www.bu.edu pasi files 2011 01 luisdelacruz1!11!1700

8/12/2019 Www.bu.Edu Pasi Files 2011 01 LuisdelaCruz1!11!1700

1/61

Oil Reservoir Simulation General Math & Num Models Computational Approach References

Computational Methods for Oil RecoveryPASI: Scientific Computing in the Americas

The Challenge of Massive Parallelism

Luis M. de la Cruz Salas

Instituto de Geofsica

Universidad Nacional Autonoma de Mexico

January 2011

Valparaso, Chile

Comp EOR LMCS 1 / 43


2/61


Mathematical and Computational Group (GMMC)

Natural Resources Dept.Dr. Ismael Herrera Revilla

Dra. Graciela Herrera Zamarron

Dr. Luis M. de la Cruz Salas

Dr. Guillermo Hernandez Garca

Dr. Norberto Vera Guzman

Students

Esther Leyva

Antonio Carrillo

Ivan Contreras

Alberto Rosas

Emilio Zavala

Ricardo Flores

Computational Group

Daniel A. Cervantes Cabrera

Alejandro Salazar SanchezDaniel Monsivais Velazquez

Renato Leriche Vazquez

Hector U. Barron Garca

Ismael Herrera Zamarron

Eduardo Murrieta Leon

http://www.mmc.geofisica.unam.mx



3/61


Table of contents

1 Oil Reservoir SimulationMotivation

2 General Math & Num Models

Axiomatic FormulationNumerical MethodsFinite Volume Method

3 Computational Approach

TUNA

4 References



4/61


Motivation

Table of contents





TUNA

4 References



5/61


Motivation

Oil Reservoir Projects

Funded by PEMEX


O l R S l G l M h & N M d l C l A h R f


6/61


Motivation


Funded by PEMEX

Collaboration with IMP and CIMAT.


Oil R i Si l ti G l M th & N M d l C t ti l A h R f


7/61


Motivation


Funded by PEMEX


1 WAG injection.


Oil R s i Sim l ti G l M th & N m M d ls C m t ti l A h R f s


8/61


Motivation


Funded by PEMEX


1 WAG injection. 2 AIR injection.




9/61


Motivation


Funded by PEMEX


1 WAG injection. 2 AIR injection. 3 SLS method.




10/61


Motivation


Funded by PEMEX


1 WAG injection. 2 AIR injection. 3 SLS method.

Oil reservoir simulation is a grand challenge.




11/61


Motivation

The major goal of reservoir simulation is to predict futureperformance of reservoir and find ways and means of optimizing

the recovery of some of the hydrocarbons under various operatingconditions.




12/61

O Rese vo S u at o Ge e a at & u ode s Co putat o a pp oac Re e e ces

Motivation


the recovery of some of the hydrocarbons under various operatingconditions.It involves four main interrelated modeling stages:




13/61

p pp

Motivation


the recovery of some of the hydrocarbons under various operatingconditions.It involves four main interrelated modeling stages:

And requires a combination of skills of physicists, mathematicians,

reservoir engineers, and computer scientists.Comp EOR LMCS 6 / 43



14/61

Motivation

Software Engineering (IEEE Comput Societys Software Eng. Body of Knowledge)

Application of a systematic, disciplined, quantifiable approach to the development,operation, and maintenance of software, and the study of these approaches; that is, theapplication of engineering to software.

Unified Process (UP, Boochet al. [2])




15/61

Motivation







16/61

Motivation




Requirements

EfficiencyAccuracyAbstraction




17/61

Motivation




Requirements

EfficiencyAccuracyAbstraction

We get a software:

ModularMantainableReliableEfficientProductive




18/61

Motivation

Oil production stages

First stage of oil reservoir production, primary recovery, the oilis extracted by natural drive mechanism.




19/61

Motivation



The reservoir pressure can be maintained, using techniques such asgas or water injection. This is known as secondary recovery.




20/61

Motivation




Tertiary or Enhanced Oil Recovery (EOR) is a generic term

that embraces several techniques used to increase the amount ofcrude oil that can be extracted from an oil field.

These techniques are based on the injection of materials notnormally present in the reservoir, and is the most advanced stage of

the exploitation of a reservoir.




21/61

Motivation




Tertiary or Enhanced Oil Recovery (EOR) is a generic term

that embraces several techniques used to increase the amount ofcrude oil that can be extracted from an oil field.

These techniques are based on the injection of materials notnormally present in the reservoir, and is the most advanced stage of

the exploitation of a reservoir.Primary recovery techniques produce 10 15 % of the reservoirsoil content. Combining the processes of secondary andtertiary recovery techniques, it is possible to produce 30

60 % of the reservoirs total oil content.




22/61

Axiomatic Formulation

Table of contents





TUNA

4 References




23/61


Extensive and Intensive Properties

In the physical sciences, intensive property(also called a bulkproperty, intensive quantity, or intensive variable), is a physicalproperty of a system that does not depend on the system size or

the amount of material in the system: it is scale invariant.Density

By contrast, an extensive property(also extensive quantity,extensive variable, or extensive parameter) of a system is directlyproportional to the system size or the amount of material in the

system.Mass




24/61


Axiomatic Formulation, (Herrera et al. [3, 4]) I

1 To find extensive Eand intensive properties :

E(t) =

B(t)

(x, t)dx

2 To establish balances:

dE

dt

= d

dt

B(t)

(x, t)dx= B(t)

q(x, t)dx+ B(t)

(x, t) ndS (1)

where q(x, t)y (x, t)are the source term in B(t)and the fluxvector through the boundary B(t), respectively




25/61


Axiomatic Formulation, (Herrera et al. [3, 4]) II

Global balance

B(t)

t + (v)

dx=

B(t)

qdx+

B(t)

dx (2)

Local balance

t + (v) =q+ (3)




26/61


Conservative form

Defining a flux function (see [5]) as f=v we get:

t

B(t)

dx+

B(t)

fdx=

B(t)

qdx (4)

and therefore

t + f=q (5)

Equivalently (4) can be written as follows

t

B(t)

dx+

B(t)

f ndS=

B(t)

qdx (6)




27/61

Numerical Methods

Table of contents





TUNA

4 References



N i l M th d


28/61

Numerical Methods

In general, the equations governing a mathematical model of areservoir cannot be solved by analytical methods.



N i l M th d


29/61

Numerical Methods


Instead, a numerical model can be produced in a form that isamenable to solution by digital computers.



Numerical Methods


30/61

Numerical Methods



Since the 1950s, numerical models have been used to predict,understand, and optimize complex physical fluid flow processes inpetroleum reservoirs.



Numerical Methods


31/61

Numerical Methods



Since the 1950s, numerical models have been used to predict,understand, and optimize complex physical fluid flow processes inpetroleum reservoirs.

Recent advances in computational capabilities have greatlyexpanded the potential for solving larger problems and hence

permitting the incorporation of more physics into the differentialequations.



Numerical Methods


32/61

Numerical Methods

1 Finite Diferences Method (FDM)

The FDM can be very easy to implement.Faster than FEM.High accuracy difference schemes can be constructed.In its basic form is restricted to handle only rectangular shapes.Introduce considerable geometrical error and grid orientation effects.Curvilinear coordinates can be used.



Numerical Methods


33/61

Numerical Methods

1 Finite Diferences Method (FDM)

The FDM can be very easy to implement.Faster than FEM.High accuracy difference schemes can be constructed.In its basic form is restricted to handle only rectangular shapes.Introduce considerable geometrical error and grid orientation effects.Curvilinear coordinates can be used.

2 Finite Element Method (FEM)The FEM cand handle complicated geometries.Reduce the grid orientation effects.Solid theoretical foundations.Can manage local grid refinement.

The quality of a FEM approximation is often higher than in thecorresponding FDM approach.FEM is the method of choice in structural mechanics.It is not easy to implement and is slower than FDM.



Numerical Methods


34/61

3 Finite Volume Method (FVM)

Values are calculated at control volumes.

Conservative method: the flux entering a given volume is identicalto that leaving the adjacent volume.Can easily be formulated to allow for unstructured meshes.Used in many computational fluid dynamics packages.FVM is in between FDM and FEM: faster and easier to implementthan FEM; and more accurate and versatile than FDM.



Numerical Methods


35/61




In oil-reservoir problems we usually require a large number of cells(105 106), therefore cost of the solution favors simpler and lowerorder approximation within each cell.



Numerical Methods


36/61




In oil-reservoir problems we usually require a large number of cells(105 106), therefore cost of the solution favors simpler and lowerorder approximation within each cell.

About 80% 90% of the total simulation time is spent on thesolution of linear systems.

Fast linear solvers are crucial to solve sparse, highly nonsymmetric,and illconditioned systems.Krylov subspace (preconditioned) algorithms are preferred.



Finite Volume Method


37/61

Table of contents





TUNA

4 References





38/61

Finite volume methods are derived on the basis of the conservativeform of the balance equations [3, 4, 5].

t

B(t)

dx+

B(t)

f dx=

B(t)

qdx or

t

B(t)

dx+

B(t)

f ndS=

B(t)

qdx

FVM is conservative.





39/61

Conservative formof general balance equation:

t

B(t)

dx+

B(t)

f dx=

B(t)

qdx

Integrating ont and taking B(t) V:

t

t

V

dV dt+ t

V

fdV dt= t

V

qdV dt

xz

y

E

W

N

S

F

B

Pee

ww

nn

ss

ff

bb

V = xyz

Notation:

NB id

P i ,j,kE i+ 1, j , k

W i 1, j , kN i, j+ 1, kS i, j 1, kF i ,j,k+ 1B i ,j,k 1

nb id

e i+ 12 , j , kw i 1

2, j , k

n i, j+ 12

, k

s i, j 12

, k

f i ,j,k+ 12

b i ,j,k 12

t n; t+ t n+ 1

t

g dt

n+1n

g dt

V

g dV

f

b

ns

ew

g dx dy dz





40/61

Approximation of the integrals:

n+1

n

t

V

dV dt

n+1P n

P

V

n+1n

V

f dV dt

n+1n

F( fnb

)dt

t

V

qdV dt

n+1n

QV dt

xz

y

E

W

N

S

F

B

Pee

ww

nn

ss

ff

bb

V = xyz

Theta scheme:

n+1

n

Fdt=

Fn+1

+ (1 )Fn

t, 0 1

Explicit (ForwardEuler) = 0 Fnt.Implicit (BackwardEuler) = 1 Fn+1t.CrankNicolson = 1/2 (Fn + Fn+1)t/2.





41/61

Recall that f =v :

F( fnb)

V

f dV =

ew

ns

f

b

(vx x)x

+(vy y)

y +

(vz z)

z

dxdydz

xz

y

E

W

N

S

F

B

Pee

ww

nn

ss

ff

bb

V = xyz

Discretized flux function:

F( fnb

) = (vx x)e (vx x)wAx+(vy y)n (vy y)s

Ay+

(vz z)f (vz z)b

Az

where Ax = y z, Ay = x z, Az = x y, represents the area ofthe faces.

Advectivev and Diffusive terms need to be approximated on the faces.





42/61

Diffusive terms

Central differences: e.g. (x)e=D

x

e

=D

P

E

xeAdvective terms

Average : (vx)e= (vx)e(E+ (1 )P), where = xexP

xe

Upstream:

Upwind

if ((vx)e >0) thene =P

else

e =Eend if

QUICK (second order

upstream)

if ((vx)e >0) thene =h(W, P, E)

else

e =h(P, E, EE)end if





43/61

Implicit, non-linear:

an

+1P n

+1P =an

+1E n

+1E +an

+1W n

+1W +an

+1N n

+1N +an

+1S n

+1S +

an+1F

n+1F

+an+1B

n+1B

+qnP

Implicit linear:

anP

n+1P

= anE

n+1E

+anW

n+1W

+anN

n+1N

+anS

n+1S

+anF

n+1F

+anB

n+1B

+qnP

Explicit:an

Pn+1

P =an

En

E +a

Wn

W+an

Nn

N+an

Sn

S +an

Fn

F +an

Bn

B +qn

P




S


44/61

Linear Systems



TUNA

T bl f


45/61

Table of contents





TUNA

4 References



TUNA

T l t U it f N i l A li ti


46/61

Template Units for Numerical Applications

Natural convection in box-shaped containers



TUNA


47/61

f1(t) =

0.5sin2(4t) for 0 t < /81 for /8 t


48/61

Navier-Stokes equations

Mass balance:

ujxj = 0

Momentum balance (Navier-Stokes):

0 ui

t

+ujui

xj =

p

xi+

2ui

xjxj+bi

Energy balance:T

t +uj

T

xj=

2T

xjxj

Equations of state:

= 0[1 (T T0)] , = 1

0

T

T=T0



TUNA

Template Units for Numerical Applications I


49/61

Template Units for Numerical Applications I

TUNA use several C++ template techniques (Blitz++).



TUNA

Template Units for Numerical Applications II


50/61

Template Units for Numerical Applications II

http://mmc.geofisica.unam.mx/(then go to my homepage).

Examples : programs that use TUNA

|--- tuna-cfd-rules.in => rules to compile the examples

|--- 01StructMesh => uniform structured meshes

|--- 02NonUniformMesh => non uniform structured meshes

|--- 03Laplace => Solution of Laplace equation

|--- 04HeatDiffusion => Solution of heat conduction problems

|--- 05ConvDiffForced => Solution of forced convection

|--- 06ConvDiff => Solution of natural convection problems

|--- 07ConvDiffLES => Solution of turbulent natural convection

|--- README.pdf => Explanation of the examples of each directory

1) Unpack TUNA and change to the TUNA dir:

% tar zxvf TUNA.tar.gz% cd TUNA

2) Blitz++: http://www.oonumerics.org/blitz/



TUNA

Template Units for Numerical Applications III


51/61

Template Units for Numerical Applications III

- Unpack with: tar zxvf blitz-09.tar.gz

- Change to blitz-0.9 with: cd blitz-0.9- Config blitz with: ./configure --prefix=$PWD/../BLITZ

- Compile and install blitz with: make install

These instruction will install Blitz in the TUNA/BLITZ directory

3) Run the examples

- Change to the Examples directory: cd Examples

- Edit the files tuna-cfd-rules.in

Change the environment variable BASE according to your paths.

(e.g. BASE = /home/luiggi/TUNA)

- Then, e.g. change to the 06ConvDiff dir:

% cd 06ConvDiff% make


52/61

Example: Natural Convection I

#include "Meshes/Uniform.hpp"#include "Storage/DiagonalMatrix.hpp"

#include "Equations/ScalarEquation.hpp"

#include "Schemes/CDS_CoDi.hpp"

#include "Equations/Momentum_XCoDi.hpp"

#include "Schemes/CDS_XCoDi.hpp"

#include "Equations/Momentum_YCoDi.hpp"

#include "Schemes/CDS_YCoDi.hpp"

#include "Equations/Momentum_ZCoDi.hpp"

#include "Schemes/CDS_ZCoDi.hpp"

#include "Equations/PressureCorrection.hpp"

#include "Schemes/Simplec.hpp"

#include "Solvers/TDMA.hpp"

#include "Utils/inout.hpp"#include "Utils/num_utils.hpp"

#include "Utils/GNUplot.hpp"

using namespace Tuna;

typedef TunaArray::huge ScalarField3D;



TUNA

Example: Natural Convection II


53/61

Example: Natural Convection II

DiagonalMatrix A(num_nodes_x, num_nodes_y, num_nodes_z);

ScalarField3D b(num_nodes_x, num_nodes_y, num_nodes_z);

StructuredMesh > mesh(length_x, num_nodes_x,

length_y, num_nodes_y,

length_z, num_nodes_z);

ScalarField3D T(mesh.getExtentVolumes());

ScalarField3D p(mesh.getExtentVolumes());

ScalarField3D u(mesh.getExtentVolumes()); // u-velocity

ScalarField3D v(mesh.getExtentVolumes()); // v-velocity

ScalarField3D w(mesh.getExtentVolumes()); // w-velocity

Range all = Range::all();T(T.lbound(firstDim), all, all) = left_wall; // Left

T(T.ubound(firstDim), all, all) = right_wall; // Rigth



TUNA

Example: Natural Convection III


54/61

Example: Natural Convection III

ScalarEquation energy(T, A, b, mesh.getDeltas());

energy.setDeltaTime(dt);

energy.setNeumann(TOP_WALL);

energy.setNeumann(BOTTOM_WALL);

energy.setDirichlet(LEFT_WALL, left_wall);

energy.setDirichlet(RIGHT_WALL, right_wall);

energy.setNeumann(FRONT_WALL);

energy.setNeumann(BACK_WALL);

energy.setUvelocity(us);energy.setVvelocity(vs);

energy.setWvelocity(ws);

energy.print();

Momentum_XCoDi mom_x(us, A, b, mesh.getDeltas());

Momentum_YCoDi mom_y(vs, A, b, mesh.getDeltas());

Momentum_ZCoDi mom_z(ws, A, b, mesh.getDeltas());

PressureCorrection press(pp, A, b, mesh.getDeltas());



TUNA

Example: Natural Convection IV


55/61

a p e Natu a Co vect o V

template

class CDS_CoDi : public ScalarEquation< CDS_CoDi< Tprec, Dim > >{

public:

typedef Tprec prec_t;

typedef typename TunaArray::huge ScalarField;

CDS_CoDi() : ScalarEquation() { }

~CDS_CoDi() { };

inline void calcCoefficients1D();



inline void printInfo() { std::cout


56/61

p

template

inline void CDS_CoDi::calcCoefficients2D() {

prec_t Gdy_dx = Gamma * dy / dx, Gdx_dy = Gamma * dx / dy;

prec_t dxy_dt = dx * dy / dt;

aE = 0.0; aW = 0.0; aN = 0.0; aS = 0.0; aP = 0.0; sp = 0.0;

for (int i = bi; i


57/61

p

for(iteration = 1; iteration tol_simplec) && (counter < 20) ) {

energy.calcCoefficients();

Solver::TDMA3D(energy, tolerance, max_iter);

errorT = energy.calcErrorL2();

energy.update();

mom_x.calcCoefficients();

Solver::TDMA3D(mom_x, tolerance, max_iter);

errorX = mom_x.calcErrorL2();

mom_x.update();



TUNA

Example: Natural Convection VII


58/61

p

mom_y.calcCoefficients();Solver::TDMA3D(mom_y, tolerance, max_iter);

errorY = mom_y.calcErrorL2();

mom_y.update();

mom_z.calcCoefficients();

Solver::TDMA3D(mom_z, tolerance, max_iter);

errorZ = mom_z.calcErrorL2();mom_z.update();

press.calcCoefficients();

Solver::TDMA3D(press, tolerance, max_iter);

press.correction();

sorsum = fabs( press.calcSorsum() );

++counter;

}

return sorsum;

}



TUNA

Example: Natural Convection VIII


59/61



References I


60/61

[1] R.A. Tapia and C. Lanius,Computational Science: Tools for a Changing World

http://ceee.rice.edu/Books/CS/chapter1/intro52.html, 2001.

[2] I. Jacobson and G. Booch and J. RumbaughPrimer,The Unified Software Development Process,Addison Wesley, 1999.

[3] I. Herrera and M. B. Allen and G. F. Pinder,Numerical modeling in science and engineering,John Wiley & Sons., USA, 1988.

[4] I. Herrera and G. F. Pinder,General principles of mathematical computational modeling,John Wiley, in press.



References II


61/61

[5] R.J. Leveque,

Finite Volume Method for Hyperbolic Problems,Cambridge University Press, 2004.


www.bu.edu pasi files 2011 01 luisdelacruz1!11!1700

Documents