difference between the numerical result and the analytical solution is substantial.

Figure 1.5 Finite difference method for one-dimensional pure convection: (A) upwind difference scheme and (B) central difference scheme.

When the central difference scheme, another basic scheme,

(1.6)

is applied, the result is Fig. 1.5B. We can see oscillatory profiles. It is well known that the finite difference method causes substantial numerical errors, which can be diffusive or oscillatory, for convection.

In the Lagrangian description, the one-dimensional pure convection problem is written as:

(1.7)

. Eq. (1.7) is equivalent to Eq. (1.1). The one-dimensional space is discretized to a finite number of particles. Each particle i . Unknown variables at new time step k+1 are calculated by using known variables at old time step k: