in Hydraulic Fracturing
A. Settari, SPE, Simtech Consulting Services Ltd.
Abstract
This paper gives a new fonnulation of fluid loss in us to introduce the effects of variable pressure, fluid
hydraulic fracturing that is much more general than the viscosity, and different fluids contacting the wall in the
classical theory while retaining its simplicity. The model filtration process, in accordance with real conditions dur
allows many parameters to vary during filtration and ing the treatment. Comparison with the experimental
can, therefore, simulate nonlinear effects. data of McDaniel et al. 8 shows that the model is capable
The model has been validated against laboratory data of exhibiting nonlinear behavior matching the laboratory
for Newtonian fluids and crosslinked gels. The results data, which cannot be explained in tenns of the previous
show that the finite length of the core, viscosity simple theory. An important feature of the model is in
screenout, and shear sensitivity are important parameters corporation of the length of the core, which produces
that. can be represented by the model. The standard nonlinear behavior and can cause large errors in
analysis gives values of leakoff coefficients that will give calculating the true value of the leakoff coefficient when
incorrect, considerably higher leakoff when applied to the simple fonnulas are used.
field conditions. The new model retaihs the simplicity of the classical
leakoff theory, although it is more comprehensive and
Introduction potentially more accurate than the simulationtype
The estimate of fluid loss is an important part of a leakoff calculations, because it is fonnulated in tenns of
hydraulic fracturing treatment design. Although the con measurable variables.
trol of fluid loss has improved with the use of modem
fracturing fluids, the size of the generated fracture areas Leakoff Models vs. Simulation
increases with the size of a job. Consequently, fluid loss The flow of fracturing fluid into the reservoir can be
can be important even in lowpenneability reservoirs for described, at least in principle, by the equations of
large treatments. multiphase flow in porous media. It would thus seem
For design calculations, fluid loss has been treated in natural that an improved treatment of fluid loss would
the past by use of the simplified theory proposed by use numerical simulation of flow in the reservoir with the
Howard and Fast, 1 which expresses the rate of filtration properties and pressure at the wall (behind the filter
perpendicular to a fracture wall as a simple function of cake) as the boundary conditions. This approach, which
leakoff coefficients. we have taken in our current work,6 is indeed more
The advantage of this approach, besides its simplicity, general. It is not restricted by the assumption of one
is that it can be directly (if not always correctly) related dimensional (10) flow, and it includes the effects of
to experimental data on fluid filtration obtained in a relative penneability and capillary pressure and handles
laboratory. Apart from the correction of the derivation of changing conditions at the fracture face.
the combined leakoff coefficient, 2,3 very little has been However, the simulation approach also has problems.
done to improve the classical theory. First, the process of fracture fluid filtration is more com
With the recent development of a simulation approach plicated than the reservoir multiphase flow. The proper
to fracturing design,4,5 it has been recognized that fluid ties of the invading fluid are greatly different from the
loss can be computed directly by solving the basic reservoir fluid and are changing with time because of
multiphase flow equations in porous media. Such an ap breakers, temperature changes, and mixing. The fluid
proach is more general and does not have many of the can be miscible with one of the resident fluids. The prop
assumptions that limit the chtssical theory. 6 However, er fonnulation would require solution of threephase
the computational cost is much higher and the data re flow (one phase being the fracture fluid) with relative
quired to describe the process are difficult to measure. penneability, capillary pressure, and viscosities chang
This paper presents a generalization of the classical ap ing with time. Even though such a fonnulation and solu
proach that includes the effect of several parameters that tion is possible, the multiphase data are almost impossi
are variable in the field. The mathematical fonnulation ble to obtain because of the nonlinearity and instability
includes the model of filtercake behavior developed by of the gels. Consequently, one must make simplifying
the author 6 and the results of the work of Biot et al. , 7 assumptions (e. g., the filtrate assumes the properties of
which improves the calculation of flow in the reservoir. the reservoir water).
The model is then fonnulated numerically, which allows On the numerical level, an extremely fine grid would
be required owing to usually very small penetration of
Copyright 1985 Society of Petroleum Engineers the fracture fluid. To avoid this, we have found it
AUGUST 1985 491
ILR' The fluid loss additives and/or highmolecular
weight polymers may form a layer of filter cake on the
fracture wall, as shown in Fig. 1. If the fluid pressure is
Pf and the original reservoir pressure P R , the total
pressure drop, tl.p=PfPR, is a result of the flow
resistance of the cake and pressure drop in the porous
p
medium.
The classical theory treats the problem as three
separate regions.
o dV w kw I1pw k wexl1pw
x U ==_._= ........... (1)
w dt ILf U w IL f V w
~I'
x and
n
one obtains
r.
uv=C)'Vt,wlthC v =
~kcf>APv
, ............ (5a)
2ttfilt , .
Pf
and
~
analytical solution to calculate the flow rate at the «......J
face. 9,10
and
::,..Pc
Ve =2Ce .Ji. .............................. (7)
x
It should be noted that to apply Eq. 5 or 7 one must
assume infinite medium. Also, the last condition for C e
cannot be physically satisfied and will be discussed later o
in relation to the work of Biot et al. 7
For calculations, one knows only Pi and P R and is in
terested in u w' Because of the incompressibility assump
tion, we have
The following mechanisms will modify T w for condi From the assumption of Darcy flow,
tions different from the test.
1. Cake erosion. This effect is well documented 2.11,12
fIL =ftJlftf·
and is measured by a minimum (steadystate) velocity,
Us, which corresponds to an equilibrium cake thickness.
These functions are currently used in the model, for lack
The value of Us obviously depends on the shear rate of of better data. Note that the filtration of the polymers
the flow along the fractllre face, but systematic data are should be a function of the permeability and pore struc
lacking in the literature. The data of McDaniel et al. 8 ture of the porous media as indicated by the theory of
can be interpreted as supporting this effect, as will be polymers 13 and laboratory evidence. 14 Therefore, the
discussed later. Equilibrium is reached when the volume C w data should be obtained with porous material
of filtrate reaches Vs =2C wlu s and the transmissibility representative of field conditions.
for V w > Vs is constant, which is obtained by using Vs Recent laboratory measurements 15 suggest that the
instead of V w in Eq. 12 .. filter cake is compressible at low pressure differentials
2. Cake compaction. In the early measurements it has and becomes incompressible at high D.p, contrary to the
been observed that filtration is approximately indepen older data.
dent of t:..p w until D.p w is decreased to 100 to 300 psia Eqs. 15 a~d 16 can be written in terms of a variable
[690 to 2068 kPa]. This can be explained by considering coefficient, C w , which by Eq. 11 is
kw in Eq. 1 a function of t:..pw such that kw lIt:..pw' In
general, we can write Cw=Twt:..pJt.
~
Z 0
0 z
...a: w
e:J w
a: II
  PRESENT MODEL
   STANDARD THEORY
0
...
...J
0
II ....,
I
II P  CASE A
v
L____ _
4
MIXING
VOLUME
..
I.
Fig. 3Representation of flow in the filtrate zone for the
general model.
2 4 II II 10 12
1l2
tim •. min
where Mmix is a mixingzone length (Fig. 3). In addi
tion, we will consider the practical case when the wall is Fig. 4Effect of step change in pressure (A) and of finite
contacted by several batches of fluids with distinct length of core (8) on leakoff.
viscosities /.t 1, /.t2 ... , as shown in Fig. 3, and arbitrary
variation of tlp v .
As derived in Appendix A, the filtration velocity ~an
then be expressed in terms of a variable coefficient, C v: If all the variables are constant, Eq. 18 will reduce to
the standard expression. However, with changing condi
tions, integration of Eq. 5a with instantaneous C v will
uv=Cvl.Jt=C v · (~)I.Jt, ............. (18) give erroneous answers. For example, the effect of a step
C avg change in tlp v and of the length of the core are shown as
Cases A and B in Fig. 4. The core and fluid data used
where C v is the instantaneous coefficient here are the same as for Test 3 discussed in the next sec
tion. Note particularly the large error that results if we
C =C .
 A [ 7r J'h . ............. (22) and
e e 3(Ic Tt:..Pve)
Un
w+1 =Tn+l
w....Apn+l
w =u ve n+1 / rt,
n+l =C ve" t ..... . (23b)
Use of Eg. 2? gives the same leakoff as Eq. 21 in the
lim!! for Ce/C v >0. In addition, we have found that use where T!+l is given by Eq. 16 and
of C e in Eq. lOa approximates very accurately the more
complicated formula (Eq. 21) in the range of practical
2C~+lC~+1
data. C~+l = _________ _
The effect of variable pressure is not as easy to correct. Cv+v'C~+4C~ ............ (24)
However, this is important only if Ce controls the filtra
tion (C e ~ Cv' C w), which is rarely the case in practice.  
In any case, our calculations presented in the following In Eq. 24, C v is given by Eq. 18 and C e by Eq. 22,
section, as well as laboratory experience, suggest that t:..p with the overall pressure difference t:..p~+l instead of
will stabilize under constant conditions. t:..p v and t:..p e. Because the terms in Eq. 23b depend on
the unknown boundary pressure, Pw, and volumes,
Limitations of the Model V n + 1 (through the constraints on V wand V v ), the solu
The model just described is limited by the assumptions tion must be obtained iteratively.
remaining in its formulation. One of the more important 3. Integrate the loss volume,
of these is the assumption of pistonlike displacement in
the invaded zone, which replaces the more realistic
model of twophase Darcy flow with capillary and
relative permeability description. To assess the accuracy
V~ + 1 = V vn
.
+ i
t
t n+1
udt.. ................... (25)
timestep, where the singularity is integrated out, which and this trend can be seen in the later part of the curve.
yields VI =2C I ill. The iterative scheme is very depend The core length alone cannot explain the large initial rate
able, and the model has since been successfully in of leakoff, which could be produced in the model only
tegrated into a generalpurpose hydraulic fracturing by assuming that initially the effective viscosity is much
simulator. lower.
Once this is recognized, a reasonable agreement can
Testing of the Model be achieved by several assumptions. Fig. 6 shows three
In the first instance, the model was tested against the different simulations: (1) initial batch of a fluid with
results obtained previously by numerical simulation, 6 Pf= 10 cp [0.01 Pa' s] followed by the test fluid; (2)
with the results being very similar. Because of the simple larger batch of Pf=26 cp [0.026 Pa' s]; and (3) mixing
nature of the examples, these tests did not involve any of zone with Pmix = 1 cp [0.001 Pa' s], because the sample
the nonlinearities built into the model. was saturated initially by water, and V mix =2Vcore '
For comparison with actual laboratory data, we have The last of these gives slightly lower filtration, but the
used the data of McDaniel el al. 8 They measured the trend ofthe curve follows the data very well. The results
filtration of linear fluid as well as crosslinked polymer clearly indicate that some kind of mixing or dilution
under various shear rates and used cores of different mechanism is present, which can be included in the
lengths. We have simulated two tests with Newtonian leakoff model in a very simple fashion.
fluid (glycerol) and one test with crosslinked natural
polymer. The pertinent data are summarized in Table 1; Test 3. The filtration is controlled in this case by the
the experimental results are those of Figs. 2 and 3 of filter cake and less so by the core, and there is a marked
Ref. 8. dependence on shear rate.
To begin with, the data for zero shear are linear
Tests 1 and 2. Because glycerol is a Newtonian fluid in .Jt,
and does not form a cake, one would expect the results to which agrees with C w theory with Us =0 (no cake ero
follow the classical theory. This is observed for Test 1 at sion). To match the test, it was necessary to assume that
ilp=I,OOO psia [6.89 MPa], which is matched by the the filtrate viscosity was completely screened out and
model with C w =0 as shown in Fig. 6. The fluid front is equal to water viscosity. We have learned subsequent
still within the core at the end of the test and the behavior ly* that the same conclusion has been reached experi
is linear in .Jt. mentally. The match gave a value of C w = 0.002 ftf
The second test was conducted with a V2in. [1cm] ..fmin, [0.00001 mf../s], which was then kept for all
core, which gives a PV, V core' of about 2.3 mL. The runs with shear.
leakoff should therefore be linear with I for V> V core, "A. Deyserkar, private communication. Dresser Titan, (Houston), 1982.
,>
/2
7.5 TEST 3
60
DATA
10
B 10 12 14 16 18 10 12 14 16 18
time,mtn'l2 1/2
ttme. mln
20
10
• 10 12 14 18 18
The behavior with shear is physically much more com tim.:min'l'l
TEST 3
150
I'f
  VARIATION OF
I
   VARIATION OF u. , §
I ~
I
1 ~.
100
              1~3 J 9
I ~
 u FLUID J I FLUID.
2000 I .02 :I
, _____________
i
41
I
r

~
~
~
50 I z
'" ~ I "'Ci:I
~ 0 SHEAR,Me"'
1
II'\~."''':::.::::::::=.::=.:,,
o 4 • . ~
ti ... ,M'fl
W
12 M
'. .. • •
Fig. 9Pressures behind cake for the two methods of mat Fig.10Example of calculated fluid loss for a multifluid
ching the f,ltration of crosslinked gel. treatment.
F I =RI(l +R) (Eq. lOb), This is an interesting result because it shows clearly the
effect of including the moving boundary between the
F 2 =2RI(I+v'1+4R 2 ) (Eq. lOa), regions. Eq. B2 shows that the error is very small if the
compressibility is small, and grows with increasing P.
~=1/2 [PI+v'(Pl)2 + (47r/3) R2 ], Because Eq. B2 gives us a correction for R+O, and
Biot's formula has the same limit for large R, one would
and expect that use of C c in Eq. lOa would give a good ap
proximation to Eq. 19. This is indeed true in the en
F3 =v'~/(l +~) (Biot et al. Eq. 21). tire range of Fig. 5 as long as P< 1, as can be verifie\~
by calculating "corrected" F 2 by substituting
The plot of these three functions is shown in Fig. 5. It [7r/3(IP)] '12 • R for R in the formula.
is apparent that Eq. lOa gives higher values of leakoff
than Eq. lOb, with the maximum difference of23% at SI Metric Conversion Factors
• CclC v 1. Biot's formula depends on the compressibili cu in. X 1.638 706 E+Ol
ty parameter P. In the limit for P=O, the curve F3 is psi x 6.894 757 E+OO
practically identical with F2 and Biot's method gives the
same result as Eq. lOa. Only for fluids with large com SPEJ
pressibility and for Cc < Cv are there large differences. Original manuscript received in the Society of Petroleum Engineers office Feb. 28,
1983. Paper accepted for publication July 19, 1984. Revised manuscript received
This is, however, a physically unlikely situation, Sept. 17, 1984. Paper (SPE 11625) first presented at the 1983 SPE Low Permeability
because to have a large compressibility and small R re Symposium held in Denver March 1416.