Int. J. Rock Mech. Min. Sci. & Geomech. Abstr. Vol. 30, No. I, pp.

37-46, 1993 0148-9062/93 $6.00 + 0.00

Printed in Great Britain. All rights reserved Copyright © 1993 Pergamon Press Ltd

Prediction of Water Flow into Rock

Tunnels: an Analytical Solution
Assuming an Hydraulic
Conductivity Gradient
L. Z H A N G t
J. A. F R A N K L I N t
A solution is obtained for predicting tunnel water inflows for the commonly
observed case of an exponential decrease of hydraulic conductivity with depth.
The analytical solution is applicable to the prediction of ubiquitous inflows from
a jointed rock mass, and the risk of major isolated inflows from singular
conductors such as faults must be separately evaluated.
The predictions are verified by finite-element modelling and compared with
published inflows for tunnels in a variety of rock conditions. The new solution
predicts the commonly observed trend that inflows at first increase then
decrease as a tunnel goes deeper, rather than continue to increase as predicted
by Goodman's solution where the hydraulic conductivity is assumed constant.
Goodman's solution can underestimate inflow by 30% or overestimate it by an
order of magnitude, depending on the depth at which hydraulic conductivity is
measured.
Upper and lower bound predictions corresponding to extreme values of
stress-independent factors such as an increase of joint spacing with depth are
examined for a range of typical tunnel characteristics. The differences are
found to be insignificant when the hydraulic conductivity gradient A is small,
but increase to 30% for A = 0.01. An estimate mid-way between upper and
lower bound values gives a maximum prediction error of about .__+15% which
is likely to be adequate in most practical applications.
The analytical and finite element models of both predict that inflows are
insensitive to tunnel diameter (maximum 2% increase in flowrate per metre
increase in diameter). This agrees with the observation that inflows during the
driving of a full-bore tunnel can be similar to those from a pilot tunnel or
exploratory drillhole ahead of the face.

INTRODUCTION ration and corrosion of reinforcement steel by salt water

in undersea tunnels.
Prediction of water inflow into a mine drift or tunnel is
Calculations help the designer to select an optimum
one of the essential tasks of underground engineering.
tunnel depth. Water inflow increases greatly as a tunnel
Inflow predictions determine pumping requirements and
approaches the more intensely fractured near-surface
the extent to which groundwater control measures will
rock, and often the choice is between going deeper, or
be needed. They also indicate the likely severity of other
spending more on rock grouting and other treatments.
water-related problems such as difficult and unsafe tun-
There is no basic difference between an undersea tunnel
nelling conditions and a slow rate of advance; difficult
and an inshore one beneath the water table, except that
blasting; instability caused by water pressure, erosion
greater precautions are warranted because of the
and swelling; lowering of the water table, draining of
catastrophic consequences of uncontrolled caving. The
surface wells, subsidence and foundation damage; in-
engineer may be justified also in taking a more conserva-
ability to bond shotcrete to wet rock; and liner deterio-
tive approach because undersea tunnels are more
difficult and expensive to investigate by drilling, and
tDepartment of Earth Sciences, University of Waterloo, Ontario, ground conditions are often less thoroughly explored.
Canada N2L 3GI. Analytical solutions for inflow prediction are more
RMMS 30/I--D 37
38 ZHANG and FRANKLIN: WATER FLOW INTO TUNNELS

convenient than numerical modelling because they allow tivity gradient, which is assumed to satisfy the exponen..
a rapid parametric study of design alternatives and rock tial form reported by Louis [6] and Thiel [7].
characteristics, which are often incompletely defined at
the feasibility and even at the design stage. The long-
term objective of the authors' research at the University HYDRAULIC CONDUCTIVITY-DEPTH RELATION
of Waterloo is to develop an expert system for inflow The hydraulic conductivity of a rock mass (Fig. 1)
prediction that combines empirical predictions of hy- commonly decreases by several orders of magnitude in
draulic conductivity from rock mass quality, with con- the range of depths encountered in engineering and
ductivity test data where available, and feeds these mining projects [6-12]. This decrease is the result of
values into analytical solutions that have been tested and several factors; the joints become tighter because of an
extended by finite element modelling. increase in the ambient stress level, and also more widely
The available analytical solutions are few, and often spaced and less persistent beneath the zone of near-
restrictive in their assumptions. Polubarinova-Kochina surface weathering and stress relief [8, 11].
[1] obtained an infinite series solution to predict flow into These factors can be grouped into stress-dependent
a horizontal drift. Goodman et al. [2] derived a solution and stress-independent categories. The stress-dependent
for steady state inflow by assuming that water table factors, notably the effect of joint aperture closure, are
drawdown is negligible, as is the case for undersea particularly important because flow is proportional to
tunnels, and also a semi-empirical solution for unsteady the cube of the joint aperture [8, 13]. They can be
state flow beneath a water table. Bello-Maldonado [3, 4] dramatically altered by the drainage effects of the tunnel,
extended Goodman's unsteady state solution by as a result of effective stress variations. Some researchers
considering the inclined initial water table common in consider the decrease in aperture to be the only signifi-
mountainous areas. cant effect [7, 14]. However, the stress-independent
The assumption of a constant average hydraulic con- factors such as joint spacing variations can also be
ductivity independent of the depth below surface, in- important under some circumstances [8].
herent in each of the above solutions, is unrealistic and Several empirical and theoretical forms of hydraulic
can give misleading predictions [5]. In this paper the conductivity--stress~lepth relation are reported in the
authors extend Goodman's steady state model by taking literature [6, 10, 12, 15, t6]. One of the most important
into account the commonly observed hydraulic conduc- and simple relations was published by Louis in 1974
[6, 7, 14]. He obtained his formula from field measure-
ments which indicated the following exponential re-
K (m/s)
lation:
i0-Ii 10 -9 10 -7 j.O-5
0 K = Ks e x p ( - Ah), (1)
where

K = hydraulic conductivity at depth h,

K~ = a constant,
A = hydraulic conductivity gradient (a constant).
20
If the hydraulic conductivities measured by in situ testing
are plotted on a logm scale vs depth then a straight line
should be formed according to this relation, which
should account for the effects of all influence factors. The
negative slope of the line divided by 0.4343 (logm e) is the
A
conductivity gradient A, and the intercept of the line on
40 the lOgl0(K) axis (h = 0) is K s.
-r-
l-- Louis from his test data for fractured amphibolites
and gneisses at a dam site in the Eau d'Olle Valley
(France) measured a conductivity variation from
10 -5 m/see near surface to about 10 -6 and 10 -s m/see in
each abutment at a depth of 60 m, giving A-values of
0.04 and 0.12/m depth. Double-packer injection tests in
6Q shales in Ontario (Canada) gave an A-value of 0.27/m
depth (Fig. 1). Tests for a proposed Northumberland
Straits tunnel in mudstones and sandstones showed an
order of magnitude decrease in conductivity over a depth
of 30 m, and an A-value of 0.08[m.
Fig. 1. Field experimentaldata for decrease of hydraulicconductivity
with depth: (A) unpublished data for Canadian Shales; (B) data by To derive a solution for tunnel inflow, the assumption
Louis [6] for French amphibolites and gneisses. is made that parameter A is linearly composed of two
 ZHANG and FRANKLIN: WATER FLOW INTO TUNNELS 39

parts to account for the stress-dependent and stress-inde- then the governing equation and boundary conditions
pendent factors, i.e. the hydraulic conductivity-depth describing the problem of water inflow can be estab-
relation can be expressed as: lished as follows [2]:

K = Ks e x p [ - Ca" - a3h] = Ks e x p [ - C(?m-?w)h - a 3 h ]

(2) OxJ+L oy
=0,
(4)
The total vertical gravitational stress a = ?mh is the H ( x , y)[, = 0 = H0, (5a)
product of the unit weight of overburden and the
depth, as found except in actively tectonic regions. The H ( x , Y ) l y ~ = Ho, (5b)
effective stress a ' is a - u, where water pressure is
assumed to be the product of the unit weight ~w and the H ( x , y)lx~ + oo = H0, (5c)
depth h [17]. where H is the hydraulic head at any given depth, Ho is
The stress-independent factors are expressed as a3h, the constant hydrostatic total head, Q is the inflow per
where a3 is a constant. Therefore the conductivity gradi- unit o f tunnel length, L is the rock cover (depth of the
ent is given by A = C(~m - ?w) + a3. If we define another tunnel below ground surface) and d is the tunnel diam-
two constants ~,wC = at and ?mC = a2, then the conduc- eter (Fig. 2).
tivity gradient A = a2 + a3 - at, and also at/a2 = ?,,/?m, The hydraulic conductivity K in equation (4) is a
Louis' hydraulic conductivity-depth relation then takes function of depth and water pressure. For steady state
the following form: tunnel inflow under conditions (5a--c) the water pressure
in the x - y coordinates of Fig. 2 can be determined
K = Ks exp(al - a2 - a3)h, (3)
directly as:
where a I can be interpreted as a coefficient related to u = ~, [H - L + y] (6)
water pressure, a 2 is related to the total stress effect and
a3 is related to stress-independent factors such as joint and the effective stress term becomes:
spacing.
The stresses are assumed isotropic, which can be far a ' = ?my - - ~ w [ H - - L + y ]
from correct for near-surface conditions where horizon- = [~s - - ?w]Y - ~.,H + ?.L, (7)
tal stresses often greatly exceed the vertical. However,
under these circumstances the near-vertical joints are substituting (7) into (2) we obtain the hydraulic conduc-
compressed and tighter than the horizontal joints, and tivity expression for this particular problem:
contribute less to the total inflow; vertical stress governs
the hydraulic conductivity of horizontal joints and is K = Ks e x p [ - Ca" - a3y] = Kt exp(--Ay)exp(al H), (8)
therefore the most important component. This is par- where K~ = Ks e x p ( - a , L).
ticularly the case for tunnels and mines in sedimentary
rocks with closely spaced horizontal bedding joints. The
assumption of stress-isotropy involves less o f an ap- FOURIER TRANSFORMATION
proximation at greater depth, irrespective of the rock Fourier transformation can be used to solve the
type. problem depicted in (4) together with boundary con-
A further practical consideration is that most hy- ditions (5a-c) (the definition and some properties of the
draulic conductivity tests are conducted from the surface Fourier transform are summarized in Appendix A).
in vertical drillholes; the hydraulic and stress an- However, first we need to eliminate the non-linear term
isotropies remain unknown and can only be estimated.
The test values, the only ones available, are calculated
7
assuming isotropic conductivity and reflect an "average" b

hydraulic conductivity of the rock mass, which, however, L

is much more influenced by near-horizontal joints /,kC'~a, ~X
perpendicular to the direction of drilling.

GOVERNING EQUATION AND BOUNDARY

y~T,~,w
(.J,.)
H
CONDITIONS
If we assume:

- - a tunnel of circular cross-section driven beneath a

sea or lake bed, or into a prolific inexhaustible
aquifer;
- - a continuous isotropic semi-infinite medium;
--isotropic hydraulic conductivity;
--isotropie stress; Fig. 2. Coordinate system and terminology.
40 ZHAN(J and FRANKLIN: WATER FLOW INTO TUNNEL,',

in equation (4) caused by the pressure-dependent hy-

draulic conductivity. To do this we need to define a new
,/~(P) . . . . . . . . . . . . . . . . . . . . . ,
variable q~ as: /p2 A 2
2 +4
~(x'Y)=Iexp(a'H°)-- exp(a'H)]exp(-AY ) (9)
.....

Substituting q~ into equation (4) and (5a-c), the govern-

ing equation and boundary conditions are simplified as
f(y)=~K,,
'l? v'x~+y " , (a,>O),

follows:
Q e x p ( ~L ) L)21
q,(x, y ) -
t- - -
8x ~ 8v 2 4 ~b +
~ exp 6(x - 0)6( v - L) = O,
(10)
• (x,y)l, =0 = 0, (1 la) -KoL~N/x'+(y + L)2 (13)

q'(x, y ) l , . ~ = 0 , (lib)
where Ko(x) is Bessers function of the second kind,
q~(x, y)l,~ +, = 0. (1 lc) order zero, which can be found from mathematical
handbooks [20].
To solve equation (10) with boundary conditions (1 la-c)
Applying the boundary condition that the hydraulic
we first apply the Fourier exponential transform for x to
head H is zero at the point x = d / 2 , y = L on the
the equation and boundary conditions giving:
opening periphery [2]:
82~ A2 +Qexp(~)6(y i L ~ ~ 0 ~ H(x = d/2, y = L) = 0
and and we have

~(~, y)l,: o = 0 (x = d/2, y = L )

~(~, y),. ~ = o

Then applying the Fourier sine transform for y:

- ~-C----C exp - ~ L
L al al

_~x2~ _ p e q ~ _ ~ + ~-~[exp T sln(PL) --- O, exp - ~ L

= [exp(a, H0) - 1] (14)
solving the transformed ordinary differential equation a~
for ~ gives:
Substituting (14) into (13), the water inflow rate (2 is
Q exp ~ - sin(PL) obtained:
= (12) 2K~ ~ [exp(al H0) - 1]exp ( - A L)]
Kt(~2+p2+A---4) " Q = (i5)
a t [ K o ( A d ) _ K°~-214+/A dZ/T--
-d 4L2) 1
the transformed variable ~ in solution (12) can be
inverted to x as [18,19]: If the tunnel diameter is very small compared with its
depth, i.e. d/L 4. 1, and noting that K~ = Ks e x p ( - a t L),
Qexp(fl~-)sin(PL)exp(-IX[~) then the solution can be further simplified as follows:
~=

2Ki p• A2
4
2K~z exp[-(A + al)L][exp(aiHo) - 1]

To invert P back to y let:

Inflow prediction should if possible be based on the
hydraulic conductivity K~ measured at tunnel depth, plus
• - 2L(P)sin(PL), the gradient A measured within several diameters above
2K]
and below this depth. The parameter Ks in solution (16)
can be replaced by the measured hydraulic conductivity
K~ to obtain:
L(P) =
A2 2KtTt exp[--alL] [exp(al H 0 ) - I]
2 -~ Q = (17)
~ 4
Q e x p ( ~ L-)
- 2f~(P)sin(PL), If the hydraulic conductivity gradient A = 0 (i.e. if the
2Kt hydraulic conductivity does not change with depth), then
 ZHANG and FRANKLIN: WATER FLOW INTO TUNNELS 41

solution (16) is simplified to Goodman's solution (see strates that for typical data, the difference between upper
Appendix B for details): and lower bound inflows is insignificant when the con-
ductivity gradient is small, less than 4% for A = 0.0001,
2K~xHo
but increases to 30% for A = 0.01. In practice, an inflow
estimate midway between upper and lower bound values
is likely to give predictions within + 15% which should
be adequate in most cases.
APPLICATION OF THE SOLUTION
INFLOW AS A FUNCTION OF DEPTH AND
Six parameters are needed when solution (16) is
TUNNEL DIAMETER
employed: H0, L, d, Ks, A and a~. The first three i.e. H 0,
L and d are known, K~ and A are measured, and a~ needs The upper bound (a 3 = 0) solution for inflow in terms
to be estimated. of Ks, (Q/Ks), and A is evaluated by equation (20) in
The constant K~ in the conductivity expression (1) is Figs 4-6 for various typical parameter values:
the surface intercept of the exponential conductivity
7w = 10 kN/m 3,
gradient, not the actual hydraulic conductivity at surface
which is a highly variable and unrepresentative quantity ~m = 26 kN/m 3,
that cannot be measured reliably. The way to obtain Ks
H o - L = 10m.
and A is by a visual or least-squares fit of a straight line
to a set of hydraulic conductivity measurements at Figure 4 gives the predicted inflows as a function of rock
different depths plotted on a log~0 conductivity vs depth depth L above tunnel centreline and conductivity
graph such as in Fig. 1. The negative slope of the line gradient A for tunnels of 4 and 16 m diameter. Each
divided by 0.4343 ( = log10 e) is the conductivity gradient of the curves except the one for A = 0 reaches a
A, and the intercept of the line on the log~0(K) axis maximum inflow at a certain depth (in some cases at a
(h =O) is Ks. greater depth than shown on the graph). The inflow
The parameter a~ can be written as follows according increases with increasing depth above this "peak point",
to the definition of the conductivity gradient: and then decreases as the depth of the tunnel becomes
even greater. Similar results are shown in Figs 6 and 7.
A = a2 + a 3 -- a I and therefore, a2 = A + aj - a 3.
The depth of maximum inflow is a function of the
Also from the definition of effective stress we have: hydraulic conductivity gradient. The peak becomes
deeper with decreasing A, until at A = 0 (Goodman's
al ~w
solution with a constant hydraulic conductivity), a
a2 ~/m monotonic increase of inflow with depth is obtained.
~w z/w( Previous parametric studies [21,22] show that Good-
al = - - a 2 = A -t-al-a3), man's solution can overestimate inflows by an order of
7m 7m
magnitude depending on the depth of hydraulic conduc-
~'w
al - - - (A - a3). (19) tivity measurements, and can underestimate them by
?m -- 7w about 30% when the calculations are based on measure-
Substituting a~ from (19) in (16) we obtain: ments at tunnel depth.

2gsTF(~m-- Tw)eXp[--( ~mA '~wa3 ~L ] F e x p ( 7w(A ~ a3)n0") - i 1

[_ \'~m--~w ~m--Tw// AL \ ])m--~)w ,,]
O (20)
A A
1
Therefore a3 which represents all stress-independent A comparison of results for the two extreme tunnel
factors needs to be determined if we are using solution diameters in Fig. 4 demonstrates that inflows are insen-
(20). Its value lies between 0 and A and may be estimated sitive to changes in the diameter of a tunnel (maximum
as about 0.25A for the usual situation where hydraulic 2% per metre of diameter). This conclusion is of some
conductivity is proportional to the cube of joint aperture practical importance and supports the observation that
and inversely proportional to joint spacing, and both the yield of water from a pilot bore or drainage gallery
aperture and spacing are decreasing exponentially with is often as great as that from the full-sized tunnel. An
depth [8, 13]. Selection of a 3 0 gives the lower-bound
= increase in tunnel diameter carries little penalty in terms
solution where joint spacing and other stress-indepen- of pumping requirements. Also, the presence of a blast-
dent factors are constant with depth. Selection of a3 --- A damaged zone around a tunnel, equivalent to an increase
gives an upper-bound solution where stresses are as- in tunnel diameter, has little effect on the inflow
sumed constant over the depth interval. Figure 3 demon- quantities.
42 ZHANGand FRANKLIN: WATER FLOW INTO TUNNELS

O/Ks (m) Q/Ks (m)

0 400 800 1200 0 20 40 60 80 100
0~_
d = 4m

d = 16m 200.
200
Depth L
(m)
I f/,e/'\ \ Upper
400~; ~ Bbund
400
Upper II/ Lower
Bound

600I Bound
600

Lower
Bound
800 800

1000 - ~ 1ooo
A = 0.0001 A = 0.01
Fig. 3. Upper and lower bound inflows for a3= 0 and a3= A.

Q/Ks (m)
Q/Ks (m)
4OO 800 1200
I I

400 800 1200

A = 0.01

200
20(

400
400
Depth L
(m)

600
600

800 800

1000
I000 (
d=4m
d=16m
Fig. 4. Effects of tunnel depth on predicted water inflow.
 ZHANG and FRANKLIN: WATER FLOW INTO TUNNELS 43

Q/Ks (_m)
1 200

1 000 J

800"
L = lO00m
600"

400- k\
._~~~~~,
/ " = - "
L--6 0 m
200"

o
0.000 0.002 0.004 0.006 0.008 0.01 0
Hydraulic Conductivity Gradient A (m-l)
Fig. 5. Effects of hydraulic conductivitygradient A on predicted inflows.

The influence of hydraulic conductivity gradient A on influence coefficient matrix technique [23] was employed
water inflow can be seen from Fig. 5. Inflow decreases to obtain the hydraulic conductivity term:
with increasing A; the rate of decrease (6Q/t3A ) is greater
K = Ks exp( - A y ) = K1 exp( - a l H ) exp ( - Ay).
when A is less than 0.002, and becomes insignificant for
A greater than 0.005. Also, the deeper the tunnel, the For purposes of evaluation, typical parameters were
more rapidly the inflow decreases. This indicates that A assigned as follows:

tunnel depth: 20 m, 60 m, 200 m, 600 m, 1000 m,

tunnel radius: 2m, 8 m
conductivity gradient: 0.0, 0.0001, 0.001, 0.005, 0.01,
surface water body depth: 10m

only has a significant effect on inflow for tunnels deeper The results obtained are plotted in Fig. 6, where the
than about 60 m. solid lines are the analytical solutions and the symbols
are the numerical results. The analytical and the numeri-
VERIFICATION BY NUMERICAL MODELLING cal solutions generally agree closely. As expected, the
solutions diverge only for large tunnels close to the
To evaluate the reliability of the analytical solution surface, i.e. when the ratio of tunnel diameter to depth
and the effect of the sink term introduced in equation (4), becomes large. This error is believed to be produced by
a finite element model was applied to simulate the same the sink term in the analytical solution. For the typical
boundary value problem: range of data employed in the example, the difference
[(Qanat - Qnum)/Qanat] is less than 6% if the ratio of rock
a-~ L T~j ~ K =o, (21) cover to tunnel diameter L / d is greater than 7, and
exceeds 20% if L / d is less than 2. Depending on the
H (x, Y )ly =o = Ho, (22a)
required accuracy of prediction, a numerical solution
H(x, Y)ly~ o~= H0, (22b) may be preferred to equation (20) for shallow tunnels
H(x, Y)Lx~± ~ = H0, (22c) with rock cover less than about two to three tunnel
diameters.
H(x, y)l(: + y: = :/4) = 0 (22d)
where all conditions are identical with those for the
C O M P A R I S O N WITH REPORTED INFLOWS
analytical solution except that the sink term is replaced
by the real hydraulic head at the tunnel wall. A linear Figure 7 compares the values predicted by our sol-
triangle grid mesh was generated automatically. The ution with some records of tunnel water inflows
hydraulic conductivity gradient was assumed as [2, 21, 24-27] presented here in Table 1. The records in
A = a 2 - al (a3 = 0). Galerkin's method was applied to nearly every case are incomplete with respect to one or
equation (21) in the discretization process, and the more of the key controlling variables such as the length
of tunnel and the length unlined, extent of grouting,
44 ZHANGand FRANKLIN: WATER FLOW INTO TUNNELS

Q/Ks (m) Q/Ks (m)

0 500 i000 o 400 800 1200
I 0 I I I I I

~Numerical
20O

~
20(

Analytica]

40( 400
Depth L
(m)

60C 600
-0 001 A=O.O00

800 800

1000 1ooo

Fig. 6. Comparison of analytical and numerical solutions for inflow.

hydraulic conductivity intercept Ks and gradient A (or unexpectedly in tunnel headings. To quote just one
the rock mass quality from which these can be esti- example, during a mucking cycle 3000 1/sec of water
mated). This points to a need for better records, but for "erupted" from the face of the Chivor II penstock tunnel
the present, the data available allow some tentative in Colombia. Sand from a very friable sandstone was
conclusions. carried to the portal some 1300 m away, and filled the
To prepare Fig. 7 some of the missing information tunnel [28].
has been estimated from general descriptions of
rock conditions. The predicted inflow curves have been Q (Lls.m)
superimposed for a conductivity gradient of A = 0.005 0 0.04 0.08 0.12
and for three values of Ks in the range 0.0004- 0 I I I I I I

0.015 m/sec. These values were selected to fit the appar- Depth L
ent trend of the data, and are for illustrative purposes (m)
only. However, they confirm that the trend of global
data with some exceptions as noted below is predicted 500
quite well by our solution with values of A and Ks similar
to those reported by Louis and others and reviewed
earlier in this paper. The comparison confirms the
\\ Ks=0015 m/s
general magnitude of A and K s values. Ks=O.O01 m/s
The data are consistent with an apparent peak inflow i000
in the depth range 200-500 m with reduced inflows at Ks=O.O004 m/s
greater depth. Notable exceptions to the general trend
include some unusually large inflows even at depths as
great as 1.5-2.5 km. These are likely to be caused by 1500
inflows from one or more major conductors such as
faults, karstic solution pipes in limestone or flow tops I'
in volcanic rocks, which remain open in spite of
high levels of ambient rock stress, and which conduct 2000
water under considerable pressure from shallower
depths. It is widely recognized that inflows originate I"
A=O.O05
from two geological sources: general seepages through
ubiquitous joints in the rock mass, and localized but
often highly conducting features that are responsible for 2500
the majority of serious incidents when encountered Fig. 7. Theoretical prediction and some are histories.
 ZHANG and FRANKLIN: WATER FLOW INTO TUNNELS 45

T a b l e 1. Some records o f w a t e r inflows into tunnels

W a t e r inflow
(1/sec. rain)
Tunnel L e n g t h (m) D e p t h (m) Max Min Ave R o c k types
Weisen-stein 3789 470 0.1060 0.0106 0.0200 m

Gotthard North 14,920 1752 0.0030 0.0010 --

Gotthard South 14,920 1752 0.0230 0.0150 -- E

Mont-Cenis 13,636 1610 0.0043 0.0037 -- Sandstone, slatey

limestone, q u a r t z i t e
Turchino 6428 386 0.0120 0.0016 -- Crystalline slates,
serpentine
Simplon 19,803 2135 -- -- 0.0510
Tauern 8551 1567 0.0070 0.0050 -- Granite, gneiss,
micaceous slate
Bosruck 4770 1300 -- -- 0.4610 Sandstone, d o l o m i t e
Wocheiner ~6334 370-1000 0.2050 0.0240 -- Chalk
Kara-wanken 7976 370-1000 -- -- 0.0110
Cremolino 3400 386 -- 0.0030 0.0360 Sandstone, c o n g l o m e r a t e
Vardo 1700 80 0.0007 -- Sandstone, shale
Karmsund 2100 180 0.0024 -- -- Greenstone, phyllite,
gneiss
Fordes-fjord 1600 170 0.0031 -- -- Gneiss
Forlands-fjord 1000 160 0.0050 -- -- Phyllite, gneiss
Hjartoy 1800 105 0.0018 -- -- Gneiss
Ellingsoy 1100 140 0.0061 -- Gneiss
Valderoy 2200 137 0.0008 -- -- Gneiss

Our solution attempts to predict ubiquitous seepages 30% for shallow tunnels in stress-relieved and weathered
only. Inflows from singular conductors are difficult if not rocks where .4 can approach 0.01. In practice, an
impossible to predict, and quantitative predictions are in estimate midway between upper and lower bound values
any case largely irrelevant because the aim is to intercept is likely to give predictions well within _+ 15% which is
and stop the flow. Thorough appreciation of tunnel adequate in most cases. Errors in measuring hydraulic
geology in relation to the probability of encountering conductivity are likely to exceed the errors of prediction.
singular conductors is thus absolutely necessary when The analytical model predicts that inflows are quite
tunnelling deep beneath the water table or under large insensitive to changes in the diameter of a tunnel (maxi-
bodies of water. Inflow is likely to be driven by a mum 2% increase of inflow per metre increase in
pressure close to the full hydrostatic head above the diameter). This agrees with the observation that inflows
tunnel, and to be transient during the period required to during the driving of a full-bore tunnel can be similar to
drain the conductor. Because pre-investigation costs are those from a pilot tunnel or exploratory drillhole ahead
so high, particularly for sub-sea tunnels, tunnelling of the face.
methods must often be adjusted to permit a pilot drift Inflow predictions can be improved with the help of
for exploration and grouting, or at least, to accommo- a more complete body of data on actual inflows during
date probing ahead of the face. tunnel driving as a function of rock conditions, and also
on the link between rock mass quality and hydraulic
DISCUSSION AND CONCLUSIONS conductivity. Reliable values for hydraulic conductivity
gradient and intercept are often unavailable when
Hydraulic conductivity under conditions of ubiqui- needed for design, particularly for undersea tunnels, but
tous seepage decreases with depth because of the com- can be estimated empirically from data on rock jointing.
bined effects of increasing stress and decreasing joint Inflow predictions and estimates of problems and costs
aperture and spacing. As a tunnel goes deeper the inflow for pumping, grouting and ground stabilization can best
at first increases and then decreases because the general be handled by an expert system approach with a
"tightening" of the rock mass overtakes the effect of the combination of empirical and analytical predictive tools.
increasing water head. This is contrary to the Goodman Separate consideration must be given to the ubiquitous
solution which predicts that inflows continue to increase inflows and to the possibility, consequences and
with the tunnel depth. treatment of singular conductors.
Some of the factors causing the decrease of hydraulic
conductivity with depth are stress-dependent and some
Accepted for publication 19 November 1992.
are not. Even though the stress-dependent factors are
usually dominant, the stress-independent factors can be
important when the conductivity gradient A is large. The REFERENCES
factors cannot readily be separated, but upper- and 1. P o l u b a r i n o v a - K o c h i n a P. Theory of Groundwater Movement,
lower-bound solutions can be found. The difference (translated from R u s s i a n by J. M. R o g e r de Wiest). P r i n c e t o n
U n i v e r s i t y press, N J (1962).
between upper and lower bound inflows is insignificant 2. G o o d m a n R., M o y e D., S c h a l k w y k A. a n d J a v a n d e l I. G r o u n d -
when the conductivity gradient is small, but increases to w a t e r inflow d u r i n g tunnel driving. Engng Geol. 2, 39-56 0 9 6 5 ) .
46 Z H A N G and FRANKLIN: WATER FLOW INTO TUNNELS

3. Bello-Maldonado A. A. Seepage towards tunnels. Proc. 2nd Int. Some operational properties are as tollows:
Cong., Int. Assoc. Eng. Geol., Vol. VII, Paper 15.1-15.10 (1974).
4. Bello-Maldonado A. A. Post-construction seepage towards tun- Fe[gm(z)] - (i~)'n~(~),
nels in variable head aquifers. Proc. Int. Congr. Rock Mech.,
Melbourne, Australia Vol. B, pp. B i l l BII7 (1983). Fe[g(z +h)]=e'h',~l~l, heR.
5. Blindheim O. T. and Ovstedal E. Water control in subsea road
tunnels in rock. Proc. 2rid Syrup. Strait Crossings (J. Krokeborg, Convolution for F,:
Ed.), pp. 223 230. Balkema, Rotterdam (1990).
6. Louis C. Rock hydraulics. Rept 74 SGN 035 AME, Bur. Geol.
Mining. Res. (BRGM), Orleans, France, pp. 14-41 and pp. 80-84
F:'tY(,)~(,)]=f '.~ f ( t ) g ( x - t)dt,
(1974).
f ( x ) = Vd'[f(a)l; g(x) = rf~[g(~)].
7. Thiel K. Rock Mechanics in Hydroengineering. Elsevier, Amster-
dam (1989). Let g: R--*R be a function with lim . . . . g(z) = 0. The Fourier sine
8. Snow D. T. Rock fracture spacings, openings and porosities. transform on the half line is defined as:
J. Soil Mech. Found. Div., Am. Soc. Civ. Engng, 94, 73-91 (1968).
9. lwai K. Fundamental studies of fluid flow through a single
fracture. P h . D . Thesis, University of California, Berkeley (1976). ~(a) = F~Lg(x)] = fo ~ g(x) sin(~x) dx.
10. Kranz R. L. The permeability o f whole and jointed Barre granite.
Int. J. Rock. Min. Sci. & Geomech. Abstr. 16, 225-234 (1979). Convolution for F~:
11. Brace W. F. Permeability of argillaceous and crystalline rocks. Int.
J. Rock Mech. Min. Sci. & Geomech. Abstr. 17, 241-251 (1980). i 2 ~ ~+'
12. Walsh J. B. Effect of pore pressure and confining pressure on
fracture permeability. Int. J. Rock. Mech. Min. Sci. & Geomech.
Abstr. 18, 469-474 (1981). g(x) = F ( ' ~ (~)1; f ( x ) = F , i []'(a )].
13. Manev G. and Avramova E. On the variation of strength and
resistance condition of the rocks in natural rock massif. Proc. 2nd
Cong. Int. Soc. Rock Mech., Belgrade. Paper 1-10, pp. 59~i8 APPENDIX B
(1970).
14. Chang X. The couple analysis of water steady state flow and its Derivations of Goodman "s Solution
stress in rock mass. Proc. 1st Symp. on Geomech. Modelling, If we ignore the depth effect on the hydraulic conductivity then the
Beijing, China. pp. 335-343 (1985). conductivity gradient A = 0, therefore the transformed differential
15. Carlsson A. and Olsson T. Storage in excavated rock caverns. equation (12) becomes:
Rock store 77. Proc. 1st Int. Symp. Storage in Excavated Rock
Caverns, Vol. 2 (1977). 7~ = Q sin(eL)
16. Wei Z. and Hudson J. Permeability of jointed rock masses. Proc. KI (~ 2 + p2)
I S R M Symp. Rock Mech. and Power Plants (Romana, M., Ed.),
Balkema, Rotterdam (1988). and applying the same principle as before we can invert ~ back to x:
17. Franklin J. A. and Dusseault M. B. Rock Engineering. McGraw-
Hill, New York (1989). = Q s i n ( e L ) e x p ( - I x ]e)
18. Bateman H. Tables of Integral Transforms. Vol. 1, McGraw-Hill, 2K 1P
New York (1954).
19. Churchill R. V. Operational Mathematics, McGraw-Hill, N e w
Q f~(p)gs(p)
York (1972). 4K I
20. Luke Y. L. Integrals of Bessel Function. McGraw-Hill, New York
(1962). e x p ( - I x JP)
f~(P) = sin(eL), g~(P)
2 I. Stabel B. A. M. Evaluation of methods for predicting tunnel water P
inflow. Diplomarbeit Thesis, Lehrstuhl fiir Felsmechanik, Univ.
Karlsruhe, Germany and Univ. Waterloo, Canada (1991). This becomes a function o f f ( P ) and g(P). Applying Fourier trans-
22. Franklin J. A., Zhang L. and Stabel B. Prediction of tunnel water forms we can obtain f ( y ) and g(y) respectively as:
inflow. Proc. 9th Annual Canada Tunnelling Conf., Montreal
(1991). In Canadian Tunnelling, pp. 97-110. Tunneling Assoc. f ( y ) = 6(y - L),
Canada, BiTech (1992). 2 y
23. Huyakorn P. S., Thomas S. D. and Thompson B. M. Techniques g(y) -
rt y2 + x 2
for making finite element competitive in modelling flow in variably
saturated porous media. Wat. Resour. Res. 20, 1099-1115 (1984). Applying the convolution principle to the equation we have:
24. Stini J. Tunnelbaugeologie. Springer-Verlag, Wien (1950).
25. Gau X. Water inflow in tunneling and its treatment. Tunnels and
Water (Serrano, Ed.). Balkema, Rotterdam (1988). • (x,y)= 6(r--L) _,j'n~dtdr'x2
26. Nilsen B. Norwegian sub-sea tunnels--a review with emphasis on
water leakages. Tunnels and Water (Serrano, Ed.), pp. 913-918.
Balkema, Rotterdam (1988). _ a ;0~o. ..,, I-x2+(Y+r!2]dr
27. Zhang L. Prediction of tunnel water inflow. M. Sc. Thesis, Univ.
Waterloo, Canada (1991). Q I'-x2 + (y + L)21
28. Marulander A. P. and Brekke T. L. Hazardous water inflows in
- 4K I n lnLx2-T~f ~ J "
some tunnels in sedimentary rocks. Proc. Rapid Excavation and
Tunnelling Conf., Vol. 1 (1981). Therefore we obtain the water inflow rate as:
4Ksndp(x, y)
APPENDIX A Q= [x2+(y+L)2]"

Fourier Transforms and Some Properties

Let g: R--*R be a function with lim:~ +_~g(z)=0. The exponential Substitutingt h e s a m e condition of ~ (X =d/2, Y = L ) = H o, we
fourier transform is defined as: obtain the Goodman solution:

g(e) = F~[g(z)] = g(z)e .... dz 2K~nHo

0
where ~ e R.