Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Look up keyword
Like this
2Activity
0 of .
Results for:
No results containing your search query
P. 1
Influence of Smoothing Length and Virtual Particles on SPH Accuracy

Influence of Smoothing Length and Virtual Particles on SPH Accuracy

Ratings: (0)|Views: 84|Likes:
Research paper for discussing the effect of optimizing the smoothing length and using virtual particles on the numerical solution accuracy.
Research paper for discussing the effect of optimizing the smoothing length and using virtual particles on the numerical solution accuracy.

More info:

Published by: Tarek Mahmoud El-Gammal on Sep 25, 2013
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as DOCX, PDF, TXT or read online from Scribd
See more
See less

09/28/2013

pdf

text

original

 
 
Abstract 
Using a well known 2-D transient heatconduction benchmark, a parametric study is conducted toidentify the best values for the smoothing length (h) as functionof particles spacing in order to minimize the error in thetemperature distribution particularly at the center of thedomain when applying the Smooth Particle Hydrodynamics(SPH) algorithm. Also the effect of the virtual particles on thenumerical solution has been investigated. The results showedthat the predefined smoothing length can have significantinfluence on the accuracy of the prediction and using theoptimum value leads to minimum error in temperatureprediction values. The results obtained with and withoutvirtual particles has been presented and compared with theanalytical solution.
Key words 
Meshless, Smoothing length, SPH, Virtualparticles
I.
 
I
 NTRODUCTION
 One of the most popular numerical meshless methods for solving the physical problems is the Smoothing ParticleHydrodynamics (SPH).
It‟s a Lagrangian adaptive methodthat doesn‟t nee
d to grid/mesh for solving the discretizeddomain. In this method the physical domain is discretizedinto unconnected particles that have physical properties (i.e.volume, mas
s, velocity….etc) of the domain
[1]. Unlike theconventional numerical methods such as Finite Difference(F.D.) the meshless method uses field functions and their derivatives at different points (particles) whose values areapproximately calculated using weighting (smoothing)kernel function. SPH is one of the recently developedmeshless methods which have acquired the reliability andtrustworthy in the last years such as the meshless element-free Galerkin (EFG) method [2], [3], [4], moving least-square methods [5], [6] and the diffuse approximationmethod [7].
Manuscript received November 30, 2012.T. El-Gammal is with the mechanical power engineering department,Cairo University, Giza, Egypt (e-mail: tarekelgammal@eng.cu.edu.eg). E. E. Khalil is with the mechanical power engineering department,Cairo University, Giza, Egypt (e-mail:khalile1@asme.org). H. Haridy is with the mechanical power engineering department, CairoUniversity, Giza, Egypt (e-mail:hatem_kayed@eng.cu.edu.eg). E. Abo-Serie was with the mechanical power engineering department,Cairo University, Egypt. He is now with Mechatronics Department,Mevalana University, Konya, Turkey (e-mail: eaboserie@mevlana.edu.tr ) 
The Lagrangian nature of SPH makes it more suitablefor simulations involving fluid free surfaces, interfaces between different material, multiphase flow and highlydeformable geometries. Simulating such applications bymesh based methods is challenging and required largememory and long computational time due to the continuouschange in the computational mesh in space and time to fitthe domain. After being developed for simulating thecosmological phenomenon [8], [9], SPH has been used insolving fluid and solid mechanics problems.The main advantages of SPH are the easiness to modelcomplex material surface behavior and complicated physics,such as solidification [10], [11] and multiphase flows [12].In SPH, The field function (Temperature, pressure,
velocity….etc.) is calculated for a discrete particle from the
field functions of the all surrounding particles weighted bythe smoothing function. The smoothing (weighting) kernelfunction takes into account the effect of the neighboring particles within a predefined radius called smoothing length(h). Many famous kernel functions use a dependent factor (K) to multiply with the smoothing length (h).In this work the value of smooth length has been variedto investigate its effect on the final result. The problemchosen for this investigation is a 2-D diffusion transient casewhich has been simulated previously using SPH techniquewith a constant predefined smooth length [13]. This previous study showed significant error particularly at thelocation where low temperature gradient exist (at the center of the plate). The current study developed the technique tominimize the error by varying the smooth length andintroducing virtual ghost particle to improve the boundaryconditions representation. The results have been comparedwith previous investigation and the analytical solution of theconduction equation.II.
 
P
ROBLEM
D
EFINITION
 The problem used here is the famous benchmark 2-Drectangular plate of lengths L x H undergoing heatconduction process. The process will be mainly cooling dueto isothermal cooling boundaries. The plate temperature will
 
Influence of Smoothing Length and Virtual Particles onSPH Accuracy
 
T. El-Gammal
1
, E. E. Khalil
1
, H. Haridy
1
, and E. Abo-Serie
2
 
1
Department of Mechanical Power Engineering, Cairo University, Giza, Egypt
2
Mechatronics Department, Mevlana University, Konya, Turkey
 
cool from initial temperature (T
i
) to final temperature (T
)which depends on the final state required to be stopped at.For 2-D heat diffusion in Cartesian coordinates, theequation for the temperature field T = T(x, y, t) is

 


(1)
where (α) is the therma
l diffusivity in (m/s
2
).An analytical solution based on separation of variablesand substitution of specified boundary conditions can bewritten in symbolic form to identify the temperature of theinterior of the plate T(x,y,t) as follows:
 





 (2)III.
 
M
ETHODOLOGY
 i.
 
G
OVERNING
E
QUATIONS USING
SPHSPH formulation starts from the integral representation
of field function using the Dirac delta function δ(x
ij
).Thefollowing relation states the one dimensional mathematicalconcept:
 
(
) (

) (

)  
(3)The main step of the derivation of SPH is substitutingthe Dirac delta function with even, limited-zone, kernelfunction W (x
ij
, h). Then we can write the functionapproximation in SPH formulation as:
 
 
〉(
) 

 
(4)Where <f(x
i
)> is the kernel approximation of the scalar fieldf(x) at particle (i).For two or three dimensions, the kernel function W (x
ij
,h) can be replaced by W (r 
ij
, h) such that (r 
ij
) is the vectorialdistance between (i) and (j). Also the differential length (dx)is updated to differential area (ds) or volume (dv).
The gradient (∂f(xi)/ ∂x) and Laplacian (∂
2
f(xi)/ ∂x
2
) of the field function are evaluated in integralrepresentation[13] to be:


 〉(
)


 
(5)


 

 
(
)




 
 (6)The second main step of SPH is after the integralrepresentation is the particle approximation. The fieldfunctions in (4), (5) and (6) are discretized in spaced,differential particles representing the domain, so theintegrations approximate to summations of functionsweighted by the kernel function.
 
〉
(
) 


(7)
 〉
(
)



(8)
 

 
(
)





(9)SPH formulation mentioned above can be substitutedinto the heat diffusion equation (1). Accordingly, the SPHrepresentation of pure heat conduction case in 2-D can bewritten as follows:
 
[
(
)]



(
)



 (10)The compact cubic spline is kernel function has beenused in this study. It was firstly proposed by Monaghan andLattanzio [14].The formulations of the cubic spline kernel function(W
ij
) and its spatial derivative can be shown in (11) and(12).
(

)

{

 
(11)









(12)Where (R=r 
ij
/h) is the ratio between the magnitude of thedistance between two neighboring particles (r 
ij
) and thesmoothing length (h). In Eq. (11), (n) is the dimension of thespace, and the coefficient (a
n
) is dependent on thesmoothing length (h) and has values:
h

h

h
 in one, two, and three dimensions, respectively.IV.
 
 N
UMERICAL
A
 NALYSIS
 ii.
 
D
OMAIN
P
ARTICLES
D
ISTRIBUTION
 Using Matlab
®
, the plate domain (L= H) is discretizedinto 41x41 particles which are classified into interior and boundary particles.The initial temperature (T
i
) is set to all the particles.Particle differential area will be calculated from its spacings
in x and y coordinates (i.e. m/ρ = ds = dx x dy). The plate
 
starts at T
i
= T (x, y, 0) = 100 ºC with isothermal cooling boundaries T
 b
=T (0, y, T) =T (L, y, T) =T (x, 0, T) =T (x, H,T) = 0 ºC.i.
 
V
IRTUAL
(
GHOST
)
PARTICLES
 The virtual or ghost particles [15], [16] are generated inevery time step as an image for the medium particles near the boundary (i.e. lie within the kernel domain of thewall/boundary particle) as in Fig. 1.
Figure 1: Real and virtual particles within the supporting zone of the boundary particle
The virtual particle gains its physical properties fromits real particle and the boundary condition of the wall. Theartificially existed particles play an important role incompleting the kernels of the particles to give accurate physical properties for all particles. Also they facilitate theapplying of the boundary conditions as in F.D. especiallyfor Neumann boundary condition.In the problem domain, the virtual particles were appliedto the four boundaries as shown in Fig. 2.
Figure 2: Plate domain discretization into real, boundary and virtual particles
ii.
 
T
IME
M
ARCHING
 Depending on implicit second order Leapfrog method,the temperature at the mid of the current time step iscalculated using the value of mid of the previous time stepand the temperature rate of change based on the values of the time step preceding.At t = dt (i.e. n=1)


|

(13)Where

|

are evaluated by initial values T
0
from theheat conduction equation.For n > 1



|
(14)Where

|
are evaluated by T
n-1
from the heatconduction equation.The time step has been chosen to satisfy the Monaghanand Cleary [17] criterion for pure heat conduction problems,
 
(15)In this case, the thermal diffusivity and time step wereconsid
ered as (α=1 cm
2
/s, dt = 0.001 s) respectively. Thevalue of (h) was then optimized to minimize the error.In a second step, the optimized SPH algorithm wascompared with no-virtual SPH to study the error due to theincomplete kernels near the boundary. Both results weretaken at the same (hopt) and the time step (dt = 0.001s).

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->