Está en la página 1de 26

Modified Formulation, e-Regularization

and the Efficient Solution of


Hydraulic Fracture Problems
Alexander M. Linkov
Rzeszow University of Technology, Eurotech (Poland)
Institute for Problems of Mechanical Engineering
(Russian Academy of Sciences)

Gennady S. Mishuris
Aberystwyth University, UK

The supports of EU Marie Curie IAPP program (Grant # 251475)


and Russian Fund for Fundamental Researches (Grant # 12-05-00140)
are gratefully acknowledged

Objective
Well

Rock

The importance of hydraulic fracturing


has grown last years because huge resources
of gas are found in low permeable shales
The key element of technology,
used in shales, is hydraulic fracturing

Rock
Analytical models and numerical simulations are important means to
increase understanding and enhance efficiency of hydraulic fracturing
(e.g. Mack & Warpinski, 2000; Adachi, Siebrits et al., 2007)

The authors emphasized the need


to dramatically speed up simulators
The objective: to develop improved strategy
for solving hydraulic fracture problems
22

Scope
(i)

Revisiting fundamentals

(ii)

Disclosing significant features of HF problem

(iii) Suggesting modified formulation


(iv) Discussion of its applications

Revisiting Fundamentals:
Reynolds Transport Theorem (RTT)

Temporal derivative of an integral over a moving volume of material particles


(x,t) is an arbitrary function
vt

V(t+t)
V(t)
S(t)

vt
dV
dS

d
(x, t )dV =
dt V ( t )

1
lim
(x, t + t )dV (x, t )dV =

t 0

t V + V

n
vnt

dV
t dV + tlim
0
V (t )
V (t )

dV = vndS

Hence,

dV = vn dS
V (t )
S (t )

lim

t 0

d
dV + vndS
(x, t)dV =

dt V (t)
V (t ) t
S (t )

Finally,
(RTT)

For any volume V(t) of moving particles, the RTT involves the
normal component vn of particle velocity at its boundary S(t).
The shear component does not contribute into the boundary change.
By the derivation, the normal component of the particle
velocity equals to the speed of the surface propagation
4

Speed Equation
When applied to the entire volume of moving fluid, the RTT implies
that the speed V* of its surface propagation equals to the normal
component vn of the particle velocity:

vn = V*

V* = vn

Speed Equation (SE)

The SE is Eqn (5) of the Kemps paper of 1989


(L.F. Kemp, SPE 18959, 1989, 287-298)

On page 289 of his paper of 1989, L.F. Kemp wrote:


Nowhere is (5) mentioned.
(5) is always present in moving boundary problem
Still the generality and significance of the SE have not been
comprehended and used systematically for further 20 years after 1989.
It is reasonable to systematically employ the SE for HF problems,
in particular, because it is the key eqn for tracing front propagation
by Level Set and Fast Marching Methods
5

Reynolds Transport Theorem


for flow of incompressible fluid in narrow channel
RTT = MC = VC
n SL dV e = v n dS
z
nS
L

dt

n+
S+

S-

S(t) = S+ + S- + SL
Sm
/ t
-:
+
At
S
v
=

u
n
z
n At S+: vn = u z / t
SL
At SL: dSL = wLdL L
n
wL

Hence for flow in narrow channel


where

w = u z+ u z

for incompressible fluid

S (t )

vn dS =

S (t )

w
t dS + wL vn dS
S (t )
L (t )
m

is the channel width (opening)

The RTT becomes:

dVe
w
=
dS + wL vn dL
dt
S ( t ) t
L (t )
m

V* = vn

RTT = MC = VC

SE

Now both the speed of propagation V* and


the particle velocity vn are averaged over the opening at L

Integral and Differential Forms of RTT


for flow of incompressible fluid in narrow channel
V* = v n
S(t)

dVe
w
=
dS + wvn dL
dt S (t ) t
L (t )

RTT = MC = VC

L(t) The Gauss theorem implies:

dVe
w
= [ + div(wv)]dS
dt S (t ) t

For distributed sinks (sources), the fluid out- (in-)flow is

dVe
= qe dS
dt
S (t )

This yields the integral form of the VC for flow in a narrow channel
w
[
+ div(wv) + qe ]dS = 0
RTT = MC = VC

S (t )

Being true for any part of moving fluid, it yields the differential form:
w
the common continuity equation (CE)
+ div(wv) + qe = 0
of hydraulic fracture problems
t

Note that the starting forms (both integral and differential) of the
MC for a flow of an incompressible fluid in a narrow channel
contain the particle velocity rather than the flux
7

Significance of Particle Velocity


V* = v n
S(t)

w
+ div ( wv ) + q e = 0
t

continuity equation (CE)

L(t) Physically, the product wv expresses the flux: q = wv


Emphasize that, similar to Poiseuille-type eqns,
the primary physical quantity is the particle velocity v,
while the flux q appears only by definition: q = wv

The particle velocity enters the SE and has the advantage over the flux:
commonly, it is a smooth function, which changes slowly in the flow region
This beneficial property has been clearly recognized and employed
in the review by M.G. Mack & N.R. Warpinski (2000). On page 6-21 of their
paper, they wrote:

Detailed numerical calculations have shown that the velocity


varies much more slowly than flow rate
The modified formulation employs the particle velocity in
(i) SE, (ii) CE, and (iii) Poiseuille-type eqn.
8

Significance of Modified Opening


We have noticed that the SE and the starting form of the CE suggest using:
1
Front
x2
(i) The particle velocity v = D( w, p)gradp
Hydraulic fracture
w
instead of the pressure p, which is commonly singular
r O x* n
at the front when neglecting the lag
O

x1

Note now that the SE, taken together with an elasticity eqn, implies
that commonly the opening has power asymptotics w = C(t)r, with 0 < < 1.
Hence its derivative is singular near the front: dw/dr , when r 0.
Thus, it is reasonable to use:

(ii) Modified opening y = w1/ instead of the opening w

Emphasize that the new variables y and v have good properties:

Modified opening is linear near the front


Particle velocity is non-zero finite smooth function up to the front
In addition to the particle velocity,
the modified formulation employs also the modified opening
9

Continuity Equation in System Moving


with Fluid Front. Its Implications
w
+ div(wv) + qe = 0
t

continuity equation (CE)

x
= V*t x

In the system, moving with the fluid front, the CE reads:


w
w
v
+ (V * v)
w + qe = 0
CE near the fluid front
t

Neglecting the lag, the opening tends to zero when approaching the front
(crack tip). In the limit w 0, the CE becomes:
[w(V * v)]
qe
+ qe = 0, hence, v = V * +

w /

This implies that the leak-off qe should be either finite at the


crack tip or its singularity less than that of w / .

Otherwise, the assumption that the flow occurs predominantly in the channel
plane is drastically violated both in the CE and Poiseuille-type eqn.

In particular, this makes dubious extending the Carter leak-off


approximation to the fluid front
10

Specific Feature of HF Problem:


Identically Satisfied Speed Equation
x

= V*t x

w
w
v
+ (V * v)
w + qe = 0
t

CE in moving system

Consider the common case of zero lag and predominantly in-plane flow
We could see that for zero lag, near the fluid front where w(t) = 0, in
q
the limit w 0, the CE becomes: v = V * + e
w /

qe
w /

0 . In the limit w 0:
v =V *
CE when w(t) = 0
We see that at any point of a fluid, where the opening is zero, the CE
degenerates to the equation, expressing the SE: CE SE
Conclusion: under common assumptions of HF modeling,
the boundary condition of zero opening
automatically leads to the
physically consistent
additional boundary condition, expressing the SE

where for predominantly in-plane flow

11

Conventional Formulation.
Its Specific Feature for Zero Lag
Continuity equation (local form)
w=0
(1)
divq + w / t qe = 0
S(t)
Poiseuille equation
Lq
(2)
q = D ( w, p )gradp
Reynolds equation (using (2) in (1) )
qn = q0
div[ D( w, p )gradp ] w / t + qe = 0 (3)
(4)
w( x,0) = 0
Initial condition (zero opening):
q n ( x ) = q 0 ( x ) (5)1
BC : at fluid pumping part
w (x ) = 0
(5)2
at propagating fluid front (zero lag)
But ! As shown, the BC (5)2 automatically yields additional BC at LC
1
p
v n =
D( w, p)
= V (6)
BC=SE !
w

dn

Thus for the elliptic (in spatial coordinates) operator


we have two, rather than one, physically consistent
boundary conditions
This indicates that there might be difficulties

For a fixed front, the boundary value problem appears ill-posed


12

Hadamard Definition
and Tychonoff Regularization
By Hadamard, a problem is well-posed when
A solution exists
The solution is unique
The solution depends continuously on the data, in a reasonable metric
Jacques Hadamard (1902), Sur les problemes aux derivees partielles et leur
signification physique, Princeton Univ. Bul. 49-52

Otherwise, a problem is ill-posed


Hadamard considered that ill-posed problems had no physical sense
A.N. Tychonoff (1943) clearly recognized significance of ill-posed
problems for applications. He was the first to suggest a means
to solve them numerically by using regularization:
A.N. Tychonoff (1963) Solution of incorrectly formulated problems
and the regularization method, Soviet Mathematics 4, 1035-1038.
[Transl. from Russian: . . , , 1963, 151, 501-504]

The SE being physically consistent, we need a proper method


of regularization for the problem of hydraulic fracturing
13

Clear Evidence that BVP is Ill-Posed:

Nordgren Problem

Well
O

h
w(x,t)

Elasticity equation for plane-strain


in vertical cross-sections p = kr w
x Reynolds equation (Newtonian fluid)
kl

x (t )

3 p w
=0
w

x
x t

In dimensionless variables, the problem becomes


y
2 w 4 w

Initial condition:
Boundary conditions:

x 2

=0

Nordgrens PDE

w(x,t0)=w0(x)

w 4

= q0

BC at inlet x = 0

x=0

w( x , t ) = 0

Speed Equation:

dx*
4 w 3
V* =
=
x = x* (t )
dt
3 x

BC at liquid front x = x*

There are three rather than two BC for the PDE of second order
in spatial variable x.
For any fixed x*, the problem is ill-posed
14

Even More Clear Evidence


that BVP is Ill-Posed

Well
O

The Nordgren problem is self-similar.


Introduce self-similar variables

x = t
x = t 4 / 5 , w( x) = t1 / 5 ( xt 4 / 5 )
x (t )
Denote y() = 3() The problem is reduced to ODE
4/5

d2y
d 2

+ a( y, dy / d, )

dy 3
=0
20

ODE

(1)

where a(y,dy/d,)=(dy/d+0.6)/(3y) is finite at fluid front =*

Boundary conditions for the ODE of second order:


dy
q
= 0.75 0
BC at inlet = 0 (2)
3 y (0)
= 0
BC at fluid front = * (3)
y( ) = 0
+ SPEED EQUATION, which is met identically by a solution of ODE
dy
= 0.6
satisfying BC (3):
SE at fluid front = * (4)

Thus, there are two, rather than one, BC at the fluid front. By Picards theorem,
the Cauchy conditions (3), (4) uniquely define y(), dy/d and consequently influx
at the inlet. Hence, a solution of BVP (1)-(3) does not exist for an arbitrary influx.

By Hadamard definition, the BV problem (1)-(3) is ill-posed15

Regularization Method
for Hydraulic Fracturing
We suggest the regularization method employing the very cause of the difficulty
Front
We have:
x2
Hydraulic fracture

r O x* n
s
x
1

w
div(D( w, p)gradp ) qe = 0
PDF
t
with two BC at a point x* of the fluid front

p(x* ) = p0 (x* )

1
p
D ( w, p )
w* ( x* )
n

(1)

Prescribed for a problem

(2)

Speed Equation

(3)

= vn
x= x

Integration of (3) and accounting for (2) yield*


p

(4)

w D ( w, p )dp v*r
p0
By using (4) we impose the BC at a small distance behind the front:
p

1
w D( w, p)dp = v*
p

(5)

The regularization method consists in using the BC (5) at a small


distance behind the front rather than the BC (2) and (3) on the front

We call this approach - regularization


It appears really efficient for solving HF problems
16

Illustration: Solution of Nordgren Problem

without regularization
Well
O

h
w(x,t)

x (t )

y
We solved both the starting and self-similar BV Nordgren problem
by finite differences
without regularization

By no means could we have more than two correct digits


Furthermore,
The results always deteriorated near the front
Using fine meshes (with the step less than 10-5x*) led to
complete deterioration of the solution in the entire region

This clearly shows that a proper regularization method


is needed to have accurate and reliable numerical results
17

Illustration: Solution of Nordgren Problem

with regularization

0
0

We have obtained that near the front:


Y (, t ) 0.75 x (t )v (t )(1 )
Hence, we may impose the BC
at the relative distance behind the front
Y ( , t ) = 0.75 x (t )v (t )

We solved both the starting and self-similar BV Nordgren problem


by finite differences with -regularization

Conclusions obtained:

The results are accurate in a wide range of (10-2 > >10-5), size (10-2 >
>10-5) and number (up to 100 000) of time steps
For ODE of self-similar formulation, there are six correct digits, at least;
For PDE, the error is less than 0.03% even for 100 000 steps
There are no signs of instability in specially designed experiments

Small time expense on a conventional laptop

Even for 100 000 steps, the time expense does not exceed 15 s

This shows that - regularization is really beneficial


The modified formulation uses -regularization, when appropriate
18

Summary on Modified Formulation


Summarizing, we come to the modified formulation of HF problem.
In contrast with the conventional formulation, it uses:
The particle velocity, as a variable smooth near the front,
instead of the pressure;
The modified opening, which is linear near the front,
instead of the opening itself;

The SE at each point of the front,


instead of the single equation of global mass balance;

-regularization to exclude solution deterioration caused by the fact that


the boundary value problem is ill-posed for a fixed position of the front
Moving spatial coordinates;
Reformulation of the common system of equations and BC in terms of
the suggested variables complimented, when appropriate, with regularization.
19

Analytical Advantages
of the Modified Formulation
Analytical advantages are evident when revisiting the classical problems

PKN model: plane-strain state in vertical cross section


Well

h
w(x,t)

x (t )

y
The conventional formulation requires involved calculations
See: Nordgren, Soc. Pet. Eng., 1972, August, 306-314

KGD model: plane-strain state in horizontal cross-section


p

x
n
Again, the conventional formulation requires involved calculations
See: Spence & Sharp, Proc. Roy. Soc. London, A, 1985, 400, 289-313;
Adachi & Detournay, Int. J. Numer. Anal. Meth. Geomech., 2002, 26, 579-604
For both problems, the modified formulation provides simple analytical solutions
See: Linkov, IJES, 2012, 52, 77-88; Linkov, Journal of Mining Science,
2013, 1; and Linkov & Mishuris, Proceedings of this Conference, 2013
20

Illustration:
Analytical Solution for Power-Type Viscosity Law
Analytical solution of Nordgren problem for a fluid with the power viscosity law
1/ n

v = k f wn +1
x

The solution shows that in the case of low-permeable formation, the


particle velocity is almost constant for any thinning fluid
in the entire flow region
21

Computational Advantages
of the Modified Formulation
The main computational advantages are:
Possibility to use well-established Level Set and Fast
Marching methods of the theory of propagating interfaces;
Avoiding deterioration of numerical solution;
Avoiding singularities at the fluid front;
Highly efficient simulators for the Nordgren and P3D
problems
See: Mishuris et al. IJES, 2012, 61, 10-23;
and Proceedings of this Conference, 2013

22

Illustration:
Efficient Solution of P3D Problem
The only difference with the Nordgren problem is the presence of the function H(y).
It is evaluated in advance, smooth and equals to the unit on the fracture front.

PDE:
IC:
BC:
SE:
-regularization:

After spatial discretization, we arrive at a dynamic system,


efficiently solved by standard solvers

This option has appeared ONLY due to employing the SE


23

Illustration: Using Standard Solver


for Nordgren Problem with Carter Leak-off
Standard MATLAB R-K solver
ode15s
is used to solve the dynamic system

= 10-5
Results of numerical experiments
(i) Guaranteed accuracy: 0.01%
(ii) Run-time to cover the time range
of 12 orders: 15 s
even in MATLAB environment
(iii) Extreme stability:
no signs of instability

Conclusion
The numerical scheme based on the modified formulation,
being accurate, robust and stable,
is an efficient means for solving P3D problems
24

Further Work
Further work may employ new options provided by the
modified formulation. They include:
Development of enhanced simulators for solving truly and P3D
problems in real time
Coupling simulation of hydraulic fracture propagation with
modeling of accompanying seismicity (in real time, as well)
The work is in progress
Further results are to be presented next year at the
International Conference
Recent Advances in Numerical Simulation of Hydraulic Fracture
(Rzeszow, Poland, 14 16 July, 2014)

The colleagues are kindly invited


25

Thank you!

26

También podría gustarte