You are on page 1of 9

WATER RESOURCES RESEARCH, VOL. 17, NO.

1, PAGES 191-199, FEBRUARY 1981

Flow Through Fractures


C. E. NEUZIL AND JAMES V. TRACY

U.S. GeologicalSurvey,Reston,Virginia 22092

Flow throughfracturesis often idealizedas flow betweentwo parallelplates(plane Poiseuilleflow).


The openingor aperturebetweenparallel plates is unambiguousand its relation to flowrate is well
known.However,fracturesin rockhaveunevenwallsand a variableaperture.A modelfor flow in a frac-
tureis proposedwhereinthe fractureis represented by a setof parallelplateopeningswith differentaper-
tures.The modelleadsto a modifiedPoiseuilleequationfor flow which includesan aperturefrequency
distributionfor the fracture.Any arbitraryaperturedistributioncanbe used;in orderto simplifycompu-
tation and demonstrate the propertiesof the modela log normalform of distributionis assumed.Even
whenan analyticalform of the distributionis assumed,two parameters,ratherthan a singlevalue repre-
senting'aperturesize'are requiredto determineflowrate.Models of aperturechangefor a fractureun-
dergoingcompression (fracturewallsdeforming)and extension(fracturewallsseparating)are developed
whichconstrainthe additionalparameterand allow calculationof flowrateas a functionof meanaper-
ture. The theoreticalrelationshipsdevelopedbetweenmean apertureand flowrate can be usedto inter-
pret publishedlaboratorydata for singlefractures.

INTRODUCTION ception of how the flow changesunder theseconditions.Some


The presence of fractures significantly affects the per- publishedexperimentaldata for flow in compressing and ex-
meability of many rocks.Fracturepermeabilityis particularly tending
fractures
is available
[Sharp,1970;Iwai, 1976].and
importantin rockssuchas granite and shalewhich are hy- can be interpreted accordingto the proposedmodel.
draulically 'tight' when intact. In theserocksflow through the
ANALYSIS
intergranular pores is often insignificantcompared to flow
through fractures. Flowbetween
uneven
walls. Complete
definition
ofthe•-
The ability of a fracture to transmit water dependsprimar- croscopicflow field betweenthe nonparallelwallsof a fracture
ily on the size of the opening,or the fracture aperture.Many presentsthe same difficulties as the analogousproblem for
workers [Baker, 1955; Huitt, 1956; Snow, 1968, 1970; Gale, flow through pores.The geometryof the openingand the re-
1977]have used flow betweensmoothparallel platesas a suitingthree dimensionalflow field are too complexfor such
model for flow in fractures. The solution to the Navier-Stokes an approachto be practical. However, the propertiesof frac-
equationfor flow betweenparallelplates,known asplane Poi- ture flow suggestthat a simplified fracture model can be of use
seuille flow, has been known to fluid mechanicians since the forcalculating
fluxthrough
fracture
s. •'•i'
nineteenth century. Experimentalevidence
obtainedbyMaini[1971
] shows
that
Plane Poiseui!leflow accuratelydescribeslaminar flow be- flow in fr.acturesoccursover paths which are anastomOS'mg
tween two parallel plates. It is the basis of the Hele-Shaw andirregular,
notstraight.
Mainiusedlucitecopies
of a frac-
analogyto porousmedia. However,fracturesin rock have un- turedrockblockto construct
an opening
whichduplicated the
even walls and a variable aperture. In fact, flow in a fracture fracture, and in which the flow behavior could be observed.
under compressivestressis possibleonly becausethe fracture Dye wasinjectedat the upstreamend of the opening.Figure 1
walls are not parallel plates; the walls must be in contact at reproducesa portion of two dyestreaksobtained by Maini in
some locations and not at others. An additional problem this manner.
ariseswhen the parallel plate analogy is applied to fractures; These experiments demonstrate that flow tends to deflect
one must define a single value representingthe 'aperture' in around narrow portions of the fracture. This resultsin the re-
an openingin which the actual aperturesizevarieswith loca- duction of aperture variability along the flowpath. Con-
tion. sequently,the collection of flowpaths in a fracture conduct
In this paper we presentan alternativemodel for flow in a flow in a manner analogousto several attached conduits of
fracture. The flow is envisionedas occurringin a set of paral- differentsizes.Work by Iwai [1976,p. 69] suggests that if the
lel plate openings,generallywith differentapertures.Thus the aperturevariation along a flowpath is not large the flow is well
model accountsfor aperture variation normal to the flow but approximated by parallel plate flow in an opening equal to
assumesthat the parallel plate analogy can be applied in the the mean flowpath aperture.
direction of flow. The new model leads to a modified Poi- Utilizing this observation an idealized fracture model can
seui!leequation. be constructedin which the flowpathsare representedby an
Two important mechanismscan mechanicallychange the opening which varies continuously normal to the flow but
aperturein a fracture;deformationof the fracturewallsdueto with constantaperturein the flow direction.The aperture var-
compressive stress(compression)and movementof the frac- iation normal to the flow directionis representedby the aper-
ture walls apart with no deformation (extension).The new ture distribution in the fracture.
model, combinedwith modelsfor aperturechangein fractures Such an idealized fracture is shownin Figure 2. In the de-
being compressedor extended,providesan alternative con- velopmentthat follows, this simplified fracture model is used
as a basisfor developinga modified Poiseuilleequation for
This
paper
isnotsubject
toU.S.copyright.
Published
in1981
by fracture
flow.
theAmerican
Geophysical
Union. The fracturemodelin Figure2 is initially chscretized
into N
Paper number 80W 10• ' 191
192 NEUZIL AND TRACY: FRACTURE FLOW

Ab by
f•,' = f•,/N
and (3) becomes

(4)

We can definea normalizedcumulativefrequencydistribu-


tion function F suchthat F(b•,) representsthe number of seg-
mentswith aperturesfalling in the kth and previousintervals.
From the definition of F it is seen that

f•,'= f(bk)- f(bk_O (5)


Fig. 1. Dyestreaksin a fracture from photographby Maini [1971].
Combining (4) and (5) gives

segments,each of length I normal to flow and each, generally, o• ¾ • f(bk ) _ f(bk_l)


with a different aperture than its neighbors.The total length Q/= Y,LJ1-•b•, Ab
k=l Ab (6)
of the openingnormal to flow is L.
Taking the limit as Ab, the discretizationof the aperturedo-
If the differencein aperturebetweenadjoining segmentsis
maM, shrinksto zero and N, the total number of segments,
not large, then the fluid to fluid shearbetweenadjoining seg-
approachesinfinity,
ments and the drag againstthe 'risers'connectingthe walls of
adjoining segmentscan be ignored. The flux through each
segmentis then given by [Lamb, 1945] limQf= LJ1--•
atto ¾b3-•-
dFdb (7)
N--•oo

Q--1-•lb3j (1) By analogy with the cumulative distribution of probability


theory the derivativeof F is simply the normalizedfrequency
distribution.Thus (7) can be written
where

gradient of piezometrichead (dimensionless); ¾ baf(b)


db (8)
segmentlength normal to flow, L;
segmentaperture, L; wheref(b) is the normalizedaperturefrequencydistribution.
dynamicviscosity,M/LT; The integral in (8) representsthe third moment about zero of
specificweightof water,M/L 2T'. the aperturefrequencydistribution.Equation (8), which will
be referred to as the modified Poiseuilleequation, combines
Equation(1) relatesthe flux throughthe segmentto its aper-
the plane Poiseuillecubic law with a descriptionof the rela-
ture and is sometimescalled the 'cubic law.' In this paper it is
also referred to as the simple Poiseui!!eequation.
tive frequencyof differentapertures, or sizesof conduits,in
the fracture.
Using (1), the total flow through the discretizedopening
A derivation analogousto that for the modified Poiseuille
idealized in Figure 2 can be written as
equationwill showthat for the fracturemodel in Figure 2 the
N
cross-sectionalareaof the openingnormalto flow is givenby
Qf=•J ¾ b,3l
,=, 1--• (2)
•l/= L bf(b) db (9)
Equation (2) sumsthe flow in all N segmentsof the opening.
An alternative approachwhich resultsin a more useful formu- and the mean aperture is thus
lation is to discretizethe aperture domain by incrementsAb.
The flow throughthe discretizedopeningcan then be approxi-
mated as
1•-- bf(b)db (10)

Useof themeanaperture/;isanunambiguous
wayto charac-
Qf= k=l
y, J 1--•
¾ O•,3f
•,l (3) terize the size of a fracture opening.It avoidsthe ambiguity
inherent in other measureswhich have been used,suchas dis-
where bk is the centralvalue of the kth interval Ab (L) and fk
is the number out of N total segmentswith apertureswhich
fall in the kth interval Ab (dimensionless).
The summationto infinity is utili•.ed becauseone does not
know a priori the range of aperture sizesin the general case.
Each fracture, of course, has a maximum aperture. The ab-
senceof larger aperturesis accomodatedby assigningzero fre-
quencyabovethat size. This analysisappliesonly if flow re-
mains laminar in the largest opening. The flow calculated
from (3) approachesthat from (2) as Ab approacheszero. Fig. 2. Conceptualfracture model used to developmodified Poi-
A normalized frequencycan be definedfor the kth interval seuille equation.
NEUZIL AND TRACY: FRACTURE FLOW 193

the distribution describing aperture frequency is crudely


Gaussianor 'bell shaped.'The nature of the problem dictates
Lb•b ,,z,
that for all b < 0 and for all b > b....
f- re) = 0
b-2E b
APERTURE where bmaxis the largestgap in the fractureor the portion of
the fracture which we are considering.
The exactform of the aperturefrequencydistributionf(b)
is difficult to determine even in the laboratory; little of this
type of data on rock fracturesis available.It may not be pos-
sible to characterize the distribution in subsurface fractures.

b In addition, the aperture frequency distribution dependson


b. APERTURE fracture orientation relative to the rock fabric, the mode of
fracturing,the stresshistory of the fracture and the size scale
being considered.It is also clear that the aperture frequency
L distribution must changespatially in singlein situ fracturesas
a result of spatial changesin stress,particularly with depth.
In the following developmentwe considera fractureor por-
b tion of a fracture which is spatially homogeneousin f(b) in
APERTURE
the directionof flow. To generalizethe development,and sim-
Fig. 3. Fracturemodelof Witherspoon et al. [1979b]for fracture plify computation,we seekan analytical function to describe
closing understresswiththeinferredbimodalaperturefrequency dis- the aperture frequencydistributionof this hypotheticalfrac-
tributions;(a) asperitiesnot in contact,(b) asperitiesin contact,and ture. The form of the function must be representativeof the
(c) elasticdeformationof asperities. form of the aperturefrequencydistributionin fractures.Func-
tions describingprobability density in probability theory are
tancebetweenasperites[Louis, 1969]or an equivalentparallel appropriatesincef(b) is then normalizedand integratesto
plate aperture [Sharp, 1970]. one. It should be pointed out that any such function used in
From (8) and (9) the averageflow velocity is simply this context is devoid of probabilisticmeaning and is simply
used to describea specifiedset of apertureswith frequencies
that are known or assumed.
b'f (b) db
O'•=J 'Y (11) The available evidencesuggeststhat the log normal func-
= of(o)
ao tion is appropriate for this purpose. Snow [1970] notes that
many argumentspoint to the suitabilityof the log normal dis-
tributionfor describingthe frequencyof fractureaperture
s in
The proposedmodel and the modifiedPoiseuilleequation rock. He studiedjoints in outcroppinggranite and found that
which it suggestsinclude other fracture models as special mean aperturesmeasuredin differentjoints were distributed
cases.Plane Poiseuilleflow is describedwhen the aperturefre-
approximately log normafly. He suggeststhat apertures are
quencydistributionmomentsaboutthe mean are zero. In that similarly distributed at depth. Sharp [1970] measured aper-
case,f(b) is a spikefunctionwith the singlevalue representing turesin a singlefracturein the laboratory.His data showa log
the parallel plate aperture. normal form with a truncated tail due to the small area avail-
The fact that flow can occurin fracturesunder compressive
able for sampling.
stressemphasizesthe simplificationsinherentin the parallel The log normal function violatesone requirementfor the
plate analogy.The flow is possibleonly becausethe fracture aperture frequency distribution. Specifically, the function
walls are not flat and only a portion of the wall area is in does not becomezero for any finite b.... However, with ap-
physicalcontact.
propriate parametersthe log normal tail decreasesrapidly
Witherspoon et al. [1979b]recognizedthisfact and proposed
enoughthat the existenceof a maximum apertureis approxi-
a simplemodel to accountfor it. Their model, shownin Fig-
mated.The rangeof log normal parameterswhich is appropri-
ure 3, consistsof flat walls with protruding flat-topped aspe- ate in this sense is discussed in a later section.
rities of height •. The resultingopeningconsistsof two aper-
Flow in fractures with log normally distributedaperture.
tures;b betweenoppositewalls and b - 2• betweenopposite
The two-parameter log normal distribution [Aitchisonand
asperities.When the walls are broughtinto contact,the aspe-
Brown, 1957] is defined by
rities 'mate' and form fixed areas of contact. As compressive
stressis applied, the asperitiesdeform elastically.The aper- f(x)--[x(2•ra2)'/:] -' exp {-(2o•) -' (In x- •):} (12)
ture frequencydistributionswe haveinferredfor thismodelat
eachstageare alsoshownin Figure 3. If f(b) is taken as this In the theoreticaldevelopmentthat follows,we assumethat
inferred bimodal distribution,the model of Witherspoonet al. the existingaperturefrequencydistributionis representedby
[1979b] is a specialcaseof the proposedmodel. a log normal densityfunctionwith the appropriateparame-
The distributionof aperturesin fractures. Selectionof ap- ters.The resultswill qualitativelyapply in fractureswith any
propriatefrequencydistributions to describenaturalphenom- similar aperturefrequencydistribution.
ena is empiricaland basedon observedfrequencydistribu- Let the aperture frequencydistributionf(b) be approxi-
tions. It is likely that the aperture size in fracturesvaries matedby a log normal densityfunctionwith parameters•' and
around a modal value and aperturesincreasinglylarger or o. The third moment about zero of a log normal densityfunc-
smallerare increasinglyrare. Further, one would expectthat tion for apertureis [Aitchisonand Brown, 1957]
194 NEUZIL AND TRACY: FRACTURE FLOW

FLOW UNDER DIFFERENT CONDITIONS


OF STRESS

Our purposein this sectionis to postulatethe behavior of


fractures,or more specifically,fracture aperturedistributions,
as they changewhen the fracture is subjectedto different con-
ditionsof stress.Combinedwith the modifiedPoiseuilleequa-
tion, it is then possibleto relate changesin flowrateto changes
in the mean aperture.
It will be recalledthat even when a log normalform of f(b)
is assumed,two parametersare required to specifythe distri-
bution. This means that in addition to specifyingthe mean
Fig. 4. Model of aperturechangedue to compressional deformation
of fracture walls. aperture,a secondconstraintis requiredin order to calculate
the flux. This additional constraintis providedby the models
of aperturechangeunder different conditionsof stress.
Effects of deformation(compression). A simple model of
fo
• bar(b)db
--[/]a
exp
(3g
+90:/2) (13) aperture change in a fracture under changing compressive
stressis depictedschematicallyin Figure 4. Accordingto the
where [/] is the unit length. In this casethe modified Poiseuille model, a changein stressnormal to the fracturewill resultin a
equation is written reversiblechangein aperture at any locationwhich is propor-
tional to the original aperture at that location. This is a sim-
¾ plification of a complexprocess.For example,it doesnot ac-
Q/--1-•p
LJ[I]3
exp(3•'
+90:/2) (14) count for the crushing of asperitiesor the creation of new
points of contact,both of which can occur in rock fractures
The cross-sectional area normal to flow is given by [Goodman,1974].However, the model shouldreasonablyap-
proximate changes in aperture between points of contact,
Af-- L [/] exp (g + 0:/2) (15)
once established.It should be applied only insofar as the de-
and the mean aperture is formation is elastic.
We assumea fracture in which the model is applicableover
/; -- [/] exp(•' + 02/2) (16) a rangeof stresses which includesstresses ,4 and B. Thus there
Thus the average fluid velocity is exists a factor C depending on the stress
increment, which re-
lates the aperture under stresses,4 and B at any location ac-
cording to
7 j[/]2exp(2•'+ 402)
Vavg--
12/• (17)
bB----CbA (18)
Equation(14) reducesto the simplePoiseuilleequationwhen If the aperturesunder the original stresshad a log normal fre-
o = 0'with
quencydistributionand (18) applies,it followsfrom the log
b = [/] exp (•) normal function that the new apertureswill also have a log
normal frequencydistribution. From (18) and a development
Accordingto (14), flow in the fracturecan be predictedif the which parallelsthe derivationof the modifiedPoiseuilleequa-
distributionparameters•' and o are known. tion it is seen that
Workers studyingfracture flow [Louis, 1969;Sharp, 1970;
Iwai, 1976;Gale, 1977; Witherspoon et aL, 1979b,and others]
havegenerallyconsidered the relationbetweenflowrateand a fo
øø
bBf(b•)
db•
=Cfoøø
bAf(b•)
db• (19)
singledescriptionof the fractureaperture,suchasmean aper- and
ture or amount of closureunder stress.However, accordingto
(14),evenwhena lognormalformof frequency
distribution
is
assumed,flux dependson two parameterswhich describethe bs3f(bs)dbB= C3 b•3f(b•) dbA (20)
log normaldistribution.The parameters•' and o can be de-
fined by the mean and the secondmomentaboutthe mean of Sincethe frequencydistributionsf are log normal, (19) and
the distribution. To define the relation between flux and mean the definition of the mean (16) lead to
aperture,it is necessary
to describethe aperturedistribution
•'• = In C + In/;• - o•2/2 (21)
by specifyingbothparameters, ratherthan describingonly an
'aperturesize.' Substituting(13) into (20) gives
Becauseof the proportionaiRybetweenflow and the cube
3rs + 90•/2 ---In [C a exp (3f• + 90•'-/2)] (22)
of aperture,flow is dominatedby the largestaperturesand
thus, in this analysis,dependslargely on the tail of the fre- Combining(21) and (22) givesthe resultthat
quencydistribution.The value of the meanapertureis more
dependenton the form of the distribution,includingthe small In [exp(3• + 90A'/2)]
3
aperturerange.This is a potentialsourceof error when sev-
eral different distributions have similar tails. Although they or
wouldyieldsimilarfluxes,the meanapertures
couMbe differ-
ent. os: = o•: (23)
NEUZIL AND TRACY: FRACTURE FLOW 195

10-2
.=

u• 10-3--

ac10-4 •"•
_

10-6
10-12 10-11 10-10 10-9
--- :;---,, ,, ,, ,, ,
10-e 10-7 10-6 10-5 10-4 10-3 10-2
Qf/JL, IN METERS2
SEC-1
Fig. 5a. The modifiedPoiscuilleequationfor a fractureundergoingcompression. Theoreticalchangesin mean aperture
and flux will follow a curve parallel to the simple Poiseuilleequation.

For a log normal distribution,the secondmoment about the creasingcompressionsuch that they follow a line parallel to
mean is defined by [Aitchisonand Brown, 1957] the simplePoiseuillecurve.That is, they will follow one of the
familyof constant
J• curvesshownin Figure5a.
M2 -- r/2/•
• (24a)
Examples of log normal distributionsrelated by (26) are
where shownin Figure 5b. These distributionsdemonstratethe theo-
retical changein aperturefrequencyin a fractureundergoing
,/• -- [exp(a2) - 1] (24b) compression.
Since,/• dependsonly on a2,whichis constant,we can define As J• increases,
the second
moment,or varianceof the log
a constant Ka such that normal distribution increases relative to the mean. When J• >
1.0, the tail of the log normal distribution starts to become
n'-= • (25) drawn out. For values much larger than 1.0 the log normal
It then follows from (24a) that distribution is probably no longer a good approximation of
any fracture aperture frequency distribution. Thus in Figure
(M2)1/2--•/• (26) 5a curvesfor • > 1.0are dashed.Aperturefrequency
distri-
butionswhich lead to displacementfrom the simple Poiseuille
Equation (26), togetherwith the definition of mean aperture
curve like that of the dashedcurvesmay exist,but would not
(16) and the modified Poiseuilleequation,definesa unique
be well approximatedby a log normal distribution.
relationship betweenmeanaperture/; and the fractureflux
Equations (19) and (20) are valid for any aperture fre-
Qf. They describea family of curvesparallel to the simple
Poiseuille curve for different values of the constant •. These quency distributionf(b). Thus if the compression
model (18)
is valid, mean aperture versus flow will plot parallel to the
curves,which are presentedin Figure 5a, expressthe theoreti-
simple Poiseuillecurve even if the aperturefrequencydistri-
cal changein mean apertureand flow in a fracture with a log
bution is not approximatelylog normal in form.
normal aperturedistributionundergoingcompression.For ex-
Effectsof extension. A model of aperture changein which
ample, a fracture under no compressivestress,but with the
the fracture walls are moved apart and no deformationof the
walls in contact, will have a mean aperture and flow which
walls takes place is shown schematicallyin Figure 6. As the
theoreticallyplots somewhereto the right of the simple Poi-
seuille equation. This initial point correspondsto a value for
R whichspecifies
theaperture
frequency
distribution
(hereas-
sumedapproximatelylog normal). If the model applies,the
mean aperture and flux will theoreticallychange under in-

1.1 I I I I I I I I I I I
1.o -
u
z
'" 0.8 •,=0.s

'" 0.6

-> 0.4
"' 0.2

0.0
0 I 2 3 4 5 6 7 8 9 10 11
APERTURE,
IN MICRONS
Fig. Sb. Theoreticalchangesin a log normal distributiondescrib-
ing aperture frequency in a fracture undergoing compression.As Fig. 6. Modelof aperturechangedueto extension
of fracture.Aper-
compresslye
stress
increases,
meanaperture/•decreases. turesat a and b changeby the sameamount.
196 NEUZIL AND TRACY: FRACTURE FLOW

10-••
I [ I I •I •I I I I I I I •I • 1
• 10-3-- _
u.i

u.• O'
4
LU

10-12 10-1• 10-•o 10-9 10-8 10-7 10-6 10-$ 10-4 10-3 10•
Qf/JL, IN METERS
2 SEC-•

Fig. 7a. The modifiedPoiseuilleequationfor a fracturewith meanaperture10-s m whenfracturewallsfirstcomeinto


contact
(0 -- 0 at • -- 10-s).Portion
ofcurveabove/•- 10-s m isbehavior
underextension,
remaining
portionis compres-
sion curve.Theoreticalsequenceof closingextendedfracture (curveAB) and compressing
fracture after walls are in con-
tact (curve BC) are shown.

fracture aperture increases,the unevennessof the walls be- simplePoiseuillecurvein Figure7a,for example, at pointB(/•
comeslessrelativeto the opening.In effect,if the aperturehas -- 10-5 m, • -- 1.0).As the wallsof the fractureare drawn
somefrequencydistribution,the geometricform of the distri- apart, the observedflux and mean aperture will theoretically
bution doesnot changewhen the mean increases.This can be follow curve BA. Reversing the changes, as the walls are
visualizedin the form of a histogram.An increasein the aper- moved together, curve AB will be followed, until, at B, the
ture shiftseachbar to the right by the sameamount;the mean walls come in contact. Application of compressivestresswill
increaseswhile the shapeof the distributionremainsexactly result in continuedchangefollowing curve BC.
the sameand the entire distributionis translatedto the right. As noted previously,the assumptionof log normal form for
The extensionmodel can be applied analyticallyto a log aperturefrequency
distribution
becomes
poorwhen• > 1.0.
normal aperturefrequencydistributionby introducinga third The extended,flat tails on log normal distributionswith larger
parameter. The new parameter, 0, representsthe amount of R valuesaccountfor the initiallysmallincrease
in flowwith
extension. In this case the modified Poiseuille equation is extension which is seenin the curves for • -- 2.0 and • = 4.0.
given by The curvesshownin Figure 7a apply when the aperturefre-
quencydistributionis log normal, but any aperturefrequency

e,= (t,+ 0)3f(t,)


,to distribution will yield silmilar curves. This is becauseas the
mean aperturein an extendingfracture increases,the aperture
variation becomessmaller relative to the opening. In effect,
¾
the fracture becomesincreasinglylike a parallel plate open-
---LJIm•[[/]g
exp(3•'
+ 902/2)
+ 30[/]
2exp(2•'
+ 202) ing. When this happensthe flow will approachthe simplePoi-
seuille curve.
+ 3t•[/] exp(g + o•/2) + ½] (27) Figure 7b shows several log normal distributions which
have the same mean but different values of R and hence dif-
and the mean aperture is ferent extension curves.
Effectsof shear. It is difficult to characterizethe effect of
• - [/] exp(•' + 02/2)+ 0 (28)
shearon the aperturefrequencydistribution in a fracture.The
In order to obtain the plot of mean aperture versusflux in specificchangeswill be governedby many factorswhich can-
this caseit is necessary to have sufficientconstraintsto specify not be determined from the aperture frequency distribution.
three parameters.To do this, we must specifythe conditions An important factor is the angle at which individual segments
(mean apertureand flux) at zero extension,i.e., when the frac- of the crack depart from the overall fracturetrend. Stressnor-
ture walls are first in contact.This is equivalentto specifying mal to the fracture at the time shearingtakes place is also im-
• in (26)whichdetermines o and•'.Thethirdparameter 0, as portant. It will partly determine whether individual asperities
noted, is simply the extension.
Curvesof flux asa functionof meanaperturefor 0 -- 0 at/•
= 10-5 m were calculatedfrom (27) and (28) and are pre- • 020
l
sentedin Figure 7a. The shapeof the curvedependsonly on z•016•x 1
the parametero• or, as definedin (26), the constanti•. The
curvescan be appliedto any initial meanaperture,that is, any
/• when0 -- 0, by simplytranslating
themparallelto the • 0041- œ'• •,•/
simple Poiseuillecurve.
An examplewill serveto demonstratethe interpretationof APErTUrE, IN MICRONS
the curves.Let us supposethat we have a fracture with walls
in contact but under no compressivestress.Its flux and mean
aperture will theoreticallyplot somewhereto the right of the will • • di•orom oxto.sio. c.•o •s show. i. Fi•.ro 7..
NEUZIL AND TRACY: FRACTURE FLOW 197

[1970] can be interpreted usingthe model proposedin this pa-


per.
Flow in a tensionfracture. Iwai [1976] conductedexperi-
APERTURE ments on flow through tension fractures in granite, marble,
and basalt. The fractures in the rocks were induced using a
'Brazilian' loading method. After splitting, the block halves
were mated with as little shear as possible.Flow during com-
pressionand extensionof the fracturewas measured.
Direct accuratemeasurementof aperturesin a compressing
fracture presentsmany technicaldifficulties.Iwai [1976] used
APERTURE a measure of fracture aperture, which he called ba, obtained
b.
indirectly by measuringstrain or closureacrossthe fracture
Fig. 8. Comparisonbetweenunshearedtensionfractureand 'un- under applied stress.His flow term contains a correction,
matched'fracture;(a) unshearedtensionfracture and diagrammatic causingit to becomezero at an arbitrary high stress.
aperturefrequencydistributionwhichwould be approximatedby a Whitherspoonet al. [1979b] reanalyzed Iwat's [1976] data.
log normaldistributionwith a relativelysmallg value,(b) fracture
with unmatchedwalls and diagrammaticaperturefrequencydistribu- They usedthe measuredflow and correctedbaby adding a re-
tion which would be approximatedby a log normal distributionwith sidual value to obtain what they called the 'true aperture.'
a largerR value. Their new measureof fracture aperture,here called b', repre-
sents the remaining portion of maximum strain, or closure,
acrossthe fracture; we assumeit is linearly related to the
are crushed or contribute to dilatation of the fracture walls. mean aperture:
The complexnature of shearmovementin fracturesis well
summarizedby Goodman[1974].Becauseof the complexities, =
no specificmodel of aperturechangedue to shearingis pro- The proportionality constant• cannot be determined but is
posed. probablycloseto 1.0 becauseof the narrow rangeof aperture
Significance
of theparameterR. Strictlyspeaking,thepa- sizeslikely to be presentin a tensionfracture.
rameter• asderivedappliesto a hypothetical
lognormaldis- Iwat's [1976] data, using the analysisof Witherspoonet al.
tribution.However,we canrelatethe• valueof a lognormal [1979b], are presentedin Figure 9. It can be seenthat the data
distribution to the actual distributionwhich it approximates. generally are well representedby the simple Poiseui!!eequa-
In thissense we mayspeakof the• valueof an aperturedis- tion for both compressionand extension.This is the behavior
tribution.
predictedby the theoreticalcurvesfor compressionand exten-
The second moment about the mean is a measure of the
'spread'or 'flatness'of the distribution;this characteristicof
the distribution is reflected in the value of •. As noted ear-
lier, an extreme example of this is a parallel plate opening,in
whichR -- 0 andthedistributiondegenerates
to a spikerepre- _Granite
(Straight
Flow)
sentingthe parallel plate aperture.
Somegeneralizations
canbe maderegarding
theR values • 10-$
which may be expectedfor different types of fractures.In a z

tensionfracture with no shearthe walls are uneven but paral- 10-3 •


-6
lel [Louis, 1969].As a result,there is little variation in aper- ß .

ture. This type of opening,which may be thoughtof as a 'bent 10-


4Z
parallelplate' openingis shownin Figure 8a. Its distribution
will be quite peaked,as shownin Figure 8a, and will have a 10-5,,,
small R value. z

If a fracture has experiencedshear or if the walls are 'un- ,,, 10-3 lO


-6•
matched' for any other reason,a relatively larger range of
aperturesizeswill be present,as shownin Figure 8b. The
• 10-4
LBasalt
(Radial
Flow) ..

aperture
distribution
will belesspeaked
andhavea larger•
value. ,,, 0-5
<

COMPARISON OF RESULTS
•10-6 10-3
Flow in fractures under different conditions of stress has
t Marble
(Radial
Flow)
been investigatedexperimentallyby Sharp [1970] and Iwai
[1976].Iwai [1976] studiedradial and straightflow through
singletensionfracturesin severalrock typesas the fractures
underwentcompression and extension.Sharp[1970]measured
flow in a singlenatural fractureundergoingextension. 10--•
10-
lO
10-
9
10-8
10-
7
10-
6
10-
5
10-
4
Iwai [1976] has interpretedhis data usinga simpleparallel Qf/JL,INMETERS
2 SEC1
plate analogy. Witherspoon et al. [1979b]reanalyzedIwai's
Fig. 9. Fractureflow data of Iwai [1976]presentedusinganalysis
data and formulated a bimodal distributionmodel (see Figure of Witherspoonet al. [1979b]. Solid lines representthe simple Poi-
3), whichwasdiscussed earlier,to explaintheir results. seuille equation; bracketsindicate approximaterange over which
The experimentalresultsof both Iwai [1976] and Sharp fractures were extended.
198 NEUZIL AND TRACY: FRACTURE FLOW

7
ture used for his experimentalwork. The histogramhe pres-
ents is a partial sampleof his many aperture measurements;
the largestapertureshownis lessthan 600 microns,and Sharp
statesthat the largestaperture measuredwas about 800 mi-
crons.The distribution,as suggestedby the histogram,is in-
complete.
Since the aperturedata are not complete,the comparison
o
o 1oo 200 300 400 500 600 700 800 between the theoreticalcurve and the experimentalflow data
APERTURE,IN 10-6 METERS is basically qualitative. However, it was found that good
Fig. 10. Aperture histogrammeasuredin laboratoryfracture by agreementcould be obtainedassuminga reasonableaperture
Sharp [1970] and distributionusedin modifiedPoiseuilleequation distributionbasedon the histogram.The histogramdata and
(dashedline).
the aperture distributionassumedfor the fracture are com-
paredin Figure 10.The calculatedflow and Sharp'sflow data
sionshownin Figures5a and7a for smallvaluesof R. As we are comparedin Figure 11. Althoughthere are only four data
have noted, unsheared tension fractures should be character- points,eachpoint is the averageof over a hundredmeasure-
izedby peakedaperture
frequency
distributions
withsmallR ments [Sharp, 1970].
values.
Flow in an 'unmatched'fracture. Sharp [1970] has con- DISCUSSION
ducted a seriesof laboratory experimentsin a singlenatural
fracture in porphyry in which the fracture walls were un- The modelsdevelopedand the laboratory data cited relate
matched,or nonparallel. to changesin the aperturedistributionand flux in a fracture
Initially, the fracture walls were seatedtogetherand then under two different external conditions. The behavior in
incrementlyopenedto changethe aperture.The mean frac- $harp's [1970] experiments was that of a tensile aperture
ture aperturein Sharp's[1970]experimentsrangedfrom 0.27 change.The walls were in contact only at the start and were
to 1.5 min. His data for small piezometricgradientsand mean then movedprogressively apart. Iwat*s[1976]experimentsbe-
aperturesbetween0.27 and 0.5 mm canbe approximatedby a gan with the fracturewalls in contactand subjectedthem to
squarelaw: progressively increasingcompressive stress.
Both types of change may occur with changesin depth
Q/ocLJb2
along in situ fractures.The work of Snow [1970] clearly dem-
As the mean aperture was increased, the exponent ap- onstratesthat mean fracture aperturegenerallydecreaseswith
proached3 and the flux becamecloseto the theoreticalPoi- depth.Sincethe stateof stressin the earth is usuallycompres-
seui!le solution. slye, the theoretical changes occurring under compression
Sharp[1970]notedthat the flow observedwith no extension would apply. However, tensile stressesand associatedfrac-
was larger than expectedbasedon the simplePoiseuillerela- tures may also occur, in which case the extension theory
tionship.He concludedthat the simpleparallel plate model would apply. The extensiontheory may also be applicableto
did not accuratelydescribehis results,statingthat 'the dis- fractureswhich are propped open.
crepancyis quitesignificant;it showsthat for roughfissures
it Effect of scale. The relative frequencyof flow channelsof
is not valid to apply a parallel plate law assuminga mean different sizesprobably has an approximatelylog normal or
opening derived from measurementson the rough fissure' similar form. Thus channels increasingly larger than the
[Sharp, 1970,p. 69]. modal size are increasingly rare. However, the dis-
Sharp[1970]extensivelysampledthe aperturesin the frac- proportionate contribution of large openingsis important

3x10-3

<

z Data
fromSharp
(1970) --

2x10-4t I , ,, ,I , , , I I I, ,I • •
10-4 10-3
Qf/JL,INMETERS2
SEC-1
Fig. 11. Experimentaldatafrom Sharp[1970]compared
to modifiedPoiseuilleequationfor extension
usingdistribution
shown in Figure 10.
NEUZIL AND TRACY: FRACTURE FLOW 199

when consideringthe effectof specimensize on flow through averageflowvelocityin fracture,L/T.


the fracture. specificweightof water,M/L2T •.
In general,if specimensizeis decreased,fewer of the larg- log normal distribution parameter,dimensionless.
est, least-frequentflow channelswill be included.Theoreti- [exp(oa) - 1], dimensionless.
cally, this is reflectedby truncationof the aperture frequency log normal distributionparameter;amount of exten-
distribution. If only the applied stressis measured,however, sion, L.
the differencewill not be apparent. Small specimensshould dynamicviscosity,M/LT.
generallyhave smallerfractureconductivitythan large speci- scalingfactor, dimensionless.
mens under the same stress. log normal distribution parameter,dimensionless.
A size effect of this type has been observedby Witherspoon
Subscripts
et al. [1979a].They attributeit in part to the variability of the
percentageof wall area in contactin smallerspecimens. Both A pertaining to distribution or stressA.
this and the effect describedabove may contribute to the ob- B pertainingto distributionor stressB.
served differences.
Conclusion. More work, particularlythe gatheringof data Acknowledgments.The analysis benefited from discussionsbe-
from careful experimentalinvestigations, is neededto provide tweenthe authorsand D. Goodmanof the U.S. GeologicalSurvey.
The authors also wish to thank John Bredehoeft and Paul Hsieh, also
a solid basis for fracture flow theory. It is hoped that this of the U.S. Geological Survey,for their helpful criticism.
studymay help providea rationalframeworkfor suchinvesti-
gations. REFERENCES
Of particularinterestis the closureof fracturesunder high
compressivestressand the ability of small openingsto trans- Aitchison, J., and J. A. C. Brown, The Lognormal Distribution,Uni-
versity Press,Cambridge, 1957.
mit water. Knowledgeof theseaspectsof the problem is nec- Baker, W. J., Flow in fissuredformations,in Proceedingsof the Fourth
essaryfor understandingthe extentof fracturepermeabilityat World PetroleumCongress,sect.II/E, pp. 379-393, Carlo Colombo
depth and its relationshipto the stateof stressin the earth. Publishers, Rome, 1955.
Gale, J. E., A numericalfield and laboratorystudyof flow in rocks
NOTATION with deformablefractures,$ci. $er. 72, Inland Waters Dir., Water
Resour. Branch, Ottawa, Ont., Canada, 1977,
Af fracturecross-sectional
area,L2. Goodman,R., The mechanicalpropertiesof joints, Proceedings of the
b fracture aperture, L. Third Congress,vol. I, part A, pp. 127-140, Int. Soc. for Rock
/• meanof aperturefrequency
distribution,
L. Mech., Denver, Colo., 1974.
Huitt, J. L., Fluid flow in simulatedfractures,AIChE J., 2, 259, 1956.
bi aperture in ith segmentof discretizedfracture model,
Iwai, K., Fundamentalstudiesof fluid flow througha singlefracture,
L.
Ph.D. thesis,208 pp., Univ. of Calif., Berkeley,1976.
b' indirect measureof fracture aperture, L. Lamb, H., Hydrodynamics,New York, 1945.
C proportionalityconstant,dimensionless. Louis, C., A study of groundwaterflow in jointed rock and its initu-
fk number of segmentswith aperturesin kth interval Ab, enceonthestability
of•rock
masses,
Res.Rep.10,90pp.,Imp.Coil.
dimensionless. of London Rock Mech., London, Sept. 1969.
Maini, T., In situ hydraulic parametersin jointed rock--Their mea-
fk' normalizedfrequency,dimensionless. surementand interpretation,Ph.D. thesis,Imp. Coll. of S½i.and
F(b) cumulativenormalizedfrequencydistributionfunction, Technol., London, 1971.
dimensionless. Sharp, J. C., Fluid flow throughfissuredmedia, Ph.D. thesis,Imp.
Coll. of Sci. and Technol., London, 1970.
f(b) normalized aperture frequencydistribution function,
dimensionless.
Snow,D. T., Rock fracturespacings,openings,and porosities,J. Soil
Mech. Found. Div. Amer. $oc. Civ. Eng., 94(SM1), 73-91, 1968.
J gradient of piezometrichead, dimensionless. Snow, D. T., The frequencyand aperturesof fracturesin rock, Int. J.
• proportionality
constant,
dimensionless. Rock Mech. Min. $ci., 7, 23-40, 1970.
I lengthof segmentnormal to flow in discretizedfracture Witherspoon,P. A., C. H. Amick, J. E. Gale, and K. Iwai, Observa-
tion of a potential size effect in experimentaldeterminationof the
model, L.
hydraulicpropertiesof fractures,Water Resour.Res.,15(5), 1142-
L--N(l), total lengthof the fracturenormalto flow, L. 1146, 1979a.
[/] unit length, L. Witherspoon,P. A., J. S. Y. Wang, K. Iwai, and J. Gale, Validity of
M2 secondmoment about the mean of the aperture fre- cubic law for fluid flow in a deformablerock fracture, Tech. Rep.
quencydistribution,L 2. 23, LawrenceBerkeley Lab., Univ. of Calif., Berkeley, 1979b.
N number of segmentsin the discretizedfracture model, (Received February 28, 1980;
dimensionless.
revisedJuly 29, 1980;
Qf fluxthroughfracture,L3/T. acceptedAugust 5, 1980.)

You might also like