Beruflich Dokumente
Kultur Dokumente
Historia del artículo: métodos de partículas son aquellos en los que el problema está representado por un número discreto de partículas. Cada partícula se mueve en
Recibido el 20 de de enero de 2009
consecuencia con su propia masa y las fuerzas externas / internas aplica en él. En este documento el método de elementos finitos de partículas basa
Recibida en forma de 27 revisada de marzo de 2009 Aceptado
en finitos funciones de forma del elemento se utiliza para resolver los fl continua uid mecánica ecuaciones en el caso de densidad heterogénea. Para
el 3 de abril de 2009 On-línea del 18 de abril de 2009
evaluar las fuerzas externas aplicadas a cada partícula, las incompresibles ecuaciones de Navier-Stokes se resuelven en cada paso de tiempo
usando una formulación de Lagrange. Toda la información contenida en el fluido se transmite a través de las partículas. Todas las clases de
densidad heterogénea fluidos y flujos multifase con interfaces internos incluyendo o no libres superficies, olas que rompen y separaciones de fluidos
palabras clave:
pueden resolverse fácilmente con esta metodología.
Método de Elementos finitos
de partículas heterogénea
fluidos Multi fluidos
2009 Elsevier Todos los derechos reservados.
formulaciones de Lagrange
Multifase flujos
Incompresible de Navier-Stokes Free-superficies
Interfaces
1. Introducción mecánica por Belytschko et al. [10] y en fl mecánica uid problemas por Oñate et al. [11-13] .
Últimamente, las ideas sin malla se generalizarse a tener en cuenta las aproximaciones de tipo fi
Durante los últimos 20 años, la simulación por ordenador de fluido incompresible flujo se ha elemento finito a fin de obtener el mismo tiempo de cálculo en la generación de malla como en la
basado en la formulación de la mecánica de Euler ecuaciones fl uido en los dominios continuos [1] . evaluación de las conectividades sin malla [14] . Este método se llama el método de elementos finitos
Sin embargo, es todavía difícil de analizar los problemas en los que la forma de las superficies libres sin malla (MFEM) y utiliza el extendido Delaunay Tessellation [15] para construir la malla en un tiempo
o interfaces internas cambia continuamente o en fl interacciones de estructura uid que están de cálculo que es lineal con el número de puntos nodales.
implicados problemas de contacto complicadas.
Más recientemente, se han utilizado métodos de partículas en el que se sigue cada partícula de Debe señalarse que los métodos de partículas se pueden usar tanto con: malla o sin malla
fluido de una manera de Lagrange [2-5] . Las primeras ideas sobre este enfoque fueron propuestos funciones de forma. La única limitación práctica es que la conectividad en los métodos sin malla o la
por Monaghan [2] para el tratamiento de problemas hidrodinámicos astrofísicas con la llamada generación de mallas en métodos con malla necesitan ser evaluados en cada paso de tiempo. Por
Hidrodinámica Método Smooth Particle (SPH). estas razones la evaluación de la conectividad no debe consumir mucho tiempo de cálculo.
0045-7825 / $ - see front matter 2009 Elsevier Todos los derechos reservados. doi: 10.1016 /
j.cma.2009.04.002
S. Idelsohn et al. / Comput. Métodos Appl. Mech. Engrg. 198 (2009) 2750-2767 2751
Heterogéneos flujos fl uido se producen con frecuencia en la naturaleza y la práctica de la formas arbitrarias. Además, la prescripción de condiciones de contorno exactas en la interfaz es
ingeniería. Sin embargo, de todos modos representan un reto importante de investigación de los dos generalmente aproximadas.
puntos, teóricos y computacionales de vista. Contrariamente a los materiales sólidos heterogéneos El PFEM pertenece a la categoría de los métodos de la interfaz de seguimiento. En cada paso
donde la principal causa de la variación en las propiedades físicas son los módulo de Young o la de tiempo se genera una nueva malla que rastrea exactamente la interfaz. Contrariamente a otras
anisotropía de los diferentes materiales involucrados, en fluidos de las principales razones de la técnicas de rastreo en una malla especial se construye alrededor de cada interfaz [30,31] , PFEM
heterogeneidad son las variaciones de la densidad y la viscosidad. reconstruye la malla completa con un generador de malla rápido. El método se vuelve competitiva
cuando las interfaces entre los diferentes fluidos son dominantes (véase Fig. 1.1 ).
dónde q = q ( X) es la densidad, u yo son los componentes cartesianas del campo de velocidad, pag la
presión, s ij el tensor desviador de tensiones, F yo el término fuente (normalmente la gravedad) y RE/
dt representa el total o
derivada en el tiempo el material de una función /.
En el procedimiento descrito en este trabajo a la aproximación de Boussinesq se introduce en Para materiales heterogéneos q es una función de la posición
todos los términos de las ecuaciones de momento: las fuentes y los inerciales. Además, el método q = q ( X). Para fl incompresible OWS la derivada de tiempo material de
permite resolver el problema de mezclar diferentes líquidos sin ningún enfoque de homogeneización. re q
dt ¼ 0. No obstante, la derivada en el tiempo espacial no es o q o t - 0 . Esta
es la razón por la cual los materiales heterogéneos se resuelven más fácilmente con formulaciones
de Lagrange.
variaciones viscosa del fluido pueden ser de nuevo debido a la variación de la temperatura. A partir de la ecuación. (2,1) , Por fl ujos incompresible la conservación de la masa
Saltos en el valor de la viscosidad entre dos o más fluidos pueden ser también participan pero no son en la formulación de Lagrange puede ser simplemente escrito como: o u yo
o X yo ¼ 0, mientras
tan importantes en general. será considerado en este trabajo sólo se heterogeneidad debido a los en la formulación de Euler conservación de la masa debe ser escrito como:
cambios o saltos en la densidad. oq
otþ u yo o q o X yo ¼ 0.
Para fl newtoniana los UID del tensor de tensiones s ij puede expresarse como una función de la
Cálculo de la interfaz entre dos líquidos inmiscibles o las superficies libres es difícil debido a que campo de velocidad a través de la viscosidad l por
ni la forma ni la posición de los dominios entre los fluidos son conocidos a priori. Hay básicamente
2 o ul
dos enfoques para el cálculo de las interfaces de este tipo de flujos: interfaz de seguimiento y s ij ¼ l o u yo þ o uj re ij : re 2: 3 Þ
métodos de interfaz de captura. El primero calcula el movimiento de las partículas de fluencia a
o Xj o X yo 3 o Xl
través de un enfoque de Lagrange, donde el dominio numérico se adapta a la forma y posición de las por incompresible o cerca flujos incompresibles
interfaces. métodos de interfaz de captura estándar consideran ambos fluidos como un único fluido o u yo o uk
o X yo ¼ 0 o o u yo o X yo o Xl
el termino:
eficaz con propiedades variables [27-29] . Las interfaces son considerados como una región de
cambio repentino en las propiedades de fluidos. Este enfoque requiere un modelado preciso del salto
2 l o u yo
¼0o 0 re 2: 4 Þ
en las propiedades de los dos fluidos, teniendo en cuenta que las interfaces pueden mover, doblar y 3 o X yo
s ij l o u yo þ o uj : re 2: 5 Þ
o Xj o X yo
Fig. 1.1. Dos de fase fluye con un fluido de densidad inferior en la parte inferior. (A) la posición inicial y la posición (b) después de 1 s.
2752 S. Idelsohn et al. / Comput. Métodos Appl. Mech. Engrg. 198 (2009) 2750-2767
El caso particular de un fl incompresible incompresible o cerca ow con viscosidad constante 2.2.1. La división de tiempo
puede ser aún más simplificada fi dada la bien conocida forma de Laplace de las ecuaciones de El tiempo de integración de la ecuación. (2,9) presenta algunas di fi cultades ya que es una
Navier-Stokes. Esta aproximación se puede utilizar en este trabajo porque entonces las ecuaciones a ecuación acoplado completo que implica cuatro grados de libertad para cada nodo. Cuando el fluido
resolver por separado en cada componente de la velocidad. Hay que señalar sin embargo que esta es incompresible o casi incompresibles ventajas se pueden tomar desde el hecho de que en la
formulación puede ser utilizado sólo para fluidos con viscosidad constante en todo el dominio. Esta ecuación. (2,9) los tres componentes de la velocidad solamente se acoplan a través de la presión. Un
aproximación no es válida para los fluidos con viscosidad o elemento variable por elemento de método fraccional a paso se ha propuesto [dieciséis] que consiste en la división de cada paso de
viscosidad constante [32] . En el Laplace formar el plazo o tiempo en dos etapas en tiempo de pseudo. En el primer paso la parte implícita de la presión se evita
que conduce a una ecuación disociadas en cada uno de los componentes de la velocidad. La parte
o o u yo o uj
s ij ¼ o l o u yo þ o uj ¼lo þlo
o Xj o Xj o Xj o X yo o Xj o Xj o Xj o X yo
o u yo o uj o u yo
du yo u norte þ 1
yo
u Ni
yo
u yo þ u yo u Ni
¼lo þlo lo : re 2: 6 Þ
¼ u norte þ 1
o Xj o Xj o X yo o Xj o Xj o Xj dt re t re t
o o s norte
ij
þ1
Entonces, las ecuaciones de momento pueden ser finalmente escribirse como: ¼? 1 pag norte þ 1 þ 1 þ F yo; re 02:11 Þ
q o X yo q o Xj
o o u yo
q du yo pag þ o s ij þ q F yo pag þ l o þ q F yo: re 2: 7 Þ dónde u yo son variables intermedias definido por la división:
dt ¼? o o X yo o Xj o X yo o Xj o Xj
o o
Como se dijo antes, la Ec. (2,7) muestran que sólo el u yo componente de velocidad está presente re UN Þ u yo ¼ u Ni þ F yo re t re t do pag norte þ re t s norte þ 1; re 02:12 Þ
q o X yo q o X j ij
en cada uno '' yo th”ecuación. Al utilizar métodos de segregación de presión como el método
re t o
fraccional a paso, esto implica que cada componente de la velocidad puede ser resuelto de forma re do Þ u norte þ 1 re pag norte þ 1 do pag norte Þ; re 02:13 Þ
independiente de los otros, lo que conduce a un ahorro considerable en tiempo de cálculo. Para la
yo ¼ u yo
q o X yo
viscosidad de la variable sin embargo los plenamente acoplados ecuaciones de Navier-Stokes (Eq. (2,2)
do es un parámetro entre 0 y 1 de fi nir la cantidad de desdoblamiento de presión.
) Se debe utilizar en todos los casos.
re q q norte þ 1 q norte
re s þ ij metropag
j þ metro yo Þ ¼ r þ Ni ¼? D s ij metro jpag metro yo Þ ¼? r Ni; u þ yo ¼ u yo en do yo;
¼ q norte þ 1 q þ qq norte
dt re t re t
donde los índices de super (+) y ( ) Representan el valor de la variable a cada lado de la interfaz. u yo þ u yo
¼? q o u norte þ 1 yo
; re 02:14 Þ
o X yo
q norte þ 1 q u yo
u Ni ¼? q o u norte þyo1 : re 02:16 Þ
q du yo q u yo re X yo; t norte þ 1 Th? u yo re X yo; t norte Þ¼ q u norte þ 1
yo
re t o X yo
dt re t re t
o o u yo
norte þ h
A partir de las ecuaciones. (2.13) y (2.16) la ecuación de masa-momento junto se convierte en:
¼? pag þ l o þ q F yo ; re 2: 8 Þ
o X yo o Xj o Xj
métodos de partículas son ventajosos para tratar problemas discretas tales como materiales
re t o u yo o granulares, sino también para tratar los problemas continuos para los que existen posibilidades de
re UN Þ u yo ¼ u Ni þ F yo re t re t do pag n) u yo ; interfaces internas, problemas de rozamiento de contacto en las interacciones UID-sólido fl o libres
q re X Þ l oo X j o Xj q re X Þ o X yo
superficies con olas que rompen.
o u yo
re segundo Þ q re X Þ ¼ o2 re pag norte þ 1 do pag norte TH) pag norte þ 1; re 02:20 Þ
re t o X yo o X 2 yo La característica más relevante de un método de partículas es que no es un dominio solución fi
re t o cado. El problemdomain se define por las posiciones de las partículas y por lo tanto, no hay un
re do Þ u norte þ 1 re pag norte þ 1 do pag norte TH) u norte þ:1
yo ¼ u yo
q re X Þ o X yo
yo
contorno de límite. Esta es la razón por la cual, cuando una ecuación diferencial es que hay que
resolver con el fin de evaluar las fuerzas, el límite debe ser identi fi cado con el fin de imponer las
Debe tenerse en cuenta que, aunque las ecuaciones. (2,20) parecen lineales no son porque son condiciones de contorno. Además, las partículas se pueden utilizar para generar un dominio discreto
todos evaluados en la posición final. El algoritmo debe ser necesariamente iterativo. Una vez u norte þ 1
dentro de la cual se resuelven las formas integrales de las ecuaciones diferenciales que regulan (ver Fig.
yo
se evalúa durante 3.1 ).
cada iteración, entonces las coordenadas se actualizan a la nueva posición, como se describe en
una sección siguiente y los tres pasos A, B y C se repiten hasta que la convergencia.
Otra característica de los métodos de partículas es que todas las propiedades físicas y
matemáticas están unidos a la misma y no de partículas a los elementos como en los métodos FE. También en cada paso de tiempo una nueva malla se genera con el fin de definir los nuevos
Para las propiedades físicas de instancia como viscosidad o densidad, variables físicas como la interfaces entre los diferentes fluidos y las funciones de forma correspondientes para resolver las
velocidad, ecuaciones diferenciales. Esta
Fig. 3.1. La mezcla de dos líquidos diferentes. El reconocimiento de las interfaces internas y libres superficies del dominio análisis y actualización de malla para las distribuciones de puntos sucesivos.
2754 S. Idelsohn et al. / Comput. Methods Appl. Mech. Engrg. 198 (2009) 2750–2767
de malla sólo es útil para la definición de las fuerzas que interactúan y se desvanece una vez que se En forma compacta:
evalúan las fuerzas (ver Fig. 3.1 ). Las funciones de interpolación son un caso particular de las
u yo ¼ norte Ti re X; t Þ u yo;
funciones de forma fi elementos finitos [14,15] . La superficie límite se define utilizando la técnica de re 3: 4 Þ
alfa-forma [16,17] . pag ¼ norte tp re X; t Þ PAG
o en forma de matriz:
264 375 u;
3.1. actualización de la posición de las partículas
con h entre 0 y 1.
The following hypothesis will be introduced: the mesh is not updated during the iterations within
each time step. This means that if a mesh update is introduced at the beginning of a time step, the
3.2. Generación de una nueva malla
same mesh (but deformed) will continue until the end of the time step. Mathematically expressed, N( X,
t n) = N( X, t n+ 1). Unfortunately, this hypothesis is not always possible to satisfy for all meshes, and this
Uno de los puntos clave para el éxito de la formulación de Lagrange descrito aquí es la
introduces small errors in the computations which are neglected in this paper.
regeneración rápida de una malla en cada paso de tiempo sobre la base de la posición de los nodos
en el dominio del espacio. En este trabajo se genera la malla utilizando el llamado Extended
Delaunay Tessellation (EDT) se presenta en [15] . El EDT permite la generación de mallas de
elementos con formas polivalentes arbitrarias (combinando triángulos, cuadriláteros y otros polígonos
en 2D y tetraedros, hexaedros y poliedros arbitrarios en 3D) en un tiempo de cálculo de orden norte, siendo
norte el número total de nodos de la malla. Detalles del procedimiento de generación de la malla y las 3.4. Discrete equations
funciones de forma de poliedros arbitrarios se pueden encontrar en [14,15] . Una vez que la nueva
malla se ha generado en cada tiempo de paso la solución numérica se halla usando el método de Using the Galerkin weighted residual method to solve the split equations, the following integral
elementos finitos estándar (FEM). forms are obtained:
Z Z Z
o
ðAÞ1 Ni q u i dV 1 Ni q u ni dV Ni fi q dV þ Z Ni c p n dV
Dt V Dt V V V o xi
Z Z
o ui
Ni l o dV Ni ð r ni s ij m j c p n m i d C ¼ 0; ð 3:6 Þ
3.3. discretización espacial V o xj o xj Cr
( )
Usando FEM, las funciones desconocidas se aproximan utilizando una interpolación orden igual q o ui o2
ðBÞZ Np pnþ1c pn dV
para todas las variables en el final con fi guración: Dt o xi o x2i
V
Z
u yo ¼ X norte l re X; t Þ u Illinois;
þ1 Np q unþ1 mi u ni þ 1 m i d C ¼ 0; ð 3:7 Þ
Dt
i
l
re 3: 3 Þ Cu
Z
q q q
ðCÞZ Ni ui þ o ð p n þ 1 c p n Þ dV ðAÞZ Ni q f i dV Ni Ni c o p n dV þ
V D t unþ1
i
o xi V D t ui V D t u ni dV þ Z V o xi
Z Z
o N i o ui n þ h
Ni ð pnþ1c pn Þ m i d C ¼ 0; ð 3:8 Þ þlZ dV ð Þ d C ¼ 0; ð 3:9 Þ
Cr V o xj o xj Cr N i r ni þ c p n m i
where the boundary conditions have been also split and V is the volume at time t n+ 1.
Fig. 6.1. Initial mesh (4305 nodes). Fig. 6.2. Relative wave height (black dots) compared to references.
2756 S. Idelsohn et al. / Comput. Methods Appl. Mech. Engrg. 198 (2009) 2750–2767
Z
1 o Np o Np o ð pnþ1c pn Þ q
ðBÞ q u i dV þ Z dV ðCÞZ Ni ui þ o ð p n þ 1 c p n Þ dV
D t unþ1 o xi
i
Dt V o xi V o xi o xi V
Z
Z
þ1 Np q unndþ C
1 ð 3:10 Þ
Dt
¼ 0; N i ð p n þ 1 c p n Þ d C ¼ 0: ð 3:11 Þ
Cu
Cr
linearity, non symmetry and non self-adjoint operators which require stabilization terms to
avoid numerical oscillations. All these problems are not present in this formulation. Only the
nonlinearity due to the unknown final positions of the particles remains.
(3) In all steps the above systems of equations are solved for
evaluation of the velocity components (steps A and C) and the pressure (step B). These
systems are scalar (only one degree of freedom by node), symmetric and positive definite.
Then, it is very easy to solve them using a symmetric iterative scheme (such as the
conjugate gradient method).
In the Eulerian form of the momentum equations, the discrete form must be stabilized in order to
Fig. 6.4. Wrong interface equilibrium with the standard stabilization method (Eq. avoid numerical wiggles in the velocity and pressure results. This is not the case in the Lagrangian
(3.22) ) which does not take properly into account the density jump at the interface. formulation where no stabilization parameter must be added in steps A and C. Nevertheless, the
incompressibility condition must be stabilized for equal-order approximations to avoid possible
pressure oscillations in some particular cases.
p ¼ 0 on C r; ð 3:12 Þ
For instance, for small pressure split ( c – 0) or for small time
step increments (Courant number much less than one) it is well known that the fractional step does
u n þ 1 m ¼ 0 on C u: ð 3:13 Þ
not stabilize the pressure waves. In those particular cases, stabilization term must be introduced in
step B in order to eliminate pressure oscillations [33] . On the other hand, examples where the c parameter
3.5. Compact form is always fixed to zero and the time increments correspond to Courant number 1 avoid all the
stabilization problems.
Using the approximations (3.4) the discrete equations become in compact form:
Pressure stabilization means to add a term into step B that in compact form represents a matrix S
ð A Þ ð M q þ D t l K Þ u ¼ M q u n þ D t F c D t B T P n; ð 3:14 Þ
in such way that:
ð B Þ LP n þ 1 ¼ 1 ð 3:15 Þ
D t ð B q u u Þ þ c LP n; ð L þ S Þ Pnþ1¼ 1 ð 3:21 Þ
D t ð B q u u Þ þ c LP n:
ð C Þ M q u n þ 1 ¼ M q u D t B T ð P n þ 1 c P n Þ; ð 3:16 Þ
Residual-based stabilization methods like Finite Calculus (FIC)
defined as projections of the pressure gradients divided by q( x). The new stabilization term becomes:
Z Z
o Np o Np o Np
B¼Z
V o x N T 1 dV; V o y N T 2 dV; V o z N T 3 dV ; Z
o
Np sop qðxÞ pi dV: ð 3:23 Þ
ui¼ Z N i u nn dþ 1C; ð 3:18 Þ V o xi o xi
Cu
The projection vector p i is obtained from its definition using the following weak expression:
Z Z
o Np o Np o Np
Bq¼ Z
V o x q ð x Þ N T 1 dV; V o y q ð x Þ N T 2 dV; V o z q ð x Þ N T 3 dV ;
Z
ð 3:19 Þ op
Ni q ð x Þ p i dV ¼ 0
V o xi
Fi¼ Z Ni fi q ð x Þ dV þ Z N i r ni d C: ð 3:20 Þ
V Cr
The Lagrangian split scheme described has two important advantages: with p i ¼ N Ti P i being P i the vector with the nodal values of p i.
Then,
Z Z
o N Tp
(1) Steps A and C are linear. The Lagrangian formulation elimi- Ni dV P Ni q N Ti dV P i ¼ 0 ) B T P ¼ M q P ) P ¼ M 1 qBT P:
V o xi V
nates the standard convection terms present in Eulerian formulations. The convection terms
ð 3:24 Þ
are responsible for non-
S. Idelsohn et al. / Comput. Methods Appl. Mech. Engrg. 198 (2009) 2750–2767
Fig. 6.5. Interface position and velocity vectors (fixed length) for two-fluid sloshing 1000:1 during the first three periods. Bottom fluid: water. Top fluid: air. 2758
The stabilization term (Eq. (3.22) ) or the new version (Eq. SP ¼ L s B sq M 1 ð 3:26 Þ
qBT P;
(3.23) ) is integrated by parts and the boundary term is neglected here. This boundary term however
may be important for preserving the total volume in free-surface flow problems [34] : where L s and B sq are the same matrices defined previously but multiplied element by element by s and
sq( x), respectively, and then assembled.
Z Z
o o Np
Np sop qðxÞ pi dV sop q ð x Þ p i dV Eq. (3.21) with the stabilization terms becomes:
V o xi o xi V o xi o xi
Z
o Np o Np
¼Z s o N Tp dV P sq ð x Þ N Ti dV P i: ð 3:25 Þ
o xi o xi o xi L þ L s B sq M 1 qBT Pnþ1¼ 1 ð 3:27 Þ
D t B q u u þ c LP n:
V V
o NT
MqC¼ Z NT q ð x Þ C ð x Þ N TT dV; L j ¼ Z j ð x Þ o N TT dV;
V V o xi o xi
Q¼Z N T Q ð x Þ dV þ Z N T q n d C:
V Cq
Boussinesq approximation for small variations in the density due to thermal dilatation of the
materials stands:
q ¼ q r ½ 1 a ð T T r Þ•; ð 4:3 Þ
where T r is a reference temperature, q r is the density of the material with temperature T r and a is a
dilatation parameter (in general a is also a function of the temperature).
The Boussinesq approximation is usually modelled by introducing a buoyancy force in the right
hand side of the momentum equations. Effectively, when the main forces of source terms in Eq. (2.1) are
the gravity forces, (e.g. f i = g i, with g i the gravity acceleration), the momentum equations with the
Fig. 6.6. Water mass variation due to remeshing for a large time period (5 s) for different mesh sizes (same time step).
standard Boussinesq approximation read:
M q P ¼ B T P; ð 3:29 Þ
For incompressible flows this implies o u i o x i ¼ 0.
ð L þ Ls Þ Pnþ1¼ 1 ð 3:30 Þ In this work, the Boussinesq approximation will be introduced in all the terms of the momentum
D t ð B q u u Þ þ c LP n þ B sq P:
1 which is evaluated ele- equations, not only in the source terms but also in the inertial terms. This is easily performed with the
The value used for s is s ¼ 1 D t þ l h2 q Lagrangian formulations where each particle will change their own density using the Boussinesq
ment by element [34] . The final algorithm reads: approximation. This means:
ð B 0 Þ M q P ¼ B T P ) P;
q r ½ 1 a ð T T r Þ Du i pþo s ij gi q r ½ 1 a ð T T r Þ•: ð 4:5 Þ
ð B Þ ð L þ Ls Þ Pnþ1¼ 1 Dt B q u u þ c LP n þ B sq P ) P n þ 1; Dt ¼ o o xi o xj
ð C Þ M q u n þ 1 ¼ M q u D t B T P n þ 1 c P n ) u n þ 1:
In problems for which the dilatation effect is large, the Boussinesq approach may be also
ð 3:31 Þ introduced in the mass conservation equation as
Steps B 0 and C may be solved explicitly with a lumped M q matrix. Step A may be solved explicitly for
small time steps and a small value of the viscosity parameter. For incompressible flows however step D q
¼D
B must be solved implicitly with the full Laplace matrix ( L + L s). Dt þ q o uoi x i Dt f q r ½ 1 a ð T T r Þ•g þ q o u i o xi
¼ q r a DT ¼ 0; ð 4:6 Þ
Dt þ q o uoi x i
4. Heat conduction equation in Lagrangian formulation
which implies:
q C DT joT þ Q; ð 4:1 Þ
Dt ¼ o o x i o xi
where T is the temperature, j the thermal conductivity, C the heat capacity and Q a possible source
terms contributions. The problem definition is completed with standard boundary conditions including
heat convection and radiation fluxes, and imposed temperature boundaries.
reads:
o ui
¼ qr ð 4:7 Þ
o xi q a DT
Dt ¼ e T :
In this case, the fluid remains incompressible due to the pressure forces but becomes
compressible for temperature dilatation. The temperature volumetric variation e T is explicitly
introduced as an initial volumetric variation at the beginning of each time step.
The density variation due to the temperature dilatation introduces a change in step B of the
fractional-step method described previously. In fact, Eq. (2.19) does not hold anymore as q n þ 1
ðxÞ– q n ð x Þ.
It may be replaced by:
Fig. 6.9. Fluid densities.
D q qnþ1qn
¼ qnþho ui ¼ qnþh eT: ð 4:8 Þ
Dt Dt o xi
After discretization, this term introduces in equations (B) a new vector E n+ 1:
Eq. (2.18) for h = 1 remains:
n o dV
qnþ1 q
ðxÞ o ui Enþ1¼ Z Np ð 4:11 Þ
eT ¼ o2 ð p n þ 1 c p n Þ: ð 4:9 Þ Dtð eT Þ
Dt o xi o x2i V
where T n+ 1 is the last temperature evaluated. nodes instead to the elements. The mesh is permanently updated and hence, it is difficult to keep
It must be noted that the case of an incompressible flow inside a closed recipient may not be physical properties at element level. In heterogeneous
Fig. 6.8. Contour plots of temperature field and mesh at different times.
S. Idelsohn et al. / Comput. Methods Appl. Mech. Engrg. 198 (2009) 2750–2767 2761
materials we can have a jump in the material density of several orders of magnitude. We must decide Both possibilities have advantages and disadvantages. Element interfaces are more stable as
where the internal interface between two different materials occurs. The most natural solution for a they do not change much when remeshing is performed. On the other hand, nodal interfaces are
particle method is to have the interface inside the elements sharing particles with different densities, more accurate because they allow representing exactly the gradient pressure gap that normally
so that in the element integration points k density takes the mean value q k ¼ 1 occurs when there is a jump in the density value (see Fig. 5.3 ).
Pnv
nv a¼1 qa
(where nv is the number of nodes of the element). We will call this possibility as interface across Two different kinds of problems will be considered:
elements ( Fig. 5.1 )
Other possibility is to impose that the interfaces between different materials are between two (1) problems in which the density varies through the domain in
elements, which will be called nodal interface ( Fig. 5.2 ). To achieve this we must accept that elements a smooth way. This is the case for instance of thermal variations of the density due to
sharing particles with two different densities have one or the other particular density value. Now the dilatations. In these cases element interface will be used (as in the numerical examples
density in the element integration points k takes the value
6.2 and 6.5).
(2) problems in presence of immiscible different materials
where there are jumps in the density properties. In these cases nodal interface will be used
if k 2 Fluid1;
qk¼ q1 (as in numerical examples
q2 if k 2 Fluid2:
6.1, 6.3 and 6.4).
The second case is more difficult to solve numerically because the solution has a discontinuity in
the pressure gradients at the
Fig. 6.10. Three different density fluids. Fig. 6.11. Impinging jet: particle positions.
2762 S. Idelsohn et al. / Comput. Methods Appl. Mech. Engrg. 198 (2009) 2750–2767
interface. For this reason immiscible heterogeneous materials will be solved using the nodal interface 6. Numerical results
algorithm in order to be able to represent exactly the pressure gradient jumps. As a consequence of
remeshing, the interface position may change suddenly from one time step to the next one introducing The capabilities of the formulation described above will be shown in several examples with two
small instabilities in the flow. Nevertheless these changes in the interface position between two time types of heterogeneities: density jump and continuous variable density. A first example is a sloshing
steps are of order h. This means that they decrease linearly with the mesh size and are of the same test for which other numerical tests are available. The example is also used to test the stabilization
order than the alpha-shape technique used for the determination of the free-surface. In spite of this proposed and the ability of the method to deal with large density jumps. In the second example the
drawback, nodal interface method improves substantially the results and becomes essential in classical Boussinesq cavity is tested and compared with existing results. Boussinesq cavity is an
immiscible fluids. interesting test for lagrangian formulations because the problem never reaches a sta-
tionary result for the particle position while the temperature remains at a stationary value. In order to condition along the interface that can be observed in Fig. 6.4 . The results shown previously have
test the ability of the method to carry out with several heterogeneous materials a three density fluid is been obtained using the new stabilization term (Eq. (3.23) ) where the density jump for heterogeneous
solved without showing any particular drawback. The fourth example consists in a strong mixing of materials has been taken into account in the definition of p.
two immiscible materials. At the beginning, both fluids mix strongly due to the initial kinetic energy.
After some time, both fluids separate due to the gravity forces. Finally, the case of two different fluids A very interesting advantage of the present formulation is to allow arbitrary large jumps in density
including thermal dilatation representing a magma chamber is presented. between the materials used. For example, in a water–air sloshing problem the density jump is 1–1000.
This kind of problems is normally treated as a single flow, neglecting the air flow. Nevertheless, in
many situations the influence of air must be taken into account, especially when bubble production is
important and the air incompressibility conditions must be considered. The present method allows
carrying out large density jumps without any drawback.
6.1. Two-fluid sloshing
This numerical test for two-fluid interface flows was first proposed by Tezduyar et al. in [30] and
deeply investigated in [31] . It consists in a closed rectangular cavity with two immiscible fluids and For instance the interface position and fluid velocity for a water–air sloshing problem are shown
small different density. The interface between both fluids is tilted in the initial configuration ( Fig. 6.1 ). in Fig. 6.5 . The density jump is 1:1000 and the gravity was set to g = 10 m/s 2. Not even from the jump
The bottom fluid has a density q = 2 kg/m 3, the top fluid has of pressure gradient nor from the stability algorithm the method has presented any difficulty.
q = 1 kg/m 3, and the dynamic viscosity is l = 10 3 Pa s for both fluids. Gravity force is set to 0.294 in the However, this problemdefined with the same geometry and initial condition than the previous test,
vertical direction. Vertical walls are considered free-slip while horizontal walls are no-slip. has now a very strong movement of all the internal particles and the interface includes the possibility
of breaking waves. Now remeshing is necessary and connectivity changes, in particular near the
This is a typical example where there is a jump in the pressure gradient on the interface between interface. During remeshing the algorithm to define the interface between the two materials (see
the two materials. This means that it is better to use nodal interfaces to allow the pressure gradients Section 5 ) introduces an error which is of order h ( the average distance between two neighbor
reproduce exactly this possibility. particles) in the volume definition of each material.
The free-slip condition is not trivial in Lagrangian methods. We have modelled it assigning l = 0 in
the elements adjacent to the free-slip walls.
In order to check the volume variation due to the remeshing process, the water mass
Fig. 6.2 shows the interface relative height at the vertical walls (side A and B) for previous works [30,31] conservation was measured. Fig. 6.6 shows the water mass for different mesh sizes and fixed time
compared with PFEM. PFEM results reproduce well both the interface amplitude and oscillation step during a large time period (5 s). We can observe that mesh sizes of order
frequency.
h = 0.005 are necessary in order to preserve the mass during all this period at a reasonable level (less
When the flow reaches the steady state, the interface height is than 5% of mass loss).
0.2998 instead of the correct value of 0.3. Notice that no mass correction algorithm was implemented.
Neither was necessary to use enhanced element integration nor shape function enrichment as in 6.2. Boussinesq cavity
interface-capturing methods. The pressure field and the interface discretization are shown in Fig. 6.3 .
This example deals with one fluid only, but now density changes as a function of temperature: q
= q r[ 1 a( T T r)]. We have applied the Boussinesq approximation as described in Section
This example is interesting to test the correct use of the stabilization term of the incompressibility
condition for equal order formulation as the presented here. Effectively, as stated before, standard 4.
stabilizations based on residual projections like FIC [34] The problem domain is a square cavity with vertical walls at
T = 20.5 and 19.5 C, and adiabatic horizontal walls. Fluid properties are the following: q = 1 kg/m 3, l = 10
propose a stabilization term based on Eq. (3.22) . This stabilization is based on the difference between 3 Pa s, g = 10 m/s 2,
the pressure gradient and the continuous pressure gradients. Nevertheless, in many cases such as a = 0.1 C 1, k = 10 3 m 2/ s, D T = 1 C, L = 1 m. It gives a Rayleigh number of Ra ¼ q g a D TL 3
density heterogeneous materials, the pressure gradients are physically discontinuous along the l k ¼ 10 6. All walls are considered no-slip.
interface. The introduction of a continuous vector p produces an error in the incompressibility This is a standard problem used for verification of thermal codes. However we must note that for
the Lagrangian formulation it is a difficult test because although the temperature distribution
Fig. 6.13. Density variation with temperature. Fig. 6.14. Viscosity variation with temperature.
2764 S. Idelsohn et al. / Comput. Methods Appl. Mech. Engrg. 198 (2009) 2750–2767
achieves very quickly a stationary state, the position of the particles is never stationary. 6.3. Heterogeneous flow with several materials
A benchmark solution can be found in [36] . Fig. 6.7 shows a good agreement between ref. [36] and The formulation proposed can be applied with no modification to any arbitrary number of fluids
PFEM’s solutions. Fig. 6.8 with different densities. In this example we check the behavior with three different fluids with different
shows the temperature field evolution to the steady state during the first 50 s. densities and same viscosity ( l = 0.1 Pa s). The computa-
tional domain is a 1 1 m cavity, with free-slip walls and gravity force g = 10 m/s 2.
There is not any comparison result to test the validity of this example. However the results show
clearly the ability of the method to deal with any kind of heterogeneous material.
Fig. 6.16. Variation of the temperature in time for four different particles.
Magmas are molten rock below the earth’s surface. Their properties depend mainly on their
The objective that motivated this work from the beginning is to be able to track the movement of
composition (SiO 2, Al 2 O 3, FeOT, H 2 O, etc.) and temperature. Typically at the beginning the chamber is
one fluid inside another. We believe that particle methods are especially well suited for this kind of
filled with one fluid only (fluid A) that cools due to the lower temperature of the surrounding crustal
problems, where the fluids fragment and the multiple interfaces become too complex for
rock. The decrease in temperature produces that some chemical components start to crystallize and
interface-capturing methods.
deposit at the bottom. The crystallization process changes the density and the composition of the
remaining melt. At some instant new magma (fluid B) enters the chamber. It is hotter, denser and less
In this example a fluid of density q = 800 kg/m 3 and temperature T = 50 C, is poured into a 2 2 m
viscous than fluid A. The difference in temperature and properties between both fluids triggers the
container filled 3/4 with water ( q = 1000 kg/m 3) at T = 20 C. Both fluids are immiscible and have the
convective transport, and A and B start to mingle. Molecular diffusion is of order 10 10 m 2/ s and can be
same viscosity, l = 10 3 Pa s. Density is considered here as constant and independent of temperature.
neglected. Fluids are considered as immiscible.
Thermal diffusivity k = j/q C has been set to k = 10 2 m 2/ s and inflow jet velocity to v = 1 m/s.
We assume in our simulation that the chamber (of radius 680 m) contains 50% of each fluid and
they are at rest at t = 0. Initial temperature of fluid A is 850 C and of fluid B, 1150 C. Walls are
considered no-slip, adiabatic and initially at 200 C. Gravity takes the usual value 9.81 m/s 2.
Fluid A Fluid B
T r = 1000 T r = 1000
q 0 = 2575 q 0 = 2725
a = 4.52e 5 a = 6.34e 5
Fluid A Fluid B
A = 4.55 A = 4.55
B = 8106.06 B = 5432
C = 162.34 C = 377.94
Fig. 6.11 shows how one of the fluids with smaller density penetrates in the other (namely water)
At temperature lower than 600 C viscosity is so high that magma would behave as solidified.
with higher density. The jet generates some vorticity and mingling, and at the end fluids separate due
However, in this temperature range we model magma as a fluid with constant viscosity (see Figs.
to their different densities. The temperature field is also represented in Fig. 6.12 .
6.13 and 6.14 ). Thermal diffusivity is set to k = 10 4 m 2/ s constant (two orders of magnitude higher
than realistic values).
6.5. Heterogeneous fluids with bouyant thermal effects Although our formulation is not complete for variable viscosity terms (see discussion on Section 2 ,
Eqs. (2.6) and (2.7) ), numerical results are quite reasonable: fluid B cools rapidly at the material
The last example represents an application of the method to the study of a magma chamber in interface and chamber boundary (see Fig. 6.15 ). Its density increases and moves downwards due to
which two different fluids with different properties and temperature mix and evolve to a more stable gravity and low viscosity. This movement drags fluid A downwards too, but fluid A is less
position.
2766 S. Idelsohn et al. / Comput. Methods Appl. Mech. Engrg. 198 (2009) 2750–2767
dense than B and pulls upwards. Convection produces fragmentation and fluid mixing. The evolution rials may be seen in Fig. 6.17 representing the density field at different time steps.
of the mixing of the two mate-
S. Idelsohn et al. / Comput. Methods Appl. Mech. Engrg. 198 (2009) 2750–2767 2767
We have tracked also the temperature of four particles along the simulation. The initial positions [8] X.Y. Hu, N.A. Adams, A multi-phase SPH method for macroscopic and
mesoscopic flows, J. Comput. Phys. 213 (2006) 844–861. [9] B. Nayroles, G. Touzot, P. Villon, Generalizing the
of the four particles can be seen in Fig. 6.15 at t = 0. Their temperature history with time is shown in Fig.
FEM: diffuse approximation
6.16 . and diffuse elements, Comput. Mech. 10 (1992) 307–318. [10] T. Belytschko, Y. Lu, L. Gu, Element free
Galerkin methods, Int. J. Numer.
Methods Engrg. 37 (1994) 229–256. [11] E. Oñate,
The capability of the Lagrangian formulation to track any physical parameter and the ability of
S.R. Idelsohn, O.C. Zienkiewicz, A finite point method in
particle methods to handle this kind of problems is demonstrated in this example. computational mechanics. Applications to convective transport and fluid flow, Int. J. Numer. Methods Engrg. 39
(1996) 3839–3866. [12] E. Oñate, S.R. Idelsohn, O.C. Zienkiewicz, R.L. Taylor, C. Sacco, A stabilized finite
point method for analysis of fluid mechanics problems, Comput. Methods Appl. Mech. Engrg. 139 (1–4) (1996)
7. Conclusions 315–346. [13] E. Oñate, C. Sacco, S.R. Idelsohn, A finite point method for incompressible flow
problems, Comput. Visual. Sci. 2 (2000) 67–75. [14] S.R. Idelsohn, E. Oñate, N. Calvo, F. Del Pin, The
The Particle Finite Element Method (PFEM) has been used to solve the incompressible
meshless finite element
Navier–Stokes equations for heterogeneous fluid flows. The results show the ability of the method to method, Int. J. Numer. Methods Engrg. 58 (6) (2003) 893–912. [15] S.R. Idelsohn, N. Calvo, E. Oñate,
deal with problems from the simple case of fluids with a single interface to the case of strong mixed Polyhedrization of an arbitrary point set,
Comput. Methods Appl. Mech. Engrg. 192 (22–24) (2003) 2649–2668. [16] S.R. Idelsohn, E. Oñate, F. Del Pin,
fluids with multiple interfaces. Problems with a big difference between the two materials were also
The particle finite element method: a
performed without showing any instability. powerful tool to solve incompressible flows with free-surfaces and breaking waves, Int. J. Numer. Methods
Engrg. 61 (7) (2004) 964–984. [17] E. Oñate, S.R. Idelsohn, F. Del Pin, R. Aubry, The particle finite element
method.
An overview, Int. J. Comput. Methods 2 (1) (2004) 267–307. [18] E. Oñate, J. Garcia, S. Idelsohn, Ship
Many of the previous works for incompressible Navier–Stokes equations have been thought for hydrodynamics, in: E. Stein, R. De Borst,
homogeneous materials. For heterogeneous fluids there are two main differences: one is the T.J.R. Hughes (Eds.), Encyclopedia of Computational Mechanics, J. Wiley, 2004.
possibility to have a discontinuity on the pressure field. The other drawback is to have discontinuous [19] S.R. Idelsohn, E. Oñate, F. Del Pin, N. Calvo, Lagrangian formulation: the only
way to solve some free-surface fluid mechanics problems, in: H.A. Mang, F.G. Rammerstorfer,
pressure gradients. The first case appears when there is a jump in the viscosity parameter (viscous
J. Eberhardsteiner (Eds.), Fifth World Congress on
heterogeneous flows). The second case is typical when there is a density jump (density Computational Mechanics, Vienna, Austria, 2002 July 7–12. [20] A. Larese, R. Rossi, E. Oñate, S.R. Idelsohn,
heterogeneous flows). In this paper we do not treat the first one (we solve problems with viscous Validation of the particle finite
element method (PFEM) for simulation of the free-surface flows, Engrg. Comput. 25 (4) (2008) 385–425.
variation but without a jump in the viscosity). For the case of density jumps however, we conclude that
standard stabilized methods such as pressure projection methods are not accurate because they all [21] S.R. Idelsohn, E. Oñate, F. Del Pin, N. Calvo, Fluid–structure interaction using
work with a continuous pressure gradient projection. To include a discontinuous pressure gradient the particle finite element method, Comput. Methods Appl. Mech. Engrg. 195 (2006) 2100–2113.
[23] S.R. Idelsohn, J. Marti, A. Limache, E. Oñate, Unified Lagrangian formulation for
elastic solids and incompressible fluids. Application to fluid–structure interaction problems via the PFEM,
Comput. Method Appl. Mech. Engrg. 197 (2008) 1762–1776.
Finally we conclude that in such cases where there is a jump in the density properties it is
[24] E. Oñate, J. Garcia, S. Idelsohn, F. Del Pin, Finite calculus formulations for finite
essential to force the interface to match the nodes (nodal interfaces). Otherwise very poor element analysis of incompressible flows: Eulerian, ALE and Lagrangian approaches, Comput. Method Appl.
approximations are obtained and cases with large density jumps cannot be solved. The disadvantage Mech. Engrg. 195 (2006) 3001–3037. [25] D. Gray, A. Giorgini, The validity of the Boussinesq approximation for
to use nodal interfaces is the lack of mass preservation during the remeshing process. Nevertheless liquids
and gases, Int. J. Heat Mass Transfer 19 (1976) 545–551. [26] A. Trykozko, W. Zijl, Complementary finite
this mass variation is order h and may be reduced decreasing the mesh size.
element methods applied to the
numerical homogenization of 3D absolute permeability, Commun. Numer. Methods Engrg. 18 (n 1) (2002)
31–41.
[27] J.A. Sethian, Evolution, implementation, and application of level set and fast
marching methods for advancing fronts, J. Comput. Phys. 169 (2001) 503–555. [28] S.J. Osher, R.P. Fedkiw,
Acknowledgements Level set methods: an overview and some recent
results, J. Comput. Phys. 169 (2001) 463–502. [29] J.-L. Guermond, L. Quartapelle, A projection FEM for
variable density
M. Mier-Torrecilla thanks the Catalan Agency for Administra-
incompressible flows, J. Comput. Phys. 165 (2000) 167–188. [30] T.E. Tezduyar, M. Behr, S. Mittal, A new
tion of University and Research Grants (AGAUR), the European Social Fund and CIMNE for their strategy for finite element
support. She also thanks F. Costa, A. Geyer and J. Martí from the Volcanology Group at the Spanish computations involving moving boundaries and interfaces. The deformingspatial-domain/space–time procedure:
II. Computation of free-surface flows, two-liquid flows and flows with drifting cylinders, Comput. Methods Appl.
National Research Council (CSIC) for their useful suggestions and comments.
Mech. Engrg. 94 (1992) 353–371.
References [32] A. Limache, S.R. Idelsohn, R. Rossi, E. Oñate, The violation of objectivity in
Laplace formulations of the Navier–Stokes equations, Int. J. Numer. Methods Fluids 54 (6–8) (2007) 639–664.
[1] J. Donea, A. Huerta, Finite Element Method for Flow Problems, John Wiley,
2003. [33] R. Codina, Pressure stability in fractional step finite element methods for
[2] J.J. Monaghan, An introduction to SPH, Comput. Phys. Commun. 48 (1988) 89– incompressible flows, J. Comput. Phys. 170 (2001) 112–140. [34] E. Oñate, A stabilized finite element method
96. for incompressible viscous flows
[3] R.A. Gingold, J.J. Monaghan, Kernel estimates as a basis for general particle using a finite increment calculus formulation, Comput. Methods Appl. Mech. Engrg. 182 (1–2) (2000) 355–370.
methods in hydrodynamics, J. Comput. Phys. 46 (1981) 429–453. [4] R.A. Gingold, J.J. Monaghan, Smoothed
particle hydrodynamics, theory and [35] E. Oñate, Derivation of stabilized equations for numerical solution of
application to non-spherical stars, Mon. Not Roy. Astron. Soc. 181 (1997) 375– advective–diffusive transport and fluid flow problems, Comput. Methods Appl. Mech. Engrg. 151 (1998)
389. 233–265. [36] G. De Vahl Davis, Natural convection of air in a square cavity. A bench mark
[5] W.K. Liu, S. Jun, Y.F. Zhang, Reproducing Kernel particle methods, Int. J. Numer.
Methods Fluids 20 (1995) 1081–1106. [6] S. numerical solution, Int. J. Numer. Methods Fluids 3 (1983) 249–264. [37] R.A. Lange, I.S.E. Carmichael,
Koshizuka, H. Ikeda, Y. Oka, Numerical analysis of fragmentation Thermodynamic properties of silicate liquids with
mechanisms in vapor explosions, Nucl. Engrg. Design 189 (1999) 423–433. [7] N. Shirakawa, H. Horie, Y. an emphasis on density, thermal expansion and compressibility, Rev. Mineral. Geochem. 24 (1990) 25–64.
Yamamoto, Y. Okano, A. Yamaguchi, Analysis of jet
flows with the two-fluid particle interaction method, J. Nucl. Sci. Technol. 38 (2001) 729–738. [38] D. Giordano, J.K. Russell, D.B. Dingwell, Viscosity of magmatic liquids: a model,
Earth Planet. Sci. Lett. (2008).