T T
x
P B
x x
u
x
u
T
x
kl ijkl
j
Ts ij
j k
l
l
k
ijkl
j
(R1)
j
w
j w
ij
i
Tw eq
i
j
j
i
ij
x
z
g
x
P
k
x t
T
t
P
G x
u
x
u
t
B
1
2
1
(R2)
( )
T
i i
T
i j
w
j w
ij
w w eq
f
x
T
x
K
x
T
x
z
g
x
P
k
C
t
T
C
(R3)
XXVI
donde
P es la presin de fluido [Pa],
T es la temperatura [C],
u
i
son los desplazamientos segn los ejes coordenados x
i
[m],
k
ij
es el tensor de permeabilidad intrnsica del medio equivalente [m
2
],
T
ijkl
es el tensor de rigidez del medio equivalente [Pa],
B
ij
es el coeficiente de Biot tensorial del medio equivalente [],
G es el mdulo de Biot [Pa],
Tw
es el coeficiente de expansin volumtrica del agua [K
1
].
Ts
es el coeficiente de expansin volumtrica del slido [K
1
].
f
T
es el trmino de fuente de calor [W/m
3
] (en nuestro caso, el flujo de calor
producido por los calentadores del experimento FEBEX)
w
es la viscosidad dinmica del agua [Ns/m
2
].
g es la gravedad [m/s
2
].
z es la elevacin sobre el nivel del mar [m].
( )
m m f f eq
+ = es la porosidad del medio equivalente []
f
,
m
son las fracciones volumtricas de fracturas y matriz rocosa
respectivamente [].
f
,
m
son las porosidades de fracturas y matriz rocosa respectivamente [] (
f
=1
para fracturas rellenas con agua).
w
,
s
son las densidades del agua y de los granos slidos respectivamente
[Kg/m
3
].
( ) ( ) ( )
s s m m w w m m f f eq
C C C + + = 1 es la capacidad calorfica
intrnseca del medio equivalente [J /m
3
K],
C
w
, C
s
son las capacidades calorficas del agua y de los granos slidos
respectivamente [J /kg K].
( ) ( )( ) ( )( )
ij Ts m m ij Tw m m f f ij T
K K K + + = 1 es el tensor de conductividad
trmica del medio equivalente [W/m K],
(K
Tw
)
ij
, (K
Ts
)
ij
son los tensores de conductividad trmica del agua y de los granos
slidos respectivamente (supuestos en nuestro caso istropos, homogneos y
constantes en el tiempo).
Propiedades contnuas equivalentes. Los coeficientes continuos equivalentes
involucrados en las ecuaciones R1, R2 y R3 se calculan mediante un procedimiento
de homogeneizacin para medios fracturados discretos, basado en los trabajos de [71] y
[1], que aplica un mtodo de superposicin de caudales (para los coeficientes
hidrulicos) o de deformaciones (para los coeficientes mecnicos), fijado un gradiente
de presin o un campo tensional global respectivamente.
El mtodo de superposicin convierte, por tanto, un medio fracturado discreto 3D en un
medio continuo equivalente por medio de la suma de las contribuciones individuales de
cada fractura. Para el clculo de la conductividad hidrulica equivalente, definimos
un bloque fracturado individual, compuesto por un prisma de matriz rocosa permeable
atravesado completamente por una fractura plana horizontal (Figura R6a), y
calculamos la solucin exacta de las ecuaciones de flujo con condiciones de contorno
lineales a trozos para la altura piezomtrica H (Figura R6b). Mediante la resolucin de
estas ecuaciones obtenemos una expresin para la conductividad hidrulica equivalente
del bloque fracturado individual, en funcin de las conductividades de la matriz y de la
fractura (de tipo Poiseuille).
XXVII
b/2
b/2
a
x
z
l
l
I
F
A
C
B
z
H
J
z
=
x
H
J
x
A
Figura R6: a. Bloque fracturado individual de un medio poroso fracturado; b.
Condiciones de contorno lineales a trozos para la altura piezomtrica H.
La expresin final de la conductividad hidrulica equivalente del bloque fracturado
individual es:
( ) ( ) ( )
( )
( )
=
+ =
+ = =
F M
H
F M A
H j i A j i ij ij
K K
K
K K K
K n n K n n K
1
1
1
;

K (R4)
donde K
A
y K
H
son las medias aritmtica y armnica respectivamente de las
conductividades de la matriz y la fractura ponderadas con la fraccin volumtrica de
fractura, , en el bloque fracturado individual. Esta solucin es generalizable a bloques
fracturados de formas cualesquiera, siempre que sean atravesados completamente por la
fractura.
La altura del bloque fracturado individual b se define de manera que el volumen total
del dominio a homogeneizar sea igual a la suma de los volmenes de cada bloque
fracturado individual. Nota: los bloques fracturados individuales as definidos,
correspondientes a un determinado dominio fracturado, por lo general se superpondrn
los unos a los otros, aunque el volumen total del dominio se conserva.
La homogeneizacin a escala del dominio fracturado se realiza ahora superponiendo las
contribuciones de cada bloque fracturado individual al flujo hidrulico global dado un
gradiente hidrulico global fijo. La expresin final de la conductividad hidrulica
equivalente del dominio fracturado es :
( ) ( ) ( ) ( )
( )
( )
<
<
+ +
=
f k
f
k i
f
k
f F
f
M
f
f
j
f
i F
f
M
f f
j
f
i ij
k
f
k i
f
k
ij
ki
ki
n A
K K
n n K K n n n A
K
2
2
1

1
1
(R5)
XXVIII
donde, para cada bloque fracturado individual f , A
k
f
es el area normal a cada una de
las direcciones coordenadas locales del bloque (igual al area de la fractura A
f
para las
bases e igual a l
k
b para la cara lateral k);
ki
es el ngulo entre los vectores unitarios
(vector normal a la cara k) y
k
n
i
u
+
+
=
2
1 1
es el tensor de flexibilidad
homogeneizado de la matriz rocosa [Pa
1
],
ijkl ijkl ijkl
G
g
F
g h
C
1 1 1
+
=
= =
N
f
f
j f i f f ijkk ij
n n F F
1
2
1
es un tensor geomtrico de 2 orden [],
( ) ( ) ( ) ( )
=
=
N
f
f l f k
f
j f i f f ijkl
n n n n F
1
2
1
es un tensor geomtrico de 4 orden [],
(
ik jl jk il il jk jl ik ijkl
F F F F G + + + =
4
1
) es un tensor geomtrico de 4 orden [],
( ) { } ( ) { } ( ) { }
1 1 1
= = =
ij ij kl ijkl kl ijkl ij ij ij ij ij
B B T B T B B B G es el mdulo de
Biot [],
kl ijkl ij
B T B = es el coeficiente de Biot [],
E es el mdulo de Young [Pa],
es el coeficiente de Poisson [],
l K h
n
es un factor de resistencia media a los esfuerzos normales [Pa],
l K g
s
es un factor de resistencia media a los esfuerzos tangenciales [Pa],
K
s
es el mdulo de rigidez normal [Pa/m],
K
n
es el mdulo de rigidez tangencial [Pa/m], y
l es la longitud media de las fracturas en el dominio de homogeneizacin [m].
XXIX
La homogeneizacin de los coeficientes ha sido llevada a cabo de dos maneras
diferentes: tomando como volumen de homogeneizacion el volumen total del medio
fracturado (70x200x70 m
3
), con lo que se obtienen coeficientes nicos para todo el
dominio (condiciones homogeneas); tomando como volumen de homogeneizacin el
Volumen Representativo Elemental (VRE) con respecto a la conductividad hidrulica
(aqul a partir del cual la conductividad hidrulica permanece invariable) y realizando
una media movil a lo largo del volumen total del medio fracturado (condiciones
heterogeneas). El VRE es un cubo de 20x20x20 m
3
, y la media mvil se ha realizado
con un salto de 10m, con lo que la homogeneizacin proporciona una coleccin de
5x18x5 valores para cada coeficiente distribuidos uniformemente a lo largo del
dominio.
Los valores de la homogeneizacin de bloquenico para los principales coeficientes
del modelo son:
(R7)
2 18
10
099 . 1 013 . 0 009 . 0
015 . 0 112 . 1 034 . 0
017 . 0 043 . 0 092 . 1
m k
ij
=
(R8) Pa T
ijkl
9
10
8336 . 0 0660 . 0 0161 . 0 0514 . 0 1589 . 0 1338 . 0
0660 . 0 5725 . 0 0777 . 0 1679 . 0 0146 . 0 0212 . 0
0161 . 0 0777 . 0 8816 . 0 3543 . 0 2142 . 0 0167 . 0
0514 . 0 1679 . 0 3643 . 0 3933 . 5 4194 . 2 2865 . 2
1589 . 0 0146 . 0 2142 . 0 4194 . 2 5982 . 3 4461 . 2
1338 . 0 0212 . 0 0167 . 0 2865 . 2 4461 . 2 1096 . 3
=
(R9)
=
9271 . 0 0163 . 0 0022 . 0
0163 . 0 9411 . 0 0186 . 0
0022 . 0 0186 . 0 9401 . 0
ij
B
(R10) Pa G
10
10 1877 . 4 =
Los valores del resto de coeficientes intermedios para esta homogeneizacin pueden ser
consultados en el anexo XI (APPENDIX XI).
Implementacin y resultados del modelo THM. El ltimo captulo de la tesis
presenta la implementacin del modelo termohidromecnico que se ha desarrollado en
el programa de elementos finitos Comsol Multiphysicsy los resultados obtenidos. El
dominio a modelar es, como ya se ha mencionado anteriormente, un bloque de granito
fracturado de 70x200x70 m
3
centrado en la galera FEBEX, con el norte geogrfico
orientado segn el eje negativo de las X. En dicho bloque se encuentran el tunel
principal de acceso al GTS, un tunel secundario correspondiente al laboratorio y
finalmente la galera FEBEX en la que se desarrolla el experimento de calentamiento
insitu. La figura R7 muestra la disposicin de estos elementos en el dominio objeto
de la simulacin.
XXX
N
FEBEX test zone
FEBEX drift
Laboratory tunnel
Main tunnel
Figura R7: Dominio de simulacin del modelo THM y nomenclatura para la frontera.
El problema se ha simulado en tres etapas:
 Equilibrio hidrolitosttico del macizo rocoso: en esta etapa no se consideran
las galeras, y se asumen condiciones de saturacin para los 365 m de roca
existentes sobre el dominio a modelar. Las cargas hidrostticas y litostticas
se han impuesto de manera gradual para la simulacin temporal. Los perfiles
de carga son funciones polinmicas y se describen en mayor detalle en el
captulo 6. La presin de fluido se calcula como presin relativa (PP
atm
) en
todos los anlisis.
 Simulacin de la excavacin de las galeras: se analiza la respuesta HM de la
roca durante la excavacin de las galeras, la cual se ha simulado haciendo
disminuir de manera gradual la presin de fluido y los esfuerzos normales en
las paredes de los tneles excavados hasta hacerlos nulos. En los perfiles de
descarga se ha utilizado el mismo tipo de funciones polinmicas que en la
simulacin anterior. Se han aplicado condiciones de contorno hidrulicas
similares a las existentes en el GTS.
 Simulacin del experimento de calentamiento: para esta simulacin se
emplea el modelo completo THM. Se ha simulado un proceso de
calentamiento de 3 aos en el interior de la parte final de la galera FEBEX
(ltimos 17 metros), en las mismas condiciones en las que se lleva a cabo el
experimento Insitu. La carga trmica aplicada corresponde a la existente
en dicho experimento, aunque se ha simulado como perfil de temperatura en
lugar de como potencia aplicada a los calentadores. La zona del ensayo est
rellena con bentonita.
Todas estas simulaciones han sido realizadas para tres tipos de condiciones de la roca:
caso istropo; caso anistropo / noorttropo homogneo y caso anistropo / no
orttropo heterogneo. El mallado de la simulacin tiene 11209 elementos, y el
problema completo THM tiene 37945 grados de libertad. Los elementos son de tipo
Lagrange cuadrtico para la parte mecnica y Lagrange lineal para las partes trmica e
hidrlulica.
XXXI
Equilibrio hidrolitosttico del macizo rocoso. En este problema se considera
nicamente un gradiente hidrulico vertical y la carga de 400 m de roca aplicada sobre
la galera FEBEX. Las condiciones de contorno se muestran en la tabla R1, y las
condiciones iniciales son P=0, u=0, v=0 y w=0.
Tabla R1: Condiciones de contorno del problema de equilibrio hidrolitosttico.
C.C. A1 A2 B1 B2 C1 C2
Trmicas      
Hidrulicas No flux No flux No flux No flux No flux P=365
w
g
Mecnicas
u=0m
u=0m
v=0m
v=0m
u=0m
v=0m
w=0m
33
=365
eq
g
El estado estacionario de esfuerzos verticales s
33
del modelo HM en la simulacin del
equilibrio hidrolitosttico se muestra en la figura R8. En dicha figura tambin se
muestra la forma deformada del dominio al final de la simulacin. Los esfuerzos
verticales del modelo HM ( max(s
33
) =1.596e7 Pa) son ligeramente inferiores a los
obenidos en el modelo puramente mecnico( max(s
33
) =1.676e7 Pa), debido al efecto
del acoplamiento de Biot.
Figura R8: Estado estacionario de los esfuerzos verticales s
33
tras el
equilibrio hidrolitosttico del macizo rocoso.
Simulacin de la excavacin de las galeras. Cuando el macizo rocoso ha alcanzado el
equilibrio hidrolitosttico, se simula la excavacin de las galeras, en la que las
condiciones hidrulicas son ms prximas a las del experimento insitu del FEBEX.
Las tablas R2 y R3 muestran la definicin de las condiciones de contorno y de las
condiciones iniciales y restricciones para este problema respectivamente.
XXXII
Tabla R2: Condiciones de contorno de la simulacin de la excavacin de las galeras.
C.C. A1 A2 B1 B2 C1 C2
Trmicas      
Hidrulicas
No
flux
No
flux
P=2.1
MPa
P=0.7
MPa
( )
100
200
7 . 0
y
P
=
( )
100
200
7 . 0
y
P
=
Mecnicas
u=0m
u=0m
v=0m
v=0m
u=0m
v=0m
w=0m
33
=365
eq
g
Tabla R3: Condiciones iniciales y restricciones de la simulacin de la excavacin de
las galeras.
Restricciones
C.I.
Zonas excavadas Zona de ensayo
Trmicas   
Hidrulicas
( )
100
200
7 . 0
y
P
= MPa
P=0 Pa P=0 Pa
Mecnicas u=0 m, v=0 m, w=0 m n
i
ii
=0 n
i
ii
=0
Las condiciones iniciales de altura piezomtrica vienen dadas por el rgimen de flujo
regional (montaa J chlistock y ro Aare), en las que existen gradiantes elevados tanto
horizontal como verticalmente, debido a las caractersticas montaosas y de baja
permeabilidad de la zona. La figura R9 muestra el estado estacionario de una seccin
horizontal del dominio a cota z =0 m, junto con los valores de altura piezomtrica
medidos en campo antes de la excavacin. Los resultados obtenidos por las
simulaciones de la Universidad Politcnica de Catalua (UPC) (figuras 3.12 y 3.13 de
[34]) son similares, con pequeas diferencias en las irregularidades locales alrededor de
la galera, que no aparecen en este modelo debido al proceso de homogeneizacin del
dominio.
a.
Figura R9: Isolneas de la altura piezomtrica en el estado estacionario de la
simulacin de la excavacin de las galeras en la seccin horizontal a cota z=0.
XXXIII
El modelo HM completo obtiene el estado estacionario que se muestra en la figura R
10. En dicha figura se muestran los esfuerzos verticales s
33
y las isosuperficies de
desplazamiento vertical w para condiciones heterogneas anistropas / noorttropas del
material. Puede observarse cmo la consolidacin sufrida en la zona de ensayo es
mayor, debido a la mayor densidad de fracturacin existente en esa zona.
b.
Figura R10: Esfuerzo vertical s
33
e isosuperficies de desplazamiento vertical w
en el estado estacionario de la simulacin de la excavacin de las galeras.
Simulacin del experimento de calentamiento. La simulacin del calentamiento
se ha llevado a cabo en dos condiciones distintas: con las galeras rellenas de un
material con propiedades equivalentes a las de la roca y con las galeras excavadas. En
ambos casos, la zona de ensayo est rellena con bentonita. Slo se presentan resultados
correspondientes al segundo caso. Las condiciones de contorno e iniciales y las
restricciones son las que se muestran en las tablas R4 y R5 respectivamente.
Tabla R4: Condiciones de contorno de la simulacin del experimento de calentamiento.
C.C. A1 A2 B1 B2 C1 C2
Trmicas T=13 C T=13 C T=13 C T=13 C T=13 C T=13 C
Hidrulicas No flux No flux No flux No flux No flux P=365
w
g
Mecnicas u=0m u=0m v=0m v=0m u=0m
v=0m
w=0m
33
=365
eq
g
Tabla R5: Condiciones iniciales y restricciones de la simulacin del experimento de
calentamiento.
Restricciones
C. I.
Zonas excavadas Zona de ensayo
Trmicas T=13 C 
2
2
14 . 1
65
100
r
T
= C
Hidrulicas Hmodel steady state  
Mecnicas Mmodel steady state  
XXXIV
El estado final del campo de temperatura se muestra en la figura R11. La densidad y la
viscosidad del agua varan con la temperatura en todas las simulaciones. La disminucin
de densidad producida en las zonas ms calientes del ensayo es de alrededor del 3% de
su valor de referencia. Por otro lado, el pico de esfuerzos trmicos es mayor en
condiciones noorttropas (10 MPa) que en condiciones istropas (6 MPa). En
cualquier caso, los esfuerzos trmicos simulados son siempre menores que los
producidos por la excavacin de las galeras. La figura R12 muestra el estado final del
campo de esfuerzos de Von Mises junto con las isosuperficies de altura piezomtrica y
la deformacin del dominio tras la simulacin.
a.
Figura R11: Seccin vertical por el eje de la galera FEBEX del campo de
temperaturas en el estado final de la simulacin del experimento de calentamiento.
Figura R12: Estado final de los esfuerzos de Von Mises, isosuperficies de
altura piezomtrica y dominio deformado en la simulacin del experimento
de calentamiento.
XXXV
Como el inters de estas simulaciones se centra en el medio fracturado, no hemos
comparado nuestros modelos con medidas en la bentonita. Sin embargo, presentamos a
continuacin algunas comparaciones de la evolucin temporal de temperatura, presin
de fluido, presin total y desplazamientos totales de la simulacin con las medidas en
los sondeos radiales de instrumentacin excavados en la zona de ensayo. Los puntos de
muestreo seleccionados son los que se muestran en la figura R13.
Figura R13: Sondeos y puntos de muestreo seleccionados para la
comparacin de temperaturas (en rojo), presin de fluido (en azul),
presin total (en verde) y desplazamiento total (en naranja) medidos y
simulados en el experimento de calentamiento (figura original de [33]).
La figura R14a muestra la comparacin de la temperatura simulada y medida en el
sondeo SF23. El punto ms cercano a la bentonita (temperatura ms alta) presenta la
mayor discrepancia entre el modelo y la medida, siendo el ajuste de los otros tres puntos
aceptable. Esta subestimacin de la temperatura en las zonas de la roca ms prximas a
los calentadores puede ser debida a las condiciones trmicas impuestas para simular el
calentamiento: se ha impuesto un perfil de temperatura en la bentonita, en lugar de
imponer la curva de potencia de los calentadores. La figura R14b presenta la evolucin
temporal de la presin de fluido (presin intersticial) en el sondeo SK1. En general se
aprecia una sobreestimacin de la presin de fluido, debida probablemente a que se han
impuesto condiciones de saturacin en los 400m de roca existentes sobre la galera
FEBEX. As mismo, la mayor presin inicial en la simulacin es debida a que se ha
comenzado con un estado estacionario de equilibrio hidromecnico tras la excavacin
de las galeras y posterior rellenado de bentonita de la zona de ensayo, no existente en
las condiciones reales del ensayo. No obstante, la curva de presin en la parte inicial del
transitorio producida por la expansin del agua se observa tanto en la simulacin como
en las medidas, y es por tanto captada por el modelo. Por ltimo, la presin total
simulada (esfuerzos de Von Mises) y medida en el sondeo SG1 se muestra en la figura
R14c. Las mismas observaciones realizadas para la presin de fluido con respecto a las
condiciones iniciales de equilibrio HM pueden ser aplicadas en el caso de la presin
total. Por otro lado, el pico de esfuerzos trmicos ocurre antes en la simulacin que en
las medidas, y casi simultneamente en todos los puntos, lo que puede ser debido a una
sobreestimacin del coeficiente de conductividad trmica.
XXXVI
Figura R14: Evolucin temporal de los datos medidos (x) y simulados () de: a.
temperaturas en el sondeo SF23; b. presin de fluido en el sondeo SK1; y c. presin
total en el sondeo SG1 en la simulacin del experimento de calentamiento.
XXXVII
Conclusiones y tareas futuras. En la primera parte de la tesis se ha aplicado una
metodologa integrada de anlisis de series temporales provinientes del ensayo en
maqueta del FEBEX, combinando las distintas tcnicas existentes en los dominios
temporal, espacial, frecuencial y de escala, con objeto de caracterizar los principales
procesos termohidromecnicos existentes en el ensayo y el funcionamiento de los
sensores. Se han presentado los resultados ms relevantes, publicados tambin en [22]
(el artculo completo se incluye en el anexo XIII (APPENDIX XIII). Se ha ofrecido una
hiptesis de la existencia de clulas de evaporacincondensacin para explicar la
disminucin de flujo de agua entrante y de humedad relativa observadas en la bentonita.
Por otro lado, el sobrecalentamiento ocurrido en la maqueta no ha causado daos o
perturbaciones irreversibles ni en el transcurso del experimento ni en el funcionamiento
de la mayora de los sensores instalados. Por ltimo, algunos sensores de presin total
cuyos datos parecan indicar un funcionamiento errneo han resultado medir
correctamente, pero la presin de fluido en lugar de la presin total por una falta de
conectividad con la bentonita. A pesar de que las tcnicas utilizadas han proporcionado
una mayor comprensin de los procesos acoplados existentes en el experimento, es
necesario profundizar en la conexin de estos resultados con las tareas de modelizacin.
En la segunda parte de la tesis se ha desarrollado una metodologa para simular un
medio fracturado 3D a partir de datos de campo. Se ha obtenido un ajuste razonable
entre la simulacin y las medidas disponibles. Esta metodologa proporciona un buen
punto de partida para el uso de mapas de trazas observados en paredes de galeras
cilndricas, frente al uso clsico en la literatura [48][91] de mapas de trazas rectilneas
en paredes planas. Por otro lado, se pueden realizar comentarios para mejoras futuras en
la simulacin: 1) podra generalizarse la heterogeneidad local de la galera a todo el
dominio de generacin, mediante el uso de procesos de Poisson no homogneos [84],
estableciendo una densidad de fracturacin para cada punto del dominio 3D (como por
ejemplo la funcin del momento de segundo orden reducido definida en [42]); 2) por
otro lado, debido al carcter estocstico del proceso de optimizacin, se debera realizar
un promedio de varias generaciones para calcular la funcin objetivo acorde con los
intervalos de confianza requeridos en la misma; 3) por ltimo, podran utilizarse los
ensayos hidrulicos realizados en la zona de la galera FEBEX existentes en la literatura
[40][41] para realizar simulaciones condicionadas hidrulicamente.
En la tercera parte de la tesis se ha desarrollado un modelo termohidromecnico en
medio continuo, y se ha definido una metodologa de homogeneizacin para estimar los
parmetros equivalentes del medio fracturado a introducir en el modelo. Se han
realizado diversas simulaciones del experimento insitu del FEBEX con dicho
modelo, y se han obtenido ajustes razonables en las principales variables observadas. Se
pueden hacer, sin embargo, algunos comentarios para tareas futuras: 1) la funcin
objetivo del proceso de simulacin del medio fracturado debera incorporar tambin
comparaciones de los resultados finales del modelo con los datos medidos, aunque este
proceso conllevara una carga computacional considerable; 2) se pueden aadir nuevos
acoplamientos y generalizaciones al modelo, como son condiciones no saturadas,
contacto bentonitaroca, comportamiento elastoplstico, produccin de nuevas fracturas,
dependencia entre la apertura de las fracturas y el campo de tensiones, etc; 3) la tcnica
de homogeneizacin definida para estimar la conductividad hidrulica (de primer
orden en las condiciones de contorno) se podra aplicar igualmente para los parmetros
mecnicos y trmicos.
XXXVIII
Por ltimo, y como conclusin final, podemos decir que la presente tesis ha desarrollado
una metodologa integrada para analizar y modelar procesos acoplados en medios
fracturados tridimensionales, con contribuciones especialemente relevantes en la
simulacin del medio fracturado mediante el uso de mapas de trazas en paredes
cilndricas y en la homogeneizacin de la conductividad hidrulica para un medio
poroso fracturado.
La lista de referencias bibliogrficas completa puede consultarse en el captulo 8 de la
tesis.
XXXIX
XL
TABLE OF CONTENTS
Acknowledgements. XI
Abstract XIII
Rsum XV
Resumen.. XVII
Resumen extendido.. XIX
Table of Contents..... XLI
List of Figures XLV
List of Tables... LI
List of Abbreviations Acronyms..... LIII
List of Symbols Nomenclature...... LV
1. Introduction. . 1
2. Description of the FEBEX Project. ...... 3
2.1. Generalities. ... 3
2.2. Insitu Experiment. 3
2.3. Mockup Test. ... 5
3. Time Series Analysis of the Mockup Test Data... 9
3.1. Description of the Data. . 9
3.2. Analysis Methodologies. ... 10
3.2.1. Correlation and Spectral Analysis. ..... 10
3.2.1.1. Simple Analysis. ...... 11
3.2.1.2. Cross Analysis. .... 12
3.2.2. Wavelets Analysis. ..... 13
3.2.2.1. Continuous Wavelet Transform. .. 13
3.2.2.2. Discrete Wavelet Transform. ... 14
3.2.2.3. Multiresolution analysis. 14
XLI
3.2.3. Matching Pursuit. 15
3.2.4. Time Evolution of Statistical Parameters. .. 16
3.3. Results of the Statistical Analysis and Discussion. ... 17
3.3.1. Physical Processes Identification. ... 17
3.3.2. Unexpected Events. .... 21
3.3.3. Sensors Reliability. ..... 25
4. Geomorphological Simulation and Reconstruction of the 3D Fractured Rock. 27
4.1. GeoMorphological Data. .. 27
4.1.1. Geology, Tunnel and Boreholes. 27
4.1.2. Fractured Network Data. 28
4.2. Reconstruction of the Fractured Medium. . 31
4.2.1. Statistical Distributions of the Fractured Network. 31
4.2.2. Optimization Methodology. 32
4.2.3. Main Fractures. ... 34
4.2.4. Nonuniform Tracemap Reproduction. .. 36
4.2.5. Optimized Fractured Medium. ... 37
4.2.6. Fracture Apertures Adjustment. . 41
5. ThermoHydroMechanical Model. .. 43
5.1. Introduction, Coupling and Upscaling. 43
5.2. Basic Assumptions and Constitutive Equations. .. 45
5.2.1. Dimensionality and Geometry. .. 45
5.2.2. Thermal Processes. . 45
5.2.3. HydroMechanical Processes. 45
5.2.4. Macroscale Constitutive Laws and Equations. .. 45
5.2.4.1. Governing Laws. . 45
5.2.4.2. Constitutive Equations. 47
5.2.5. System of Equations. .. 48
5.3. Equivalent Continuum Properties. 49
5.3.1. Introduction and Generalities. 49
5.3.2. Hydraulic Equivalent Coefficients. 49
5.3.2.1. Upscaled Conductivity of Individual Fractured Blocks...... 49
5.3.2.2. Domain Upscaling: Superposition Approach for Discharge Rates. 58
5.3.3. Mechanic and HydroMechanic Equivalent Coefficients. . 64
XLII
5.3.4. Implementation and Results of the Upscaling. .. 67
5.3.4.1. REV Study and Moving Average. ... 67
5.3.4.2. Oneblock Homogenization. 69
5.3.4.3. Moving Average Homogenization... 72
6. Implementation and Results of the THM Model. . 79
6.1. Domain and Problem Definition. .. 79
6.2. HydroLithostatic Equilibrium of the Rock Mass. 86
6.3. Drifts Excavation Simulation. ... 91
6.4. Heating Experiment Simulation. ... 96
7. Conclusions and Future Work... 115
8. References. ... 119
9. Appendices 127
APPENDIX I: Fractal Characterization of the FEBEX Tracemap 129
APPENDIX II: Orientation Angles for a Planar Fracture in 3D Space. 131
APPENDIX III: Intersection of a Circular Fracture with a Cylindrical Tunnel 133
APPENDIX IV: Detailed Results of the Fractured Medium Optimization... 137
APPENDIX V: PseudoSpectral Method for the 1D AdvectionDiffusion Equation.. 141
APPENDIX VI: DualContinuum Model for Fractured Rock (Illustrative
Examples).. 149
APPENDIX VII: Temperature Dependence of Water Viscosity.. 153
APPENDIX VIII: Matricial Form of the 2
nd
and 4
th
rank tensor equations.. 155
APPENDIX IX: Upscaling the Basic Fractured Block Flux Density by the Method
of Vectorial Surface Flux... 161
APPENDIX X: Solid rotations and their matrix representation in 2D and 3D .... 163
APPENDIX XI: Full Results of the Fractured Medium THM Upscaling.. 167
APPENDIX XII: Comsol MultiphysicsReport of the THM Simulations... 171
APPENDIX XIII: Full article of the reference [22] (Preprint).. 181
APPENDIX XIV: Full article of the reference [3] (Preprint).... 203
APPENDIX XV: Full article of the reference [23] (Preprint)... 211
XLIII
XLIV
LIST OF FIGURES
Spanish extended abstract figures
Figura R1: Anlisis de la evolucin del campo de humedad relativa en la seccin
vertical longitudinal del ensayo en Maqueta.
Figura R2: a. Temperatura de la bentonita (sensor T_A5_1_1) durante el incidente de
sobrecalentamiento (figura superior) y b. Reconstruccin de la componente de ruido de
la seal (figura inferior).
Figura R3: Situacin de la galera FEBEX en el Grimsel Test Site (de [73]) y dominio
de simulacin del medio fracturado.
Figura R4: a. Medio fracturado simulado; b. Mapa de trazas medido en la galera
FEBEX; c. Mapa de trazas simulado en la galera FEBEX.
Figura R5: Principales procesos acoplados en un sistema termohidromecnico.
Figura R6: a. Bloque fracturado individual de un medio poroso fracturado; b.
Condiciones de contorno lineales a trozos para la altura piezomtrica H.
Figura R7: Dominio de simulacin del modelo THM y nomenclatura para la frontera.
Figura R8: Estado estacionario de los esfuerzos verticales s
33
tras el equilibrio hidro
litosttico del macizo rocoso.
Figura R9: Isolneas de la altura piezomtrica en el estado estacionario de la
simulacin de la excavacin de las galeras en la seccin horizontal a cota z=0.
Figura R10: Esfuerzo vertical s
33
e isosuperficies de desplazamiento vertical w en el
estado estacionario de la simulacin de la excavacin de las galeras.
Figura R11: Seccin vertical por el eje de la galera FEBEX del campo de
temperaturas en el estado final de la simulacin del experimento de calentamiento.
Figura R12: Estado final de los esfuerzos de Von Mises, isosuperficies de altura
piezomtrica y dominio deformado en la simulacin del experimento de calentamiento.
Figura R13: Sondeos y puntos de muestreo seleccionados para la comparacin de
temperaturas (en rojo), presin de fluido (en azul), presin total (en verde) y
desplazamiento total (en naranja) medidos y simulados en el experimento de
calentamiento (figura original de [33]).
Figura R14: Evolucin temporal de los datos medidos (x) y simulados () de: a.
temperaturas en el sondeo SF23; b. presin de fluido en el sondeo SK1; y c. presin
total en el sondeo SG1 en la simulacin del experimento de calentamiento.
Main text figures
Figure 1: General layout of the Insitu experiment of the FEBEX project.
Figure 2: General layout of the mockup experiment at CIEMAT.
Figure 3: Distribution of the instrumentation sections, levels and angular positions in
the Mockup test.
Figure 4: Functions available in the Correlation and Spectral Analysis.
XLV
Figure 5: Study of the evolution of statistical parameters by moving window.
Figure 6: Evolution analysis of the spatial distribution of the data for the relative
humidity sensors in the Mockup.
Figure 7: Crosscorrelation between the bentonite temperature sensors of section A2
and the relative humidity sensors of section A3 (time period analyzed: 1997 data).
Figure 8: Evolution of the data of relative humidity sensor V_A3_4 (upper figure) and
Matching Pursuit analysis of the time series (lower figure).
Figure 9: Evolution of the bentonite temperature sensors of section A5 in the Mockup
experiment (plotted period: 28/12/9926/9/01).
Figure 10: Temperature of the bentonite (sensor T_A5_1_1) before (a.) and after (b.)
the overheating incident (upper figures) and Multiresolution Analysis (lower figures).
Figure 11: a. Temperature of the bentonite (sensor T_A5_1_1) during the overheating
incident (upper figure) and b. Reconstruction of the noise component of the signal
(lower figure).
Figure 12: Total pressure (sensor PT_A6_3) during the overheating incident (upper
figure) and Continuous Wavelet Transform analysis of the signal (lower figure).
Figure 13: Evolution of the autocorrelation function of the bentonite temperature
sensors of section A5 in the Mockup experiment (analysed period: 28/12/9926/9/01).
Figure 14: Simple correlograms of total pressure sensors (upper figures) and fluid
pressure sensors (lower figures) showing similarities in their behaviour (time period
analyzed: 1997 year data).
Figure 15: Alpine structures in the Central Aar Massif according to [82].
Figure 16: Location of the FEBEX drift within the GTS general layout (from [73]) and
fractured medium generation domain.
Figure 17: Pole diagram of the fractures in boreholes FBX95001 y FBX95002.
Figure 18: Map of traces on the wall of the FEBEX drift, divided into five different
zones according to their geological features [73].
Figure 19: a. Cumulative histogram of trace length of the FEBEX drift tracemap; and
b. Cumulative histogram of 3D trace chord of the FEBEX drift tracemap.
Figure 20: Families classification of the fracture data of boreholes FEBEX95001 and
FEBEX95002.
Figure 21: Fracture aperture frequency in the GTS tunnel.
Figure 22: Geometric relations of the 2D trace (a.) to infere the 3D dip and plunge of a
single fracture (b.) from the trace map.
Figure 23: Pole diagram of the large discrete fractures of the FEBEX drift.
Figure 24: Comparison of the FEBEX traces map (upper figure) with the traces of the
simulated big fractures (lower figure).
Figure 25: Algorithm of the optimization process to simulate the fractured medium.
Figure 26: a. Evolution of the objective function by averaging 2 realizations of the
generation algorithm to get each value of the objective function. b. Evolution of the
objective function for 750 realizations with the optimum parameter values.
XLVI
Figure 27: a. Cumulated distribution function of trace lengths on tunnel ( observed; 
 fitted); b. Cumulated distribution function of chord lengths on tunnel ( observed; 
 fitted); c. FEBEX drift observed tracemap ; d. FEBEX drift fitted tracemap.
Figure 28: Whole view of the reconstructed fractured medium with 2906474 fractures.
Figure 29: Fraction of the reconstructed fractured medium inside the domain.
Figure 30: Evolution of the OF in the apertures adjustment.
Figure 31: Coupled processes in a thermohydromechanical system.
Figure 32: Individual fractured block of a fractured porous medium.
Figure 33: Piecewise linear B.C. for the individual fractured block.
Figure 34: possible prismatic configurations for a valid fractured block fulfilling eq.
(79).
Figure 35: Example of the A
FLOW
matrix: projection of the outgoingflux surface of the
block in the normal plane to each component direction of the flux.
Figure 36: Results of the global discharge rate (eq. 77) and the equivalent hydraulic
conductivity (eq. 81) for some particular cases.
Figure 37: REV determination for K
ij
in the simulated fractured medium.
Figure 38: Algorithm of the upscaling process.
(*)
The algorithm of fracture intersections
with the homogenization subdomain is showed in the next figure.
Figure 39: Algorithm of fractured medium intersections with the homogenization
subdomain.
Figure 40: Equivalent intrinsic permeability ellipsoid for the oneblock homogenization
of the fractured medium.
Figure 41: Equivalent reduced stiffness tensor ellipsoid for the oneblock
homogenization of the fractured medium.
Figure 42: Equivalent Biot coefficient ellipsoid for the oneblock homogenization of
the fractured medium.
Figure 43: Hydraulic and hydromechanic equivalent coefficients for the moving
average homogenization. a. Equivalent intrinsic permeability k
ij
; b. Equivalent stiffness
tensor T
ijkl
(only T
ij
with i,j=1,2,3); c. Equivalent Biot coefficient B
ij
.
Figure 44: Equivalent intrinsic permeability for the five Xlayers of the moving
average.
Figure 45: Equivalent reduced stiffness tensor for the five Xlayers of the moving
average.
Figure 46: Equivalent Biot coefficient for the five Xlayers of the moving average.
Figure 47: Equivalent volumetric fracture density for the five Xlayers of the moving
average.
Figure 48: Domain of the THM model and boundaries nomenclature.
Figure 49: Two heating profiles for the experiment simulations: a. Exponential
function; b. Polynomial function.
XLVII
Figure 50: Comparison between the effects of a hydraulic load at the top boundary in
the bottom boundary for two different loading times.
Figure 51: Meshgrids used in the THM model.
Figure 52: Crosssectional features to show output results of the models.
Figure 53: Fluid pressure field in the AA (left side) and the BB (right side) cross
sections for the time t = 9.5e5s 11 days: a. Isotropic/homogeneous anisotropic
conditions; b. Heterogeneous anisotropic conditions.
Figure 54: Steady state of the s
33
stress field in the mechanical model with three
different rock mass stiffness conditions: a. Isotropic conditions; b. Nonorthotropic
homogeneous conditions; c. Nonorthotropic heterogeneous conditions.
Figure 55: Steady state fluid pressure field after hydrolithostatic equilibrium of the
rock mass.
Figure 56: Steady state vertical stress s
33
field after hydrolithostatic equilibrium of the
rock mass.
Figure 57: Time evolution of the fluid pressure (a. and c.) and vertical stress (b. and d.)
at the point R and through the crossline LL respectively.
Figure 58: Crosssection AA of the temperature field in the steady state (a.) and a
closer detailed view of the test zone (b.).
Figure 59: Time evolution of the fluid pressure in the drifts excavation simulation. Four
time instants are showed: a. Time t = 0 years; b. Time t = 22 days; c. Time t = 45 days;
d. Time t = 3.17 years.
Figure 60: Hydraulic head isolines at steady state: a. Horizontal cross section at z=0;
b.: Vertical cross section AA.
Figure 61: Vertical stress steady state for the isotropic mechanic model.
Figure 62: Vertical stress s
33
and vertical displacement isosurfaces steady state for the
HM drifts excavation simulation: a. Homogeneous anisotropic/nonorthotropic
conditions; b. Heterogeneous anisotropic/nonorthotropic conditions.
Figure 63: Fluid pressure and water flow lines steady state for the HM drifts excavation
simulation with heterogeneous anisotropic/nonorthotropic conditions.
Figure 64: Crosssection AA of the temperature field in the final state (a.) and a closer
detailed view of the test zone (b.).
Figure 65: Time evolution in the point R (lefthand side) and in the vertical crossline
LL (righthand side): (a.) and (b.) temperature; (c.) and (d.) water density.
Figure 66: Vertical stresses s
33
in the crossline LL for different rock conditions: a.
isotropic stiffness tensor; b. homogeneous nonorthotropic stiffness tensor.
Figure 67: Final state of the fluid pressure. Flow at z=0 is also showed (only horizontal
components).
Figure 68: Final state of the Von Mises stresses, hydraulic head isolevels and deformed
shape of the domain.
Figure 69: Final state of the displacements u (a.), v (b.) and w (c.).
XLVIII
Figure 70: Detailed view of the THM final state of temperature (a.), fluid pressure (b.)
and Von Mises stress (c.).
Figure 71: Detailed view of the THM final state of normal stresses s
11
(a.), s
22
(b.) and
s
33
(c.) and shear stresses s
23
(d.), s
13
(e.) and s
12
(f.).
Figure 72: Detailed view of the THM final state of normal strains
11
(a.),
22
(b.) and
33
(c.) and shear strains
23
(d.),
13
(e.) and
12
(f.).
Figure 73: Detailed view of the THM final state of displacements u (a.), v (b.)
and w (c.).
Figure 74: Selected boreholes and sampling points for the temperature (red), intersticial
pressure (blue) total pressure (green) and total displacements (orange) comparisons in
the THM analysis (original figure from [33]).
Figure 75: Time evolution of measured (x) and simulated () temperatures in boreholes
SF23 (a.), SF14 (b.) and SB22 (c.) for the THM analysis.
Figure 76: Time evolution of measured (x) and simulated () fluid pressure in boreholes
SK1 (a.) and SJ 5 (b.).
Figure 77: Time evolution of measured (x) and simulated () total pressure in borehole
SG1: a. Points from 01 to 05; b. Points from 06 to 10.
Figure 78: Time evolution of measured (x) and simulated () total pressure in borehole
SG2: a. Points from 01 to 05; b. Points from 06 to 10.
Figure 79: Time evolution of measured (x) and simulated () total displacements in
boreholes SI1 (a.) and SI2 (b.).
APPENDICES figures
Figure A1: Algorithm to estimate the fractal dimension of the FEBEX fractured area:
a. Original image of the traces map of FEBEX drift; b. Bidimensional MRA of the
image; c. Modulus, phase and 95% of highest modulus values; d. Estimation of fractal
dimension from the modulus.
Figure A2: Fractal dimension estimation for the five different zones of the FEBEX
drift.
Figure A3: Angles criteria for the 3D planar fractures used in the thesis.
Figure A4: a. Disk fracture intersection with a cylindrical tunnel in 3D and b. Trace
formed in the tunnel wall developed in 2D.
Figure A5: Different types of intersections between a disk fracture and a cylindrical
tunnel in 3D depending on the number of solutions of the equations system: a.
Uncomplete trace (two extreme points); b. Uncompleted trace (four extreme points); c.
Complete trace (zero extreme points); and d. No trace (zero extreme points).
Figure A6: Evolution of the objective function (OF) in the first step of the
optimization process.
Figure A7: First step optimization: a. Cumulated distribution function of trace lengths
on tunnel ( observed;  fitted); b. Cumulated distribution function of chord lengths
on tunnel ( observed;  fitted); c. FEBEX drift observed tracemap; d. FEBEX drift
fitted tracemap; e. Observed tracemap detail; and f. Fitted tracemap detail.
XLIX
Figure A8: Evolution of the objective function (OF) in the second step of the
optimization process.
Figure A9: Second step optimization: a. Cumulated distribution function of trace
lengths on tunnel ( observed;  fitted); b. Cumulated distribution function of chord
lengths on tunnel ( observed;  fitted); c. FEBEX drift observed tracemap; d.
FEBEX drift fitted tracemap; e. Observed tracemap detail; and f. Fitted tracemap detail.
Figure A10: Discretization of a function f(x)R in Z.
Figure A11: Time evolution of the advectiondiffusion equation for a triangular
function with constant coefficents.
Figure A12: Time evolution of the advection equation for a sinusoidal function with
space dependent coefficient.
Figure A13: Time evolution of the diffusion equation for a triangular function with
time and space dependent coefficient.
Figure A14: Time evolution of the advectiondiffusion equation for a complex
function with time and space dependent coefficients.
Figure A15: Time evolution of the rock matrix fluid pressure for an example of the
dualcontinuum model in a 2D fractured medium.
Figure A16: Time evolution of the fractures fluid pressure for an example of the dual
continuum model in a 2D fractured medium.
Figure A17: Steady state of the fractures fluid pressure for an example of the dual
continuum model in a 3D fractured medium.
Figure A18: Water dynamic viscosity values (xmarked points) and fitted polynomial
(solid line) used in the models.
Figure A19: Anticlockwise rotation of degrees in 2D.
Figure A20: Relation between the anticlockwise rotation of degrees and the fracture
normal vector in 2D.
Figure A21: Anticlockwise rotation of degrees over the X
3
axis followed by a
clockwise rotation of degrees over the X
2
axis in 3D.
Figure A22: Relation between the anticlockwise rotation of degrees over the X
3
axis
followed by a clockwise rotation of degrees over the X
2
axis and the fracture normal
vector in 3D.
L
LIST OF TABLES
Spanish extended abstract tables
Tabla R1: Condiciones de contorno del problema de equilibrio hidrolitosttico.
Tabla R2: Condiciones de contorno de la simulacin de la excavacin de las galeras.
Tabla R3: Condiciones iniciales y restricciones de la simulacin de la excavacin de
las galeras.
Tabla R4: Condiciones de contorno de la simulacin del experimento de
calentamiento.
Tabla R5: Condiciones iniciales y restricciones de la simulacin del experimento de
calentamiento.
Main text tables
Table 1: Sensors installed in the Mockup test and associated parameters.
Table 2: Distribution of fractures in borehole FEBEX95001.
Table 3: Distribution of fractures in borehole FEBEX95002.
Table 4: Parameters of the orientation distributions for each family (dip and plunge,
both in degrees).
Table 5: Aperture frequencies according to the fracture family (qualitative
classification).
Table 6: Main characteristics of the fixed fractures of the simulated network.
Table 7: Number of fractures and intersections of the fractured medium generated by
the optimization process.
Table 8: Numerical experiments plan.
Table 9: Boundary conditions of the hydrolithostatic equilibrium simulation.
Table 10: Initial conditions and constrains of the hydrolithostatic equilibrium
simulation.
Table 11: Boundary conditions of the drifts excavation simulation.
Table 12: Initial conditions and constrains of the drifts excavation simulation.
Table 13: Boundary conditions of the heating experiment simulation.
Table 14: Initial conditions and constrains of the heating experiment simulation.
APPENDICES tables
Table A1: Main characteristics of the optimum fractured medium obtained in the first
step of the optimization process.
Table A2: Main characteristics of the optimum fractured medium obtained in the
second step of the optimization process and comparison with the measured values.
LI
Table A3: B.C. and I.C. for the 3D example of the dualcontinuum model.
Table A4: Excel dataset and polynomialfunction fitted values of water dynamic
viscosity for the temperature interval [0, 100].
Table A5: Kelvin notation for the 2
nd
rank stress tensor.
Table A6: Kelvin notation for the 2
nd
rank strain tensor.
Table A7: Kelvin notation for the 4
th
rank stiffness tensor (only the first row showed as
an example).
Table A8: Kelvin notation for the 2
nd
rank Biot coefficient tensor.
Table A9: Kelvin notation for the 2
nd
rank intrinsic permeability tensor.
LII
LIST OF ABBREVIATIONS ACRONYMS
1D, 2D, 3D Onedimensional, twodimensional or threedimensional respectively.
AA Crosssection longitudinal to the FEBEX axis (see Figure 52).
AGP Almacenamiento Geolgico Profundo (Deep Geologic Disposal).
BB Crosssection transversal to the FEBEX axis (see Figure 52).
BMT3 BenchMark Test phase 3.
CIEMAT Centro de Investigaciones Energticas, MedioAmbientales y Tecnolgicas.
COMSOLMultiphysics, Finite Element Method software for coupled modeling.
CWT Continuous Wavelet Transform.
DECOVALEX
DWT Discrete Wavelet Transform.
ENRESA Empresa Nacional de REsiduos radiactivos S.A.
EURATOM
FEBEX Fullscale Engineered Barrier EXperiment in crystalline hostrock.
FEMLABOld COMSOL Multiphysics.
GTS Grimsel Test Site.
HM, HM HydroMechanical.
INPT Institut National Polytechnique de Toulouse.
LL Vertical crossline passing through the FEBEX gallery (see Figure 52).
MATLABScientific software.
MRA MultiResolution Analysis.
NAGRA
OF Objective Function
PDE Partial Differential Equation.
REV Representative Elementary Volume.
SB22, SF14, SF23, SG1, SG2, SI1, SI2, SJ 5, SK1, SK2 Radial boreholes of the
FEBEX drift (see Figure 74).
THM, THM ThermoHydroMechanical.
TM, TM ThermoMechanical.
UMFPACK Direct solver of COMSOL Multiphysicsused in the simulations.
UPM Universidad Politcnica de Madrid.
VAF Volume Averaged Flux.
VSF Vectorial Surface Flux.
XYZ Reference coordinated system.
LIII
LIV
LIST OF SYMBOLS NOMENCLATURE
Chapter 3: Analysis of the mockup test data
{x
i
} Time series (i=1,2,,n, with n the length of the series).
x Mean value of the time series {x
i
}.
k Time difference (or lag) between two values of the time series being
compared.
r
k
Autocorrelation function for a time lag k.
m Truncature of the correlation analysis (k varies from 0 to m).
C
0
Spectral variance of the time series.
S
F
Spectral density function for a frequency coefficient F.
j Imaginary number.
D
k
Filtering window for the spectral analysis.
r
+k
, r
x,y
(k) Crosscorrelation function between series {x
i
} and {y
i
} for positive
values of k.
r
k
, r
y,x
(k) Crosscorrelation function between series {x
i
} and {y
i
} for negative
values of k.
S
x
2
, S
y
2
Spectral variances of series {x
i
} and {y
i
} respectively.
C
x,y
(k), C
y,x
(k) Spectral covariances between series {x
i
} and {y
i
} for a lag k.
a,b
Wavelet function of dilation a and translation b.
C
f
(a,b) Wavelet coefficients of the continuous wavelet transform of function f
with wavelet
a,b
.
c
jk
Wavelet coefficients of the discrete wavelet transform of function f with
wavelet
jk
.
V
m
Multirresolution analysis (set of spaces of L
2
(R) successively enclosed
verifying some properties)
O
m
Set of spaces of L
2
(R) orthogonal to V
m
.
mk
Smoothing function (to build orthogonal projection of f in V
m
).
mk
Wavelet function (to build orthogonal projection of f in O
m
).
D Redundant dictionary of oscillatory functions (atoms) for the Matching
pursuit analysis.
k
v
(t), k
a,b,
Gabor function of dilation a, translation b and modulation .
k
v0
(t) Atom (or vector) taken from the dictionary D.
R(t) Residual vector of the Matching pursuit analysis of function (t).
Standard deviation of time series {x
i
}.
Chapter 4: Simulation of the 3D fractured medium
S
1
, S
2
, S
3
, S
4
/K
4
, K
2
/L, K
1
, K
3
, S
5
, ZK
1
, and ZK
2
Alpine structures in the Central
Aar Massif according to [82].
x
cf
, y
cf
, z
cf
Coordinates of the center of fracture f in the absolute reference system.
x
ct
, y
ct
, z
ct
Coordinates of the center of fracture f in the FEBEX tunnel local
reference system.
R
min
Minimum radius of the fracture size Pareto distribution.
R
max
Maximum radius of the fracture size Pareto distribution.
b Exponential coefficient of the fracture size Pareto distribution.
g() Histogram of trace lengths of the simulated fractured medium in the
FEBEX drift wall.
LV
h() Histogram of 3D trace chords of the simulated fractured medium in the
FEBEX drift wall.
nbins Resolution or number of elements of the histograms defined above.
ntrtun() Number of traces (intersections) produced in the FEBEX drift.
ntrb
1
() Number of traces (intersections) produced in borehole FEBEX95001.
ntrb
2
() Number of traces (intersections) produced in borehole FEBEX95002.
M Penalization factor in the objective function of the fractured medium
optimization.
D Diameter of the FEBEX drift.
m Distance between the intersections of the trace with the right and left wall
lines of the drift measured in the 2D tracemap.
l Distance between the intersections of the trace with the floor and roof
lines of the drift measured in the 2D tracemap.
dip of the fracture according to APPENDIX II angles criteria.
plunge of the fracture according to APPENDIX II angles criteria.
p
21
Areal fracture density (trace length / intersecting plane surface).
p
32
Volumetric fracture density (fracture intersecting surface / intersecting
object volume).
maxp
21
(.) Maximum (measured) p
21
for each zone of the FEBEX drift wall.
N Number of fractures of the fractured medium.
min_aper Adjusted aperture for fractures classified as open fractures.
max_aper Adjusted aperture for fractures classified as wet fractures.
Chapter 5: Thermohydromechanical model
x
1
, x
2
, x
3
Absolute coordinated system [m].
x, y, z Local coordinated system to the individual fractured block [m].
M, m Matrix.
F, f Fracture.
P Fluid (water) pressure [Pa].
T Equivalent medium temperature [C].
u
i
Displacements [m].
Fluid production (net variation of volume of fluid by unit volume of the
equivalent medium) [].
q, q
i
Darcy velocity or flux density [m/s].
Q, Q
i
Discharge rate [m
3
/s].
ij
Stress tensor [Pa].
ij
Strain tensor [].
g Gravity [m/s
2
].
z Elevation over the see water level [m].
,
w
Water dynamic viscosity [Ns/m
2
].
Water kinematic viscosity [m
2
/s].
E Youngs modulus [Pa].
Poissons ratio [].
w
,
s
,
eq
Density of water, solid grains and equivalent medium respectively
[Kg/m
3
].
f
,
m
Volumetric fractions of fractures and matrix respectively [].
f
,
m
,
eq
Porosities of fractures, matrix, and equivalent medium respectively
(
f
=1 for waterfilled fractures) [].
LVI
C
w
, C
s
Specific heat capacities of water and solid grains respectively [J /kg K].
(C)
eq
Intrinsic specific heat capacity of the equivalent medium [J /m
3
K].
(K
Tw
)
ij
, (K
Ts
)
ij
, (K
T
)
ij
Tensorial thermal conductivities of water, solid grains and
equivalent medium respectively [W/m K].
Tw
,
Ts
Volumetric thermal expansions of water and solid respectively [K
1
].
w0
Water density at the reference temperature T
0
=5C.
f
T
Heat source term [W/m
3
].
k
ij
Tensorial intrinsic permeability of the equivalent medium [m
2
].
K, K
ij
Tensorial hydraulic conductivity of the equivalent medium [m/s].
T
ijkl
Tensorial stiffness coefficient [Pa].
B
ij
Tensorial Biot coefficient [].
G Biot modulus [Pa].
K
n
Fracture (or crack) normal stiffness coefficient [Pa/m].
K
s
Fracture (or crack) shear stiffness coefficient [Pa/m].
l
H
Homogenization scale [m].
Individual fractured block domain (with
A
and
C
as matrix
subdomains and
B
as fracture subdomain).
Whole fractured medium domain.
F
,
I
Exterior faces and interface boundaries of the individual fractured block
respectively.
a Fracture aperture [m].
b Individual fractured block height [m].
l, l
i
Individual fractured block length [m].
H Hydraulic head [m].
j, J
i
Hydraulic gradient [].
I Identity matrix.
K
M
Scalar hydraulic conductivity of the matrix [m/s].

F
K Scalar hydraulic conductivity parallel to the fracture plane [m/s].
F
K Scalar hydraulic conductivity normal to the fracture plane [m/s].
K
A
, K
H
Hydraulic conductivity arithmetic and harmonic means respectively as
defined by equation (58).
[]
*
Any variable referred to the local coordinated system of the individual
fractured block.
[]
+
, []

Any variable in the positive or negative side of the individual fractured
block interface respectively.
=
+
= =
k n
i
k i i k
k
k
x x x x n C with
C
C
r (1)
where {x
0
, x
1
, ..., x
n
} are the discrete values of the time series; x is the mean value; r
k
is
the value of the correlogram for a lag k, varying from 0 to m; and m is the truncation
point. The factor C
0
is the spectral variance of the series.
The spectral density function. The spectral density function corresponds with the change
from a time domain (time series space) to a frequency domain by change of variables
(Fouriers transform of the correlogram). This transformation is interesting because it
permits a better understanding of the time series components, as they are well separated
in the frequency domain. On the other hand, this analysis computes the variance
decomposition in frequency, i.e., how much each frequency contributes to the whole
variability of the signal. The expression used is that proposed by [47]:
(2)
+ =
=
n
i
k k F
Fk r D S
1
2 cos 2 1 2
where k is the lag and F=j/2m (j imaginary number; m truncature point). The factor
D
k
is a window that filters the signal in order to decrease the relative importance that the
noise acquires in the high frequency band. There are several types of valid windows, but
after a wide testing experience with the Mockup data it appears that the best windows
is that of Tukey [22].
In correlation and spectral analysis, the choice of lag k and truncation m is critical point
because it determines the observation window (time interval in which the analysis is
carried out). Moreover, all the series information with characteristic times smaller than
2k can not be noticed. As for m, it must always be inferior to n/2 (at least two values are
necessary to obtain any averaging quantity). Indeed, to have more easily interpretable
results, we have taken m smaller than n/3.
3.2.1.2 Cross analysis
In the crossanalysis the time series is considered to be the response of the
system to another time series at the input (causeeffect relation). To carry out the
crossanalysis, similar tools to those of the simple analysis are used, with some specific
considerations:
The crosscorrelogram. The crosscorrelogram establishes the inputoutput relation. If
the input series is random, the crosscorrelogram corresponds to the impulse response of
the system. When the input cannot be considered a random or quasirandom series, the
12
crosscorrelogram can still provide information about the response of the system: causal
or noncausal relation between input and output, kind of correlation (directly or
inversely proportional), importance of the correlation, etc. The crosscorrelogram is
obtained with the following expression [47]:
( ) ( )
( ) ( )
( ) ( )
=
+
=
+
+
= =
=
= =
=
= =
k n
i
i y
k n
i
i x
k n
i
k i i x y
y x
x y
x y k
k n
i
k i i y x
y x
y x
y x k
y y n S x x n S
x x y y n k C
S S
k C
k r r
y y x x n k C
S S
k C
k r r
1
2
1 2
1
2
1 2
1
1
,
,
,
1
1
,
,
,
and with and
) ( with
) (
) (
) ( with
) (
) (
(3)
where x is the input signal, y the output signal, S
x
2
and S
y
2
are the spectral variances
respectively, and C
x,y
(k) is the covariance for the time lag k.
3.2.2 Wavelets analysis
3.2.2.1 Continuous Wavelet Transform (CWT)
This technique appears as an alternative to avoid the limitations of the classical
frequency analysis of the Fourier transform [16][36], and in particular, its limitation to
characterize nonstationary aspects of the time series as discontinuities
[27][54][59][67][74][77][85][90]. Lets consider a basis function (t) with zero
integral over R and with a well localized Fourier transform (rapidly decreasing towards
zero). A family of functions of two parameters a (dilation) and b (translation), denoted
a,b
(t), can be constructed as follows [27]:
( ) ( ) R R , , ,
1
) (
*
,
= b a
a
b t
a
t
b a
(4)
Function (t) oscillates hence around 0 and vanishes rapidly. The name of wavelet
comes from this quick attenuation. The corresponding wavelet coefficients are defined
by (continuous wavelet transform):
(5)
+
= dt t t f b a C
b a f
) ( ) ( ) , (
,
In this approach, function (t) is called the mother wavelet, and must satisfy the
following conditions:
i)
( )
< =
+
K d
=
Z j Z k
jk jk
t c t f ) ( (8)
with
+
= = dt t t f t t f c
jk jk jk
) ( ) ( ) ( ), ( (9)
and
) 2 ( 2 ) (
2
k t t
j
j
jk
= (10)
This decomposition is the discrete wavelet transform. This transform lets to optimally
minimize the number of necessary coefficients for the characterization of f(t). However,
the problem of choosing correctly the mother wavelet arises.
3.2.2.3 MultiResolution Analysis (MRA)
This kind of analysis searches orthogonal projections of a function into
successive subspaces corresponding to different resolutions of the function. It is,
therefore, a multiscale representation. Lets define mathematically the multirresolution
analysis V
m
in L
2
(R) [59].
A multirresolution analysis is defined as a set of spaces V
m
of L
2
(R) successively
enclosed, i.e., mZ, V
m
V
m+1
, which verify the following properties:
1) If v V
0
, then kZ, v (tk) V
0
.
2) There exists a function g V
0
such that v V
0
, {v
k
}
kZ

. Then, the functions {
+
=
=
k
k
k t g v t v ) ( ) (
kg
}
kZ
={g (tk)}
kZ
are a basis
of V
0
.
3) is dense in L
m
Z m
V
2
(R).
4) { } 0 =
m
Z m
V .
5) f (t) V
m
if and only if f (2t) V
m+1
.
6) If f (t) V
m
, then f (tk/2m) V
m
kZ.
Orthonormal basis of successive subspaces O
m
and V
m
can be built by a wavelettype
method. An exact decomposition of function f can be created by the expression:
( ) ( ) ( ) ( ) ( ) t f F t f F t f
O V
+ = (11)
where
( ) ( )
+
=
=
k
mk mk
V
t t t f t f F ) ( ) ( ), ( is the orthogonal projection of f in V
m
, and (12)
14
( ) ( )
+
=
=
k
mk mk
O
t t t f t f F ) ( ) ( ), ( is the orthogonal projection of f in O
m
. (13)
The procedure to build functions and is well explained in [36].
3.2.3 Matching pursuit
The principle of this method, introduced by [60], consist in searching throughout
a redundant dictionary D of oscillatory functions those which best approximate the time
series, yielding to a basis of functions representing the series. It is an adaptive method,
as the representation basis is determined along the decomposition process.
The dictionary D used is composed by a family of timefrequency functions (or atoms),
generated by dilation, translation and modulation of a unique real analyzing function
k(t) L
2
(R).
The atoms in the dictionary are defined by the Gabor functions:
( )
t i
e
a
b t
k
a
t k
=
1
(14)
where a >0 is the dilation scale, b the translation parameter and is the modulation in
frequency. Lets define =(a, b, ) as the atom index in the dictionary. The factor
a 1 let us normalize k
(t) to 1 under the L
2
norm.
It consists therefore on choosing a unique vector k
0
from the dictionary D, in such a
way that the scalar product ( ) ( ) t k t
0
,
is maximized. This atom will be the most
approximate to the series. The time series
(t) is then decomposed in:
( ) ( ) ( ) ( ) ( ) t R t k t k t t
+ =
0 0
,
(15)
where R
(t) is the residual vector after the approximation of
(t) in the direction of
k
0
. As k
0
(t) is orthogonal to R
(t), it is verified that:
( ) ( ) ( ) ( )
2
2
0
2
, t R t k t t
+ =
. (16)
The main idea of the method is to decompose once again the residual vector R(t),
finding a second vector k
1
(t) that will approximate it most, as it made with the original
series
(t). The procedure is then repeated iteratively with the respective residual
vector:
15
( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) ( ) t R t k t k t R t R
t R t k t k t R t R
t R t k t k t t
n
n n
n n
1
2
1 1
0 0
,
...
,
,
+
+ =
+ =
+ =
(17)
It is easy to determine a convergence criterium for this algorithm, by examining the
decreasing of the residual vector norm. Finally, the signal is decomposed in the
following way:
( ) ( ) ( ) ( )
=
=
0
,
i
i i
i
t k t k t R t
(18)
where atoms k
i
are the ones who best approximate the signal
(t). An ordered set of
atoms (k
0
(t), k
1
(t), , k
n
(t)) is obtained, and it is able to represent the energetic time
frequency distribution of the signal.
3.2.4 Time evolution of statistical parameters
The Statistics provides us with several statistical parameters which are able to
characterize the nature and structure of any given data series. In the case of data
obtained from measuring processes, the correct behaviour of the sensor is usually
characterized by the constancy of those statistical parameters along the time.
The statistical technique used in this study consists in the analysis of the evolution of
several statistical parameters of data series. For that purpose, the series is multiplied by
a finite window of length w and the statistical parameters are calculated in that portion
of data; then, the window moves progressively along the series with a time increment
t, and the parameters are calculated for each instant; finally, a representation of each
parameter vs time is obtained (see Figure 5).
16 186 357 528 698 869 1040 1210 1381
32
34
36
38
S
t
a
t
i
s
t
i
c
p
a
r
a
m
e
t
e
r
time
Statistics calculated
Prob. distributio
AR coefficients
Standard deviati
t
1
w(tt
1
)
t
2
w(tt
2
)
t
3
w(tt
3
)
Time series
16 186 357 528 698 869 1040 1210 1381
32
34
36
38
S
t
a
t
i
s
t
i
c
p
a
r
a
m
e
t
e
r
time
Statistics calculated
Prob. distributio
AR coefficients
Standard deviati
t
1
w(tt
1
)
t
1
w(tt
1
)
t
2
w(tt
2
)
t
2
w(tt
2
)
t
3
w(tt
3
)
t
3
w(tt
3
)
Time series
Figure 5: Study of the evolution of statistical parameters by moving window.
16
The
 Autocorrelation function
statistics analyzed in this study by this technique have been:
: described above.
Probability distribution  : establishes the number of data in the series for each
 Standard deviation
range of values (histogram), divided by the total number of data.
: measures the dispersion of the data around the mean. It is
defined as:
( )
2
1
2
1
1
=
=
n
i
i
x x
n
, (19)
where x represents the mean of the series.
ach computation time yields to a function for the autocorrelation and for the
.3 Results of the statistical analysis and discussion
E
probability distribution, whereas it gives a single value for the standard deviation. Thus,
animations have been obtained for the two first statistics, and a function of time for the
last one.
3
.3.1 Physical processes identification
The Mockup test operational phase started in February 1997 with an initial
basic tool in signal analysis is the general scanning of original data. That would
the heaters) start to be more separated than the ones in the inner parts (day 1011). This
3
inundation of the air gaps left between the bentonite blocks. Then, the heating phase
started. A maximum temperature condition of 100 C in the heaters surface was
imposed in the heaters power control system, and a constant injection pressure of 5 bars
permitted the progressive hydration of the clay. Around the day 900 of the heating
phase, a decrease in the water intake of the experiment with respect to the predictions of
the models was observed. This decrease induced also a decrease in the hydration rate
(less relative humidity than expected for the four levels of instrumentation), and a
decrease in the total pressure measured in the Mockup. The problem of determining
whether a new physical process was starting in the experiment or the physicochemical
models were not taking into account all the coupled processes correctly since the
beginning of the heating phase arose. Data analysis has been carried out to try to answer
that question.
A
consist, in this experiment, in spatially locating the sensors exactly in their position
within the Mockup in a 2D longitudinal section and plotting the rough data along time.
This is the best way to have an approximate idea of how the system behaves, and can
provide very useful information that is not always easy to see in the usual time series
plots. Figure 6 shows the time evolution of the relative humidity field. After the initial
inundation mentioned above, the first year of evolution is characterized by a uniform
gradient from heaters towards the external surface of the Mockup (day 339). Later on,
the gradient lines in the outer parts of hot sections (socalled the ones which include
17
Figure 6: Evolution analysis of the spatial distribution of the data for the relative
humidity sensors in the Mockup.
nalysis technique offers, therefore, the evidence of a decrease in the relative humidity
radient in the external part of the bentonite. That gradient decrease would yield to a
ent decrease, a crosscorrelation
nalysis between the bentonite temperature sensors and the relative humidity ones of
a
g
decrease in the water intake of the Mockup structure.
To look for the possible causes of the humidity gradi
a
two consecutive hot sections has been made. The coupled THM processes are
characterized by correlations between the different parameters involved on them. This is
the case of the temperature and the relative humidity, which are two parameters
mutually correlated by the constitutive laws of the bentonite. In Figure 7, the cross
correlograms of the analysis bentonite temperature of section A2 relative humidity
of section A3 for the four levels of sensors are presented. In levels 2 and 4 of the
bentonite, the natural correlation described above between those parameters can be
observed. However, in levels 1 and 3 that correlation disappears. This fact indicates that
there might be some other process that would be perturbing the relation. That process
could be identified as the evaporation of the water phase into water vapor in the first
level of bentonite (the closest to the heater) and a later condensation of the water vapor
in the third level. This outward water vapor flux would act as a barrier against the
hydration water inflow, causing the slowing down in the hydration rate.
18
19
Zone A
# = 0
# = 1
# = 2
# = 3 # =5
# = 6
# = 7
T_A2_4_#
T_A2_2_# T_A2_1_#
T_A2_3_#
# = 4
Seccin 2
bentonite temperature
Section 2
Section 3
relative humidity
1
2
3
4
Evap.
Cond.
Heater
Surface
(water
inject
1,0
0,8
0,6
0,4
0,2
0,0
0,2
0,4
0,6
0,8
1,0
80,0 60,0 40,0 20,0 0,0 20,0 40,0 60,0 80,0
Lag (days)
C
r
o
s
s

C
o
r
r
e
l
a
t
i
o
n
T_A2_1_4/ V_A3_1
T_A2_1_4/ V_A3_2
T_A2_1_4/ V_A3_3
T_A2_1_4/ V_A3_4
1,0
0,8
0,6
0,4
0,2
0,0
0,2
0,4
0,6
0,8
1,0
80,0 60,0 40,0 20,0 0,0 20,0 40,0 60,0 80,0
Lag (days)
C
r
o
s
s

C
o
r
r
e
l
a
t
i
o
n
T_A2_2_4/ V_A3_1
T_A2_2_4/ V_A3_2
T_A2_2_4/ V_A3_3
T_A2_2_4/ V_A3_4
1,0
0,8
0,6
0,4
0,2
0,0
0,2
0,4
0,6
0,8
1,0
80,0 60,0 40,0 20,0 0,0 20,0 40,0 60,0 80,0
Lag (days)
C
r
o
s
s

C
o
r
r
e
l
a
t
i
o
n
T_A2_3_4/ V_A3_1
T_A2_3_4/ V_A3_2
T_A2_3_4/ V_A3_3
T_A2_3_4/ V_A3_4
1,0
0,8
0,6
0,4
0,2
0,0
0,2
0,4
0,6
0,8
1,0
80,0 60,0 40,0 20,0 0,0 20,0 40,0 60,0 80,0
Lag (days)
C
r
o
s
s

C
o
r
r
e
l
a
t
i
o
n
T_A2_4_4/ V_A3_1
T_A2_4_4/ V_A3_2
T_A2_4_4/ V_A3_3
T_A2_4_4/ V_A3_4
Figure 7: Crosscorrelation between the bentonite temperature sensors of section A2
and the relative humidity sensors of section A3 (time period analyzed: 1997 data).
19
20
To determine whether the humidity gradient decrease has started at some point of the
experiment or it has occurred since its beginning, a comprehensive matching pursuit
analysis has been made on the relative humidity signals. This type of analysis is
appropriate to discover new frequency components or the disappearance of old ones.
Figure 5 shows an example of the matching pursuit technique applied in a relative
humidity sensor of level 4 (V_A3_4). In this case, the time period of the analysed series
goes from J anuary 1998 to October 2000. New components associated with the
desaturation of the fourth level of bentonite (decrease of relative humidity from day 900
aprox.) do not clearly appear from some point on the time series (see Figure 5 below). It
yields, therefore, to supposing that the evaporationcondensation cellules causing the
decrease in the humidity gradient have existed since the beginning of the experiment,
taking probably a greater relevance from day 900 on.
Figure 8: Evolution of the data of relative humidity sensor V_A3_4 (upper figure) and
Matching Pursuit analysis of the time series (lower figure).
20
21
3.3.2 Unexpected events
In November 2000 the 29th, an overheating incident occurred in the Mockup,
due to a failure in the heating control system that implied an increasing of the
temperatures near the heaters to more than 200C (see Figure 9). A question about the
scientific interest in continuing the experiment arose, and to evaluate quantitatively the
short and longterm consequences of that incident became a priority.
Figure 9: Evolution of the bentonite temperature sensors of section A5 in the Mockup
experiment (plotted period: 28/12/9926/9/01).
The wavelet transform has been used to determine the influence of the overheating in
the sensors installed inside the Mockup and in the long run behaviour of the
experiment. The Discrete Wavelet Transform (DWT) has been applied to the bentonite
temperature sensors, in a smaller time period from one month before the incident to one
month after it. Figure 10 shows the DWT analysis of the worstcase located sensor
(T_A5_1_1), which is one of the first level temperature sensors (the closest to the
heater) in section A5 (central section of the heater A). Left column corresponds with the
data series before the overheating and right column with the one after the incident. A
similar response across every scale can be appreciated, that is, there has not been any
important change in the frequency components of the signal due to the high
temperatures reached in the overheating. We deduce, therefore, that sensors and heating
control system continue working correctly. In the other hand, to estimate approximately
the duration of the perturbation produced by the overheating, the highfrequency
component (noise) of the wavelet analysis has been isolated and reconstructed. The
lower curve of Figure 11 shows the reconstructed noise of the detailed temperature
signal (upper curve). In this case, the noise is rapidly attenuated and the perturbation
produced is not longer than 75 hours in any of the stages of the incident. This result
confirms the fast recovering of temperature sensors to their normal regime.
21
22
a. Before the overheating. b. After the overheating.
Figure 10: Temperature of the bentonite (sensor T_A5_1_1) before (a.) and after (b.)
the overheating incident (upper figures) and Multiresolution Analysis (lower figures).
a.
b.
Figure 11: a. Temperature of the bentonite (sensor T_A5_1_1) during the
overheating incident (upper figure) and b. Reconstruction of the noise
component of the signal (lower figure).
22
23
The Continuous Wavelet Transform (CWT) has also been applied. The representation of
the wavelet transform is continuous but redundant, so reconstruction of the signal is not
exact, but it allows to have much better visual information of frequency components
that are present on it, and to detect the nonstationarities. An example in the application
of the continuous wavelet transform is the analysis made to the total pressure sensor
PT_A6_3 in Figure 12. In this case, the total pressure data series analysed includes the
overheating incident (around day 1385, upper graph of Figure 12). However, the most
interesting event in this time series is located some days before, more precisely around
day 1365, in which a slight cooling of the experiment room caused the decreasing of the
total pressure. In the continuous wavelet transform, an increase in the high frequency
components (log2(1/a) = 8) can be appreciated, what means that a noise appears
probably due to the cooling of the experiment room. That result evidences the
sensibility of the Mockup experiment to the external temperature.
1357 1361 1365 1369 1373 1377 1381 1385 1389 1393 1397
4.0
5.0
6.0
7.0
time (days)
T
o
t
a
l
p
r
e
s
s
u
r
e
(
b
a
r
)
time (days)
l
o
g
2
(
1
/
a
)
1357 1361 1365 1369 1373 1377 1381 1385 1389 1393 1397
2
3
4
5
6
7
8
Figure 12: Total pressure (sensor PT_A6_3) during the overheating incident (upper
figure) and Continuous Wavelet Transform analysis of the signal (lower figure).
23
24
With the available data, we can say that the overheating incident did not cause
substantial failures neither in the sensors behaviour nor in the experiment development.
But it becomes interesting to try to determine approximately how long has the
perturbation of the experiment results lasted from the correlation point of view. Results
of this analysis are shown in Figure 10 for the bentonite temperature sensors (section 5
of heater A). Figure 13a (day 1295) presents the normally behaving system
correlogram right before the overheating, in which the daily periodicities can be
appreciated (amplified for the outer levels of bentonite). The influence of the heating
control system over the level 1 of temperature sensors (red curves) is also seen. Figure
13b (day 1388) corresponds to the overheating discontinuity, where that characteristic
shape of the correlogram for all the temperature sensors is completely lost. The instant
in which the characteristic shape is recovered could be associated to the end of the
system perturbation due to the overheating, and in this case it has been set around the
day 1397 (see Figure 13c). Thus, the correlation perturbation lasts, or equivalently, the
recovering time is about 9 days for the temperature sensors in section A5.
a. 28/8/00 (day 1295) b. 29/11/00 (day 1388)
Bentonite temperature sensors
Of section A5
c. 8/12/00 (day 1397)
Figure 13: Evolution of the autocorrelation function of the bentonite temperature
sensors of section A5 in the Mockup experiment (analysed period: 28/12/9926/9/01).
level 1
level 2
level 3
level 4
level 1
level 2
level 3
level 4
24
25
3.3.3 Sensors reliability
Another classical application of the Correlation analysis is to predetect failures
in sensors. An analysis of the 1997 data for all the total pressure sensors installed in the
Mockup has been made. Some sensors (PTA102, PTB42 and PTB102) showed an
anomalous behaviour in their autocorrelation function, what could be firstly interpreted
as a faulty behavior (Figure 14, upper graphs). A deeper analysis shows that those
sensors behave similarly to the fluid pressure ones, from an autocorrelation point of
view (Figure 14, lower graphs). That means total pressure anomalous sensors work
normally, but they are not properly connected to the solid phase (bentonite), so what
they were actually measuring was just the fluid phase pressure. Later on, some of those
unconnected sensors got connected to the bentonite and started to measure the same
total pressure value than the surrounding ones (CIEMAT, 2002).
25
26
a. Section 10A b. Section 4B c. Section 10B
0,2
0,0
0,2
0,4
0,6
0,8
1,0
0
,
0
3
,
8
7
,
5
1
1
,
3
1
5
,
0
1
8
,
8
2
2
,
5
2
6
,
3
3
0
,
0
3
3
,
8
3
7
,
5
4
1
,
3
4
5
,
0
4
8
,
8
5
2
,
5
5
6
,
3
6
0
,
0
Paso (das)
A
u
t
o
c
o
r
r
e
l
a
c
i
n
PT_B4_2
PR_B4_2
PZ_B4_2
0,2
0,0
0,2
0,4
0,6
0,8
1,0
0
,
0
3
,
4
6
,
8
1
0
,
1
1
3
,
5
1
6
,
9
2
0
,
3
2
3
,
6
2
7
,
0
3
0
,
4
3
3
,
8
3
7
,
1
4
0
,
5
4
3
,
9
4
7
,
3
5
0
,
6
5
4
,
0
5
7
,
4
6
0
,
8
Paso (das)
A
u
t
o
c
o
r
r
e
l
a
c
i
n
H_B4_2
A
B
0,2
0,0
0,2
0,4
0,6
0,8
1,0
0
,
0
3
,
5
7
,
0
1
0
,5
1
4
,0
1
7
,
5
2
1
,
0
2
4
,5
2
8
,0
3
1
,5
3
5
,
0
3
8
,5
4
2
,0
4
5
,
5
4
9
,
0
5
2
,
5
5
6
,0
5
9
,5
Paso (das)
A
u
t
o
c
o
r
r
e
la
c
i
n
PT_B10_2
PR_B10_2
PZ_B10_2
PT_B10_3
PR_B10_3
PZ_B10_3
0,2
0,0
0,2
0,4
0,6
0,8
1,0
0
,
0
3
,
5
7
,
0
1
0
,5
1
4
,0
1
7
,5
2
1
,0
2
4
,5
2
8
,0
3
1
,5
3
5
,0
3
8
,5
4
2
,0
4
5
,5
4
9
,0
5
2
,5
5
6
,0
5
9
,5
Paso (das)
A
u
t
o
c
o
r
r
e
la
c
i
n
H_B10_2
H_B10_3
0,2
0,0
0,2
0,4
0,6
0,8
1,0
0
,
0
3
,
5
7
,
0
1
0
,5
1
4
,0
1
7
,5
2
1
,0
2
4
,5
2
8
,0
3
1
,5
3
5
,
0
3
8
,5
4
2
,0
4
5
,
5
4
9
,0
5
2
,5
5
6
,0
5
9
,5
Paso (das)
A
u
t
o
c
o
r
r
e
la
c
i
n
PT_A10_2
PR_A10_2
PZ_A10_2
PT_A10_3
PR_A10_3
PZ_A10_3
0,2
0,0
0,2
0,4
0,6
0,8
1,0
0
,
0
3
,
5
7
,
0
1
0
,
5
1
4
,
0
1
7
,
5
2
1
,
0
2
4
,
5
2
8
,
0
3
1
,
5
3
5
,
0
3
8
,
5
4
2
,
0
4
5
,
5
4
9
,
0
5
2
,
5
5
6
,
0
5
9
,
5
Paso (das)
A
u
t
o
c
o
r
r
e
l
a
c
i
n
H_A10_2
H_A10_3
Figure 14: Simple correlograms of total pressure sensors (upper figures) and fluid pressure sensors (lower figures) showing similarities in their
behaviour (time period analyzed: 1997 year data).
26
4. GEOMORPHOLOGICAL SIMULATION AND RECONSTRUCTION OF
THE 3D FRACTURED ROCK
4.1 Geomorphological data
4.1.1 Geology, tunnel and boreholes
The GTS is located in the southern part of the Central Aar Massif, around 400m
below the surface. The rocks in this area are almost exclusively granitic, and are
intruded by sets of lamprophyres and, to a lesser extent, by aplites. Different
discontinuity systems have been described (see Figure 15): ductile systems (S
1
, S
2
and
S
3
), brittle systems (S
4
/K
4
, K
2
/L=lamprophyre direction, K
1
, K
3
and S
5
) and tensile
systems (ZK
1
and ZK
2
). For a detailed description of the fracture systems present in the
GTS see [8][18][49][72][82]. FEBEX experiment area is located in the northern part of
the Laboratory tunnel of the GTS (Figure 16), where a marked water discharge was
encountered. Two exploratory boreholes where made: FEBEX95001 and FEBEX
95002 [40][41][73]. The aim of these boreholes was to detect and recognize the main
structures of deformation and their importance from the hydrogeologic point of view,
and to determine the azimuth and dip of the drift to perform the FEBEX experiment.
After the FEBEX drift excavation, a series of radial boreholes were drilled out of the
test zone (last 17 m of the drift), to characterize the presence of fractures around it [34].
However, no information on fractures orientation was registered. The data coming from
the radial boreholes have not been used for this simulation. The generation domain for
the fractured network simulation is a block of 70x200x70 m
3
centered in the FEBEX
drift (Figure 15).
Figure 15: Alpine structures in the Central Aar Massif according to [82].
27
Figure 16: Location of the FEBEX drift within the GTS general layout (from [73])
and fractured medium generation domain.
LEGEND
Test areas
BK Flow tests in fracture systems
MI Migration experiments
US Seismic tests
VE Ventilation tests
WT Heating tests
BOS Boreholes sealing
EDZ Excavation disturbed zones
EP Excavation in shear zone MI
FEBEX Engineered barriers experiment
TOM Seismic tomography development
TPF Biphasic flow
CP Connected porosities
ZPK Biphasic flow in fracture networks
ZPM Bphasic flow in rock matrix
Tests i n phase IV (19941996)
LEGEND
Test areas
BK Flow tests in fracture systems
MI Migration experiments
US Seismic tests
VE Ventilation tests
WT Heating tests
BOS Boreholes sealing
EDZ Excavation disturbed zones
EP Excavation in shear zone MI
FEBEX Engineered barriers experiment
TOM Seismic tomography development
TPF Biphasic flow
CP Connected porosities
ZPK Biphasic flow in fracture networks
ZPM Bphasic flow in rock matrix
Tests i n phase IV (19941996)
4.1.2 Fractured network data
Data collected from the two exploratory boreholes present the number of
fractures and number of open fractures every core sample 4m long. Tables 2 and 3 show
those data, along with a selection of highly conductive zones within the boreholes.
Figure 17 shows the orientations and families of all the intersecting fractures in a pole
diagram or stereonet.
Table 2: Distribution of fractures in
borehole FEBEX95001 (from [73]).
Table 3: Distribution of fractures in
borehole FEBEX95002 (from [73]).
Depth
(m)
Total #of
fractures
#of open
fractures
Fracture
zones
04 20 9 Zone 1.1
48 7 1
812 23 20 Zone 1.2
1215 10 0
1518 17 0 Zone 1.3
1831 11 1
3134 3 3 Zone 1.4
3467 52 2
6770 11 3 Zone 1.5
7077 1 0
Depth
(m)
Total #of
fractures
#of open
fractures
Fracture
zones
024 50 7
2431 40 37 Zone 2.1
3138 19 12 Zone 2.2
3839 1 0 Zone 2.3
3952 11 2
5262 56 32 Zone 2.4
6270 31 14 Zone 2.5
7074 8 1
7479 23 14 Zone 2.6
7990 36 9
90109 77 5 Zone 2.7
109128 30 1
128133 29 11 Zone 2.8
28
1
0,8
0,6
0,4
0,2
0
0,2
0,4
0,6
0,8
1
1 0,
8
0,
6
0,
4
0,
2
0 0,2 0,4 0,6 0,8 1
FBX95001
FBX95002
Figure 17: Pole diagram of the fractures in boreholes
FEBEX95001 y FEBEX95002.
A general cartography of the drift was developed (Figure 18), where the most important
lithologic and structural features can be appreciated. Five different zones, with different
structural characteristics, can be distinguished along the main axis of the drift:
 The first zone, between 0.0 and 14.0 m, is characterized by homogeneous granite
with low fracturing. There are some quartzfeldspar veins of little hydraulic
relevance.
 The second zone, between 14.0 and 25.5 m, is characterized by a high fracture
density. There are some breccified zones (breccias) with more than 10
fractures/m. Granite is highly altered in this zone, and water flow is
approximately 70 liter/day overall.
 The third zone is located between 25.5 and 53.0 m, and has characteristics very
similar to the first zone. There can be found also two thin dikes of lamprophyre
at 40.0 m.
 The fourth zone is the one located between 53.0 and 63.0 m. It is characterized
by a lowtomedium wall outflow rate; also by the presence of quartzfeldspar
veins; and by the existence of three lamprophyre dikes of various widths
(0.20 m, 1.50 m and 0.04 m respectively). The second dike is especially
important because water outflow can be observed there, not only at the
lamprophyre/granite contact, but also in the interior of the dike (approximately
3.7x10
4
to 7.0x10
4
l/min/m), where there are many open fractures without
sealing materials.
 Finally, the fifth zone extends from 63.0 to 70.0m, and presents a high degree of
fracturing (opened and closed fractures) with various sealing materials and
29
quartzfeldspar veins. However, the outflow in this zone is not, in general, very
significant (on average about 2.0x10
4
l /min/m).
Tracemap of the FEBEX drift has been digitalized to obtain some statistics from it. Two
cumulative distribution functions have been obtained from the digitalized tracemap and
have been used for optimization purposes: cumulative histogram of the trace length and
cumulative histogram of the 3D trace chord, the last obtained by reconverting the two
dimensions trace coordinates into three dimensions in the cylindrical tunnel. Figures
19a and 19b show the two histograms so obtained. The fractal dimension of each zone
has also been determined [29], although it was not used for the fractured medium
simulation process. APPENDIX I presents these results.
Figure 18: Map of traces on the wall of the FEBEX drift, divided into five different
zones according to their geological features (from [73]).
a. b.
Figure 19: a. Cumulative histogram of trace length of the FEBEX drift tracemap; and
b. Cumulative histogram of 3D trace chord of the FEBEX drift tracemap.
30
4.2 Reconstruction of the fractured medium
4.2.1 Statistical distributions of the fractured network
A synthetic fractured medium has been generated from the field data. First, the
following fracture parameters and statistics were defined (and if possible preevaluated)
by using the abovedescribed geologic information:
Fracture location. A homogeneous Poisson process is used to define the
coordinates (x
cf
, y
cf
, z
cf
) of the fracture centers (f denotes each individual
fracture). Accordingly, the three coordinates (x
cf
, y
cf
, z
cf
) are uniformly
distributed random variables within the bounds of the rectangular box domain.
However, the nonuniform pattern of the traces in the drift wall has also been
simulated, so that the distribution of centres in the proximities of the drift
becomes nonuniform. A more detailed description of this local simulation is
given further below.
Fracture orientation. Four different families of fractures have been defined
according to both morphological (stereonet of Figure 17) and genetic (families
classification of [82] in Figure 15) criteria. Uniform distributions within angle
intervals have been used for the dip (maximum slope direction) and the plunge.
Figure 20 shows the stereonet of our families classification, and Table 4 the
parameters of these orientation distributions. Relations between the fracture
direction, dip, plunge and pole (vector normal to the fracture plane) are
presented in APPENDIX II.
Fracture density. There are several measures of fracture density. In our case,
the p
21
(trace length / intersecting plane surface) of the tracemap of the
experiment drift has been adjusted. Moreover, an anisotropic p
21
has been
obtained indirectly by considering the five different zones of the trace map
presented above.
Fracture aperture. Data on fracture aperture are only qualitative.
Measurements in boreholes FEBEX95001 and FEBEX95002 only distinguish
between filled fractures, open fractures and wet fractures (Figure 21). Increasing
apertures of 1e8m, 1e5m and 1e2m have been assigned according to this
division (Table 5). However, in a later stage they have been adjusted to fit the
model results with hydraulic measurements (see chapter 4.2.6 below).
Fracture size. The power law distribution has been used. There are three
parameters in this distribution: R
min
, R
max
and the exponential coefficient b. In a
preliminary optimization the sensitivity of the R
max
with respect to the
minimization function happened to be negligible, so we set a fixed value equal
to half the maximum dimension of the generation domain (R
max
=100m). The
other two parameters have been optimized so that the synthetic medium fits the
geologic data.
31
1
0,8
0,6
0,4
0,2
0
0,2
0,4
0,6
0,8
1
1 0,
8
0,
6
0,
4
0,
2
0 0,2 0,4 0,6 0,8 1
S1+S2
K4
K2+L
S3
K1+K3
S4
Non clas.
3
1
2
4
1
2
0 50 100
S1
S2
S3
K1
K2
K3
K4
L
ZK(1+2)
F
r
a
c
t
u
r
e
s
y
s
t
e
m
# of discont inuit ies
Open and wet fractures
Open fractures
Filled fractures
Figure 20: Families classification of the
fracture data of boreholes FEBEX95001
and FEBEX95002.
Figure 21: Fracture aperture frequency
in the GTS tunnel (from [49]).
Table 4: Parameters of the orientation
distributions for each family (dip and
plunge, both in degrees).
Table 5: Aperture frequencies according
to the fracture family (qualitative
classification).
Aperture
Family
% wet
frac.
% open
frac.
% filled
frac.
FAMILY 1
(S1+S2, K4)
0,27 0,51 0,22
FAMILY 2
(K2+L)
0,62 0,18 0,20
FAMILY 3
(S3, K1+K3, S4)
0,10 0,31 0,58
FAMILY 4
(ZK1, ZK2)
0,40 0,44 0,16
180 ; 10 180 ; 90
FAMILY 4
UNIFORM
180 ; 90 180 ; 90
FAMILY 3
UNIFORM
12 ; 11 28 ; 16
FAMILY 2
UNIFORM
10 ; 11 143 ; 16
FAMILY 1
UNIFORM
Standard
deviation
Mean pole
(dip; plunge)
180 ; 10 180 ; 90
FAMILY 4
UNIFORM
180 ; 90 180 ; 90
FAMILY 3
UNIFORM
12 ; 11 28 ; 16
FAMILY 2
UNIFORM
10 ; 11 143 ; 16
FAMILY 1
UNIFORM
Standard
deviation
Mean pole
(dip; plunge)
4.2.2 Optimization methodology
A Montecarlo algorithm has been implemented to reconstruct stochastically the
synthetic fractured medium [2][6][10][28] [39][87]. An optimization procedure based
on simulated annealing [66] has been used to adjust fracture size distribution so as to
minimize the discrepancy between synthetic fractured medium and real fractured
medium, according to the geologic data available. Many studies have developed
methodologies to infere the size distribution out of planar [48][53][91][95][97] or more
recently curved [64][65] tracemaps, although there are some precautions to be taken
when doing that [58]. As it was explained above, the power law or Pareto distribution
has been used to characterize the size of the fracture network, and only the R
min
and b
parameters have been optimized. Simulated annealing is an optimization technique that
tends to reproduce the cooling processes that occurs in the minerals sequential
crystallization on the volcanic magma. A progressively decreasing temperature
parameter T controls the acceptance or rejection criteria of new points in the search of
the optimum. For that, an exponential expression that considers temperature and
32
discrepancy between the optimum reached up to that instant and the new value of the
objective function is used [66]. A new approach for adapting the search interval for
possible new points which takes into account number of rejected and accepted points
over a number of iterations has been implemented [38]. This approach obtains global
minimums with a lower computational cost (less number of objective function
evaluations).
We define the objective function (OF) to be minimized as follows:
( ) ( ) ( ) ( ) ( ) ( ) ( )
( ) ( )
( )
( ) ( )
( )
( ) ( )
( )
( ) ( )
( )
2
21
21 21
2
2
2 2
2
1
1 1
2
1
2
1
2
...
...
+ + =
= =
Y p
X p Y p
M
Y ntrb
X ntrb Y ntrb
M
Y ntrb
X ntrb Y ntrb
M
Y ntrtun
X ntrtun Y ntrtun
M X h Y h X g Y g X F
nbins
m
m m
nbins
m
m m
(20)
where
Y is the variable that represents the FEBEX drift experimental data,
X is the variable that represents the generated fractured medium,
g() is the histogram of trace lengths,
h() is the histogram of 3D trace chords,
nbins is the resolution or number of elements of the histograms,
ntrtun() is the number of traces (intersections) produced in the FEBEX drift,
ntrb
1
() is the number of traces (intersections) produced in the borehole
FEBEX95001,
ntrb
2
() is the number of traces (intersections) produced in the borehole
FEBEX95002,
p
21
() is the areal density defined before (total trace length divided by drift wall
area) and
M is the penalization factor (in our case M=1).
The two first terms of the objective function correspond to the
2
error measure of the
discrepancy between the observed and simulated tunnel traces histograms (both trace
length and 3D chord length), whereas the last four terms are penalization terms (the
bigger the penalization factor M is, the more the objective function gets a higher value
for a given discrepancy). To compute the histograms, the exact trace of each generated
fracture left on the cylindrical drift wall must be calculated. APPENDIX III presents the
analytical solution for this problem, used in the optimization algorithm.
Each iteration, the Montecarlo routine generates fractures sequentially, until a stopping
criterion decides when the generation process ends. In our case, the total number of
observed features (ntrtun+ntrb
1
+ntrb
2
) acts as stopping criterion. Once that number is
reached, the generated medium is completed.
Due to the stochastic nature of the objective function, it is highly recommendable to
compute each value as the average of a certain number of evaluations (i.e., generations
of the fractured medium). This number of evaluations has to be determined according to
stability criteria on the objective function value, for example with the Tstudent test,
which sets the minimum number of evaluations in order to assure that the average lies
into a confidence interval of the mean value (in our case, the 90% confidence interval).
For our objective function, the mean of a minimum of 25 evaluations, i.e., 25 generated
media, is necessary to reach the 90% confidence on the objective function value.
33
There is an elevated computational cost to calculate each evaluation of the objective
function. Therefore, an intermediate solution must be used and a twostep optimization
process has been obtained:
 Firstly, an optimization process in which each value of the OF is obtained as the
average of 3 evaluations.
 As the optimized medium is the result of an average over 3 fractured media, to
finally get the one to be used in the THM model we have chosen the best
realization out of 750 obtained with the optima parameters found in the
optimization process.
Before showing the resulting optima parameters and the generated medium, lets do
some additional remarks about the optimization algorithm.
4.2.3 Main fractures
The most important features in the fractured system have been imposed, in order
to preserve some geometric and hydraulic consistency in the model. A total of 17
fractures with complete or near complete intersecting section with the drift have been
selected. To obtain their dip and plunge from the fracture traces, geometric relations
of the 2D curvilinear trace left on the wall and the 3D real dip and plunge of the planar
disk fracture have been used (see Figure 22):
( )
= =
+ = =
sin
arctan arctan
arctan
2
l
D
L
D
plunge
m
D
dip
(21)
where
D is the diameter of the FEBEX drift,
m is the distance between the intersections of the trace with the right and left
wall lines of the drift measured in the 2D tracemap,
l is the distance between the intersections of the trace with the floor and roof
lines of the drift measured in the 2D tracemap, and
and are the dip and plunge of the fracture according to APPENDIX II
angles criteria.
Figure 23 shows the polar diagram of the selected fractures with the so obtained
directions and plunges, and Figure 24 presents the comparison between the real trace
map and the traces obtained from the large discrete fractures, labeled from 1 to 17. In
blue we represent the opened fractures, in yellow the filled fractures (closed), and in
green we represent the lamprophyre dykes. This classification will be used to further
estimate the fracture apertures (see section 4.2.6). The sizes of the fractures have been
estimated taking into account the Grimsel geology reports [49][73] and the correlations
between the different boreholes of the FEBEX area: there are fractures crossing
completely the domain and some others with diameter of the order of the domain size.
However, to keep the stochastic nature of the network, fracture radii have been
34
estimated with a uniform distribution between [1.14/sin(), 100], being the dip of
each fixed fracture. The minimum value of the distribution, different for each fracture,
assures that the intersection of the fracture with the FEBEX drift is complete, as
observed in the tracemap. The maximum value of the distribution has been the same as
the maximum value for the power law used for the rest of the fractured medium.
The main characteristics of the fixed fractures are given in Table 6. Fracture center
coordinates (x
ct
, y
ct
, z
ct
) are given with respect to the tunnel local reference system.
Lamprophyres apertures (fractures 5, 12 and 14) have been set to 1.5max_aper for the
two first and 2max_aper for the last one (see Chapter 4.2.6 for fracture aperture
adjustment). Fractures have been assigned to a family from those of Figure 20
according to their orientation.
a. b.
a.) to infere the 3D dip and plunge of
) from the trace map.
Figure 22: Geometric relations of the 2D trace (
a single fracture (b.
Open fractures
Filled fractures
Lamprophyre
Figure 23: Pole diagram of the large discrete fractures
of the FEBEX drift.
35
Table 6: Main characteristics of the fixed fractures of the simulated network.
Frac. # dip () plunge () x
ct
(m) y
ct
(m) z
ct
(m) R
f
(m) aperture (m) family
1 326.80 82.52 18.45 0 0 52.1082 6.65e6 1
2 149.44 61.70 14.25 0 0 34.2575 1e8 1
3 331.57 88.79 12.45 0 0 43.9373 6.65e6 1
4 147.93 73.27 9.60 0 0 23.5164 1e8 1
5 234.98 83.11 4.10 0 0 58.4632 2.48e5 2
6 91.31 10.83 1.40 0 0 77.4904 1e8 3
7 215.00 74.21 8.45 0 0 53.5393 6.65e6 2
8 188.82 75.53 12.00 0 0 64.4759 6.65e6 2
9 223.75 57.17 15.85 0 0 21.9800 6.65e6 2
10 238.20 60.88 17.65 0 0 38.7911 6.65e6 2
11 290.96 e6 3 17.64 20.45 0 0 79.1480 6.65
12 51.93 86 5 2 .63 19.20 0 0 68.4490 2.48e
13 231.93 73.61 20.0 46.7499 6.65e6 2 0 0 0
14 209.81 71.15 24.60 0 0 57.3035 3.31e5 2
15 98.43 86.38 30.35 0 0 79.6561 6.65e6 3
16 2 234.98 61.47 30.95 0 0 7.1390 6.65e6
17 200.96 69.33 32.75 0 0 60.7708 1e8 2
F re 2 parison of the FEBEX traces ma (upper figure) with the traces of the
mul g fr ures ( er
4.2.4 Nonuniform tracemap reproduction
In addition to the objective function criteria, we have tried to reproduce the n
un m p of f e d alo the F EX racem do so he
fo ing sion rith be implemented inside the red network
eneration algorithm:
 1. Initially set the p
21
on each of the five zones of the FEBEX drift to zero.
 2. Generation of fracture f and computation of p
21
on each zone in case that
intersection occurs.
 3. Addition of the p
21
of fracture f to the previous p
21
.
1 3 5 6 7 8 11 1 15 16 2 4 9 10 2 13 14 17
igu 4: Com p
si ated bi act low figure).
no
ifor attern ractur ensity ng EB drift t ap. To , t
llow exclu algo m has en fractu
g
36
 4. Identify zones where p
21
surpasses maxp
21
. If not all the zone densities have
go from step 5 to 8 (if all the zon
been surpassed, e densities are already
surpassed, the trace is added normally).
select a zone with the p
21
lower than the
corresponding limit given by the experimental data.
components of the translation vector to move the fracture trace
long the drift up to that zone. The component is computed by subtracting the
oves
to a higher zone and to the maximum coordinate if the trace will move to a
8. Move the fracture to that zone summing up the translation vector components
Fig
anneali
4.2 O
Figure 26 shows the evolution of the objective function (OF) on the two stages
of t t
those iterations where some enhancement on the minimizing process was produced. At
the
stage (2
The op
R
max
=1
Figures 27, 28 and 29 and Table 7 below show the optimum solution for the
syn t
in term ntersections with the different sampling objects.
he multiple figures below (Figure 27abcd) display the fracture trace characteristics
red in
e FEBEX drift, as stated before, in which the X direction follows the Geographic
orth. Figures 28 and 29 shows the reconstructed 3D fracture network with
etails on the results of the optimization processes
e APPENDIX IV.
 5. Generation of an aleatory number to
 6. Generation of an aleatory coordinate within the interval of the selected zone.
 7. Calculate the
a
aleatory coordinate of step 6 to the minimum coordinate of the trace if it m
lower zone.

to the coordinates of the trace.
ure25 presents the whole fracture network generation algorithm. For the simulated
ng algorithm of the optimization process, see [66] and the variant of [38].
.5 ptimized fractured medium
he optimization process described above. The plots show the values of the OF a
first stage (26a), a minimum value of 0.559 in the OF is obtained. At the second
6b), the best realization out of 750 gives a better value of the minimum of 0.465.
timized parameters of the power law size distribution are: R
min
=0.1985m,
00m, and b=3,3048.
theically generated medium. Table 7 lists the main fractured network characteristics
s of number of fractures and i
T
resulting from the optimization procedure (trace lengths, chords, and intersections on
the tracemap). The generation domain consisted of a block of 70x200x70m
3
cente
th
N
N =2906474 disc fractures. For more d
se
37
Generate fracture:
Fracture family: randomly selected by prob. freq.
Fracture size (predetermined pdf)
Fracture centre: Poisson proc. (up to maxnumtraces)
Fracture orientation: randomly selected frompdfs
Fracture aperture: randomly selected by prob. freq.
Calculate tunnel and/or boreholes traces analytically
Calculate traces statistics (tracelength, chordlength,
numtraces=numtracestun+numtracesbor(j), equivalent
diameter, dmin and dmax of pseudoellipse, area of
pseudoellipse, etc) and fracture area.
INITIALIZE stopping criterias:
generation: maxnumtraces (boreholes(j)+tunnel, j=1,2)
optimization: maxiter, maxerror, p
21
in 5 tunnel zones
Do we reach
maxnumtraces ?
Calculate value of Objective Function:
2
error of tracelength and chordlength histograms with
penalizations on numtracestun, numtracesbor(j) andp
21
Do we reach maxiter OR
error is <maxerror ?
Set parameters of statistical distributions:
Frac. size (power law: R
min
, b) OPTIM, R
max
FIXED
Frac. family (4): given by prob. frequencies FIXED
Frac. center (hom. Poiss. Proc. in domain ), FIXED
Frac. orientation (unif. in and cos ) FIXED
Frac. aperture (3): given by prob. frequencies FIXED
Store GENERATED FRACTURED MEDIUM and
statistical parameters of the generation and STOP
Revise parameters
R
min
and b
according to
optimisation
process (SA)
Randomselection of the random seed
(for the generation process)
EXCLUSION TEST:
Can the fracture intersect the
tunnel or boreholes?
numfrac =
numfrac+1
numtraces =
numtraces+1
numfrac =
numfrac+1
Do we surpass trace density
p
21
in tunnel zone i?
Randomly
move center to
unfilled zones
All trace
densities p
21
(i)
are surpassed
yes
no
yes
no
no
yes
yes
no
Figure 25: Algorithm of the optimization process to simulate the fractured medium.
38
a. b.
Figure 26: a. Evolution of the objective function by averaging 2 realizations of the
generation algorithm to get each value of the objective function. b. Evolution of the
objective function for 750 realizations with the optimum parameter values.
Table 7: Number of fractures and intersections of the fractured medium generated by
the optimization process.
#of
fractures
#of tunnel
traces
#of intersect. in borehole
FEBEX95001
#of intersect. in borehole
FEBEX95002
2906474 800 144 234
a. b.
c.
d.
Figure 27: a. Cumulated distribution function of trace lengths on tunnel ( observed;
 fitted); b. Cumulated distribution function of chord lengths on tunnel ( observed;
 fitted); c. FEBEX drift observed tracemap ; d. FEBEX drift fitted tracemap.
0
5
10
15
20
25
30
35
40
0 100 200 300 400
OF
iteration
0,4
0,45
0,5
0,55
0,6
0,65
0,7
0 100 200 300 400 500 600 700
OF
iteration
39
igure 28: Whole view of the reconstructed fractured medium with 2906474 fractures.
X
Y
Z
F
Figure 29: Fraction of the reconstructed fractured medium inside the domain.
X
Y
Z
40
4.2.6 Fracture apertures adjustment
fit
) with
the
s.
thod
een
(m
Fracture apertures where initially set with approximate but arbitrary values.
After obtaining the optimized fractured medium, a new Montecarlo process is run to
the discrete threevalue aperture distribution (filled, open and wet fractures
experimental hydraulic measurements. Following the data given in [33],
homogenized rock surrounding the FEBEX drift must have an equivalent hydraulic
conductivity of about 510
11
to 810
11
m/s to give the measured values of flow coming
into the entire test zone (last 17m of the drift). Therefore, an optimization process has
been run, also based on the simulated annealing method used before, to calculate the
apertures that best fit the averaged equivalent hydraulic conductivity, i.e., 6.510
11
m/
For the computation of the equivalent hydraulic conductivity, a homogenization me
described in the next chapter has been used. The set of filled fractures has b
considered to have the same hydraulic conductivity as the rock matrix. The results of
the optimization process have yielded to the following optimized apertures:
min_aper =6.649810
6
m for the open fractures, and
max_aper =1.654710
5
m for the wet fractures.
Figure 30 shows the evolution of the discrepancy between both hydraulic conductivities
easured vs calculated).
Figure 30: Evolution of the OF in the apertures adjustment.
41
42
5 THERMOHYDROMECHANICAL MODEL
5.1 Introduction, coupling and upscaling
Briefly, the modeling approach considered in this work is based on an equivalent
continuum model for fractured rock, capable of representing the coupled processes
involved in the FEBEX experiment in its different phases (excavation, heating, and
steady state). The model yields a fully coupled set of termohydromechanical (THM)
equations. Several works have been consulted and compared to yield to the final form of
the equations [4][7][44][57][68][70][75][78][94]. The final THM continuum equations
are then solved numerically using a continuum finite element code (COMSOL
Multiphysics).
The coupled continuum equations resulting from our model can be summarily described
as a combination of Darcy's law and of Biot's poroelasticity equations for saturated
medium, together with the classical conservation equations for mass, momentum and
energy, with the Darcy and Biot laws cast in their most general, anisotropic form.
Compressibility and thermal expansion of fluid are also considered. The continuum
coefficients involved in these constitutive laws are determined based on a
homogenization approach specifically developed for fractured rocks, essentially an
adaption from [71], [1], and several other works like the BMT3 benchmark exercise
within the DECOVALEX project and others [5][24][67][83].
The coupled THM equations resulting from the above approach take into account to a
certain extent the structural complexity of the discontinuous, fractured rock mass. The
equivalent continuum is hydraulically nonisotropic and mechanically nonorthotropic.
A twoway coupling between pressure variations and stress variations is directly taken
into account in the equations, in addition to a two way coupling of temperature on stress
and pressure. Figure 31 presents an example of the main coupled processes given in a
thermohydromechanical system. In our model, the only process not taken into account
is the variation of fracture apertures and matrix porosity due to stresses.
Water pressure influence
on effective stress
H
T
M
and fracture apertures
Changes in rock porosity
Figure 31: Coupled processes in a thermohydromechanical system.
43
The advantage of odeling coupled
HM processes in the presence of many, variously oriented fractures, while a discrete
actures approach would become rapidly untractable as the number of fractures and
Starting from a known rock matrix with a given distribution of fractures, the
constit
the continuum approach is that it can be used for m
T
fr
their geometrical complexity increases.
utive laws of the equivalent continuum are obtained by a linear superposition
approach, based on the methods developed by several authors, in particular [81] for
hydraulics, and [71] for hydromechanics. The continuum equations are formulated for
a 2D or 3D fractured rock made up of an elastic permeable matrix, and irregularly
distributed, waterfilled elastic fractures. Here, we focus on certain essential features of
the coupled continuum model, rather than on detailed simulation results.
Remark: Before selecting the final equations and the numerical method to solve them,
revious studies of other possible numerical methods and modeling approaches have
the periodic boundary
conditions needed for the convergence of the method. Some results and the
nge coefficient
(reactive term) between them [4][35][50][51]. The concentration of each
species in this model would represent the fluid pressure on each of the
continuum media. This is a powerful approach to couple flow in the matrix with
flow through the fractures. However, results obtained for our simulated fractured
medium indicate that exchange occurs much faster than diffusionadvection
flow. Thus, a single continuum approach would yield to equivalent results.
APPENDIX VI presents some results and the full formulation of this dual
continuum approach.
 Waveletbased numerical methods to solve PDEs: wavelets are not only used
for timescale analysis as seen in chapter 3. Bibliography referring applications
of wavelets and multirresolution analysis to numerical modeling and
homogenisation has been studied [9][12][19][25]. However, due to the
complexity of the proposed methods, there has not been enough time to further
consider these approaches for THM modeling.
p
been developed. We do not expose in a detailed manner those previous studies in this
thesis report, to keep clarity and simplicity, but let us summarize here the main aspects
and conclusions obtained from those studies:
 Pseudospectral method for EDPs resolution: a 1D model for the advection
diffusion equation with timespace variable coefficients (flux velocity and
diffusion coefficient) has been developed. This model is solved by a pseudo
spectral method based in the Fourier transform described in [88]. This is a quite
fast and robust method, whose main limitation lies in
detailed formulation of the advectiondiffusion equation solved by this method
are presented in APPENDIX V.
 Dualcontinuum approach for the hydraulic coupling: this approach is based in
the equations describing reactivediffusive systems such as mixtures of several
reactive species in a fluid medium. In our case, the fractured medium is
considered as a superposition (mixture) of two continuum media (species):
fractured medium and rock matrix medium, with an excha
44
5.2 Basic assumptions and constitutive equations
5.2.1 Dimensionality and geometry
In 3D space, the fractured rock is assumed to be made up of intact porous rock
and planar disk fractures or fractures, with known lengths, orientations, and apertures.
5.2.2 Thermal processes
Full thermal coupling is modeled, including heat convection and conduction.
owever, isotropic equivalent thermal conductivity is assumed. Contrary to hydraulics,
5.2 H
constan
Pa/m).
concep
conven
assume law, which can be viewed as an approximation
the full Navier Stokes equations (neglecting inertial terms, transient effects, and non
plan
permea
and iso
5.2.4
5.2.4.1
H
no tensorial upscaling has been made for thermal coefficients, and only the
consideration of the fracture volumetric fraction has been used to compute the
equivalent coefficients. Nevertheless, for future stages of the research the same kind of
tensorial upscaling (see chapter 5.3 below) could be done.
.3 ydromechanical processes
The fracture normal and shear stiffness coefficients, K
n
and K
s
, are taken
t, independent of stress and same for all fractures (K
n
10
11
Pa/m, K
s
10
10
However, a constant aperture of fractures is assumed. Terzaghis effective stress
t [86] is used for the coupling between hydraulics and mechanics, with a
tion of negative compressive stress. The hydraulic behavior of fractures is
d to be governed by Poiseuille's
to
ar flow components within each fracture) [11][17][78]. The intact rock matrix is a
ble elastic medium, satisfying Hooke's law. Both homogeneous/heterogeneous
tropic/nonorthotropic cases are studied.
Macroscale constitutive laws and equations
Governing laws
Equations for thermohydromechanical phenomena are derived by writing
ation equations for energy, mass and momentum:
Mass balance for f
conserv
luid:
i
i
x
q
t
where
is the fluid production term [] (net va
(22)
riation of volume of fluid by unit
volume of the equivalent medium).
q
i
is the Darcy velocity [m/s].
45
Momentum balance for fluid (Darcys law):
=
i
q
j
w
j w
ij
x
z
g
x
P
k
(23)
k
ij
is the upscaled tensorial intrinsic permeability of the equivalent
] (see chapter on upscaling).
P is the fluid (water) pressure [Pa].
where
medium [m
2
w
is the water density [Kg/m
3
].
w
is the water dynamic viscosity [Ns/m
2
].
g is the gravity [m/s
2
].
z is the elevation over the see water level [m].
Momentum balance for the equivalent medium:
0 =
j
eq
j
ij
x
z
g
x
(24)
where
ij
are the stresses [Pa].
eq
is the density of the equivalent medium [Kg/m
3
], given by the
expression:
( ) ( )
s m m w m m f f eq
+ + = 1
where
(25)
f
,
m
are the volumetric fractions of fractures and matrix
f m
and matrix respectively []
=1 for waterfilled fractures).
s
is the density of the solid grains [Kg/m
3
].
:
respectively [].
, are the porosities of fractures
(
f
Heat energy balance for the equivalent medium
( ) ( )
T
j
ij T
i i
i w w eq
f
x
T
K
x x
T
q C
t
T
C +
(26)
th temperature of the equivalent medium [C],
where
T is e
(C)
eq
is the intrinsic specific heat capacity of the equivalent medium
[J /m
3
K], given by the expression:
46
( ) ( ) ( )
s s m m w w m m f f eq
C C C + + = 1 (27)
where
C
w
, C
s
are the specific heat
grains respectively [J /kg K].
(K
T
)
ij
is the tensorial thermal conductivity of the equivalent medium
m K], given by the expression:
capacities of water and solid
[W/
( ) ( )( ) ( )( )
ij Ts m m ij Tw m m f f ij T
K K K + + = 1 (
28)
the tensorial thermal conductivities of
tively (supposed in this case
niform in time).
heat flow produced by the
5.2.4.2 Constitu
where
(K
Tw
)
ij
, (K
Ts
)
ij
are
water and solid grains respec
, homogeneous and u isotropic
f
T
is the heat source term [W/m
3
] (in our case,
EBEX experiment heaters) F
tive equations
To reduce the conservation equations to a sy
dependent variables u
i
(solid displacements), p (fluid pressu.re) and T (equivalent
medium temperature), we use the following constitutive relations:
Fluid production:
stem of equations in terms of the
T P
G
B
Tw eq ij ij
+ = (
1
29)
where
B
ij
is the tensorial Biot coefficient [],
ij
are the strains [],
G is the Biot modulus [Pa],
eq
is th o
e p rosity of the equivalent medium [], given by the expression:
( )
m m f f eq
+ = (30)
Tw
is t
1
].
mal stress):
he volumetric thermal expansion of water [K
Equivalent medium stresses (Biot equation +ther
T T P B e T
Ts kl ijkl ij kl ijkl ij
= (31)
where
ijkl
is the tensorial stiffness coefficient [Pa],
Equ
T
Ts
is the volumetric thermal expansion of solid [K
1
].
ivalent medium strains:
47
i j
=
j
i
ij
x
u
x
u
2
1
(32)
where
Water density:
u
i
are the displacements on each coordinated direction (i =1, 2, 3) [m].
previous relations assume a dependency
on the temperature and pressure variations:
of the water density
( ) ( ) ( )
0 0 Pw Tw w0
1 P P T T
w
= + + (33)
As we assume tha
the
frac
cau nce density has been used:
t compressibility of water is already taken into account in
Biot modulus G, through the normal and shear stiffness of a single
ture (filled with water), we will only consider explicitly variations of
w
sed by T. The following refere
( ) ( )
3
0 0 0
/ 1000 5 m Kg T
w w
= = .
Water dynamic viscosity: there is a dependency of the water dynamic
viscosity on temperature detailed in the APPENDIX VII. The fitted model
used for the water dynamic viscosity as a function of temperature is:
+ 10 3.3185 10 7.0620
5 
2 7 3 9 4 12
T
T T T
w
=
5.2.5 System
The go ulated and combined with the constitutive
equations to ob
are temperature, T, fluid pressure, P, and the displacements in the three directions, u , u
nd u
3
, as functions of time t and space coordinates (
equations is:
10 5.6591 T
(34)
( )
0.0017 + 10 4.4658
of equations
verning laws can be manip
tain the reduced system of equations, in which the dependent variables
1 2
a x
1
, x
2
, x
3
). The reduced system of
( ) ( ) 0 =
T T
x
P B
x x x
T
kl ijkl
j
Ts ij
j k
l k
ijk
2
1
u u
x
l
j
(35)
l
+
z
g
P
B (36)
j
w
j w
ij
i
Tw eq
i
j
j
i
ij
x x
k
x t
T
t
P
G x
u
x
u
t
1
2
1
( )
T
i i
T
i j
w
j w
w
x x x x x
ij
w eq
f
T
K
T z
g
P
t
T
C
=
(37)
This system of equations can be converted into a pseudomatricial form which enters
directly n to be used (Comsol Multiphysics ). This
conversion is developed in APPENDIX VIII.
k
C
i the THM model software
48
5.3 Equivalent continuum properties
5.3.1 Introduction and generalities
We no s
pertur
based
whole dom he size of the homogenization
bdomain, or homogenization scale l
H
, will be that of the Re
olume (REV). This point will be discussed later.
The superp
equivalent
singular fr
solution fo posed by only
one fractu and a portion of rock matrix above and below the fracture. Indeed, to
perform the homogenization within each subdomain, only the portion of fractures lying
inside the subdomain has to be considered. Thus, in general, the intersection of disk
fractur
will be tak
the superp
mechanical stiffness are com matical
and physical developments considers conductivity and stiffness as tensors in the 3D
ace.
a new expression for the equivalent hydraulic conductivity
ed on previous works by [71] and [1], by considering a permeable matrix instead of
is consideration, fractured network degree of connectivity
3][55][71] is not needed for correcting the upscaled conductivity.
.3.2 Hydraulic equivalent coefficients
5.3.2.1 Upscaled conductivity of individual fractured blocks
w consider an arbitrary set composed of N fractures having variou
a es, lengths, and orientations. Equivalent homogenized properties are determined
on a linear superposition approximation, which may be applied either to the
ain, or more generally, to a subdomain. T
su presentative Elementary
V
osition approach yields to convert the discrete 3D fractured medium into an
continuum by summing up all the individual contributions due to each
acture [5]. Therefore, the first step in the upscaling is to calculate the exact
r the flow equations into an individual fractured block, com
re
es with parallelepipedshaped subdomains will yield to polygonal fractures. This
en into account in the oneblock exact calculation of the coefficients. Then,
osition is applied and coefficients such as hydraulic conductivity and
puted. The treatment followed during all the mathe
sp
In hydraulics, we develop
bas
an impervious one. Due to th
[4
In mechanics, the original expressions of those works have been used, although some
remarks are also made for future developments similar to those made for hydraulics.
5
Continuum Darcys law is stated in terms of the intrinsic permeab
However, as we consider water density and viscosity locally constant at the scale of the
individual fractured block, this is equivalent to compute the hydraulic conductivity
tensor, and convert it later into intrinsic permeability by multiply
proportionality factor. Relationship between both quantities is given by the expression:
ility tensor.
ing by the appropriate
ij
w
w
ij
k
g
K
(38)
49
We must start by decomposing the fractured porous medium into a number of building
locks (individual fractured blocks). Here we assume that it is sufficient to consider
nly the case of porous blocks traversed by a single planar fracture parallel to the axes
32, where the main dimensions of the fractured
lock are also stated. However, this type of building block could be supplemented in the
ture
b
o
of the block, as illustrated in Figure
b
fu with other prototype configurations, as for example a block with two
perpendicular fractures.
F
z
b/2
b/2
a
x
B
l
l
I
rested in studying
n auxiliary subscale problem, namely : what is the upscaled permeability (tensor) of a
ed
any kind of prismaticshaped block with irregular bases, i.e., the block resulting from
e polygonalshaped intersection of the disk fractures with the homogenization
to particularize for 2D (rectangular basic
actured building block).
nated system (individual
actured block), we use a different notation for coordinates (x, y, z) to distinguish from
the glo
RACTURED BLOCK
We produce an exact analysis of the solution
y designing an experiment such that the lowest order distribution of heads (fluxes) at
ass conservation,
Figure 32: Individual fractured block of a fractured porous medium.
Here, given the individual fractured block of Figure 32, we are inte
a
3D rectangular block consisting of a porous matrix (M) traversed by a single planar
fracture or fault (F). All the results obtained in this subscale problem can be appli
to
th
subdomain. These results are also easy
fr
Remark: as this derivation is made over a local coordi
fr
bal one (x
1
, x
2
, x
3
) used in the previous chapter. Also, compacted vectorial and
matricial notations are used instead of subscripts.
EXACT LOW ORDER ANALYSIS OF FLOW THROUGH AN INDIVIDUAL
F
for flow in a fractured porous block
b
the boundaries and inside the domain is obtained, while all the m
50
continuity conditions, and Darcys law are satisfied everywhere locally (both inside the
block and on its boundary). We develop the Darcy flux equations for the case of
tensorial conductivities. We will obtain an equivalent conductivity tensor by setting
the appropriate boundary conditions for the hydraulic head and by defining the matrix
fracture interface equations, given a fixed local hydraulic gradient
*
0
j . From now on in
this chapter, the subscript * denotes the local coordinated system (x, y, z) of the
individual fractured block, whom unitary vector is denoted by u. As will be seen, the
west order exact solution obtained for the basic block of Figure 32 is a piecewise
linear distribution of total pressure head H(x, y, z), or equivalently, a piecewise constant
distribution of the flux vector and the gradient vector.
It is emphasized that the exact flow distribution thus obtained depends strongly on the
boundary conditions of the experiment. The boundary conditions are chosen here such
that H(x, y, z) is of the lowest possible order : H(x, y, z)C
0
(continuous functions) on
R
3
, which are not necessarily differentiable. In our case, more precisely, H(x, y, z) is
not differentiable in the direction transverse to matrixfracture interface. For the case of
interest here, let us generalize the 2layer system described in [76] to a 3layer block
(matrixfracturematrix), with the following permeability tensors for each medium:
 Isotropic matrix: (39)
 Anisotropic fracture: (40)
here is the conductivity parallel to the fracture, here the Poiseuille conductivity,
lo
I K
M
M
M
M
M
K
K
K
K
=
=
0 0
0 0
0 0
F
F
F
F
K
K
K
0 0
0 0
0 0


K

F
K w
and
F
K is the quasiinfinite conductivity transverse to the fracture.
Given the 3D geometry defined in Figure 32, we look for a local flux density
( ) ( )
k
k
z y x
q q q
=
* * * *
, , q , with (k =A, B, C), such that:
( )
( ) ( )
I k
z
y
x
q
q
x
q
div
= 0 q
z y x z y x
H grad
z y
= =
, , , , ,
) ( K j K q
(41)
with B.C.:
( ) ( ) ( )
F z y x
z y x z J y J x J z y x H + + = = , ,
* * *
x j , ,
*
0 0 0 0
(42a)
gradient).
Figure 33 shows the piecewise linear type of B.C. for the hydraulic head (piecewise
constant for the hydraulic
51
z
H
J
z
A
x
H
J
x
=
Figure 33: Piecewise linear B.C. for the individual fractured block.
In the interface there must be fulfilled continuity of the flux density transverse to the
fracture and continuity of the hydraulic head and gradient parallel to the fracture:
( ) ( )
( ) ( )
( )
I
y y
x x
z z
z y x
J J
H H
x x
+
+
y y
J J
H H
y x H y x H
q q
= =
+
+
+ +
, ,
, ,
n q
(42b)
Lets solve equation 2a) and (42b) for the three different
omains of the fractured block defined in Figure 32:
 Subdomain
=
+
n q
(41) given the conditions of (4
d
A
: from eq. (41):
( ) ( ) ( )
( ) ( )
( ) ( )
( ) ( )
=
=
=
= =
A
K q
A
A
A A
A A
A A
z M z
y M y
x M x
M
M
M
M
J K q
J K q
J K q
K
K
K
* *
* *
* *
* * *
0 0
0 0
0 0
j j (43)
Particularizing for the boundaries:
( ) ( ) ( )
( ) ( )
*
0
*
*
0
* *
0
*
: 2 / ; 2 /
;
: 2 /
A
b z =
z z A
y y x x
J J l y l x
J J J J
A
A
A
= = =
= =
(44)
so we get finally:
52
( ) ( )
*
0
* *
0
*
j q j j = =
M
K
A A
(45)
which also fulfils the condition div(q
*
)=0.
 Subdomain
C
: analogously we find the same solution for this subdomain:
( ) ( )
*
0
* *
0
*
j q j j = =
M
K
C C
(46)
 Subdomain
B
: the flux density within the fracture can be determined with
eqs. (45) and (46) and the conditions given by (42b):
o Continuity of the flux density normal to
AB
(or equivalently
BC
) gives:
( ) ( )
( ) ( )
( ) ( )
*
0
* *
0
*
* *
z
M
z z M z F
J
K
J J K J K
B B
A B
= =
n q n q
(47a)
o Continuity of the hydraulic head H and gradient
=
* *
z z
q q
A B
=
F
K
u
H
parallel to
AB
(or
BC
):
( ) ( ) ( )
( ) ( ) ( )
*
0
* * *
*
0
* * *
y y y y
x x x x
J J J J
J J J J
B A B
B A B
= =
= =
(47b)
and applying Darcys law inside the subdomain
B
:
( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( )
= = =
= =
= =
* *
x F x
J K q
B B
*
0
* *
0
* *
*
0
 * *  *
*
0
 * 
z M z z
F
M
F z F z
y F y y F y
x F x
J K q J
K
K
K J K q
J K q J K q
J K q
B B B
B B B
B
(48)
As (q
*
)
A
and (q
*
)
C
can be identified with the matrix flux density q
*
M
and (q
*
)
B
is
al result can be expressed as follows: the fracture flux density q
*
F
, the fin
53

0 0
M
K
,
( )
( )
*
0
 *
0 0 j q
F F
K
*
0
*
0 0
0 0
0 0
0 0
j q
M
F
M
M
M
K
K
K
K
( )
( )
0 0
M
K
K
=
=
*
0
*
*
0
*
0 1 0
0 0 1
j j
j j
F
F
M
(49, 50)
UPSC
ous section to the individual
fractured b k e d
over the bl .
ALING TO THE BLOCK SCALE
To upscale the exact equations defined in the previ
loc scale, we have to define some kind of av rage of the flux an gradient
ock We search, thus, to define a blockupscaled equation of the type:
( ) ( ) ( )
=
* * *
j K q (51)
There are several ways to proceed with this average. Here we consider the Volume
Averaged Flow (VAF). However, in APPENDIX IX we present the approach of the
The VAF it is defined as follows:
Vectorial Surface Flux (VSF) [50][51], in order to compare both methods if desired.
( )
( )
d
d
*
*
q
q (52)
where d is the volume differential element.
the individual fractured block:
This expression can be applied to
( )
( ) ( ) ( ) ( ) ( )
( )
( )
( )
+
*
F
F
V
V
V V
q (53)
Lets denote the volumetric fraction of fracture within the block by
=
+
=
*
* *
*
M
M F F M M
V V V
q
q q
q
, and the
olumetric fraction of matrix would therefore be ( ) 1 . v Using eq. (49) yie
0 1 0
0 0 1
j q
+
+
=
F M
F M
K K
K K
(54)
On the other hand, the global gradient j over the block would be, similarly:
lds to :
( )
( )
( )
*
0


*
0 0
M
K
*
( )
( )
( ) ( ) ( ) ( ) ( )
( )
( )
( )
+ =
+
= =
* *
* *
*
*
F
F
M
M F F M M
V
V
V
V
V
V V
d
d
j j
j j
j
j
(55)
54
and using eq. (50) and the volumetric fractions within the block yields to
( )
( )
*
0
*
1 0 0
0
0
1 0
0 1
j j
+
=
F
M
K
K
(56)
eplacing in (54) with its value obtained from (56)
*
0
j R
( )
( )
( ) ( )
+
+
=
* 

*
0 1 0
0 0 1
j q
F M
K K
K K
(57)
( )
+
1
0 0
F M
1
F M
K K
Finally, we obtain the equivalent blockupscaled hydraulic conductivity ( )
K :
( )
( ) ( )
( )
M
K
=
+ = =
F
F M ij ij
K
K
K K K
K
K K
K
1
1
1 ; 0 0
0 0
*
33
 *
22
* *
22
*
11
*
K (58)
e can identify K
11
*
and K
22
*
to the arithmetic mean, denoted K
A
, of
weighted by the corresponding volumetric fractions and K
33
*
to the harmonic mean,
enoted K , of and K weighted similarly by the volumetric of fracture and matrix
Equation (58) represents the equivalent hydraulic conductivity of an individual fractured
block in the reference frame of the block: it gives the response of the block (in terms of
ux density) to any imposed head gradient (to be used now in larger scales
To obtain the final expression of the equivalent conductivity of the block
+ =
F M
K K 1
 *
11
K 0 0
*
33
W

and K
F
K
M
d
H
F
K
M
with respect to the block.
fl ).
( )
it is
( ) necessary to express ( )
*
q and
*
*
T
T T
(60)
where the equivalent conductivity tensor is finally defined by:
( ) ( )
T
A K A K =
*
(61)
Lets develop this expression by using the expression of the rotation matrix given in
APPENDIX X and the conductivity tensor
( )
2
3 3 2 3 1
3 2 2 2 1
2
3 3 2 3 1
3 2 2 2 1
1
and the equivalent conductivity tensor yields in Euler indices to:
H A H A H A
A H A A H A A
H A H A A H A A
A
A
T
K n n n n
n n n n n
K n n n n
n n n n n
K n K n K n n K n n K n n K n n
n K n n K n K
n
n
K
n
n n
K n n K
n
n n
K
n
n n n
K n n K n n K n n K
n
n n
K
n
n n n
K n K
n
n
K
n
n n
n
n n
n
n
n n
n
n n
K
K
n
n n n
n
n
n
n
n n
+ + +
+ + + +
+ + + +
=
= =
2
3 1 2 1
2
1
2
3 1 2 1
2
1
2
3
2
12 3 2 3 2 3 1 3 1
3 2
2
2
2
12
2
1
2
12
2
3
2
2
2 1
2
12
2 1
2
12
2
3 2 1
3 1 3 1 2 1 2
12
2 1
2
12
2
3 2 1 2
1 2
12
2
2
2
12
2
3
2
1
3
1 2
12
12
3 2
12
3 1
2
1 3 2
1
12
2
12
3 1
*
1
1
0 0 0
0 0
A K A
(62)
K
H
n n
n n
K
n n
n n
2 1
12 12
3 12
12 12
0 0
0
H
K n =
=
3 2
( ) ( ) ( )
H j i A j i ij ij
K n n K n n K + = =
K
Development of equations (3863) can be equally applied to any kind of fr
(63)
acture and
lock shape without any loss of generality, providing that matrix portions above and
exa tion
nal fractures, as the one showed in the third example below, to
hich we finally apply equation (63).
b
below the fracture cover completely its horizontal surface. Figure 34 shows possible
valid geometries for this ct solu . In our case we have fracture disks, represented
as equalarea regular polygons, which will be cut in the homogenization algorithm
while intersecting the homogenization domain. As a consequence of the intersection we
get irregular polygo
w
56
Figure 34: possible prismatic configurations for a valid fractured block fulfilling eq. (63).
Remark: it is important to note the result that appears when particularizing for the case
of a 2D impermeable fractured medium with isotropic Poiseuille K
F
(i,j =1,2; K
M
=0;
K
F

=K
F

):
( ) ( )
( ) ( ) ( )
( )
( ) ( ) ( )
( )
( )
j i ij F
F
j i F j i ij
F M
j i F M j i ij
H j i A j i ij ij
n n K
K
n n K n n
K K
n n K K n n
K n n K n n K
=
+ +
=
+ + =
= + =
0
1
1
0 1
1
1
1

(64)
which corresponds to the equivalent conductivity of [5] for a 2D fracture F .
edium, fractures intersect with others in an irregular way. Taking
at into account, the superposition of the individual fractured
section would also lead to matrix/fracture blocks intersecting with each other. To be
EVALUATION OF INDIVIDUAL FRACTURED BLOCK SIZE
In a fractured m
th blocks of the previous
consistent from a mass conservation point of view, we choose the block size of each
fractured block in an appropriate manner, such that the total volume of the fractured
domain is preserved. Indeed, we set the block volume associated to the fracture f,
denoted by
f
V
m
A
m
A (65)
57
if the block is prismatic, the transversal dimension of the block b can then be identified
with the fraction
m
m
A
V
ogenization dom
. Knowing that the specific surface of a fracture with respect
to the hom ain can be defined, neglecting the lateral faces of the
fracture, as
=
V
A
f
f
2
, we can also express the block height b in terms of this
parameter as follows:
=
m
m
b
2
(66)
cal; its transverse thickness b depends on the fracturing density of the total volume V
.
ecall that is the homogenization domain.
s inside , so that the hydromechanical parameters to calculate can be
liable (see figure in [56] for inhomogeneous Poisson networks).
Lets denote by
the sum of the specific surfaces of the fractures within the
domain, .From this definition of the block volume, the volumetric fractions of
fracture, , and matrix, (1), inside each prismatic block can be derived:
Equation (65) gives a local volume of each block. However, only the area of the block is
lo
R
Remark: the size of should be chosen such that fracture density is relatively
homogeneou
re
m
m
( )
2
2
= =
= =
f f
f
f f
F
a a
b A
a A
V
V
(67)
5.3.2.2 Domain upscaling: superposition approach for discharge rates
. In the
ischarge
previous section, we
have c
The objective here is to homogenize up to a macroscale domain
superposition approach, contributions of each fractured block to the global d
rate are summed up, given a known frozen head gradient J. In the
omputed the blockscale equivalent hydraulic coefficient, which describes the
hydraulic response of the block to any given global head gradient. Now, to upscale we
have to add up contributions of each individual fractured block of some measurable
quantity, such as the discharge rate. First, lets calculate the discharge rate Q flowing
out from an individual block from the flux density q obtained in the previous section,
that is:
( )
( )
( )
=
q
q
Q
2
FLOW
A
ds
F
(68)
ds is the surface differential element and A
FLOW
the projection of the outgoingflux surface of the block in the normal plane to each
component of the flux density q. Note that q is not a net flux but a flowthrough flux.
where is a diagonal matrix that express
58
Figure 35 shows an example of twocomponent flow through an inclined block and the
corresponding out x surfaces for each component.
goingflu
b
n
=
1
0
J
J
3
J
3
u
5
n
5
n
1
u
2
n
1
u
l
1
l
2
l
4
l
3
2
u
3
u
1
u
4
n
3
u
A
FLOW
matrix: projection Figure 35: Example of the
of the outgoingflux surface of the block in the
normal plane to each component direction of the
flux.
The outflow surface A
FLOW
is computed as follows for a prismatic block, whatever the
shape of the polygonal fracture is:
( )
<
=
f f f
n A A (69)
2
ki
k
k i k FLOW
where, for fractured block f , A
k
=A
f
for the base faces and A
k
=l
k
b for the lateral
ces of the prismatic block ;
ki
is the angle between unitary vectors and . Note
integral over the
teral perimeter of the block. Indeed, due to symmetry of a prismatic block face
project
k
n
i
u
fa
that for other fracture shapes, the sum for the lateral faces becomes an
la
ions into negative or positive sense for a give direction, the matrix
f
FLOW
A can be
initially computed independently of the direction of the flux (more on this in the remark
further below).
The superposition can now be carried out by adding up the individual contributions of
each block to the global discharge rate:
=
m
Q Q
m
(70)
Substituting the results obtained in the previous equations into (70), the global discharge
te yields to: ra
( ) ( ) ( ) ( )
( )
+ + +
=
<
m
j
F M
m
j
m
i F M
m m
j
m
i ij
k
k i
m
k i
J
K K
n n K K n n n A Q
ki

1
2
(71)
m m
m m
1
1
59
To convert into hydraulic conductivity we have to divide by the total outflow surface on
each direction. The equivalent hydraulic conductivity finally yields to:
( ) ( ) ( ) ( )
( )
( )
<
<
+ +
=
f k
f
k i
f
k
f F
f
M
f
f
j
f
i F
f
M
f f
j
f
i ij
k
f
k i
f
k
ij
ki
ki
n A
K K
n n K K n n n A
K
2
2
1

1
1
(72)
Remark: In expression (72), the equivalent hydraulic conductivity seems to depend
indirectly on the direction of the global hydraulic gradient, by terms of the outgoing
flux area A
FLOW
defined in equation (69). However, if flow goes in the contrary
direction, it is easy to show that the sum of the area projections in the normal plane is
exactly the same as computed for the other sense. Therefore, the equivalent hydraulic
conductivity tensor, or more precisely, the corresponding equivalent intrinsic
permeability, def .e., is really an
intrinsic property arge rates by
multiplying for flux direction
appears.
ARTICULAR CASES
t general expression of the equivalent hydraulic
onductivity for the domain defined in Figure 32. Lets see some particular cases to
heck the behaviour of this formula to approximate frac
ples, we distinguish between different configurations of the fractured
1
ses coincides with the domain block, and b=l
3
. Also, the equivalent
an appreciate the effect of the flow direction:
me for the three cases,
discharge rates vary according to the flow d
 Inclined single fracture (Figure 36d): corresponds to a fractured medium in
ined here is independent of the flow direction, i
of the medium. Afterwards, when computing the disch
the hydraulic gradient, then the dependence on the
P
Equation (72) is the mos
c
c tured media.
In the first exam
network. The domain is a cubic block of V
=l
1
x l
2
x l
3
crossed by fractures completely.
Fractures are identified by the cardinals F=1,2, etc; fracture area by A , A
2
, etc. Both
cases with K
M
0 and K
M
=0 are showed:
 Parallel single fracture (Figures 36a, 36b, and 36c): correspond to a fractured
medium with only one fracture, parallel to the domain faces. This is exactly the
same configuration as the individual fractured block used above for the exact low
order flow computation. Therefore, the corresponding individual fractured block
for these ca
conductivity tensor is the same as the exact conductivity tensor of the loworder
computation. However, here we c
although the equivalent hydraulic conductivity is the sa
irection on each case.
which the only existing fracture is inclined. Thus, the individual fractured block
for the upscaling will also be inclined, as showed in the figure, and
2
3
2
1 3 1
l l l l A V b
f
+ = =
.
60
 Simplest cartesian network (Figures 36e): corresponds to a 3D crossshaped
medium , with three fractures perpendicular to each other and intersecting in the
center of the domain. In this case, ( )
1
1
3
1
2
1
1
+ + = = l l l A V b
f
.
We can also consider other type of particular cases related to the nature of the fracture
itself. Equation (72) corresponds to the case of a permeable medium with a faultlik
fracture, that is, a fracture filled with porous material surrounded by the porous matrix
If the filling of the fracture is isotropic, then . However, this expression can
lso be particularized to other cases according to the nature of the fracture: for example,
e
.

F F
K K =
a
if the fracture is opened, then =
F
K and
.  Pois
F F
K K = , where
. Pois
F
K is the Poiseuille
conductivity given by the expression:
( )
=
12
2
.
f
Pois
F
a g
K . Some particular cases may be:
 Impermeable faultlike fractured medium ( 0 =
M
K ):
( ) ( ) [ ]
( )
<
=
f k
f
k i
f
k
ij
ki
n A
K
2

<
f
F
f f
j
f
i ij
k
f
k i
f
k
ki
K n n n A
2

(73)
Permeable fractured medium (
. 
;
Pois
F F F
K K K = =
):
( ) ( ) ( ) ( )
( )
( )
<
<
=
f k
f
k i
f
k
f k
ij
ki
ki
n A
K
2
2
 Impermeable fractured medium
+ +
Pois f f f f f f
n K K n n n A 1
.
f
M f
j
f
i F M j i ij k i k
K
n
1
(74)
(
. 
; ; 0
Pois
F F F M
K K K K = = =
):
( ) ( ) [ ]
( )
<
<
=
f k
f
k i
f
k
f
Pois
F
f f
j
f
i ij
k
f
k i
f
k
ij
ki
ki
n A
K n n n A
K
2
2
.
rk: Developing, again, the case of a 2D impermeable fractured medium (BMT3
ields to:
(75)
Rema
case)
y
( ) ( )
=
F
j j i ij
F
F
F
j j i ij i
J n n
g
a J
a g
A
a L
n n a Q
2
3
2
12 12
e expression obtained in [5].
(76)
which corresponds exactly with th
61
62
FRACTURED DOMAIN K
M
0 K
M
=0
=1
a.
F
1
n
= J
A
FLOW
K A Q
Q
Q
H
=
=
3
0
0
J
3
J
1 3
2
1
0
0 =
ij
K
K
K
K
0
0
0 0
=
H
A
A
0
0
3
2
1
=
=
0
0
0 =
Q
Q
Q
=
ij
K
0 0 0
0 0
0


F
F
K
K 0
=1
F
1
n
= J
A
FL
b.
0
0
1
J
OW
0
0
3
2
1 2 1
=
=
=
Q
Q
b l Q
ij
K
K
K
K
0 0
0
0 0

1
=
=
=
Q
Q
K l
F
0 0
0
0
J K
A
=
H
A
A
0
1 1 2
J a
=
ij
K
0
0
3
2
Q
0
0
0


F
F
K
K
2
1
J
=
3
J
J J
A
FLO
1
n
F=1
W
c.
2 1 2
1 2 1
J K A
J K b l Q
J K b l Q
H
A
A
=
=
3 1 3
Q
=
=
H
A
A
ij
K
K
K
K
0 0
0
0 0
2
1
=
=
l Q
l
F
0 0
0 0
0

F
K
0
3
2

2 1
1

1 2
=
Q
J K a
J K a
F
=
ij
K
0
Q
0
0

F
K
Fig 36: Results of the global discharge rate (eq. 68) and the equivalent aulic conductiv eq cular ity ( . 72) for some parti hydr ure cases.
F=1
n
=
3
2
1
J
J
J
J
A
FLOW, 1
A
FLOW, 2
A
FLOW, 3
b
d.
J
K
A
+ +
=
+ =
+
+
+
=
1 3
2
3
2
1
1
1
2
3
2
1
3
2 3
2
2
3
2
1 2
3 1
2
3
2
1
1
2
3
2
1
1
2 1
2 2
2 2
K K K
l l
l
l l
l
l b Q
J K l l b Q
J
K K
J
K K
l
A
l l
l
l b Q
H A H
A
A H A H
+ J A
3
l
l
+
=
2
0
2
0 0
2
0
2
A H A H
A
A H A H
ij
K K K K
K
K K K K
K
( )
( )
1 3
1 1
3 2
2
3
2
1
3
2 l l
F
+
=
1

2
 2
3
2
1 1 2
3 1
3 1
1 2
2
3
2
1
1
2
J J
b
l A
l l
K a
Q
J K l l a Q
J J
b
l A
l l
l l
a
Q
F
+ =
+
+
=

1
K
F
=
2
0
2
0 0
2
0
2
 

 
F F
F
F F
ij
K K
K
K K
K
b
F=1
1
n
=
3
2
1
J
J
J
J
A
FLOW, 1
2
n
3
n
F=2
F=3
A
FLOW, 2
A
FLOW, 3
e.
( ) [ ]
( ) [ ]
( ) [ ]
3 2 1 3 3
2 3 1 2 2
3 2 1
J K b l l K A Q
J K b l l K A Q
J K b l l Q
A H
A H
A H
+ + =
+ + =
1 1
K A + + =
( )
( )
( )
( )
( )
( )
( ) j i K
b l l A
K b l l K A
b l l A
K b l l K A
=
2
b l l
K b l l K A
K
ij
A H
A H
A
ii
=
+ +
+ +
+ +
+ +
+ +
+ +
0
2 1 3
2 1 3
3 1 2
3 1
3 2 1
3 2 1
A
H
( )
( )
( )
( )
( )
( )
3

3 2 1
3 3 2 1
3
2

3 2 1
2 2 3 1
2
1

3 2 1
1 1 3 2
1
A
Q
=
J K
A A A
A a l
Q
J K
A A A
A a l l
Q
J K
A A
A a l l
F
F
F
+
+
=
+ +
+
=
+ +
+
l
+
( )
( ) ( ) ( )
( )
( ) ( ) ( )
( )
( ) ( ) ( )
( ) j i K
A A A b l l A
K A a l l
A A A b l l A
K A a l l
A A A b l l A
+
1
K A a l l
K
F
F
F
ii
+ + + +
+
+ + + +
+
+ + +
+
=
3 2 1 2 1 3

3 3 2 1
3 2 1 3 1 2

2 2 3 1
3 2 1 3 2

1 1 3 2
ij
= 0
Figure 36(cont.): Results of the global discharge rate (eq. 68) and the equivalent hydraulic conductivity (eq. 72) for some particular cases.
63
64
5.3.3 Mechanic and hydromechanic equivalent coefficients
A strainbased superposition approach was developed in [71] to obtain
equivalent hydromechanical laws for an elastic rock containing many fractures. The
individual fractures or fractures were assumed to behave elastically or quasilinearly
under compression and shear [71], and to satisfy Terzaghi's effective stress
approximation [86].
Our implementation assumes linear elastic laws with constant coefficients. The mean
strain, due to the imposed global stress tensor
ij
, is calculated by linear superposition of
the local displacements occurring throughout the intact rock matrix and the discrete
fractures, keeping the global stress constant. This leads to linear hydromechanical laws
coupling solid stress and fluid pressure to solid strain and fluid strain (or fluid
production), similar to the poroelastic laws developed earlier by Biot [13][14].
The homogenized hydromechanical relations developed by [71] are summarized below.
These relations were obtained by applying a superposition principle to the
displacements (and strains) occurring, respectively, in the fractures and in the rock
matrix between the fractures, for a given stress field (assumed the same
throughout the rock mass), similarly to the approach followed for the hydraulic
homogenization. Odas relations are expressed in terms of probability distribution
integrals, and BMT3 work adapted them to discrete sums of empirical distributions over
the whole set of fractures. These are the expressions we have used for our work.
The mean (homogenized) fractured rock strain tensor is related to total (global) stress
and to fluid pressure by the following relation (we use overbars to designate the
averages or homogenized quantities coming from Oda's work):
p B T
ij kl ijkl ij
+ = (77)
with:
ijkl ijkl ijkl
C M T + =
ijkl
T =Total homogenized compliance tensor
ijkl
M =Compliance tensor of the isotropic rock matrix
ijkl
C =Compliance tensor due solely to fractures
ij
B = Homogenized strainpressure coupling coefficient (barred Biot
coefficient)
The 'matrix' compliance  characterizing the perfectly elastic, isotropic, intact rock
matrix  is given by the 4
th
rank tensor:
( )
kl ij jk il jl ik ijkl
E E
M
+
+
=
2
1 1
where E is Young's modulus (pressure units) and is the Poisson ratio (dimensionless).
'global'
(78)
The 'fractures' compliance  characterizing the specific contribution of the fracture
stem to the total compliance of the fractured rock mass  is given by the 4
th
rank
tensor:
sy
ijkl ijkl ijkl
G
g
F
g h
C
1 1 1
+
= (
79)
re, no
onnectivity issues are needed in this approach [55][71].
In this equation, F
ijkl
and G
ijkl
are geometric tensors independent of elastic properties of
the matrix or fractures, and related only to fracture density and to the distributions of
fracture orientations, lengths, and apertures. The precise expressions of these tensorial
parameters will be given below, after [71]. It should be emphasized right away that
these socalled 'geometric' tensors do not depend on any other configuration parameter
of the fracture network; in particular, they are not explicitly related to the spatial
locations of individual fractures, of fracture intersections, etc. Therefo
c
The parameters ( , ) h g are directly related to the mean elastic properties of the
fractures, namely, their shear stiffness modulus (K
s
) and their normal stiffness modulus
(K
n
). Here we assume that the individual fractures have constant shear stiffness and
normal stiffness moduli, K
s
and K
n
, and we use an approximation of Oda's expressions,
which can be expressed as follows:
l K h
n
(80)
l K g
s
where l is the mean length of fractures over the homogenization domain. In our case,
r planar circular fractures, l is the mean diameter of fractures, R 2 . fo
s original formulation, the fracture stiffness coefficients were taken to be linear
nctions of the ambient effective stress (projected on the fracture's normal vector).
da's model leads to increasing (shear and normal) stiffnesses for a fracture that is
decreasing stiffness for a
fracture th tensile normal stresses. Future
developmen ude dependencies as the
ones used i
The socal
Geom F
In Oda'
fu
O
being closed by compressive normal stresses, and also,
at is being opened or reopened by
ts of the hydromechanical upscaling could incl
n [71].
led 'geometric tensors', F
ijkl
and G
ijkl
, are given below:
etrical Tensors
ijkl
and F
ij
(Empirical discrete sums instead of Odas integral
expressions):
( ) ( ) ( ) ( )
( ) ( )
=
=
= =
=
N
f
f
j f i f f ijkk ij
N
1
f
f l f k
f
j f i f f ijkl
n n F F
n n n n F
1
2
1
2
(81)
1
65
Geometrical Tensor G
ijkl
(Derived from F
ijkl
):
( )
ik jl jk il il jk jl ik ijkl
F F F F G + + + =
4
1
(82)
ations, represents the length of fracture f (fracture diam
In these equ eter), n
f
the unit
vector normal to the fracture f, and the specific area of fracture f with respect to the
f
homogenization domain, which was defined earlier.
Finally, we now give an explicit expression for the remaining tensorial coefficient, B ,
that couples strain to fluid pressure in Oda's strain equation (77). According to [71],
this 2
nd
rank tensor is related to the 4
th
rank fracture compliance tensor C , as follows:
ijkk ij
C B = (83)
A more explicit result can be obtained after calculating the contracted fracture
compliance tensor (on the righthand side). To obtain the final result, we note first that
the contracted tensor G
ijkl
yields F
ij
, based on equation (215). Using this identity, it can
then be seen from equation (211) that the contracted tensor
ijkl
C yields h F
ij
. In short,
the final result is:
ij ij
F
h
B
1
= (84)
On the other hand, Oda's relations lead to the pressure/strain constitutive law of fluid
production given in eq. (5), neglecting here the thermal term. This relation can be
restated as follows:
( ) =
kl kl
e B G P
(
which is exactly of the same form as the corresponding anisotropic Biot equation. This
law involves the scalar Biot modulus G and the 2
nd
rank tensorial coefficient of Biot
B
85)
ij
. For a fractured rock mass, G is given by any of the following formulas (they are all
equivalent):
( ) { }
( ) { }
1
ij ij ij
( ) { }
1
= B B T
1
=
=
ij ij kl ijkl
kl ijkl ij ij
B T B
B B G
(86)
while B
ij
and the associated (barred) quantity are given by:
h F B B T B
ij ij kl ijkl ij
= = ; (87)
The full derivation of these formulas for the Biot coefficients can be found in [5].
66
5.3.4. Implementation and results of the upscaling
.3.4.1 REV study and moving average
5
The volume of the homogenisation subdomains has been determined by means
m
omogenization formula (72).
bdomain. Figure 37 shows the result of this calculation, where the block size of 20
eters has been set to be the REV of the fracture
ogenization blocks containing any portion of these fractures will
roduce a somehow smoothed profile of hydraulic conductivity in the transversal
irection. This effect has coherence with the na
faultylike damaged zone around them that decreases the conductivity gradually towards
e rock matrix.
Figure 37: REV determination for K
ij
in the simulated fractured medium.
of the Representative Elementary Volume (REV). It is defined as the minimu volume
where computed parameters can be considered to be ergodic or statistically stationary.
In our case, the minimum block volume has been estimated by with respect to the
quivalent hydraulic conductivity resulting from the h e
To determine the REV of our fractured medium, we have computed the mean square
root of the diagonal components of the hydraulic conductivity tensor (square root of the
sum of squares) for increasing homogenisation sizes in different points of the
su
m d medium.
It is important to note that the 17 fixed fractures have been taken out from the REV
study, because they have been created with different statistics than the rest of the
fractured medium (uniform distributions instead of power law for the fracture size pdf).
Afterwards, the big fractures will be included in the homogenisation with the chosen
REV, so that these important features are taken into account for the hydraulic model.
oing so, the hom D
p
d ture, as the big fractures use to have a
th
P=(0,0,0)
P=(0,65,0)
P=(0,65,0)
P=(0,30,0)
P=(0,30,0)
P=(15,0,0)
P=(15,0,0)
67
Once the optimized fractured medium has been obtained and the upscaling formula has
een defined, the computation of the equivalent parameters can be achieved. This has
indow with a window size equal to the REV of
e discrete jump of the window has been set to
0m, i
rnal
b
been done through a moving average w
he fractured medium, that is, 20m. Th t
1 n order to be approximately of the same order of magnitude than the finite
element mesh mean size. The resulting moving average will lead to matrices of
equivalent coefficients of size 5x18x5 for each of the homogenized parameters.
Implementation of the upscaling routine has been done in MATLAB. Due to the
normous size of the fracture data file, and to memory limitations, an inte e
subroutine to compute intersections by subsets of the fractured data has been developed.
The algorithms followed to compute the upscaled coefficients are presented in Figures
38 and 39.
INITIALIZE variables:
k T B G
f
ij
,
ijkl
,
ij
, , ,
F
ijkl
, F
ij
, G
ijkl
, C
ijkl
, M
ijkl
, T
ijkl
For
1: Subdomain
For
1: N
subset
: N
fractures
Compute Fracture
Intersections (*)
Load fracture data
Add intersecting fractures to
subdomain intersections array
Compute equivalent upscaled
coefficients:
k
ij
, T
ijkl
, B
ij
, G,
f
,
SAVE computed variables:
k
ij
, T
ijkl
, B
ij
, G,
f
,
F
ijkl
, F
ij
, G
ijkl
, C
ijkl
, M
ijkl
, T
ijkl
Figure 38: Algorithm of the upscaling process.
The algorithm of fracture intersections with the homogenization subdoma
(*)
in is
showed in the next figure.
68
LOAD fracture data matrix:
(node coordinates of regular polygon
approximating circular fractures)
Exclude fractures outside
the subdomain
Compute intersection of fractures
bigger than the subdomain
Compute intersection of the
rest of fractures
SAVE fracture intersection matrix
Compute intersections of
lines with subdomain faces
Choose fracture center or
interior point close to it
Build lines parallel to the
axes passing through the
interior point
Compute intersections
of lines with
subdomain edges
Compute intersections of
fracture polygon edges
with the subdomain faces
Build lines in the fracture
plane passing through the
face intersection points.
Build intersecting irregular
polygon coordinates matrix
Figure 39: Algorithm of fractured medium intersections with the homogenization
subdomain.
5.3.4.2 Oneblock homogenization
A oneblock homogenization has been performed. Equivalent coefficients for the whole
domain have been computed. Results show a quasi isotropic equivalent medium in
hydraulic conductivity and a quasiorthotropic medium for the mechanical stiffness
tensor. The equivalent intrinsic permeability of the fractured medium is:
(88)
with the following eigenvalues and eigenvectors of the principal axes directions:
0 3157 . 0
2163 . 0 7708 . 0 5495 . 0
4242 . 0 6269 . 0 7735 . 0
; 10
107 . 1 0 0
0 142 . 1 0
0 0 055 . 1
2 18
v m k
ij
with
Figu sic
permeability.
2 18
10
099 . 1 013 . 0 009 . 0
015 . 0 112 . 1 034 . 0
017 . 0 043 . 0 092 . 1
m k
ij
=
8794 . 0 1133 .
re 40 shows four different views of the ellipsoid of equivalent intrin
69
Figure 40: Equivalent intrinsic permeability ellipsoid for the oneblock
homogenization of the fractured medium.
The equivalent mechanic stiffness of the fr red medium, in reduced matricial form,
is:
9
10
0514 . 0 1679 . 0 3643 . 0 3
1589 . 0 0146 . 0 2142 . 0 4194 . 2 5982 . 3 4461 . 2
1338 . 0 0212 . 0 2865 . 2 4461 . 2 1096 . 3
=
e approximation of the stiffness tensor to orthotropic medium would be:
0 0 0 4194 . 2 5982 . 3 4461 . 2
0 0 0 2865 . 2 4461 . 2 1096 . 3
he first quarter of the matrix (we denote it T
ij
) can be represented as the permeability,
actu
0167 . 0
393 . 5 4194 . 2 2865 . 2
T Pa
ijkl
8336 . 0 0660 . 0 0161 . 0 0514 . 0 1589 . 0 1338 . 0
0660 . 0 5725 . 0 0777 . 0 1679 . 0 0146 . 0 0212 . 0
0161 . 0 0777 . 0 8816 . 0 3543 . 0 2142 . 0 0167 . 0
Th
Pa T
l
9
10
0 0 0 3933 . 5 4194 . 2 2865 . 2
=
ijk
8336 . 0 0 0 0 0 0
0 5725 . 0 0 0 0 0
0 0 8816 . 0 0 0 0
T
by computing its principal axes (eigenvectors) and principal components (eigenvalues)
(see Figure 41):
70
There is one order of magnitude of difference between the minimum and the maximum
stiffness principal components. The direction of anisotropy could be influenced by the
big fractures orientation.
Figure 41: Equivalent reduced stiffness tensor ellipsoid for the oneblock
homogenization of the fractured medium.
The Biot coefficient for the homogenized medium is:
9271 . 0 0163 . 0 0022 . 0
0022 . 0 0186 . 0 9401 . 0
with the following principal components and directions:
lipsoid.
= 0163 . 0 9411 . 0 0186 . 0
ij
B
These values of the Biot coefficient imply that there exist a high coupling between
hydraulics and mechanics, and hydraulic pressure is transmitted to the matrix in form of
stresses in more than a 90% for all directions, almost isotropic. Figure 42 shows the
orresponding el c
71
Figure 42: Equivalent Biot coefficient ellipsoid for the oneblock homogenization of
the fractured medium.
The homogenized Biot modulus G is:
he rest of intermediate 2
nd
and 4
th
rank tensors of the oneblock homogenization can b
ound
.3.4.3 Moving average homogenization
Pa G
10
10 1877 . 4 =
T
f
e
in the APPENDIX XI.
5
ogenization yield to a 5x18x5
parameter values spatially located within a partition of the domain. Main features
of the fractured medium can be appreciated in the heterogeneity of the parameters along
the domain.
tridimensional view of the intrinsic permeability ellipsoids is displayed in Figure 43a.
plane for the five Xlayers of
rientations mostly NNWSSE
amily 2) influences the heterogeneous distribution of permeability. Thus, fractures
crossing the FEBEX drift at the test zone provide a preferential NNWSSE flowpath. In
the rest of the domain, the permeability heterogeneities are less remarkable, although
clear differences among permeability ellipsoids in contiguous subdomains can still be
ppreciated. Big fractures also have an effect in the equivalent stiffness tensor (Figure
32
is also represented in Figures 47ae as spheres.
The results of the moving average process of hom
set of
A
Figures 44ae show projections of the ellipsoids in the YZ
omogenization. The presence of big fractures with o h
(f
a
43b and Figures 45ae), which gets smaller (softer medium) as the fracture density
increases. This will imply a higher deformation in the FEBEX area than in the rest of
the domain. However, Biot coefficient (Figure 43c and Figures 46ae) and Biot modulus
are much more homogeneous within the domain partitions. The volumetric fracture
ensity (scalar values) d
72
a.
b.
c.
Figure 43: Hydraulic and hydromechanic equivalent coefficients for the moving
average homogenization. a. Equivalent intrinsic permeability k
ij
; b. Equivalent
stiffness tensor T
ijkl
(only T
ij
with i,j=1,2,3); c. Equivalent Biot coefficient B
ij
.
73
a.
b.
c.
d.
e.
igure 44: Equivalent intrinsic permeability k F
ij
for the five Xlayers of the moving
average: (k
ij
)
min
=1.0910
18
m
2
, (k
ij
)
max
=6.9110
18
m
2
.
74
a.
b.
c.
d.
e.
Figure 45: Equivalent stiffness tensor T
ijkl
(only T
ij
with i,j=1,2,3) for the five X
the moving average: (T layers of
mi x
10
10
Pa.
ij
)
n
=8.1810
8
Pa, (T
ij
)
ma
=3.21
75
a.
b.
c.
d.
e.
Figure 46: Equivalent Biot coefficient B
ij
for the five Xlayers of the moving
average: (B
ij
)
min
=0.73 , (B
ij
)
max
=1.
76
a.
b.
c.
d.
e.
Figure 47: Equivalent volumetric fracture density for the five Xlayers of the moving
average: (
f
)
min
=1.5410
5
, (
f
)
max
=2.5310
5
.
77
78
6 IMPLEMENTATION AND RESULTS OF THE THM MODEL
6.1 Domain and problem definition
The domain is a block with the Geographic North oriented towards X. The dimensions
of the block are 70m x 100m x 70m, with the origin of coordinates in the center of
block. There are three connected drifts: the Main tunnel, the Laboratory tunnel and the
FEBEX drift, the last one being centered in the origin of coordinates. In the test zone of
the FEBEX drift there exist a heating process defined here by a temperature gradient
from 100C at r=0m to 35C at r=1.14m, being r the radial direction in the FEBEX drift.
Figure 1 shows the domain considered, along with the nomenclature for the boundaries.
N
FEBEX test zone
FEBEX drift
Laboratory tunnel
Main tunnel
Figure 48: Domain of the THM model and boundaries nomenclature.
The problem has been simulated in three different stages:
 Hydrolithostatic equilibrium of the rock mass: at this stage, there is no drift
and a fully saturated 365m rock mass is assumed to be lying over the upper
boundary of the domain. Both the hydrostatic and the lithostatic loads have
been imposed gradually for the timedependent analysis. The load profiles
used are polynomial functions and are described further below. Relative fluid
pressure is computed in all the models (PP
atm
).
 Drifts excavation simulation: the HM response of the fractured rock is
analyzed during the excavation of the drifts, which has been modeled by
gradually decreasing both the normal stresses and the fluid pressure in the
boundaries of the excavated tunnels. The same kind of polynomial functions
of the previous stage has been used for the decreasing profiles. Real
hydraulic conditions existing in the Grimsel Test Site have been used this
time.
79
 Heating experiment simulation: at this stage, the full THM model is used. A
3year heating process has been simulated around the FEBEX test zone (last
17 m of the FEBEX drift). Heat load profile is determined by the FEBEX
ions. Two possible cases have been considered:
the test zone is filled with bentonite [79][92][93] and the access drifts are not
filled with any material; the test zone is filled with bentonite and the access
Insitu experiment condit
drifts are filled with some material similar to the surrounding rock. Only
results corresponding to the first case are presented here.
Three different HM conditions of the rock have been compared in the analyses (the
thermal conditions are in all the cases isotropic and homogeneous):
HMisotropic homogeneous material: with simplified values coming from
the 1x1x1 homogenisation of the fractured rock domain.
Hanisotropic and Mnonorthotropic homogeneous material: with the full
set of values of the 1x1x1 homogenisation of the fractured rock domain.
Hanisotropic and Mnonorthotropic heterogeneous material: with the
5x18x5 homogenisation of the fractured rock domain.
The thermal transient is imposed and consists in a 3year heating according to the
FEBEX insitu experiment specifications. The thermal load is gradual, following the
temperature profile of the experimental measurements. Two different loading functions
have been tested (Figures 51a and 51b). The first one is an exponential function (Figure
49a), but it leads to convergence problems in the coupling with hydraulics. Thus, a
polynomial function (Figure 49b) with zero derivative in t=0 has been used for the TH
and THM models. At the maximum temperature point, the functions are defined as:
 Exponential function:
=
0
1 100
t
t
e T
 Polynomial function:
>
0
100 t t if
s 510
= 0 0
3 2 100
t
t
t
t
T
0
2 3
t t if
Where t
0
i
loading tra
The hidro hydrostatic and lithostatic
stabilisation of the rock mass due to the 365m rock weight existing over the domain
(FEBE d
both hydra
has to be
reflected in
loading tim
each of th
loadings is the same as the polynomial function used for the thermal loading.
6
s =57.87 days, approximately equal to the two month of the heat
nsient of the FEBEX experiment to reach the 100C.
mechanical transient has consisted in the
X rift is ~400m deep). A gradual loading has been used, as stated before, in
ulic and mechanic models to assure convergence. However, the loading rate
sufficiently fast so that the coupled hidromechanical processes can be
the results without the influence of the loading conditions. Therefore, the
e has to be at least 10 times smaller than the characteristic diffusion time for
e processes. The function shape used for both hydraulic and mechanic
80
a. b.
Fig
A quick a the
characteristic time for the hydraulic process and for the coupled hydromechanic process.
The ori a
ure 49: Two heating profiles for the experiment simulations: a. Exponential
function; b. Polynomial function.
nalysis of the hydraulic equation can yield to the determination of
gin l hydraulic equation of our system is:
j
w
j w
ij
i
Tw eq
i
j
j
i
ij
x
z
g
x
P
k
x t
T
t
P
G x
u
x
u
t
B
1
2
1
Let us now consider only the terms in fluid pressure and constant values of the
parameters for all the subdomain (1block homogenization):
j i w
ij
P
k
P 1
x x
=
t G
We can e
th n write:
w
ij
H
j i
H
k G
R
x
P
x
R
t
P
with
here R
H
can be seen as a characteristic diffusion rate for hydraulics.
block homogenization,
H
yields to:
w
Considering the values of the parameters corresponding to the 1
R
046 . 0
10 1
10 1 . 1 10 1877 . 4
6
18 10
=
H
R m
2
/s
and dividing some characteristic length of the domain, in our case L=70m, by the
hydraulic rate R
H
, we get the characteristic time of the hydraulic process:
h 29 s 10 06 . 1
046 . 0
70
5
2 2
= = = =
H
R
L
t
H
81
To illustrate this concept, we plot the hydraulic pressure in two points at the top and
bottom boundaries, for two different loading time scales. In the first case, the hydraulic
loading time (t
0
=5e7s) is faster than the characteristic time of hydraulics, and the
loading process at the bottom boundary (Figure 49b) is mainly due to the diffusion of
the pressure introduced at the top boundary (Figure 49a). In the second case, the
hydraulic loading time (t
0
=1e10s) is slower than the characteristic time, and loading
process at the bottom (Figure 49d) reproduces the loading at the top (Figure 49c). We
have taken the same t
0
than in the thermal transient, i.e., t
0
=5e6s, as loading time scale
for the hydraulic problem.
ble 2 resumes the numerical experimen p ned for the model. Both the final steady
te analysis and the time dependent analysis have been carried out for each case:
Ta ts
a
lan
st
b. a.
Figure 50: Comparison between the effects of a hydraulic load at the top boundary in
the bottom boundary for two different loading times.
Table 8: Numerical experiments plan.
HOMOGENEOUS
ISOTROPIC
HOMOGENEOUS
ANISOTROPIC
NO
ORTHOTROPIC
HETEROGENEOUS
ANISOTROPIC
NON
ORTHOTROPIC
homogeniz.)
N
(1x1x1 homogeniz.) (5x18x5
1) Hydrolithostatic rock
t=1e9s=31.7years
Hydrolithostatic rock
t=1e9s=31.7years
Hydrolithostatic rock
t=1e9s=31.7years
HM
mass equilibrium up to mass equilibrium up to mass equilibrium up to
2)
HM
Drifts excavation and
stabilisation up to
t=1e8s=3.17years
Drifts excavation and
ilisation up to
t=1e8s=3.17years
Drifts excavation and
stabilisation up to
t=1e8s=3.17years
stab
3)
THM
Heating up to
t=1e8s=3.17years after
Heating up to
t=1e8s=3.17years after
Heating up to
t=1e8s=3.
hidrolithostatic
equilibrium
hidrolithostatic
equilibrium
17years after
hidrolithostatic
equilibrium
82
The following constant values have been used for the different properties of the
materials:
Constants
 Gravity:
2
/ 81 . 9 s m g =
Matrix/fracture properties
Fracture Matrix
5
 Volumetric fraction: 10 94 . 1 =
f
; 99998 . 0 1 = =
f m
 Porosity: 1 =
f
; 008 . 0 =
m
Water/solid grains properties
Water Solid
 Density:
3
0
/ 1000 m Kg
w
= ;
 Specific heat capacity:
3
/ 2350 m Kg
s
=
kgK J C
w
/ 4180 = ; kgK J C
s
/ 850 =
mK W K
Tw
/ 58 . 0 = ; mK W K
Ts
/ 1 . 2 =  Thermal conductivity:
1 4
10 421 . 4
= K
Tw
;
1 5
10 1 . 2
= K
Ts
 Thermal expansion coef.:
Equivalent porous medium
 Equivalent porosity:
410 m Kg
s f f m m w f f m m eq
=
 Equivalent th
f Tw
3
10 02 . 8 +
= =
f f m m eq
 Equivalent density:
2.3 ))  (1 + )  (1 ( + ) + ( =
3 3
/
ermal conductivity:
mK W K
Ts
/ 088 . 2 )) = K K
f m m
 (1 + )  (1 ( +
f f m m Teq
) + ( =
 Equivalent heat capacity ter
) ) + ( ) ( =
H c iz e .
Simplification n used for the isotropic case:
B T
ij IJ
10
0 1 . 1
0 0
0 0
9 . 0
0 0 0
0
3 0
0 3
0
=
m:
K m J C
3 6
/ 10 015 . 2 ) = C  (1 + )  (1 ( + C
s s f f m m w w f f m m eq
ydrauli and mechanic homogen
s of them have bee
ed coefficients are thos presented in Chapter 5
G m
2 18
10 4 ; 10
1 . 1
=
k
ij
0 0
0 ;
9 . 0 0 0
0 9 .
Pa
9
; 10
0 0 0
0 0 0 0
0 0 0 0
0 0
1 . 1 0 0
0 0
0
0
0
0 0 0
0 0 0 0
Pa
83
Constrains for the excavated drifts and the heating zone have been defined with
of logical e
the help
xpressions identifying the different geometrical features of the domain, and
ependent of the spatial coordinates. Logical expressions have been constructed with
e following local cylindrical coordinates:
d
the aid of th
( )
2 2
48 2 y MAINr = z +
x MAINx =
( ) ( ) ( ) ( ) ( 15 sin 2 x LABr = )
2 2
48 15 cos 35 z y + + +
( ) ( ) ( ) ( ) 48 15 sin 35 15 cos + = x LABx y
( ) sin 2 FBXr = ( ) ( )
2 2
80 cos 80 z y x + +
( ) ( ) y x FBXx = 80 sin 80 cos
 Installation zone in the FEBEX drift:
( ) ( ) ( ) 3 . 15 35 14 . 1 2
2
< = FBXinst FBXx FBXx FBXr
BEX drift connecting with the Laboratory tunnel:  Auxiliary zone in the FE
( ) ( ) ( ) 35 162366 . 41 14 . 1 2
2
< = FBXx FBXx FBXr FBXaux
 Test zone in the FEBEX drift:
( ) ( ) ( ) 35 18 14 . 1 2
2
< = FBXx FBXx FBXr FBXtest
 eaters location with H in the test zone in the FEBEX drift:
( ) ( ) ( ) [
( ) ( ] 425 . 32 885 . 27
865 . 26 325 . 22 14 . 1 2
2
< +
=
FBXx FBXx
FBXx FBXx FBXr FBXheat
)
 Grimsel main access tunnel:
( ) ( ) ( ) 35 35 75 . 1 2 = MAINr MAINtun
2
MAINx MAINx
Grimsel laboratory tunnel:

( ) ( ) ( ) ( ) 15 cos 70 0 75 . 1 2
2
= LABx LABx LABr LABtun
he mesh of this problem has 11209 elements, from which 2620 are boundary elem T
and 800 are edge elem
ents
ents. The element type is Lagrange quadratic for mechanical
variables and Lagrange linear for thermal and hydraulic variables. The minimum
element quality is 0.0282, due to the sharp corners produced by the drifts intersections.
Mesh is showed in Figure 51a. Simplyfied meshes have been used for the problems
without drifts (Figures 51b and 51c). There are 37945 degrees of freedom for the THM
problem.
84
a.
b. c.
Figure 51: Meshgrids used in the THM model.
The time dependent solver used is a direct solver called UMFPACK. More details on
the Comsol Multiphysicsparameters of the model can be found in the APPENDIX
XII.
Results are presented in different cross sections, lines and points. In addition to common
crosssecti eatures related to the FEBEX drift orientation
have been used. Figure 52 displays the relative location of those features within the
domain
 Point R
ons parallel to the axis, other f
:
: middle point of the second heater, of coordinates:
( ) ( ) 0 ), 80 sin( 155 . 30 ), 80 cos( 155 . 30 , , =
P P P
z y x
 Cross line LL: vertical cross line passing through the point R, from z=35
to z=35.
 Cross section AA: vertical cross section along the axis of the FEBEX drift.
 Cross section BB: vertical cross section transversal to the axis of the
FEBEX drift, and passing through the point R.
85
A
A B
B
L
L
R
Figure 52: Crosssectional features to show output results of the models.
6.2 Hydrolithostatic equilibrium of the rock mass
Only a vertical hydraulic gradient and the lithostatic pressure due to 400 m of
rock over the FEBEX drift are applied in this problem. The boundary conditions are
given in Table 11, and initial conditions and constrains are given in Table 12.
Table 9: Boun m simulation.
B.C. A1 A2 B1 B2 C1 C2
dary conditions of the hydrolithostatic equilibriu
Thermal      
Hydraulic No flux No flux No flux No flux No flux P=365
w
g
u=0m
u=0m
v=0m
v=0m
u=0m
v=0m
33
=365
eq
g Mechanic
w=0m
Table 10: Initial conditions and constrains of the hydrolithostatic equilibrium
simulation.
Constrains
I.C. Excavated zones
(FBXinst, FBXaux,
Test z FBXheat)
one (
LABtun, MAINtun)
Thermal   
Hydra ic ul P=0 Pa  
Mech ic an u=0 m, v=0 m, w=0 m  
86
HMODEL RESULTS
We have performed a hydraulic timedependent analysis with the different
conditions for the rock mass, to analyse the differences in the response of the rock.
Although the steady state is the same for all the conditions (hydrostatic equilibrium), an
intermediate time lets appreciate the differences between them: Figure 53 shows the
cross sections AA and BB for the homogeneous anisotropic conditions (Figure 53a)
and the heterogeneous anisotropic conditions (Figure 53b) for the intrinsic permeability
of the rock at time t=9.5e6. The homogeneous anisotropic conditions yield to similar
results than the isotropic ones, not showed here, with almost horizontal isosurfaces for
the fluid pressure. However, the heterogeneous conditions lead to a high permeability
flow path in the FEBEX test zone, where the big fractures traverse the domain.
a.
b.
Figure 53: Fluid pressure field in the AA (left side) and the BB (right side) cross
e t =9.5e5s 11 days: a. Isotropic/hom
io er a c .
MMODEL RESULTS
The mechanical model, in the absence of thermal or hydraulic effects and
with to a
horizontal lithostatic profile for isotropic conditions. Figure 54a presents the
tridimensional field for the s
33
stress in the left hand side osssection in the
center an the BB cro ction in the right hand side. Nonorthotropic homogeneous
conditions add torsional and shear effec
Yaxis extremes in the lower bounda igu
orthotropic heterogeneous conditions (F much irregular pattern of
vertical stresses along the domain. In the heterogeneous case, we can appreciate high
around t g test zone.
sections for the tim ogeneous anisotropic
onditions condit ns; b. Het ogeneous nisotropic
out excavations or other geometrical features perturbing the stress field, yields
, the AA cr
d ssse
ts yielding to the concentration of stresses at the
ry of the domain (F
igure 54c) yield to a
re 54b). Finally, non
stresses state he heatin
87
88
a.
b.
c.
Figure stress field in chanical model with three different rock mass stif ic
orthotropic homogeneous conditions; c. Nonorthotropic heterogeneo
fness conditions: a. Isotrop
us conditions.
the me 54: Steady state of the s
33
conditions; b. Non
89
HMMODEL RESULTS
The full HM model yields to the steady state showed in Figures 55 for fluid
pressure and 56 for vertical stresses s
33
. The irregularly deformed domain is also
showed in those figures. Vertical stresses (negative stress means pointing downwards)
are slightly smaller than the ones reached in the mechanic model only, due to the Biot
coupling effect. Thus, the maximum negative stress is 1.596e7 Pa, whereas the
heterogeneous mechanical model yielded to a maximum of 1.676e7 Pa. This can be
appreciated in Figure 57b, where the time evolution of s
33
in the point R is presented.
After the decrease of vertical stress due to the rock mass weight load, s
33
starts
increasing due to the fluid pressure increase (Figure 57a) through the Biot coefficient.
Figures 57c and 57d display the vertical cross line LL of fluid pressure and vertical
stress s
33
respectively for different time instants. Fluid pressure initial condition is
Pa, and increases up to the hydrostatic profile. Vertical stresses initial condition is the
linear lithostatic profile of the domain, and weight loading together with
heterogeneous stiffness coefficient yields to the heterogeneous profile at equilibrium
Figure 55: Steady state fluid pressure field after hydrolithostatic equilibrium of the
rock mass.
P =0
the
.
Figure 56: Steady state vertical stress s
33
field after hydrolithostatic equilibrium of
the rock mass.
a. b.
c. d.
Figure 57: Time evolution of the fluid pressure (a. and c.) and vertical stress (b. and
d.) at the point R and through the crossline LL respectively.
90
6.3 Drifts excavation simulation
Once the fractured rock mass is stabilized in the hidrolithostatic steady state, we
simulate the drifts excavation process. However, hydraulic conditions have been
modified to approximate the real conditions existing at the Grimsel Test Site. On the
other hand, although the FEBEX drift was excavated much later than the Grimsel access
tunnels, we have simulated all the excavations simultaneously for simplicity. The
excavation process has been simulated by gradually modifying the boundary conditions
on the drift walls (called constrains in the tables) from the rock steady state conditions
to the tunnel conditions (P =0 and s
33
=0). Boundary conditions are defined according
to the nomenclature for the boundaries given in Figure 48, and with the help of the
above logical expressions for the geometrical features in the domain. Tables 13 and 14
show the boundary conditions and the initial conditions and constrains respectively:
Table 11: Boundary conditions of the drifts excavation simulation.
B.C. A1 A2 B1 B2 C1 C2
Thermal      
Hydraulic
No
flux
No
flux
P=2.1
MPa
P=0.7
MPa
( )
100
200
7 . 0
y
P
=
( )
100
200
7 . 0
y
P
=
M
u=0m
w=0m
echanic u=0m u=0m v=0m v=0m v=0m
33
=365
eq
g
Table 12: Initial conditions and constrains of the drifts excavation simulation.
Constrains
I.C. Excavated zones
(FBXinst, FBXaux,
LABtun, MAINtun)
Test zone
(FBXheat)
Thermal   
Hydraulic
( )
100
200
7 . 0
y
P
= MPa
P=0 Pa P=0 Pa
Mechanic u=0 m, v=0 m, w=0 m n
i
ii
=0 n
i
ii
=0
HMODEL RESULTS
The initial conditions for the hydraulic head (Figure 58a) are given by the
regional regime (J chlistock mountain and Aare river), which has very high gradients
both vertical and horizontal, corresponding to the mountainous characteristics of the
area and to the low permeability of the medium. The different sizes of the initial
conditions flux arrows depend on the heteroteneous distribution of the intrinsic
permeability. The heterogeneous anysotropic hydraulic model yields to the hydraulic
head steady state of Figure 58b after the excavations for the crosssection X=4.6m.
Typical hydrostatic pressure profile is strongly modified due to the imposed conditions
in the excavated zones (P=0). As a consequence, flux is directed towards the drifts.
91
a.
b.
Figure 58: Crosssection AA of the te steady
closer detailed view of the te zone (b.).
the fluid pressure is presented in Figures 59ad, in which th final
and two intermediate times are showed. Pressure below zero may be due to numerical
Figure 60 plots the isolines of the hydraulic head at steady state: Figure 60a plots the
n z=0, and Figure 60b plots the vertical crosssection AA
th respect to previous figures). Hydraulic head values obtained
in hy
mperature field in the state (a.) and a
st
Evolution of e inicial,
errors.
horizontal crosssectio
(oppositely oriented wi
draulic tests made in boreholes FEBEX95001 and FEBEX95002 before the
excavation of the drift are also showed in red. The observation points closest to the
FEBEX drift are the ones which present the highest discordance, due to the effects of
the excavation on the hydraulic head field around the excavated zone. Results obtained
by the UPC simulations (figures 3.12 and 3.13 of [34]) are similar, with slight
differences in the local irregularities around the FEBEX drift. This is due to the
homogenization performed in this model, alternatively to the inclusion of discrete
fractures in the UPC models.
92
93
Fig fluid rifts excavation simulation. Four time instants are showed: a. Time t = 0 years; b.
Time day Time t = 45 days; d. Time t = 3.17 years.
pressu
t = 22
re in the d
s; c.
ure 59: Time evolution of the
94
a.
b.
Figure 60: Hydraulic head isolines at steady state: a. Horizontal cross section at z=0;
b.: Vertical cross section AA.
MMODEL RESULTS
The isotropic case yields to the steady state for vertical stresses
section AA plotted in Figure 61. Accumulation of vertical compressi
at the top and bottom of the excavated drifts. Vertical displacement occurs uniformly in
the domain.
s
33
along cross
ve stresses occurs
Figure 61: Vertical stress steady state for the isotropic mechanic model.
HMMODEL RESULTS
The excavation of the drifts yields in the coupled HM model to the steady state
showed in Figures 62 and 63. Both homogeneous and heterogeneous anisotropic/non
orthotropic conditions are compared. Figure 62a presents the vertical stress field along
with the vertical displacement isosurfaces for the homogeneous case. Figure 62b
presents the same results for the heterogeneous case. We can appreciate that
consolidation in the FEBEX test zone (last 17 m of the FEBEX drift) is higher, due to
the higher fracture density in this zone. Figure 63 displays the fluid pressure for the
heterogeneous anisotropic case. Water flow towards the drifts can be also seen.
a.
b.
Figure 62: Vertical stress s
33
and vertical displacement isosurfaces steady state for
e HM drifts excavation simulation: a. Homogeneous anisotropic/nonorthotropic th
conditions; b. Heterogeneous anisotropic/nonorthotropic conditions.
95
Figure 63: Fluid pressure and water flow lines steady state for the HM drifts
excavation simulation with heterogeneous anisotropic/nonorthotropic conditions.
6.4 Heating experiment simulation
The simulation of the heating experiment has been performed for two different
cases: with excavated drifts and without excavated drifts (drifts filled up with concrete
or some other material of similar conditions than the rock mass). The real conditions for
the FEBEX experiment are with excavated drifts. In both cases, however, the test zone
has been filled up with bentonite. We only show results of the simulation w h
excavations.
The boundary conditions of this new problem are defined for each case similar to those
imposed in the previous problems. Initial conditions correspond to the corresponding
steady states reached in the excavation simulation for each of the cases. Table 16 gives
the initial conditions and the constrains of the problem.
Table 13: Boundary conditions of the heating experiment simulation.
B.C. A1 A2 B1 B2 C1 C2
it
Thermal T=13 C T=13 C T=13 C T=13 C T=13 C T=13 C
Hydraulic No flux No flux No flux No flux No flux P=365
w
g
Mechanic u=0m u=0m v=0m v=0m u=0m
v=0m
w=0m
33
=365
eq
g
96
Table 14: Initial conditions and constrains of the heating experiment simulation.
Constrains I.C.
Excavated zones
(FBXinst, FBXaux,
LABtun, MAINtun)
Test zone (FBXheat)
Thermal T=13 C 
2
2
14 . 1
65
100
r
T
= C
Hydraulic Hmodel steady state  
Mechanic Mmodel steady state  
TMODEL RESULTS
In the heating problem the three types of analyses (isotropic, homogeneous
anisotropic/nonorthotropic and heterogeneous anisotropic/nonorthotropic) give the
same result, as thermal coefficients are isotropic for all cases.
The t 6 a,
and m
mperature surpasses slightly the profile imposed (up to 100C), surely due to the
umerical method used and mesh interpolation errors.
hermal steady state result through the cross section AA is given in Figure
a detail of the nearfield of the heaters is given in Figure 64b. The maximu
4
te
n
a.
b.
Figure 64: Crosssection AA of the temperature field in the final state (a.) and a
closer detailed view of the test zone (b.).
97
The temperature evolution in the middle point of the second heater, R, is showed
in Figure 65a. Profile LL for different times is also showed in Figure 65b. Density
varies with temperature in all the ther eating conditions
assumed in our problems the decrease zon
reference value. The Rpoint evolution of d
Figures 65c and 65d.
mal models, an
of density in the hot
and LL profile
d for the h
es is about 3% of the
ensity are showed in
a. b.
c. d.
Figure 65: Time evolution in the point R (lefthand side) and in the vertical crossline
LL (righthand side): (a.) and (b.) temperature; (c.) and (d.) water density.
TMMODEL RESULTS
Thermal stresses are produced in the heating zone of the drift. Figures 66a and
66b show the LL profile of stresses at different times for the cases of isotropic and
homogeneous nonorthotropic stiffness coefficients respectively. Both isotropic and
nonorthotropic approximations produce increasing vertical stresses near the heaters.
The stress jump due to thermal stresses is higher for the nonorthotropic conditions
(10 MPa) than for the isotropic ones (6 MPa). In all cases, thermal stresses are much
smaller than the stresses produced by the excavation of the drifts.
98
b.
a.
Figure 66: Vertical stresses s
33
in the crossline LL for different rock conditions: a.
isotropic stiffness tensor; b. homogeneous nonorthotropic stiffness tensor.
THMMODEL RESULTS
Test zone of the FEBEX drift has been filled up with bentonite and the 3year heating
has been simulated. The steady state of the fluid pressure is showed in Figure 67, along
with the horizontal components of the water flow at z =0. Figure 68 presents the Von
Misses stress, the hydraulic head isolevels and the deformed shape of the domain. Note
the thermal stresses around the test zone and the influence of the boundary conditions
for displacements on the cummulation of stress in the lower boundary. The
displacements at steady state are also shown for the whole domain in Figure 69:
horizontal displacements u (Figure 69a) and v (Figure 69a) and the vertical
displacement w (Figure 69c).
A detailed view of the FEBEX drift wit in e AA crosssection is plotted for the
different variables of the coupled processes: temperature T (Figure 70a), fluid pressure
P
7
shear
rains
23
,
13
,
12
(Figures 72d, 72e and 72f respectively) and displacements u, v, w
c respectively). In the Von Mises stress plot (Figure 70c),
uled not only around the heaters, but also in the bentonite
ck c
boreholes drilled out
from the test zone of the FEBEX drift.
h th
(Figure 70b), Von Misses stress (Figure 70c), normal stresses s
11
, s
22
, s
33
(Figures 71a,
1b and 71c respectively), shear stresses s , s , s (Figures 71d, 71e and 71f
23 13 12
respec
st
tively), normal strains
11
,
22
,
33
(Figures 72a, 72b and 72c respectively),
(Figures 73a, 73b and 73
hermal stresses are cumm t
ro ontact, the last being due mainly to shear forces. In the crosssections of the
displacements (Figures 73a, 73b and 73c) we can appreciate how the bentonite expands
both longitudinaly (displacement v) and radially (displacement w for this section) due to
the heating process.
As our interest here is in the simulation of the fractured rock, we have not explicitely
compare our model with measurements taken in the bentonite. However, we do present
some comparisons of the model with data coming from the radial
99
Figure 67: Final state of the fluid pressure. Flow at z=0 is also showed (only horizontal
components).
Figure 68: Final state of the Von Mises stresses, hydraulic head isolevels and deformed
shape of the domain.
100
a.
b.
c.
of the displacements u Figure 69: Final state (a.), v (b.) and w (c.).
101
a.
b.
c.
Figure 70: Detailed view of the THM final state of temperature (a.), fluid pressure (b.)
and Von Mises stress (c.).
102
a.
b.
c.
Figure 71: Detailed view of the THM final state of normal stresses s (a.), s (b.) and
ear stresses s (d.), s (e.) and
11 22
s
33
(c.) and sh
23 13
s
12
(f.).
103
d.
e.
f.
Figure 71 (cont.): Detailed view of the THM final state of normal stresses s
11
(a.), s
22
and s (b.) f.).
33
(c.) and shear stresses s
23
(d.), s
13
(e.) and s
12
(
104
a.
b.
c.
Figure 72: Detailed view of the THM final state of normal strains
11
(a.),
22
(b.) and
33
(c.) and shear strains
23
(d.),
13
(e.) and
12
(f.).
105
d.
e.
f.
Figure 72 (cont.): Detailed view of the THM final state of normal strains
11
(a.),
22
nd (b.) a f.).
33
(c.) and shear strains
23
(d.),
13
(e.) and
12
(
106
a.
b.
c.
Figure 73: Detailed view of the THM final state of displacements u (a.), v (b.)
and w (c.).
107
We have compared our results with field measurements of temperature, fluid pressure,
total pressure and total displacements. Points of selected boreholes where data have
been taken for comparison are showed in Figure 74:
Temperature (red points): the selected boreholes are: a borehole drilled in a hot
section (closed to the second heater), SF23; a borehole drilled in a cold section
(far from heaters), SF14; and a borehole at the end of the FEBEX drift, SB22.
Fluid (or interstitial) pressure (blue points): the selected boreholes are the SK1,
inclined and crossing longitudinally the test zone, and the SJ 5, at the end of the
test zone, and the sampling points are showed in blue in Figure 74.
Total pressure (green points): the selected boreholes are the two ones where this
variable has been measured, SG1 and SG2.
Total displacement (orange points): the selected boreholes are the two ones
where this variable has been measured, SI1 and SI2.
Figure 74: Selected boreholes and sampling points for the
temperature (red), intersticial pressure (blue) total pressure (green)
and total displacements (orange) comparisons in the THM analysis
(original figure from [33]).
Figure 75 shows the comparison of the measured and the simulated temperature in
borehole SF23. The point closest to the bentonite (the hotest one) presents the higher
discrepancy, whereas there is a good agreement for the other three points. Indeed, when
considering the cold boreholes SF14 (Figure 76) and SB22 (Figure 77), a good
agreement between measured and simulated temperatures is found in all the sampling
nts. This is due to the imposed conditions to simulate the heating transient: instead of
imposing a source term with a fixed heat flux, we have fixed a temperature profile
with
temperature field in the nearest rock around the heaters.
poi
in the bentonite based on measured data. This probably underestimates the
108
a.
b.
c.
Figure 75: Time evolution of measured (x) and simulated () temperatures
in boreholes SF23 (a.), SF14 (b.) and SB22 (c.) for the THM analysis.
109
Figure 76 displays the time evolution of the measured and the simulated intersticial (or
fluid) pressure for the borehole SK1 (Figure 76a) and the borehole SJ 5 (Figure 76b). As
full saturation is assumed in the model, and may not be so in the nearfield, some
overestimation was expected to occur. However, both the large overestimation and the
flat shape observed in the simulated fluid pressure are due to the initial conditions
imposed in the simulation: full hydromechanical stabilisation of the rock mass and the
bentonite filling the test zone were assumed. Doing so, the fluid pressure starts and ends
at the same steady state value. No stabilisation should have been imposed, to lead the
fluid pressure at the bentoniterock mass contact start at zero (i.e. at the atmospheric
pressure). Nevertheless, the transient peak observed in the meassured pressure around
t=0.810
7
s is also reflected in the simulation. On the other hand, the fluid pressure
measured at the borehole SJ 5 is higher than the one measured at the borehole SK1, what
confirms the coherence of the hydraulic gradient conditions imposed in the simulation.
a.
b.
Figure 76: Time evolution of measured (x) and simulated () fluid pressure
in boreholes SK1 (a.) and SJ 5 (b.).
110
Total pressure has been measured in two boreholes, SG1 (Figure 77) and SG2 (Figure
78), vertical and horizontal respectively, located at the middle section of the FEBEX
drift test zone. There, sensors labelled from 01 to 05 are the furthest ones from the
gallery (Figures 77a and 78a), and sensors from 06 to 10 are the closest ones (Figures
77b and 78b). We have compared the measured total pressure with the Von Mises
stresses obtained in the model. Von Mises stress overestimates slightly the total
pressure, although it gets the same order of magnitude. The thermal stresses resulting
from the thermomechanical coupling occur fastly in the model than in the reality (see
Figure 77b). This is surely due to a too high thermal conductivity in the rock mass, the
same that caused the low temperatures in the sampling points closest to the heaters (see
Figure 75a). On the other hand, the negative trend of the Von Mises stress at the long
term, together with the high initial value, are due to the hydrolithostatic steady state
used as initial condition, as it was noticed for the fluid pressure. A new time analysis
ith the appropriate initial and boundary conditions would be convenient to clarify the
behavior of the model in this aspect.
a.
w
b.
igure 77: Time evolution of measured (x) and simulated () total pressure in
: a. Points from 01 to 05; b. Points fr
F
borehole SG1 om 06 to 10.
111
a.
b.
Figure 78: Time evolution of measured (x) and simulated () total pressure in
borehole SG2: a. Points from 01 to 05; b. Points from 06 to 10.
Finally, total displacement measurements have been compared with the modulus of the
displacement vector (u, v, w) in the direction of each selected borehole. The order of
magnitude of the displacements is well simulated, but not the transient behaviour.
Further studies are necessary to better interpret the total displacement measurements
(they get negative values, for instance), and to be able to compare them correctly with
the outputs of the simulation.
112
a.
b.
Figure 79: Time evolution of measured (x) and simulated () total displacements in
boreholes SI1 (a.) and SI2 (b.).
113
114
7 CONCLUSIONS AND FUTURE WORK
In the first part of the thesis, a comprehensive statistical methodology to analyse
the time series coming from an experiment of coupled thermohydromechanical
processes has been applied. Several techniques covering the time, space, frequency and
scale frames of the measured data have been used, to obtain a qualitative and
quantitative characterization of the physical processes occurring, unexpected events and
general performance of the sensors installed within the experiment. Some of the most
relevant results obtained in our study have been presented here, and are already
published in [22] (full article in APPENDIX XIII).
Firstly, concerning the physical processes occurring in the Mockup test, an insight to
the unpredicted slow down of the water intake and hydration rate of the bentonite has
been offered: a decrease in the relative humidity gradient has been observed, and the
hypothesis of the development of an evaporationcondensation cell has been analysed to
explain the unpredicted behaviour.
Secondly, a deep study of the overheating incident in the Mockup has been done. No
important consequences or irreversible perturbations of the processes or the sensors
have been evidenced from the results obtained. Moreover, a quick recovery of the
temperature normal measurements has been observed, being that of the humidity and
pressure sensors a bit longer.
Finally, a systematic study of the performance of all the sensors has been made, and
results concerning an abnormal behaviour of some total pressure sensors has been
presented. There has been seen that those sensors work correctly, but they measure the
fluid pressure instead of the total pressure due to a lack of connectivity with the solid
phase of the bentonite.
The statistical methodology used in this work has demonstrated to be very useful to
understand and characterize coupled processes and sensors reliability within this kind of
experiments. Statistical techniques should be considered as an essential analysis tool,
complementary to any modeling process, not only in the nuclear waste engineering, but
also in any complex multidisciplinary engineering experiment. However, further work
is necessary to better link the results of the statistical analyses with the modeling tasks.
In the second part of the thesis, we have developed a methodology to simulate a 3D
fractured network that fits optimally a cylindrical tunnel tracemap. The optimization
procedure searches for the best parameters of the size distribution to minimize the
discrepancies between measured and simulated trace length and 3D trace chord
histograms and number of intersections with the tunnel and two exploratory boreholes.
A good agreement between both fractured mediums has been found in the results. This
methodology provides a good starting point for the use of cylindrical tracemaps to
simulate geological 3D fractured networks, and completes the classical use of flat wall
tracemaps that is more extensively developed in the literature.
115
Although satisfactor omments for future
ork:

ature, because it depends on the
generation of a fractured medium obtained with statistical distributions. This
average has to be determined according to the
confidence interval needed in the objective function.
the third part of the thesis, a thermohydromechanical model has been developed and
ropic and heterogeneous homogenized coefficients have been
sed for hydraulic, mechanic and thermal submodels. Variation of water properties with
 As stated before, the stochastic nature of the fractured medium generation
this would require an enormous computational effort, such that it would only be
as, for instance: non
saturated or partially saturated conditions, better coupling in the bentoniterock
contact, elastoplastic model, new cracks development, dependency of the
fracture aperture on stress, etc.
y results have been obtained, here are some c
w
The nonhomogeneous fractured network simulated near the gallery could be
assumed to be the same in the overall domain, so the statistical functions
producing the fractured medium would be extended everywhere. To do that,
nonhomogeneous Poisson processes should be used [84], and some kind of
density measure must be defined for every point in the 3D space. The reduced
second moment function [42] of the tracemap could be used for this purpose.
 The objective function has a stochastic n
means that, for the same set of parameters R
min
, R
max
, and b, we can obtain
different values of the objective function for each realization. Therefore, an
average of the objective functions of several realizations should be used to
obtain a more reliable value given a set of parameters. The number of
realizations to use in that
 For a hydrological validation of the fractured medium, hydraulic and transport
tests available in the site should be used, with hydraulically conditioned
fractured networks.
In
implemented in Comsol Multiphysics, and an upscaling methodology has been
defined to feed the model with homogenized coefficients. Several oneway and twoway
couplings have been incorporated to the model: Biot hydromechanical coupling,
thermal stresses, thermal expansion of water, and heat convection and conduction. Fully
anisotropic / nonorthot
u
temperature has also been considered.
The FEBEX experiment has been simulated with this model, with a good agreement
between measured and simulated values of the main variables. Some remarks can be
also made in this part for future work:
process makes necessary to compute it as an average of several realizations.
Indeed, the objective function should be chosen not only based in morphological
aspects of the fractured network, as it is made in this work, but also such that
minimizes the discrepancies between the THM timedependent model outputs
and the field measurements as compared in the last part of chapter 6. However,
possible with the use of computer clusters or powerfull workstations.
 Future enhancements of the model should incorporate new couplings between
the processes and generalize the material conditions
116
 The upscaling performed in the hydraulic coefficients could be generalized for
the mechanical and the thermal coefficients as well. By following a similar
Fina
model
in the
coeffic
reasoning, we could fix the global stress field (constant) and compute the strain
by superposition of the individual contributions of each fracture to the total
strain, given a loworder boundary condition for stresses. Similarly, in the
thermal model we could upscale the coefficients by weighting with the fracture
volumetric fraction of each homogenization subdomain, as the presence of
fractures influences substantially the heat transport by conduction.
lly, in this work we have developed an integrated methodology to analyse and
coupled processes in tridimensional fractured media, with special contributions
simulation of the fractured network and in the upscaling of the hydraulic
ients.
117
118
8 REFERENCES
[1] Ababou R. Approaches to Large Scale Unsaturated Flow in Heterogeneous,
Stratified, and Fractured Geologic Media. Report NUREG/CR5743, U.S.
Nuclear Regulatory Commission, Washington D.C., USA. Government Printing
Office, pp.1150, 1991.
[2] Ababou R. & L. W. Gelhar. SelfSimilar Randomness and Spectral
Conditioning: Analysis of Scale Effects in Subsurface Hydrology, Chapter XIV
in Dynamics of Fluids in Hierarchical Porous Media, J . Cushman editor,
Academic Press, New York, pp. 393428. 1990.
[3] Ababou, R, Caamn, I., Elorza, F.J . ThermoHydroMechanical Simulation of a
3D Fractured Porous Rock: Preliminary Study of Coupled MatrixFracture
Hydraulics. Proceedings of the COMSOL Multiphysics Conference 2005, pp.
193198. Ed. J . M: Petit and J . Daluz. France. November 2005.
[4] Ababou R., A. Millard, E. Treille, M. Durin, & F. Plas. Continuum Modeling of
Coupled ThermoHydroMechanical Processes in Fractured Rock.
Computational Methods in Water Resources (CMWR'94 Heidelberg), Kluwer
Acad. Publishers, A. Peters et al. (eds.), Vol.1, Chap.6, pp.651658. 1994.
[5] Ababou R., A. Millard, E. Treille, & M. Durin. Coupled ThermoHydro
Mechanical Modeling for the Near Field Benchmark Test 3 (BMT3) Of
Decovalex Phase 2. CEA/DMT Progress Report 93/488. May 1994.
[6] Ababou R., A.C. Bagtzoglou, & E.F. Wood. On the Condition Number of
Covariance Matrices Arising in Kriging, Estimation, and Simulation of Random
Fields. Mathematical Geology, Vol. 26, No. 1, pp. 99133. 1994.
[7] Alonso E., J . Alcoverro, F. Coste, L. Malinsky, V. MerrienSoukatchoff, Kadiri,
T. Nowak, H. Shao, T.S. Nguyen, A.P.S. Selvadurai, G. Armand, S.R. Sobolik,
M. Itamura, C.M. Stone, S.W. Webb, A. Rejeb, M. Tijani, Z. Maouche, A.
Kobayashi, H. Kurikami, A. Ito, Y. Sugita, M Chijimatsu, L. Brgesson, J .
Hernelind, J . Rutqvist, C.F. Tsang & P. J ussila. The FEBEX Benchmark Test:
Case Definition and Comparison of Modelling Approaches. International
Journal of Rock Mechanics & Mining Science. Ed. Elsevier. May 2005.
[8] Amiguet, J .L. Grimsel Test Site. Felskennwerte Von Intaktem Granit.
ZusammenStellung Felsmechanischer Laborresultate Diverser Granitischer
Gesteine. NAGRA, NIB 8508, September 1985.
[9] Bacry, E, Mallat, S. & Papanicolaou, G. A Wavelet Based SpaceTime Adaptive
Numerical Method for Partial Differential Equations. Mathematical Modelling &
Numerical Analysis 26: pp. 793834. 1992.
[10] Baecher G.B., Einstein H.H. & Lanney N.A. Statistical Description of Rock
Properties and Sampling. Energy Resources and Excavation Technology;
proceedings, 18th U.S. Symposium On Rock Mechanics. Golden: Colo. Sch.
Mines Press. Wang FD, Clark GB, editors. pp. 5C1.18. 1977.
[11] Barenblatt G., Zelthov I. & Kochina I. Basic Concepts in the Theory of Seepage
of Homogeneous Liquids in Fissured Rocks, Journal of Applied Mathematics 24,
213240. 1960.
119
[12] Beylkin, G & Coult, N. A rategy for Reduction of Elliptic
PDEs and Eigenvalue Problems. Program in Applied Mathematics. University
[13]
[15]
ood Cliffs, New J ersey, 1994.
[17]
dia 11, 263280.
[18]
[22]
& Image
[24]
Multiresolution St
of Colorado at Boulder. 1998.
Biot, M.A. Theory of Deformation of a Porous Viscoelastic Anisotropic Solid.
Journal of Applied Physics 27: pp. 459467. 1956a.
[14] Biot, M.A. General Solutions of the Equations of Elasticity and Consolidation for
a Porous Material. Journal of Applied Mechanics 23: pp. 9196. 1956b.
Box, G.E.P., J enkins, G.M. & Reinsel, G.C. Time Series Analysis: Forecasting
and Control. Third Edition, Prentice Hall, Englew
[16] Bracewell, R.N. The Fourier Transform and its Applications. McgrawHill, Inc.
USA. 1986.
Braester C. & Zeitoun D. Pressure Transient Response of Stochastically
Heterogeneous Fractured Reservoirs, Transport in Porous Me
1993.
Bruer V., Kilger B. & Pahl A. Grimsel Test Site. Engineering Geological
Investigations for the Interpretation of Rock Stress Measurements and Fracture
Flow Tests. NAGRA, NTB 8837E, April 1989.
[19] Brewster, M.E. & Beylkin, G. A Multiresolution Strategy for Numerical
Homogenization. Applied & Computational Harmonic Analysis 2: pp. 327349.
1995.
[20] Brockwell, P.J . & Davis, R.A. Introduction to Time Series and Forecasting.
SpringerVerlag New York, Inc. 1996.
[21] Caamn, I. & Paredes, C. The Applications of Wavelet Transform to Edge
Detection and Fractal Analysis on Digital Terrain Elevation Models. Proceedings
of the International Workshop on Fractal Mathematics Applied to Soil and
Related Heterogeneous Systems PedoFract04, pp. 1718. Spain. 2004.
Caamn, I., Elorza, F.J ., Mangin, A., Martn, P.L. & Rodrguez, R. Wavelets
and Statistical Techniques for Data Analysis in a MockUp HighLevel Waste
Storage Experiment. International Journal on Wavelets, Multiresolution
Processing 2(4): pp. 351370. December 2004.
[23] Caamn, I., Elorza, F. J ., Ababou, R. 3D Fracture Networks: Optimal
Identification and Reconstruction. Accepted for publication in the proceedings of
the XIth IAMG Annual Congress. Belgium. September 2006.
Chan T., R. Christiansson, G.S. Boulton, L.O. Ericsson, J . Hartikainen, M.R.
J ensen, D. Mas Ivars, F.W. Stanchell, P. Vistrand & T. Wallroth. DECOVALEX
III BMT3/BENCHPAR WP4: the ThermoHydromechanical Responses to a
Glacial Cycle and Their Potential Implications for Deep Geological Disposal of
Nuclear Fuel Waste in a Fractured Crystalline Rock Mass. International Journal
of Rock Mechanics & Mining Science. Ed. Elsevier. May 2005.
[25] Charton, P. & Perrier, V. A PseudoWavelet Scheme for the TwoDimensional
NavierStokes Equations. Matemtica Aplicada e Computacional 15: pp. 139
160 1996.
120
[26] CIEMAT. FEBEX (FullScale Engineered Barriers Experiment in Crystalline
Host Rock): 21 Data Report. MockUp Test. Technical Report 70ImaL9101.
Spain. 2002.
[27] Daubechies, I. Ten Lectures on Wavelets. CBMSNSF Series in Applied
Mathematics. SIAM. 1992.
[28] Dershowitz W.S & Einstein H.H. Characterizing Rock J oint Geometry with J oint
System Models. Rock Mechanics & Rock Engineering 21(1):2151. 1988.
.
[30] tal Analysis of J oint Patterns in Granite. International Journal of
[31]
tividad (AGP). Diseos Conceptuales Genricos. Technical Report Num.
[32]
. Spain. 1996.
Num. 03/98. Spain. 1998.
rme de Sntesis. Technical Report Num.
[35]
[37]
tures. Tectonophysics (226): pp.
[38]
pp. 65100.
[40]
al Characterization and Modeling. UPC. 70UPCM01001. 1998.
[29] Dverstorp, B. & Andersson, J . Application of the Discrete Fracture Network
Concept with Field Data: Possibilities of Model Calibration and Validation
Water Resources Research 25(3): pp. 540550. March 1989.
Ehlen J . Frac
Rock Mechanics & Mining Science 37(6): pp. 909922. 2000.
ENRESA. Almacenamiento Geolgico Profundo de Residuos Radiactivos de
Alta Ac
11/95. Spain. 1995.
ENRESA. FEBEX (FullScale Engineered Barriers Experiment in Crystalline
Host Rock): Test Plan, MockUp Test. Version 1
[33] ENRESA. FEBEX (FullScale Engineered Barriers Experiment in Crystalline
Host Rock). Diseo Final y Montaje del Ensayo Insitu en Grimsel. Technical
Report
[34] ENRESA. FEBEX (FullScale Engineered Barriers Experiment in Crystalline
Host Rock): Etapa Preoperacional. Info
09/97. Spain. 1997.
Fadili A. & R. Ababou. Dual Homogenisation of Immiscible Steady TwoPhase
Flow in Random Porous Media. Water Resources Research, Vol.40, W01513,
DOI:10.1029 / 2003WR002465, 2004.
[36] Gasquet C. & Witomski, P. Analyse de Fourier et Application. Masson. Paris.
1995.
Gillespie, P.A., Howard, C.B., Walsh, J .J ., & Watterson, J . Measurement and
Characterization of Spatial Distributions of Frac
112141. 1993.
Goffe, Ferrier and Rogers, 1994. Global Optimization of Statistical Functions
with Simulated Annealing. Journal of Econometrics (60, 1/2):
[39] Goodman R.E. Methods of Geological Engineering in Discontinuous Rocks. St.
Paul: West Publishing Co. 1976.
Guimer J ., Ortuo F., Vzquez E., Ruiz B., Martnez L., Carrera J . & Meier P.
Pulse Tests at in Drift Boreholes. Performance and Evaluation. UPC, 70
UPCL01001. 1996.
[41] Guimer J ., Carrera J ., Martnez L., Vzquez E., Ortuo F., Fierz T., Blher C.,
Vives L., Meier P., Medina A., Saaltink M., Ruiz B. & Pardillo J . FEBEX
Hydrologic
121
[42] Hanisch, K.H. & Stoyan, D. Remarks on Statistical Inference and Prediction for a
HardCore Clustering Model. Math. Operationsf. Statist., ser. statist. (14): pp.
559567. 1983.
Fracture Networks Based on Regular Lattice
per J ., Alonso E.
n. Report no EUR 19147 EN. 2000.
[48] . Inference of Discontinuity Trace Length
[49] in J ., Schuler P. & Buletti M. Grimsel Test Site. Geology.
[50]
d Mass Exchange Coefficient. ASME
[51] M., R. Ababou, B. Noetinger & M. Quintard. Coefficient Dchange
e Mcanique 332 (2004), pp. 679686, Elsevier. 2004.
nal Of
[53] nd Simulation of J oints for Rock Engineering.
[55] of
Interconnection to Permeability in Fractured Networks. Journal of Geophysical
Research 90 (B4): pp. 30873098. 1985.
[43] Hestir, K. & Long, J .C.S. Analytical Expressions for the Permeability of Random
TwoDimensional Poisson
Percolation and Equivalent Media Theories. Journal of Geophysical Research 95
(B13): pp. 2156521581.
[44] Hudson J .A., O. Stephansson & J . Anderssond. Guidance on Numerical
Modelling of ThermoHydroMechanical Coupledprocesses for Performance
Assessment of Radioactive Waste Repositories. International Journal of Rock
Mechanics & Mining Science. Ed. Elsevier. May 2005.
[45] Huertas F., FuentesCantillana J .L., J ullien F., Rivas P., Linares J ., Faria P.,
Ghoreychi M., J ockwer N., Kickmaier W., Martinet M.A., Sam
& Elorza F.J . Full Scale Engineered Barriers Experiment for a Highlevel
Radioactive Waste in Crystalline Host Rock (FEBEX Project). Final Report.
European Commissio
[46] J enkins, G.M. Practical Experiences with Modeling and Forecasting Time
Series. Gwilym J enkins & Partners Ltd. J ersey. 1979.
[47] J enkins, G.M. & Watts, D. G. Spectral Analysis and its Applications. Holden
Day. San Francisco. 1968.
J imnezRodrguez R., N. Sitar
Distributions Using Statistical Graphical Models. International Journal of Rock
Mechanics & Mining Science. Ed. Elsevier. February 2006.
Keusen H.R., Gangu
NAGRA, NTB 8714E. 1989.
Kfoury M., B. Noetinger, M. Quintard, & R. Ababou. Upscaling in Fractured
Heterogeneous Media: Permeability an
Journal of Applied Mechanics (JAM). 2005.
Kfoury
MatriceFracture en Milieu Poreux Fractur : Changement Dchelle Par
Approche Stochastique. C.R.A.S.  Comptes Rendus de l'Acadmie des Sciences
de Paris : C.R. d
[52] La Pointe, P.R. Derivation of Parent Fracture Population Statistics from Trace
Length Measurements of Fractal Fracture Populations. International Jour
Rock Mechanics & Mining Science (39): pp. 381388. 2002.
La Pointe P.R. Pattern Analysis a
Comprehensive rock engineering. Volume 3rock testing and site
characterization. Oxford: Pergamon Press. Hudson J A, editor. pp. 215239.
1993.
[54] Lewalle, J . Tutorial On Continuous Wavelet Analysis of Experimental Data.
Siracuse University. 1995.
Long, J .C.S. & Witherspoon, P.A. The Relationship of the Degree
122
[56] Long, J .C.S. & Billaux, D. From Field Data to Fracture Network Modeling: an
Example Incorporating Spatial Structure. Water Resources Research 23(7): pp.
12011216. J uly 1987.
[58]
s Theorem Apply? Mathematical Geology 35(1): pp.
[59]
[61]
sity. Technical Report.
[62] ystmes Hydrologiques a
[68]
aste Repositorya
[69] n T.S., A.P.S. Selvadura & G. Armand. Modelling the FEBEX THM
[57] Long J .C.S., Remer J ., Wilson C. & Witherspoon P., Porous Media Equivalents
for Networks of Discontinuous Fractures, Water Resources Research. 1982.
Lyman G.J . Stereological and Other Methods Applied to Rock J oint Size
EstimationDoes Crofton
923. 2003.
Mallat, S. A Theory for Multiresolution Signal Decomposition: the Wavelet
Representation. IEEE Transactions on Pattern Analysis & Machine Intelligence
11 (7): pp. 674693. 1989.
[60] Mallat, S. & Zhang, Z. Matching Pursuit in a TimeFrequency Dictionary. IEEE
Transactions on Signal Processing 41, pp. 33973415. 1993.
Mallat, S. & Hwang, W.L. Singularity Detection and Processing with Wavelets.
Courant Institute of Mathematical Science. N.Y. Univer
March 1991.
Mangin, A. Pour une Meilleure Connaissance des S
Partir des Analyses Corrlatoire et Spectrale. Journal Of Hydrology, Amsterdam,
N 67 pp. 2543. 1984.
[63] Martn, P.L. Desarrollo y Diseo de Experimentos Aplicados a la Validacin del
Concepto Espaol de Almacenamientos de Residuos Radiactivos de Alta
Actividad. Ph.D. thesis. Universidad Politcnica de Madrid. 2004.
[64] Mauldon M. Estimating Mean Fracture Trace Length and Density from
Observations In Convex Windows. Rock Mechanics & Rock Engineering
31(4):20116. 1998.
[65] Mauldon M. & Mauldon J .G. Fracture Sampling On a Cylinder: From Scanlines
to Boreholes and Tunnels. Rock Mechanics & Rock Engineering 30(3):12944.
1997.
[66] Metropolis, N., Rosenbluth, A., Rosenbluth, M., Teller A., and Teller, E.
Equation of State Calculations by Fast Computing Machines. Journal of
Chemical Physics. (21): pp. 10871092. 1953.
[67] Meyer, Y. Wavelets: Algorithms and Applications, SIAM. 1993.
Min K.B., J . Rutqvist, ChinFu Tsang & Lanru J ing. Thermally Induced
Mechanical and Permeability Changes around a Nuclear W
FarField Study Based On Equivalent Properties Determined by a Discrete
Approach. International Journal of Rock Mechanics & Mining Science. Ed.
Elsevier. May 2005.
Nguye
Experiment Using a State Surface Approach. International Journal of Rock
Mechanics & Mining Science. Ed. Elsevier. April 2005.
[70] Noorishad, J ., Tsang C.F., Stephansson O. & J ing L. Coupled thermohydro
elasticity phenomena in variably saturated fractured porous rocks : Formulation
and numerical solution. Developments in Geotechnical Engineering: Coupled
123
ThermoHydroMechanical Processes of Fractured Media (O. Stephansson, L.
J ing & CF. Tsang, eds.) 79: pp. 93134. Elsevier. 1996.
Oda. M. An Equivalen [71] t Continuum Model for Coupled Stress and Fluid Flow
[72] ock
[73]
X (GrimselSuiza). CIEMAT, 70IMAM 201. 1997.
[76]
ater Resources 20(56): pp. 253278. 1997.
[79] ck. Published
[80]
Olivella, S. (directors).
[81]
ch 5: pp. 12731289. 1969.
[83] d, A. Thoraval, & R. Ababou.
ses of Fractured
[84] W.S. & Mecke, J . Stochastic Geometry and its Applications.
[85]
Analysis in Fractureed RockMasses. Water Resources Research, 22/13/, 1986.
Pahl A., Heusermann S.T., Bruer V., Glggler W. Grimsel Test Site. R
Stress Investigations. NAGRA, NTB 8839E. 1989.
Pardillo J ., Campos R. & Guimer J . Caracterizacin Geolgica de la Zona de
Ensayo FEBE
[74] Polikar, R. The Wavelet Tutorial. Rowan University, College of Engineering
(Http://Engineering.Rowan.Edu/~Polikar/Wavelets/Wttutorial.Html). 2001.
[75] Quintard M. & Whitaker S., Transport in Chemically and Mechanically
Heterogeneous Porous Media: Theoretical Development of Region Averaged
Equations for Slightly Compressible SinglePhase Flow, Advances in Water
Resources 19, 2947. 1996.
Renard, P. & de Marsily, G. Calculating Equivalent Permeability: a Review.
Advances in W
[77] Resnikoff, H. L. & Wells, R. O. J r. Wavelet Analysis (The Scalable Structure of
Information). Springer Verlag New York, Inc. 1998.
[78] Rutqvist, J . & Stephansson, O. The Role of Hydromechanical Coupling in
Fractured Rock Engineering. Hydrogeology Journal 11: pp. 740. 2003.
Sahimi M. Flow and Transport in Porous Media and Fractured Ro
by VCH. 1995.
Sanchez, M. J . ThermoHydroMechanical Coupled Analysis in Low
Permeability Media. Ph.D. thesis. Gens, A. &
Departament D'enginyeria Del Terreny, Cartogrfica I Geofsica. UPC. 2004.
Snow, D.T. Anisotropic Permeability of Fractured Media. Water Resources
Resear
[82] Steck, A. Die Alpidischen Strukturen in den Zentralen AareGraniten des
Westlichen Aarmassivs. Eclogae geol. Helv. (61): pp. 1948. 1968.
Stietel A., A. Millard, E. Treille, E. Vuillo
Continuum Representation of Coupled HydroMechanical Proces
Media: Homogenisation and Parameter Identification. Developments in
Geotechnical Engineering: Coupled ThermoHydroMechanical Processes of
Fractured Media (O. Stephansson, L. J ing & CF. Tsang, eds.), 79:135164,
1996, Elsevier.
Stoyan, D., Kendall,
J . Wiley & Sons. 1987.
Strang G. & Nguyen, T. Wavelets and Filter Banks. Wellesey Cambridge
Press. 1997.
[86] Terzaghi, V.K. The Shearing Resistance of Saturated Soils and the Angle
between the Planes of Shear. First International Conference of Soil Mechanics,
Vol. 1: pp. 5456. Harvard University. 1936.
124
[87]
tor. Water Resources
[88] 
[89] rated Conditional Global
[92] Mechanical Characterization of a Bentonite from
International Meeting on
T3. INERIS Report, 1993.
6.
[96] vior of naturally fractured reservoirs. Society
[97] ship Between
Tompson A.F.B., R. Ababou, L.W. Gelhar. Implementation of the Three
Dimensional Turning Bands Random Field Genera
Research, 25(10), pp. 22272243. 1989.
Trefethen, L.N. Pseudospectra of Linear Operators. SIAM Review 39, pp. 383
406.
Tran, N.H., Chen, Z. & Rahman, S.S. Integ
Optimisation for Discrete Fracture Network Modelling. Computers &
Geosciences 32: pp. 1727. 2006.
[90] Vetterli, R., O. & M. Wavelets and Signal Processing, IEEE Signal Processing
Magazine, P.1438. 1991.
[91] Villaescusa E. & Brown E.T. Maximum Likelihood Estimation of J oint Size
from Trace Length Measurements. Rock Mechanics & Rock Engineering
25(2):6787. 1992.
Villar M.V. ThermoHydro
Cabo de Gata. A Study Applied to the Use of Bentonite Sealing Material in High
Level Radioactive Waste Repositories. Ph.D. Dissertation and ENRESA
Technical Publication 01/2002, Madrid. 2002.
[93] Villar M.V. & Lloret A. Temperature Influence on the Hydromechanical
Behavior of a Compacted Bentonite. Proceedings of the
Clays in Natural and Engineered Barriers for Radioactive Waste Confinement,
Reims. 2002.
[94] Vuillod E. & A. Thoraval. ThermoHydroMechanical Coupled Modeling in the
Near Field  BM
[95] Warburton P.M. Stereological Interpretation of J oint Trace Data: Influence of
J oint Shape and Implications for Geological Surveys. International Journal Of
Rock Mechanics & Mining Science Geomech Abstracts 17(6): pp. 30531
1980.
Warren, J .E. & Root P.J . The beha
of Petroleum Engineers Journal, pp. 245255. Trans., AIME, 228. 1963.
Zhang L., Einstein H.H. & Dershowitz W.S. Stereological Relation
Trace Length and Size Distribution of Elliptical Discontinuities. Gotechnique
52(6): pp. 41933. 2002.
125
126
9 APPENDICES
127
128
129
APPENDIX I: Fractal Characterization of the FEBEX Tracemap
This APPENDIX characterizes the fractal dimension of the FEBEX tracemap, as
an alternative to build the fractured medium from a fractal model, instead of the Poisson
model used in the thesis (for the fracture centers).
Fractal dimension of the FEBEX area has been estimated by using the traces map of the
FEBEX drift. We have implemented an algorithm of edge detection based on the
wavelet transform [61]. This algorithm extracts the lineaments of an image through
different scales by using the 2D multiresolution analysis, and we compute the fractal
dimension with a Boxcounting process on each image. For more details on the
description of this technique see [21]. In our case, the fractal dimension has been
estimated in D=1.70 (2.70 translated into 3D). Figure A1 shows the different stages of
the fractal dimension estimation algorithm and the final result of the estimation.
Same estimation of fractal dimension has been performed separately in the five different
zones in which geological reports divided FEBEX drift (see Figure 18). The two last
zones, which were chosen to install the FEBEX experiment, have the highest values of
fractal dimension, due to their higher fracture densities. This may be used for future
generations of the fractured medium with fractal models. Figure A2 shows the results
of those estimations.
a.
b.
c.
d.
Figure A1: Algorithm to estimate the fractal dimension of the FEBEX fractured
area: a. Original image of the traces map of FEBEX drift; b. Bidimensional MRA of
the image; c. Modulus, phase and 95% of highest modulus values; d. Estimation of
fractal dimension from the modulus.
ZONE 1 ZONE 2 ZONE 3 ZONE 4 ZONE 5
D
1
=1.6162 D
2
=1.6800 D
3
=1.4383 D
4
=1.8205 D
5
=1.762
Figure A2: Fractal dimension estimation for the five different zones of the FEBEX
drift.
However, it is still an opened question whether if the fractal dimension of a curve image
(cylindrical surface of the FEBEX drift developed in a plane) is equivalent to that of a
planar image or not, or if it exists any relation between them that could be used to
convert the calculated value.
130
APPENDIX II: Orientation Angles for a Planar Fracture in 3D Space
e
referred to the Geographi rred to the FEBEX drift
cal solution of the in planar
disk fracture and a cylindrical tunnel requires the use of several coordinated systems.
direction, both m respect to aphic North
In the simulation of the fractured medium of Chapter 4, some field data ar
c North, whereas some other are refe
local coordinated system. Additionally, analyti tersection of a
This APPENDIX clarifies the notations followed for the fracture vectors and angles.
A planar fracture F in 3D is defined geographically by its direction and plunge. We
can alternatively define the dip (direction of the maximum slope), instead of the fracture
easured with the Geogr N
. In our case, N
c re
geom r
normal to the plane containing that fracture. Taking the center of the fracture as the
oincides with the X direction. On the other hand, we define the same planar fractu
etrically by means of its center (a 3coordinates point in the space) and the vecto
n
F
origin of the coordinate reference system, the 3 components of the normal vector are
defined in terms of the angles formed with the coordinate axes. The following relations
hold for the different angles defining a planar fracture in 3D:
=
=180
(A1)
where
dip, angle formed by the projection of
F
n
F
n
F
u
y
x
z
n
F
cossin
n
F
= sinsin
cos
N
u
F
v
F
N Geographic North
= (N, u
F
) dip
= (u
F
, v
F
) plunge
n
F
v
F
Figure A3: Angles criteria for the 3D planar fractures used in the thesis.
131
132
APPENDIX III: Intersection of a Circular Fracture with a Cylindrical Tunnel
The analytical solution of the intersection of a planar disk fracture with a
cylindrical excavation is presented here, as it is needed to perform the
optimization/reconstruction of the fractured network of Chapter 4.2.
To define analytically the equation of a trace produced by the intersection of a circular
fracture on a cylindrical tunnel, we have to solve the system of equations formed by the
cylindrical tunnel equation and the circular fracture (disk) equation in 3D. Three
different coordinated systems have been considered:
o Absolute coordinated system XYZ: in which the Geographical North points
towards the direction X.
o Tunnel relative coordinated system X
t
Y
t
Z
t
: we have identified the origin of
this coordinated system with the previous one for simplicity, even if in the
FEBEX project it is different. The +X
t
axis forms an angle of 80 with the
+X axis (FEBEX drift direction of N260E).
o Fracture relative coordinated system X
f
Y
f
Z
f
: the X
f
points towards the dip
direction within this system.
Figure A4a shows schematically these three reference systems for a given fracture.
There could be considered an additional coordinated system X
tr
Y
tr
, as a result of the
development of the cylinder in a 2D plane, the so called tracemap, which is shown in
Figure A4b. The system of equations mentioned above is:
2 2 2
2 2 2
,
2 2
, 0
t t
t t t t
f f f f
L L
y z R x
x y R z
+ =
+ =
(A2)
being
R
t
tunnel radius,
L
t
tunnel length and
R
f
fracture radius.
To solve the system of equations (A2), we have to express all the equations in terms of
the same coordinated system. Using the angles and defined in APPENDIX II for
fracture f, and being (x
fc
, y
fc
, z
fc
) the coordinates of the fracture center, lets write the
equations in terms of the tunnel coordinated system with the aid of the corresponding
rotation matrices (see APPENDIX VIII):
ROTATION FROM (X, Y, Z) TO (X
f
, Y
f
, Z
f
)
cos cos sin cos sin
, sin cos cos sin sin
sin 0 cos
f
f
f
x x
y A y A
z z
= =
(A3)
133
RO
TATION FROM (X
t
, Y
t
, Z
t
) TO (X, Y, Z)
( ) ( )
( ) ( ) , sin 260 180 cos 260 180 0
0 0
t
t
t
y B y B
z z
= =
(A
cos 260 180 sin 260 180 0
1
x x
4)
Adding up the two rotation matrices (3.2) and (3.3), yields to
ROTATION FROM (X
t
, Y
t
, Z
t
) TO (X
f
, Y
f
, Z
f
)
(cos cos cos80 (cos cos sin80
sin
cos sin sin80) cos sin cos80)
t t
x x
(sin cos80 (sin sin80
, 0
f
f t
f
x
y A B y C y C
z
cos sin80) cos cos80)
( sin cos cos80
sin s
t
t t
z z
+
+
+
= =
(A5)
+
a.
b.
Figure A4: a. Disk fracture intersection with a cylindrical tunnel
in 3D and b. Trace formed in the tunnel wall developed in 2D.
134
Considering also the translation and rotation of the fracture center (origin of the fracture
coordinated system), we can express the fracture coordinates as:
( ) ( ) ( )
( ) ( ) ( )
11 12 13
21 22 23
f t fc t fc t fc
f t fc t fc t fc
x c x x c y y c z z
y c x x c y y c z z
= + +
( ) ( ) ( )
31 32 33 f t fc t fc t fc
z c x x c y y c z z
= + +
(A6)
= + +
Introducing (3.5) into (3.1) yields to
( ) ( ) ( ) ( ) ( ) ( )
2 2
2 2
11 12 13 21 22 23 f f t fc t fc t fc t fc t fc t fc
x y c x x c y y c z z c x x c y y c z z
+ = + + + + +
=
(A7)
where
)
+ + + +
2 2 2
0
t t t t t t t t t t t t
Ax By Cz Dx y Ex z Fy z Gx Hy Iz J = + + + + + + + + +
( )
( )
( )
( ) ( ) ( )
( ) ( ) ( )
2 2
11 21
2 2
12 22
2 2
13 23
11 12 21 22
11 13 21 23
12 13 22 23
2 2
11 21 11 12 21 22 11 13 21 23
2 2
11 12 21 22 12 22 12 13 22 23
11 13
2
2
2
2
2
2
fc fc fc
fc fc fc
A c c
B c c
C c c
D c c c c
E c c c c
F c c c c
G c c x c c c c y c c c c z
H c c c c x c c y c c c c z
I c c
= +
= +
= +
= +
= +
= +
= + + + + +
= + + + + +
= + ( ) ( ) ( )
( ) ( ) ( ) ( ) (
2 2
21 23 12 13 22 23 13 23
2 2
11 13 21 11 13 21 23 21 22 23
fc fc fc
fc fc fc fc t t fc t fc
c c x c c c c y c c z
J c x c y c z c x c c c c x c y y c z z
= + + + + + + +
(A8)
o the final system of equations to be solved is
S
( ) ( ) ( )
2 2 2
31 32 33
2 2 2
,
2 2
0
0 Ax
t t
t t t t
t fc t fc t fc
t t t t t t t t t t t t
L L
y z R x
c x x c y y c z z
By Cz Dx y Ex z Fy z Gx Hy Iz J
+ =
+ + =
+ + + + + + + + +
(A9)
he solution of this system of equations yields to the analytical equation defining the
ace, and if we solve it with equality in the third equation we get the extreme points of
e trace. Plane circular fractures intersecting with a cylindrical wall produce complete
r uncompleted ellipticalshaped traces. Depending on the number of solutions of (A9)
hich depend in fact of the size and orientation of the fracture with respect to the
nnel), four ferent general cases of intersection can be found (see Figure A5).
T
tr
th
o
(w
tu dif
135
a.
A
B
b.
A
B
C
D
c. d.
Figure A5: Different types of intersections between a disk fracture and a
cylindrical tunnel in 3D depending on the number of solutions of the
equations system: a. Uncomplete trace (two extreme points); b. Uncompleted
trace (four extreme points); c. Complete trace (zero extreme points); and d.
No trace (zero extreme points).
136
APPENDIX IV: Detailed Results of the Fractured Medium Optimization
We present in this APPENDIX a more detailed version of the results of the
fractured medium optimization presented in Chapter 4.2. We rec that the optimization
process was com sed by 2 succesive steps (see Chapter 4.2.2).
STEP 1: OPTIMIZATION AVERAGES OF 3 EVALUATIONS
Due to the stochastic nature of the fractured media generated with statistical
distribution functions, a given set of values for these functions can yield to different
fractured networks depending on the alleatory seed used to generate them. For this
reason, on each iteration of the optimization process (a fixed set of parameters), we
compute the objective function (OF) value as the average of the values obtained for 3
different fractured media (different alleatory seeds). Doing so, we minimize the
dependence of the results on the alleatory seed. The initial values for the Pareto
distribution set of parameters, PARETO
ini
=(R
min
, R
max
, b)
ini
, and the corresponding
search interval for each of those parameters, SINT
ini
, have been:
PARETO
ini
=(1, 100, 2)
SINT
ini
=(1, 0, 1) (A10)
Figure A6 presents the evolution of the OF in the optimization process, where only the
enhanced values have been plotted: 368 it tions have been necessary to reac the
optim m. Table A1 shows the values of th ptimum set of parameters found, as well
as th O igures
A7 sh one:
Figur the
cumu the
tracem
Figure A6: Evolution of the objective function
(OF) in the first step of the optimization process.
all
po
WITH
era h
e o u
e F value and the number of fractures of the optimized fractured medium. F
ow a comparison of the simulated fractured medium and the measured
e A7a plots the cumulative histograms of the trace lengths; Figure A7b plots
lative histograms of the 3D trace chords; and Figures A7c and d display
aps generated on the drift wall.
0
5
10
15
20
25
30
35
40
0 100 200 300 400
OF
iteration
137
Table A1: Main characteristics of the optimum fractured medium obtained in the
first step of the optimization process.
# of fractures R
min
R
max
b OF
2813731 0,19851 100 3,3048 0,55917
b.
a.
c.
d.
e.
f.
Figure A7: First step optimization: a. Cumulated distribution function of trace lengths
on tunnel ( observed;  fitted); b. Cumulated distribution function of chord lengths
on tunnel ( observed;  fitted); c. FEBEX drift observed tracemap; d. FEBEX drift
fitted tracemap; e. Observed tracemap detail; and f. Fitted tracemap detail.
138
STEP 2: BEST REALIZATION OF 750 WITH OPTIMA PARAMETERS
In the second step of the f 750 realizations of the
fractured medium with the optimum arameters found in step 1, changing the alleatory
s , jective function value has been ameliorated with respect to the
f
FO =0.4651 =0.0525 +0.0493 +0.0931 +0.0050 +0.1843 +0.0810 (A11)
(decomposition of the OF in the different contribution terms is described in chapter
4.2.2)
The 35elements long vector of the alleatory seed corresponding to the optimum
fractured medium is also given for results reproduction purposes:
Seedopt =(0.0453, 0.2314, 0.7191, 0.5697, 0.0413, 0.1733, 0.1024, 0.8156, 0.4372, 0.2802,
0.4669, 0.6666, 0.6525, 0.2039, 0.8377, 0.6420, 0.9931, 0.4040, 0.0103, 0.7557, 0.2186, 0.3339,
0.5347, 0.0721, 0.8249, 0.6506, 0.7555, 0.8487, 0.8015, 0.3614, 0.4406, 0.7144, 0.000, 0.000, 0.000)
S ilarly to step 1, Figure A8 presents in the optimization
process, where only the enhanced values have been plotted: the realization 554 gets the
best result of the OF. Table A2 shows a comparison of the characteristics of the
optimum fractured medium and the field measurements. Figures A9 show a
c arison of the simulated fractured medium and the measured one: Figure A9a plots
the cumulative histograms of the trace lengths; Figure A9b plots the cumulative
histograms of the 3D trace chords; and Figures A9c and d display the tracemaps
generated on the drift wall.
Table A2: Main characteristics of the optimum fractured medium obtained in the
second step of the optimization process and comparison with the measured values.
Measured Simulated
optimization we obtain the best out o
p
eed. In this step the ob
irst one:
im the evolution of the OF
omp
Number of fractures  2906474
Number of tunnel traces 614 800
Number of intersections with
borehole FEBEX95001
155 144
Number of intersections with
borehole FEBEX95002
410 234
Volumetric density
32
 2.9816
Areal density
21
1.4933 1.9182
Areal density
21
in zone 1
1.1544 1.4301
Areal density
21
in zone 2
2.0964 2.5298
Areal density
21
in zone 3
0.7961 1.2622
Areal d
21
ensity in zone 4
2.0825 2.6892
Areal density
21
in zone 5
3.0774 3.3651
139
Figure A8: Evolution of the objective function
(OF) in the second step of the optimization process.
a. b.
c.
d.
e. f.
Figure A9: Second step optimization: a. Cumulated distribution function of trace
lengths on tunnel ( observed;  fitted); b. Cumulated distribution function of chord
lengths on tunnel ( observed;  fitted); c. FEBEX drift observed tracemap; d.
FEBEX drift fitted tracemap; e. Observed tracemap detail; and f. Fitted tracemap detail.
0,4
0,5
0,7
0,65
0,45
0,55
0,6
0 100 200 300 400 500 600 700
OF
iteration
140
APPENDIX V: PseudoSpectral Method for the 1D AdvectionDiffusion Equation
This APPENDIX describes an efficient numerical method to solve the
advectiondiffusion equation the based on the Fourier transform, which was thought
initially as a way to further use the multiscale waveletbased algorithms for solving the
PDEs of the THM model of Chapter 6.
1D MODEL FOR THE ADVECTIVEDIFFUSIVE EQUATION WITH VARIABLE
COEFFICIENTS
A 1D m del solving the advectiondiffusion equation with time and space
variable coefficie t en developed. The
same algorithm c epending on the
lution). The equ ed in the Fourier
ansform.
The advectiondifusion equation with variable coefficients can be written:
o
ns (flux velocity and diffusion coefficient) has be
an be applied for the nonlinear case (coefficients d
ation is solved by a pseudospectral method [88] bas so
tr
2
2
) , (
) , (
) , (
) , (
) , (
x d
t x u d
t x D
x d
t x du
t x v
dt
t x du
+ = (A12)
where ( ) t x u , is the dependent variable (for instance, concentration), x is the position
vector, and t the time. The first term of the right hand side represents the advective
transport, with advective velocity ( ) t x v , , and the second term represents the diffusive
transport, where the diffusion coefficient is given by the tensor ( ) t x D , . This equation
has been applied to a unidimensional periodic domain x =[0, L]. Thus, for an initial
value u
0
we have:
=
=
+ =
0
2
2
) 0 , (
) , ( ) , 0 (
) , (
) , (
) , (
) , (
) , (
u x u
t L u t u
dx
t x u d
t x D
dx
t x du
t x v
dt
t x du
(A13)
PSEUDOSPECTRAL METHOD FOR THE RESOLUTION OF PDEs
The pseudospectral method applied to the resolution of Partial Differential
Equations (PDEs) is based in the Fourier interpolation concept. For a given function
u(x) R, we can define an algorithm to interpolate it in an infinite domain by the
following steps:
 Considering its orthogonal projection {v
j
} on Z, which can be considered as a
discretization of the real line (see Fi A10). gure
 Performing the Fourier transform ( ) v of the discretized function {v
j
}:
( )
+
=
j
x j i
v e x v
(A14)
141
u(x
j
) =v
j
(x
j
=jx)
Figure A10: Discretization of a function f(x)R in Z.
 Substituting in the exact inverse Fourier transform of u(x) the values of its
transform ( ) u by the approximated values ( ) v , which correspond to the u(x)
transform performed uniquely in the discretization points x
j
:
Exact value: ( ) ( )
=
x
x
d u e x u
x i
2
1
(A15)
Approximated value: ( ) ( )
=
x
x
d v e x u
x i
2
1
~
(A16)
Remark: this approximation can also be considered as an interpolation of the u(x
j
)
values, because ( ) v is a continuous function although it only contains information of
the {x
j
} points.
Substituting the values of ( ) v in the previous integral expresin:
( )
( )
( )
( )
( )
( )
+
+
=
=
=
=
=
=
j
j
j
x x i
x x i
j
x x i
j
j
x i
x i
x x
x
x x
x
v
e
v
x
d e v
x
d v e e
x
x u
x
x
j
j
x
x
j
x
x
j
2
sin 2
2
2
2
~
(A17)
We obtain finally:
( ) ( ) ( )
( )
( )
x
x
x
x
x
x
x S v x x S x u
j j
= =
c sin
sin
where ,
~
(A18)
x
142
This function sinc(x/x) provides an exact interpolator of function u(x) in the points
x
j
}.
{
Correspondingly, an analogue expresin can be obtained for the Fourier interpolator in
the case of a periodic finite domain of N+1 points, as the one we are loking for:
( ) ( ) ( )
( )
2
tan
2
2
x
x
x
N
N
= =
sin
where ,
~
2
x
x S v x x S x u
N j j N
(A19)
Once an interpolator of function u(x) has been defined, we can approximate the different
differential operators present in the EDPs for their resolution. To do that, it is enough
to apply the operators to the approximated function ( ) x u
~
. In the advectiondiffusion
quation, tho e se operators are:
Advective operator:
( ) ( ) x u
dx
d
u u
dx
du
x x
~ ~
= =
(A20)
Substituing ( ) ( )
+
=
j j
v x x S x u
~
we have:
( ) ( ) ( ) ( )
+
+
=
=
j j k j j k k x
v x x S
dx
d
v x x S
dx
d
x u
~
(A21)
And performing the corresponding derivative we get the final result:
( ) { } ( )
( )
( )
2
tan 2
=
=
=
j k
j k
j k
D
v D x u
j k
j k N
j j k N k x
if
1
if 0
with
~
,
,
(A22)
Where(D
N
)
k,j
is a Toeplitz matriz with the following structure:
143
1 1
0 ................
2
2tan 2tan
2 2
............................................................................
1
2tan
2
.......
1 1
...
2
2tan 2tan
2 2
N
x x x
D
x x
=
......................
1 1
0 ...
2
2tan 2tan
2 2
.........................................................................................................
1 1 1
................
x x
2
2tan 2tan 2tan
2 2 2
x x x
0
(A23)
D
iffusive operator:
( ) ( ) x u
dx
d
u u
dx
u d
xx xx
~ ~
2
2 2
2
= = (A24)
Computing the corresponding second derivative, we obtain, in an
way, the final expression for this
(
analogue operator:
( ) { }
( )
( ) =
+
v D x u
j j k N k xx
~
,
2
)
)
( )
(
( )
j k
x j k
j k
x
D
N
if
2
sin 2
if
6
1
3
2
2
2
(A25)
Where (D
N
(2)
)
k,j
is also a Toeplitz matrix (remark that D
N
(2)
(D
N
)
2
):
=
+ + j k
j k
1
with
1
,
2
( )
2
2
2 2
2
1 1 1 1
...... .............
2 3 6
2sin 2sin 2tan
2 2 2
...................................................................................................................
N
x x x x
D
=
2
2
2 2 2 2
...........
1 1 1 1 1
... ...
2 2 3 6
2sin 2sin 2sin 2sin
2 2 2 2
.........................................................................................
x x x x x
..............
2
2
2 2 2
.......................
1 1
..................
2 3 6
2sin 2sin 2sin
2 2 2
x x x x
(A26)
1 1
144
TIME DISCRETIZATION FOR THE ADVECTIONDIFFUSION EQUATION
For the time discretization of the advectiondiffusion equation we will adopt two
ifferent implicit schemes: the Middle Point Rule for the advective term and the
Advective term: we start from the advective equation with variable
coefficients:
d
Advanced Euler for the diffusive term. Each of those schemes leads to convergence of
the method for the corresponding term of the equation:
( )
( )
( ) ( )
( ) ( t x u t x v
dt
t x du
dx
t x du
t x v
dt
t x du
x
,
~
,
, ,
,
,
= = ) (A27)
the following way (by simplicity we will denote
u(x,t)=u(t)):
where du/dx has been substituted by the approximation computed with
the Pseudospectral method explained above. The Middle Point Rule can
be applied in
( ) ( ) ( )
( ) (t u t x v
t
t u t u
dt
t x du
x
~
,
2
1 1 ,
=
)
+
= (A28)
( ) ( ) ( ) ( ) t u t x v t t u t u
x
~
, 2 1 1 + = + (A29)
rm: we start from n equation with variable
oefficients:
Diffusive te the difusi
c
( )
( )
( ) ( )
( ) ( ) t , (A30) x u t x D
dt
t x du
dx
t x u d
t x D
t x du
xx
~
,
, ,
,
,
2
2
= =
by
e Pseudospectral method explained above. The Advanced Euler
scheme can be applied in the following way:
dt
where d
2
u/dx
2
has been substituted by the approximation computed
th
( ) ( ) ( )
( ) (t u t x D
t
t u t u
dt
t x du
xx
~
,
1 ,
=
)
+
= (A31)
( ) ( ) ( ) ( ) t u t x D t t u t u
xx
~
, 1 + = + (A32)
hole equation: we start with the advectiondiffusion equation with
W
variable coefficients:
( )
( )
( )
( )
( )
( )
( ) ( ) ( ) ( ) t x u t x D t x u t x v
dt
t x du
dx
t x u d
t x D
dx
t x du
t x v
dt
t x du
xx x
,
~
, ,
~
,
,
,
,
,
,
,
2
2
+ =
+ =
(A33)
145
where du/dx y d
2
u/dx
2
have been substituted by the approximations
computed with the Pseudoespectral method. The previous schemes of
time discretization can be applied in a joint way as follows:
( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( ) t u t x D t u t x v
t
t u t u
t
t u t u
dt
t x du
xx x
~
,
~
,
1
2
1 1 ,
+ =
+
= (A34)
( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( 1 2 1 2 1 )
~
,
~
, 2 + + = + t u t u t u t u t x D t u t x v t
xx x
(A35)
( ) ( ) ( ) ( ) ( ) ( ) [ ] 1 2
~
,
~
, 2
3
1
1 + + + = + t u t u t u t x D t u t x v t t u
xx x
( ) ( ) (A36)
APPLICATION OF THE MODEL: EXAMPLES
The described model has been implemented in Matlab 6.0. That implementation
ermits the resolution of the unidimensional advection and difussion equations
separatedly, and the resolution of the complete equation. It also permits the selection of
ariable coefficients, both in time and space. In the sequel we show some simple
amples of application of the model. In the following figu
nction values in different time instants, but also the init
nction value considering only advection and the time and/or space varying functions
of the coefficie
 Figure A11 shows an example of the time evolution of the advection
diffusion equation obtained for a simple triangularshaped function. In this
example, we consider constant coefficiens v=1 m/s and D =0.1 m/s
2
in all
the domain.
 Figure A12 presents an example of the advection equ
sinusoidal function, in which the advection coefficient varies in space
(v=2m/s in the interval [2,0] and v=1m/s in the rest of the domain), but
constant in time.
ays an example of the diffusion equation, applied to a
tria /s in
the interval [,0) and D=0.4m/s in the interval [0,]) and time
(D(x,T)=4*D(x,0)).
 Finally, Figure A14 shows a real application example. In this example, the
complete advectiondiffusion equation is applied to a real function (a one
dimensional field of some physical variable), and time and space variable
coefficients are used.
p
v
ex res we present, not only the
fu ial function value, the final
fu
nts v(x,t) and D(x,t):
ation, applied to a
 Figure A13 displ
ngular function, with diffusion coefficient varying in space (D=0.1m
146
b. a.
c. d.
Figure A11: Time evolution of the advectiondiffusion equation for a
triangular function with constant coefficents.
a.
b.
c. d.
Figure A12: Time evolution of the advection equation for a sinusoidal function
ependent coefficient.
with space d
147
b. a.
c.
d.
Figure A13: Time evolution of the diffusion equation for a triangular function
with time and space dependent coefficient.
a. b.
c.
d.
Figure A14: Time evolution of the advectiondiffusion equation for a complex
function with time and space dependent coefficients.
148
APPENDIX VI: DualContinuum Model for Fractured Rock
(Illustrative Examples)
A dualcontinuum hydraulic model for fractured media has been developed
and implemented in Comsol Multiphysics 2.3 (previously called Femlab) [3] (full
article in APPENDIX XIV) as an alternative to the singlecontinuum model used for
hydraulics in the THM simulations. In this model, the fractured medium is considered
as a superposition of two continuous media (fractures and rock matrix), in which an
exchage coefficient relates the continuum va edia (in this case fluid
pressure).
This model is based in equations describing the reactivediffusive systems, such as
mixtures of several reactive chemical components in a common fluid medium. The
equation describing the concentration for each component i of such a system is:
riables of both m
( )
i i i
i
dl
R c D
t
c
u = +
(A37)
where
c
D
i
R
i
reaction rate of species i.
u
dl
velocity profile of species i.
APPROXIMATION EQUATIONS FOR A DUALCONTINUUM
To apply the equations of a diffusivereactive system to a fractured medium, we
consider the rock matriz and the fractures as the two only species existing in the system,
in such a way that they are present all over the domain, as two continuous media
(contrary to the classic discrete fracture network models). Concentration for each of the
species in this model represents the fluid pressure on each media, the velocity profile is
equivalent to the capacity of the media, and the reaction terms are characterized by an
exchange coefficient between the rock m ix and the fractures.
The final system of equations would be set as follows:
i
concentration of species i.
diffusion coefficient of species i.
atr
( ) ( )
( ) (
+ = +
= +
F M F F
F
F
F M M M
M
M
P P P K
t
P
C
P P P K
t
P
C
)
(A38)
where
P
M
, P
F
fluid pressure in matrix nd fractures respectively.
C
M
, C
F
capacity of matrix and fractures respectively.
The exchange coefficient is based on the model described in [5]. In this model, a
discrete partition of the domain in different blocks is considered, and an estimation of
a
K
M
, K
F
diffusion coefficients for matrix and fractures respectively.
exchange coefficient between matrix and fractures.
149
the value of the coefficient for each of them is made out of: the specific surface of
each matrix block, the porosity and the diffusion coefficients K
M
, K
F
.
This concept is then applied in the domain homogenization process at different scales,
and a different exchange coefficient is assigned for each block of each partition of the
domain within each considered scale. In such a way, the scale and space invariance
between rock matrix and fractures can be estimated.
EXAMPLES OF THE MODEL APPLICATION
In the first example considered here, a time dependent exchange analysis has
been carried out in a very simple 2D sinthetic medium with three fractures. For the
estimation of , the finest case for blocks partition of the domain has been used, in
which each pixel is considered as a block. If the pixel belongs to a fracture, the
exchange coefficient will be equal to 1 in that pixel, and if it belongs to the matrix, the
exchange coefficient will be equal to zero. Thus, no upscaling algorithm has been
necessary in this example. The inicial and boundary conditions are the following:
I.C.: P
M
=0, P
F
=1 (A39)
B.C.:
x 0
F
P
(Neumann no flux conditi
x
n
n
0
M
P
ons in ) (A40)
Figure A15 displays the fluid pressure in the rock matrix at different times, starting at
zero pressure and increasing with time at those places where the exchange coefficient is
equal to one (presence of a fracture). At the same time, pressure propagates towards the
interior of the matrix due to diffusion. Figure A16 displays the fluid pressure in the
fractures continuum at different times. In this case, the inicial pressure is equal to one
all over the domain, and decreases almost uniformly in all the domain due to the high
diffusion coefficient imposed for fractures (K
F
=10
4
K
M
here). However, slight
differences can be appreciated at those zon is
equ to 1, in which pressure dissipates faste
In the second example, a time dependent flow & exchange analysis of a homogenized
3D fractured medium has been carried out. An extrait of about 19000 fractures of the
fractured medium simulated in the thesis has been homogenized in a 3x3x3 partition of
the domain. The inicial and boundary conditions are the following (Table A3):
Table A3: B.C. and I.C. for the 3D example of the dualcontinuum model.
igure A17a displays the steady state of a case with low exchange coefficient, in which
the flow process is dominant. Figure A17b displays the steady state of a case with high
xchang coefficient, in which the exchange process is the dominant one.
Medium B.C. (Neumann) I.C.
es in which the exchange coefficient
r. al
A
1
: flux=10
14
m
3
/s A
2
:no flux o flux
F
e e
A
3
:n
m
B : flux=10
14
m
3
/s B :no flux B :no flux
Pm=0 Pa atrix
1 2 3
A
1
: flux=10
14
m
3
/s A
2
:no flux A
3
:no flux
fra
B
1
: flu
A1
B1
B2
A2
A3
B3
cture
x=10
14
m
3
/s B
2
:no flux B
3
:no flux
Pf=10 Pa X
Y
Z
150
151
a. b.
c.
d.
Fig atrix fluid pressure for an example of
the dualcontinuum model in a 2D fractured medium.
ure A15: Time evolution of the rock m
b. a.
c. d.
Figure A16: Time evolution of the fractures fluid pressure for an example of the
dualcontinuum model in a 2D fractured medium.
a.
Flow direction
b.
Flow direction
Figure A17: Steady state of the fractures fluid pressure for an example of the
dualcontinuum model in a 3D fractured medium.
152
APPENDIX VII: Temperature Dependence of Water Viscosity
An Excel application with functions of water temperature dependent properties
has been used in Chapter 6.1 for the computation of the water viscosity as a function of
temperature in the THM model. These functions are defined for the interval of
temperature [0, 100]. Water dynamic viscosity values have been obtained, and a 4
th
degree polynomial has been fitted to the dataset. To assure the continuity of the
polynomial fitting beyond the interval, artificial values have been assigned for some
higher temperatures.
Figures A18a and b show the data and the fitting polynomial for two different
temperature scales. Table A4 presents the data set used for the fitting.
a.
0 10 20 30 40 50 60 70 80 90 100
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
x 10
3
Temperature (C)
W
a
t
e
r
s
c
o
s
i
t
y
(
N
s
/
m
2
)
b.
d
y
n
a
m
i
c
v
i
100 50 0 50 100 150 200
0
0.002
0.004
0.006
0.008
0.01
0.012
0.014
0.016
Tempe e (C)
W
a
t
e
r
d
y
n
a
m
i
c
v
i
s
c
o
s
i
t
y
(
N
s
/
m
2
)
ints)
and fitted po
ratur
Figure A18: Water dynamic viscosity values (xmarked po
lynomial (solid line) used in the models.
153
Table A4: Excel dataset and polynomialfunction fitted values of water dynamic
viscosity for the temperature interval [0, 100].
Temperature
[C]
Dynamic
viscosity
[Ns/m
2
]
Fitted
dynamic
viscosity
[Ns/m
2
]
Temperature
[C]
Dynamic
viscosity
[Ns/m
2
]
Fitted
dynamic
viscosity
[Ns/m
2
]
0 0.001787 0.0017191 51 0.000538 0.00052106
1 0.0017283 0.001675 52 0.0005293 0.00051215
2 0.0016722 0.001632 53 0.0005208 0.00050356
3 0.0016187 0.0015901 54 0.0005125 0.00049528
4 0.0015677 0.0015493 55 0.0005044 0.00048731
5 0.001519 0.0015095 56 0.0004966 0.00047964
6 0.0014726 0.0014708 57 0.0004889 0.00047224
7 0.0014283 0.0014331 58 0.0004814 0.00046512
8 0.001386 0.0013964 59 0.0004741 0.00045826
9 0.0013456 0.0013606 60 0.000467 0.00045165
10 0.001307 0.0013259 61 .00046 0.00044528 0
11 0.0012701 0.001292 62 0 004532 0.00043915 .0
12 0.0012348 0.0012591 63 0.0004465 0.00043323
13 0.0012011 0.0012271 64 0.00044 0.00042754
14 0.0011688 0.001196 65 0.0004336 0.00042204
15 0.001138 0.0011657 66 0.0004274 0.00041675
16 0.0011084 0.0011363 67 0.0004213 0.00041164
17 0.0010801 0.0011077 68 0.0004154 0.00040671
18 0.001053 0.00108 69 0.0004096 0.00040194
19 0.001027 0.0010531 70 0.000404 0.00039734
20 0.001002 0.0010269 71 0.0003985 0.0003929
21 0.000978 0.0010015 72 0.0003932 0.00038859
22 0.0009549 0.00097685 73 0.000388 0.00038443
23 0.0009326 0.00095294 74 0.000383 0.00038039
24 0.0009112 0.00092974 75 0.0003781 0.00037648
25 0.0008906 0.00090725 76 0.0003732 0.00037269
26 0.0008707 0.00088545 77 0.0003685 0.000369
27 0.0008516 0.00086432 78 0.0003639 0.00036541
28 0.0008331 0.00084385 79 0.0003594 0.00036192
29 0.0008153 0.00082401 80 0 0355 0.00035851 .00
30 0.000798 0.00080481 81 0. 03507 0.00035519 00
31 0.0007813 0.00078621 82 0.0003464 0.00035194
32 0.0007651 0.00076821 83 0.0003422 0.00034876
33 0.0007495 0.00075079 84 0.0003381 0.00034564
34 0.0 0.00034258 007343 0.00073393 85 0.0003341
35 0.00 0.00033957 07197 0.00071763 86 0.0003301
36 0.0007055 0.00070187 87 0.0003262 0.00033661
37 0.0006917 0.00068664 88 0.0003224 0.00033369
38 0.0006784 0.00067191 89 0.0003187 0.00033081
39 0.0006655 0.00065768 90 0.000315 0.00032795
40 0.000653 0.00064394 91 0.0003114 0.00032512
41 0.0006409 0.00063067 92 0.0003079 0.00032232
42 0.0006291 0.00061785 93 0.0003044 0.00031953
43 0.0006178 0.00060549 94 0.000301 0.00031676
44 0.0006067 0.00059355 95 0.0002977 0.00031399
45 0.000596 0.00058203 96 0.0002944 0.00031123
46 0.0005856 0.00057092 97 0.0002912 0.00030847
47 0.0005755 0.00056021 98 0.0002881 0.00030571
48 0.0005657 0.00054988 99 0.000285 0.00030294
49 0.0005562 0.00053992 100 0.000282 0.00030016
50 0.000547 0.00053032
154
APPENDIX VIII: Matricial Form of the 2
nd
and 4
th
rank tensor equations
will use the Kelvin notation for 2 rank and 4 rank tensors. For
xample, equation (31) can be written differently:
In this APPENDIX we will explain how to get the reduced system of equations
(3537) of Chapter 5.2.5 out from the governing laws and the constitutive equations and
we will convert it into the pseudomatricial form used in Comsol Multiphysics 2.3
Coefficient Form, in order to be able to clearly identify each coefficient involved in the
system of equations.
For this derivation we
nd th
e
 Euler notation: T T P B e T
Ts kl ijkl ij kl ijkl ij
= , (i, j, k, l =1, 2, 3)
 Kelvin notation: T T P B e T
Ts K IK I K IK I
= , (I, K =1, 2, 6)
Table A5: Kelvin notation for the 2
nd
rank stress tensor.
ij
11
22
33
23
13
12
I
1
2
3
4
5
6
Table A6: Kelvin notation for the 2
nd
rank strain tensor.
ij
11
22
33
2
23
2
13
2
12
I
1
2
3
4
5
6
Table A7 elvin notation for the 4
th
rank stiffness tensor (only the first row showed
as an example).
T
ijkl
t
1111
t
1122
t
1133
t
1123
t
1113
t
1112
: K
T
IJ
t
11
t
12
t
13
t
14
t
15
t
16
Table A8: Kelvin notation for the 2
nd
rank Biot coefficient tensor.
B
ij
b
11
b
22
b
33
b
23
b
13
b
12
B
I
b
1
b
2
b
3
b
4
b
5
b
6
Table A9: Kelvin notation for the 2
nd
rank intrinsic permeability tensor.
k
ij
k
11
k
22
k
33
k
23
k
13
k
12
k
32
k
31
k
21
k
I 1
k
2
k
3
k
4
k
5
k
6
k
7
k
8
k
9
k
155
Let us start with the set of equations (35) (three equations), which came from the
e momentum balance of the equivalent medium
combination of th , (24), and the
edium stress and strains equati (3 ion (3 be
ma cons e inhe ries s, as
t
t
t
t
t
t
12
13
23
1112
3 1113
2312 3 1123
3 1133
2212 3 1122
3 1111
(A41)
where the shear stresses
ij
( noed, in eral, as eep e
notation of
ij
f e of y.
The qua e e uivalent medium, ons
equivalent m
expressed in
ons, (31) and
rent symmet
2). Equat
of tensor
1) can
follows:
33
b
b
b
b
b
b
T
33
t
1212
t t
1312 2312
t t
3312 2212
t t t t t
t t t
231 2323
t t
3323 2223
t t t t t
t t t
221 2223
t t
2233 2222
t
1112
t t
1133 1122
Ts
23
13
23
22
11
12
12
13
22
11
2
2
33
22
11
13
23
33
22
11
12
1312 131 2313 3313 2213
3312 331 3323 3333 2233
111 1123
ij) are de t gen
ij
. We k , though, th
or the shak simplicit
momentum balance e tions for th q i.e. equati (24), are:
3 3
33
2 1
x x
23 13
+
2 3
23
1 3
13
x
z
g
x
x
z
g
x
x
z
g
x
= +
= +
= +
(A42)
Substituting equ 1) yields, in lvin not
2 1
x
22 12
2 1
x x
12 11
+
x
+
eq
eq
eq
15
t
6
56 5 55 4
t t + +
3 35 2 5
t + +
3
36
t T
Ts Ts
16
t
6 66 5 56 4
t t + +
3 36 2 26
t + +
2
x
13 Ts 6 16 5 15 4 14 3 13 1 11
t t + t t + t T T t +
x
g
eq
z
x
( )
( )
( )
2
34
T t T
Ts Ts
24 4 6 44 2 1 14
22 2 6 24 2 1 12
26 16 6 4 2 26
1
t T P b t t t
t T P b t t t
T t t b t t
x
Ts Ts
= + +
+ + +
+ + +
14
t
Ts
46 5 45 4
t t + +
3 34 2 4
t + +
3
23
T t T
Ts Ts
12
t
Ts
26 5 25 4
t t + +
3 23 2 2
t + +
2
x
36 Ts 6 66 5 56 46 3 36 1 16
t + t t + t P T T t +
x
g
eq
z
x
156
( )
( )
( )
3
33 23 13 3 6 36 5 35 4 34 3 33 2 23 1 13
3
34 24 14 4 6 46 5 45 4 44 3 34 2 24 1 14
2
x
z
g T t T t T t P b t t t t t t
x
T t T t T t P b t t t t t t
x
eq Ts Ts Ts
Ts Ts Ts
= + + + + +
+
+ + + + + +
(A4
35 25 15 5 6 56 5 55 4 45 3 35 2 25 1 15
1
T
x
Ts Ts Ts
3)
ow we have to introduce the equation (32), defining the deformations in terms of
T t T t t P b t t t t t t + + + + + +
N
equivalent medium displacements:
( ) ( [ ) ]
( ) ( ) [ ]
( ) ( ) [ ]
1
35 25 15
3
5
1
2
56
2
1
56
3
55
1
55
3
45
2
45
3
35
2
25
1
1
15
3
36 26 16
2
6
1
2
66
2
1
1 3 2 3 3 2 1
1
16
2
13 12 11
1
1
1
1
2
3
1
15
2
3
14
3
2
14
3
3
11
1
x
g T t t t P b
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
T t t t
x
P b
x
u
t
x
u
x
u
t
x
T t t
x x
x
u u u
x
u
t
x
u
t
x
u
t
x
u u u
x
Ts
= + +
+
+ +
+
+ + +
16
2
1
16
1
3
15 13
2
2
12
1
1
t
x
t
x
t t
x
t
x
t
+ +
t P b
66
3
56
1
56
3
46
2
46
3
36
2
26
t
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
+
+
Ts
2
x
1 3 2 3 3 2
z
x
Ts
eq
3
x
( ) ( ) [ ]
( ) ( ) [ ]
( ) ( ) [ ]
2
34 14
3
4
3
1
2
46
2 1 3 2 3 3 2 1 3
23 22 12
2
2
2
2
26
2
1
1
3
25
1
2
2
3
3
2
24
3
23
2
2
1
1
12
1
2
66
2
66
1
3
3
1
56
2
46
3
2
3
3
36
2
26
1
1
1
x
z
g T t t t
x
P b
x
u
t
x x
T t t t
x
P b
x
u
t
x
u
t
x
u
t
u
t
x
t
x
t
u
t
x
u
t
x
u
t
x
u
t
x
t
x
u
x
u
t
x
t
x
u
x
u
t
x
t
x
u
x
Ts
= + +
+
+ +
+
+
3
u
1
u
2
u
16
t
46
t
56
t
2
x
+
22
+
24
+
36 26 16
1
6
1
u u
T t t t
x
P b
x
Ts
+ + +
26
+
3
5
x
3
x
1
x
+
1
46
3
45
1
45
3
44
2
44
3
34
2
24
1
14
x
u
t
x
u
t
x
u
t
u
t
x
u
t
x
u
t
x
u
t
x
u
t +
eq
Ts
24
x
157
( ) ( ) [ ]
( ) ( ) [ ]
( ) ( ) [ ]
3
33 23 13
3
3
3
x
g T t t t
x
P b
x
eq Ts
= + +
1
2
36
2
1
36
1
3
35
3
1
35
2
3
34
3
2
34
3
3
33
2
2
23
1
1
13
3
34 24 14
2
4
2
1
2
46
2
1
46
1
3
45
3
1
45
2
3
44
3
2
44
3
3
34
2
2
24
1
1
14
2
1 1
z
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
T t t t
x
P b
x
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
x x
Ts
+
+ + +
(A44)
35 25 15 5
1
2
56
2
1
56
1
3
55
3
1
55
2
3
45
3
2
45
3
3
35
2
2
25
1
1
15
1
T t t t P b
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
u
t
x
Ts
+ + +
x
z
,
0
2
=
x
z
, and
1
3
=
z
x
and writing the equation in matrix form we get:
t t t
t t t
P b
b
u t t t
t t t
u t t t
t t t
u t t t
t t t
t t t
P
b
b
b
u
t t t
t t t
t t t
u
t t t
t t t
t t t
u
t t t
t t t
t t t
eq Ts
Ts
Ts
+ +
+ +
+
+
+ +
+ +
T t t t
t t t
=
+ +
+ +
13 12 11
0
g T
t t t
t t t
t t t
P
b
b
b
u
t t t
t t t
t t t
u
t t t
t t t
t t t
u
t t t
t t t
t t t
T
t t t b t t t t t t t t t
+
+ +
33 23 13
34 24 14
35 25 15
3
4
5
3
33 34 35
34 44 45
35 45 55
2
34 23 36
44 24 46
45 25 56
1
35 36 13
45 46 14
55 56 15
34 24
23 22 12
36 26 16
2
6
3 23 24 25
36 46 56
2 24 22 26
46 26 66
1 25 26 12
56 66 16
35 25 15
36 26 16
5
6
1
3
35 45 55
36 46 56
13 14 15
2
45 25 56
46 26 66
14 12 16
1
55 56 15
56 66 16
15 16 11
0
14 4 34 44 45 44 24 46 45 46 14
(A45)
Now we develop equation (36), coming from the combination of the mass
balance for fluid (22), the Darcys law (23) and the fluid production state equation (29):
3 3
33
2 2
23
1 1
13
3
3 3
23
2 2
22
1 1
12
2
3 3
13
2 2
12
1 1
11
1
1
2
2
1
12
1
3
3
1
13
2
3
3
2
23
3
3
33
2
2
22
1
1
11
1
x
z
g
x
P k
x
z
g
x
P k
x
z
g
x
P k
x
x
z
g
x
P k
x
z
g
x
P k
x
z
g
x
P k
x
x
z
g
x
P k
x
z
g
x
P k
x
z
g
x
P k
x t
T
t
P
G
x
u
x
u
t
b
x
u
x
u
t
b
x
u
x
u
t
b
x
u
t
b
x
u
t
b
x
u
t
w
w
w
w
w
w
w
w
w
w
w
w
w
w
w
w
w
w
Tw eq
(A46)
Considering here that
w
is constant in space, we have in Kelvin notation:
b
( ) ( ) ( ) 0
1 1
3
4
5
3 4 5
4 2 6
5 6 1
3
3
4
5
2
4
2
6
1
5
6
1
=
k
k
k
g P
k k k
k k k
k k k
t
T
t
P
G
u
t
b
b
b
u
t
b
b
b
u
t
b
b
b
w
w
Tw eq
(A47)
158
And finally, we develop equation (37), which came from the combination of the
heat energy balance, (26), and the Darcys law, (23):
( )
( ) T K f
x
T
x
z
g
x
P k z
x
P k
x
z
g
x
P k
w
w w w
+
23 13
2 3 3 2 2 1 1
x
g
x
T
x
z
g
x
P k
x
z
g
x
P k
x
z
g
x
P k
x
T
x
z
g
x
P k
x
z
g
x
P k
x
z
g
x
P k
C
t
T
C
T T w
w
w
w w
w w w
w
w
w
w
w
w
w w eq
= +
+
+
3 3 3
33
2 2 1 1
23 22 12
1 3 3
13
2 2
12
1 1
11
(A48)
where we have assumed a fully isotropic tensor (K
T
)
ij
(as it depends on fracture and
matrix volumetric fractions, it is space dependent and has to stay inside the space partial
derivative). Again, we have assumed also independency of
w
in space, and applying
that
0
1
=
x
z
,
0
2
=
x
z
, and
1
3
=
x
z
. The final form of equation (37), in Kelvin notation,
is:
( ) ( ) 0
3
4
5
2
3 4 5
4 2 6
5 6 1
= +
T K f
k
k
k
gC P
k k k
k k k
k k k
C
T
t
T
C
T T w w
w
w w
eq
(A49)
159
160
A NDIX IX: Upscaling the Basic Fractured Block Flux Density by the Method PPE
of Vectorial Surface Flux
The Vectorial Surface Flux (VSF) method is presented in this APPENDIX as an
alternative to the Volume Averaged Flux (VAF) method used in Chapter 5.3.2.1 of the
thesis. For a vector flux density q at the scale of an individual fractured block, the VSF
is defined as follows:
*
* F
F
(A
where ds is the length differential element in 2D and the surface differential element in
3D, and
ds
ds
q
q
50)
lying this equation to the domain defined in Figure 32 yields to:
F
is the sum of the external surfaces of the block.
App
( )
( )
2
*
2
*
0
2 4 4
2
2 4 4
2 2
0 0 0 0
1 0 0 0 0
2 2 0 0 0 0
M M F
M
M F
M F
M M
l
b a
l l b a a l
a
l l
l l b a a l
b b
K K
a a
K K
l l
b b
K K
+
+ +
= =
+ +
+ +
= +
+ +
q q q
q q
j
F
+ = q
(A51)
calling
2
a
l
b
=
+
we can write:
0
( )
( )
* *
0
1 0
0 1 0
0 0
M F
M F
M
K K
K K
K
+
= +
q j
(A52)
On the other hand, the global gradient j
*
over the block would be, similarly:
*
* F
F
ds
ds
j
j
(A53)
( )
( )
( )
2
* *
0 2
1 0 0
2 4 4
1 0 1 0
2 4 4
0 0
M M F
M
F
l l b a a l
l l b a a l
K
K
+ +
= = +
+ +
j j j
j I j
(A54)
161
* *
0
1 0 0
0 1 0
0 0 1 1
M
F
K
K
=
j j
(A55)
and by direct substitution of (A55) into (A52) we can identify terms with those of eq.
(60):
( )
( )
* * *
0 0
1 0 0
0 1 0
0 0
1 1
M F
M F
M
M
F
K K
K K
K
K
K
+
= +
q
*
= j K j
(A56)
where
( )
( )
*
*
11
* * * *
22
*
*
33
1
0 0
0 0 ; 1
0 0
1
1
M F
xx
yy ij ij M F
zz
M F
K K
K
K K K K
K
K
K K
= +
= = =
K
K +
(A57)
Note that terms K
11
*
and K
22
*
correspond to some kind of arithmetic mean of and
K
M
weighted by and (1) respectively. These weighting factors are in
specific external surfaces of fracture and matrix with respect to the block, given by
= S
F
/S
block
and (1) = S
M
/S
block
respectively. On the other hand, term is a
harmonic mean of and K
M
weighted by the same factors mentioned above.
F
K
fact the
K
33
*
F
K
162
APPENDIX X: Solid rotations and their matrix representation in 2D and 3D
Solid rotations are used in Chapter 4.2 for the optimization of the fractured
medium
ROTATION MATRIX IN 2D
We aim to write the rotation matrix in 2D in term
e unitary normal vector to a fracture plane defin
* *
, where computing the exact analytical solution of the intersection of a planar
disk fracture with a cylindrical excavation (developed in detail in APPENDIX III)
becomes necessary to calculate the cumulative histograms of trace length and 3D trace
chord.
s of the vector components of
th ed within an elementary rock block.
There are two coordinated systems to consider: (X
1
, X
2
) is the fracture relative
coordinated axis, also know here as the (hydraulically speaking) principal axes; and the
absolute coordinated system (X
1
, X
2
).
In general, an anticlockwise rotation from the (X
*
, X
*
) coordin
1 2
ated system to the
(X
1
, X
2
) one (see Figure A19) is defined by the rotation matrix:
cos sin
sin cos
A
=
(A58)
with the following properties:
i) A A I A A
T T
= =
ii)
(A59)
x A x x A x
T
= =
* *
(A60)
X
1
*
X
2
*
X
2
X
1
Figure A19: Anticlockwise rotation of
degrees in 2D.
163
The Darcys Law in the principal axes (X
1
*
, X
2
*
) is:
* * *
J K q = (A61)
nd knowing that a
J A J q A q
T T
= =
* *
; (A62)
we have in the absolute axes (X
1
, X
2
):
J A K A q J A K q A
T T T T
= =
* *
(A63)
The rotation matrix A can be expressed in terms of the unitary normal vector
components (see Figure A20):
cos
sin
2
1
n
n
n
=
2 1
1 2
n n
n n
A (A64)
X
2
*
X
2
X
1
*
X
1
n
2
n
1
n
Figure A20: Relation between the
anticlockwise rotation of degrees and the
fracture normal vector in 2D.
ROTATION MATRIX IN 3D
We aim to write the rotation matrix in 3D in terms of the vector components of
the unitary normal vector to a fracture plane defined within an elementary rock block.
As in the 2D case, there are two coordinated systems to consider: (X
1
*
, X
2
*
, X
3
*
) is the
fracture relative coordinated axis; and the absolute coordinated system (X
1
, X
2
, X
3
).
164
In ed
tation of degrees (dip) around the X
3
*
axis followed by a righthanded rotation of
egrees (plunge) around the X
*
axis (see Figure A21). The rotation described is
defined
this case, we define a composed rotation which corresponds to a righthand
ro
d
2
by the following composed matrix:
cos sin 0 cos 0 sin
sin cos 0 0 1 0
A A A
= =
0
0 1 sin 0 cos
= =
cos cos sin cos sin
sin cos cos sin sin
sin 0 cos
=
(A
65)
ith the following properties: w
i) A A I A A
T T
= = (A66)
ii) x A x x A x
T
= =
* *
(A67)
X
1
*
X
2
*
X
3
X
2
X
1
X
3
*
=n
1
st
2
nd
Figure A21: Anticlockwise rota on of degrees over the X
3
axis
followed by a
2
axis in 3D.
he Darcys Law in the principal ax
1 2 3
*
) the same as (A6.13.2), but with
D vectors and matrices, and applying properties i) and ii) in a similar way that in the
te axes (X
1
, X
2
, X
3
):
ti
clockwise rotation of degrees over the X
es (X
*
, X
*
, X T
3
2D case, we get the Darcys Law in the absolu
165
J A K A q J A K q A
T T
= =
*
T T
*
(A68)
ote the similarity with equation (A63). The rot N ation matrix A can also be expressed
in terms of the unitary normal vector components (see Figure A22). To do it we use the
llowing geometrical relations: fo
1
12
n
n
1
2
2
12
3
2 2 2
1 2 3
2 2
12
12 1 2
cos
sin cos
sin sin
sin
cos
1
sin
n
n n
n n
n
n
n
n n n n
n
n
n n n
n
=
=
=
=
= =
= + + =
= +
= =
69)
:
3
cos n n =
3
12
n
(A
so the rotation matrix yields to
1 3 2
1
12 12
2 3 1
2
12 12
12 3
0
n n n
n
n n
n n n
A n
n n
n n
(A70)
X
1
*
X
2
*
X
3
X
2
X
1
N=X
1
X
3
*
n
n
3
n
2
n
12
n
1
Figure A22: Relation between the anticlockwise rotation of
degrees over the X
3
axis followed by a clockwise rotation of
degrees over the X
2
axis and the fracture normal vector in 3D.
166
APPENDIX XI: Full Results of the Fractured Medium THM Upscaling
This APPENDIX presents the values of the full set of homogenized or upscaled
oefficients for the oneblock homogenization of Chapter 5.3.4.2.
ONE BLOCK HOMOGENIZATION
Equivalent 2nd rank intrinsic permeability tensor
c
kij = 1.0e017 *
0.1092 0.0043 0.0017
0.0034 0.1112 0.0015
013 0.1099
0.7735 0.626
0.5495 0.7708 0.2163
0.3157 0.1133 0.8794
eigenvalues = 1.0e017 *
0.1055 0 0
0 0.1142 0
0 0 0.1107
Equivalent 4
0.0009 0.0
eigenvectors =
9 0.4242
th
rank stiffness tensor
Tijkl = 1.0e+009 *
3.1096 2.4461 2.2865 0.0167 0.0212 0.1338
2.4461 3.5982 2.4194 0.2142 0.0146 0.1589
2.2865 2.4194 5.3933 0.3643 0.1679 0.0514
0.0167 0.2142 0.3643 0.8816 0.0777 0.0161
0.0212 0.0146 0.1679 0.0777 0.5725 0.0660
0.1338 0.1589 0.0514 0.0161 0.0660 0.8336
eigenvectors =
0.7553 0.4310 0.4937
0.6546 0.5328 0.5363
0.0318 0.7283 0.6846
eigenvalues = 1.0e+009 *
0.8933 0 0
0 2.2700 0
0 0 8.9378
167
Equivalent 2nd rank Biot coefficients tensor
22
0.0186 0.9411 0.0163
0.0022 0.0163 0.9271
eigenvectors =
0.3839 0.7094 0.5911
0.6374 0.2596 0.7255
0.6681 0.6553 0.3525
eigenvalues =
0
0 0.9313 0
0 0 0.9642
Equivalent Biot modulus
Bij =
0.9401 0.0186 0.00
0.9129 0
G = 4.1877e+010
2nd rank geometric tensor
Fij =
12.8851 1.7267 0.4056
1.7267 7.4709 1.5788
0.4056 1.5788 3.3025
2
nd
rank complementary??? tensor
Bij_prime = 1.0e009 *
0.1844 0.0247 0.0058
0.0247 0.1069 0.0226
0.0058 0.0226 0.0473
4
th
rank geometric tensor
Fijkl =
8.4255 3.3614 1.0981 0.6930 0.4323 0.7664
3.3614 3.3866 0.7229 0.3898 0.0256 0.7374
1.0
0.6
0.4
981 0.7229 1.4815 0.4960 0.0523 0.2229
930 0.3898 0.4960 0.7229 0.2229 0.0256
323 0.0256 0.0523 0.2229 1.0981 0.6930
168
0.7664 0.7374 0.2229 0.0256 0.6930 3.3614
4
th
rank geometric tensor
12.8851 0 0 0 0.2028 0.8634
0.7894 0 0.8634
0 0 3.3025 0.7894 0.2028 0
.4317 0.1014
0.2028 0 0.2028 0.4317 4.0469 0.3947
4 0 0.1014 0.3947 5.0890
Gijkl =
0 7.4709 0 
0 0.7894 0.7894 2.6934 0
0.8634 0.863
Fractures equivalent 4
th
rank compliance tensor
0.7587 0.4329 0.1414 0.0892 0.0267 0.0248
0.4329 0.6329 0.0931 0.0628 0.0033 0.0286
0.1414 0.0931 0.2818 0.0491 0.0357 0.0287
0.0892 0.0628 0.0491 0.2923 0.0331 0.0112
0.0267 0.0033 0.0357 0.0331 0.4377 0.0328
0.0248 0.0286 0.0287 0.0112 0.0328 0.2953
Matrix equivalent 4
Cijkl_prime = 1.0e009 *
th
rank compliance tensor
Mijkl_
0.2000 0.0600 0.0600 0 0 0
0 0
0.0600 0.0600 0.2000 0 0 0
0 0.1300 0 0
0 0 0 0 0.1300 0
prime = 1.0e010 *
0.0600 0.2000 0.0600 0
0 0
0 0 0 0 0 0.1300
Fractured medium equivalent 4
th
rank compliance tensor
Tijkl_prime = 1.0e009 *
0.7787 0.4389 0.1474 0.0892 0.0267 0.0248
0.4389 0.6529 0.0991 0.0628 0.0033 0.0286
0.1474 0.0991 0.3018 0.0491 0.0357 0.0287
0.0892 0.0628 0.0491 0.3053 0.0331 0.0112
0.0267 0.0033 0.0357 0.0331 0.4507 0.0328
0.0248 0.0286 0.0287 0.0112 0.0328 0.3083
169
170
AP HM Simulations PENDIX XII: Comsol Multiphysics Report of the T
sent in this APPENDIX the report automatically generated by the
COMSOL Multiphysics software, in order to give all the necessary details to
reprodu esis.
Applic
o PDE, Coefficient Form
1. MODEL PROPERTIES
Propert
Model name THM3DwithExcavations
Author
Compan
Departm
Referen
Saved d
Creatio
FEMLA
2. CON
Name Expressi Value
thetam
thetaf
row0
ros
g
betaTgr
betaTw 4.421e4 4.421e4
Cw
Cs 850 850
KTw 0.58
KTgran 2.1 2.1
sin80
cos80
sin15
cos15
R
T0
2.7182818284 2.718282
uw0 1e3 0.001
uw0 1e6 1e6
We pre
ce the THM models developed in the Chapter 6 of the th
ation modes and modules used in this model:
Geom1 (3D)
o PDE, Coefficient Form
o PDE, Coefficient Form
y Value
Israel Caamn Valera
y U.P.M., E.T.S.I.Minas
ent D.M.A.M.I.
ce 
ate Aug 4, 2006 9:48:23 AM
ate Aug 23, 2005 11:11:25 AM n d
B version FEMLAB 3.1.0.157
STANTS
on
0.008 0.008
1 1
1000 1000
2350 2350
9.81 9.81
anite 3*7e6 2.1e5
4180 4180
0.58
ite
sin(80*pi/180) 0.984808
cos(80*pi/180) 0.173648
sin(15*pi/180) 0.258819
cos(15*pi/180) 0.965926
1.14 1.14
13 13
e
m
n
171
3. GEOMETRY
Number of geometries: 1
Point Mode Edge Mode
Boundary Mode Subdomain Mode
4. GEOM1
Space dimensions: 3D
4.1. Scalar Expressions
Name Expression
FBXinst (FBXr2=35)*(FBXx<=15.30)
FBXaux (FBXr2<=1.75^2)*(FBXx>=41.162366)*(FBXx<=35)
FBXtest (FBXr2<=R^2)*((FBXx>=18)*(FBXx<=35))
FBXheat (FBXr2<=R^2)*((FBXx>=22.325)*(FBXx<=26.865)+
(FBXx>=27.885)*(FBXx<=32.425))
MAINtun ((y48)^2+z^2<=1.75^2)*(x>=35)*(x<=35)
LABtun (((sin15*(x+35)+cos15*(y48))^2+z^2)<=1.75^2)*((cos15*(x+35)sin15*(y
48))>=0)*((cos15*(x+35)sin15*(y48))<=70/cos15)
FBXr2 ((sin80*x+cos80*y)^2+z^2)
FBXx (cos80*xsin80*y)
H P/(row*g)+zp
Hx Px/(row*g)
Hy Py/(row*g)
Hz Pz/(row*g)+1
q1 (k1*Hx+k6*Hy+k5*Hz)
q2 (k6*Hx+k2*Hy+k4*Hz)
q3 (k5*Hx+k4*Hy+k3*Hz)
q q1^2+q2^2+q3^2
E11 ux
172
E22
33 wz
23 0.5*(vz+wy)
+t26*E12
13+t36*E12
s23 4*E23+t45*E13+t46*E12
s13 5*E23+t55*E13+t56*E12
s12 6*E23+t56*E13+t66*E12
on_Mises
33s11*s33+3*s23^2+3*s13^2+3*s12^2)^0.5
erp(4,x,y,z) +(1heter)*2e5
+(fim*(1thetam)+fif*(1thetaf))*ros
KTw+(fim*(1thetam)+fif*(1thetaf))*KTs
if*thetaf)*row*Cw+(fim*(1thetam)+fif*(1thetaf))*ros*Cs
+(1heter)*3e9)*GFactor
) +(1heter)*1e11*nuw0/g)*kiFactor
+(1heter)*1e11*nuw0/g)*kiFactor
,x,y,z) +(1heter)*1e11*nuw0/g)*kiFactor
ter*coeffinterp(423,x,y,z) +(1heter)*0)*kiFactor
erp(413,x,y,z) +(1heter)*0)*kiFactor
(h rp(412,x,y,z) +(1heter)*0)*kiFactor
(h finterp(432,x,y,z) +(1heter)*0)*kiFactor
(heter oeffinterp(431,x,y,z) +(1heter)*0)*kiFactor
(h finterp(421,x,y,z) +(1heter)*0)*kiFactor
(h finterp(811,x,y,z) +(1 0.2)*biFactor
(h finterp(822,x,y,z) +(1 0.2)*biFactor
(h interp(833,x,y,z) +(1 *biFactor
(h interp(823,x,y,z) +(1heter)*0)*biFactor
(h finterp(813,x,y,z) +(1heter)*0)*biFactor
(h coeffinterp(812,x,y,z) +(1heter)*0)*biFactor
(h oeffinterp(1111,x,y,z) +(1heter)*4e9)*TijFactor
(h (1122,x,y,z) +(1heter)*2e9*0)*TijFactor
(h (1133,x,y,z) +(1heter)*2e9*0)*TijFactor
(h 1123,x,y,z) +(1heter)*0)*TijFactor
(h (1113,x,y,z) +(1heter)*0)*TijFactor
(h 1112,x,y,z) +(1heter)*0)*TijFactor
(h oeffinterp(2222,x,y,z) +(1heter)*4e9)*TijFactor
(h (2233,x,y,z) +(1heter)*2e9*0)*TijFactor
(h (2223,x,y,z) +(1heter)*0)*TijFactor
(h oeffinterp(2213,x,y,z) +(1heter)*0)*TijFactor
(h oeffinterp(2212,x,y,z) +(1heter)*0)*TijFactor
3 (heter*coeffinterp(3333,x,y,z) +(1heter)*4e9)*TijFactor
t34 (heter*coeffinterp(3323,x,y,z) +(1heter)*0)*TijFactor
vy
E
E
E13 0.5*(uz+wx)
E12 0.5*(uy+vx)
s11 t11*E11+t12*E22+t13*E33+t14*E23+t15*E13+t16*E12
s22 t12*E11+t22*E22+t23*E33+t24*E23+t25*E13
s33 t13*E11+t23*E22+t33*E33+t34*E23+t35*E
t14*E11+t24*E22+t34*E33+t4
t15*E11+t25*E22+t35*E33+t4
+t4 t16*E11+t26*E22+t36*E33
(s11^2+s22^2+s33^2 V
s11*s22s22*s
fif heter*coeffint
fim 1fif
roeq (fim*thetam+fif*thetaf)*row
thetaeq fim*thetam+fif*thetaf
KTeq (fim*thetam+fif*thetaf)*
roCeq (fim*thetam+f
G (heter*coeffinterp(3,x,y,z)
(heter*coeffinterp(411,x,y, k1 z
(heter*coeffinterp(422,x,y,z) k2
k3 (heter*coeffinterp(433
k4 (he
k5 (heter*coeffint
eter*coeffinte k6
k7 eter*coef
k8 *c
eter*coef k9
b1 eter*coef heter)*
b2
b3
eter*coef
eter*coe
heter)*
ff
eter*coeff
heter)*0.2)
b4
b5 eter*coef
b6 eter*
t11 eter*c
t12 eter*coeffinterp
eter*coeffinterp t13
t14 eter*coeffinterp(
t15 eter*coeffinterp
t16 eter*coeffinterp(
t22 eter*c
t23 eter*coeffinterp
eter*coeffinterp t24
t25 eter*c
t26 eter*c
t3
173
t35 (heter*coeffinterp(3313,x,y,z) +(1heter)*0)*TijFactor
interp(3312,x,y,z) +(1heter)*0)*TijFactor
tor
t45 (heter*coeffinterp(2313,x,y,z) +(1heter)*0)*TijFactor
t46 (heter*coeffinterp(2312,x,y,z) +(1heter)*0)*TijFactor
t55 (heter*coeffinterp(1313,x,y,z) +(1heter)*1e9*0)*TijFactor
t56 (heter*coeffinterp(1312,x,y,z) +(1heter)*0)*TijFactor
t66 (heter*coeffinterp(1212,x,y,z) +(1heter)*1e9*0)*TijFactor
TijFactor (1)
GFactor (1)
kiFactor (10.99*(FBXtest))
biFactor (1)
heter 1
muw muw0+(0.00074.4658e5*T+5.6591e
7*T^23.3185e9*T^3+7.0620e12*T^4)
row row0*(1tempdependentrowmuw*betaTw*(T5))
HboundB1 3*0.7e6/(row*g)+zp
HboundB2 0.7e6/(row*g)+zp
zp z+1700
4.2. Mesh
ded mesh
37945
r of edge e 800
f bound
f elemen
elemen
4.3. Application
n mo
ion mo
pplicati
ty Value
ault element ty uadratic
e extension
k constraints
.2. Variables
endent varia
ependent var
ape functions lag(2,'T')
ior bounda not active
t36 (heter*coeff
t44 (heter*coeffinterp(2323,x,y,z) +(1heter)*1e9*0)*TijFac
KTs KTgranite*(1 0.6*(FBXtest))
betaTs betaTgranite*(1+4*(FBXtest))
4.2.1. Exten
Number of degrees of freedom
4.2.2. Base mesh
Numbe lements
Number o ary elements 2620
Number o ts 11209
Minimum t quality 0.0282
Mode: PDE, Coefficient Form
Applicatio de type: PDE, Coefficient Form
Applicat de name: T (thermal model)
4.3.1. A on Mode Properties
Proper
Def pe Lagrange  Q
Wav Off
Wea Off
4.3
Dep bles: T, T_t2
Ind iables: x, y, z
Sh : shlag(1,'T'), sh
Inter ries
174
4.3.3. Settings
Point Se Edge Settings
Edge 161
term (weak) weak term (weak) 0
k term (dwe
tr term (cons
constr
e functions (
al value (wcin
Boundary Settin
ary 14, 14, 25
Neumann boundary conditionDirichlet boundary condition
term (weak)
rm (dwe 0
term (cons
kconstr
pe functions (
gration order
ial value (wcin
domain Sett
domain
pe functions (s hlag(2,'v') shlag(2,'w')
gration order
straint order
k term (weak)
ak term (dwe
str term (cons Xr2/R^2)))*(FBXheat)
coefficient (c) KTeq
Absorption coefficient (a) 0
Source term (f) 0
Mass coefficient (da) roCeq
Conservative flux convection coeff.
(al)
{{0;0;0}}
Convection coefficient (be) {{'(row*Cw/muw)*(k1*Px+k6*Py+k5*Pz)(row^2)*g*Cw*k5';'
(row*Cw/muw)*(k9*Px+k2*Py+k4*Pz)(row^2)*g*Cw*k4';'
(row*Cw/muw)*(k8*Px+k7*Py+k3*Pz)(row^2)*g*Cw*k3'}}
Conservative flux source term (ga) {{0;0;0}}
weakconstr 1
Subdomain initial value 1, 3
T T0
ttings
Point 140
weak 0
dwea ak) 0 dweak term (dweak) 0
cons tr) '0' constr term (constr) '0'
weak 1 weakconstr 1
Shap wcshape) [] Shape functions (wcshape) []
Initi it) {0;0} Integration order (wcgporder) 2
Initial value (wcinit) {0;0}
gs
Bound 613, 1519
Type
weak 0 0
dweak te ak) 0
constr tr) 0 0
(q) 0 0
(h) 1 1
(g) 0 0
(r) 0 13
wea 1 1
Sha wcshape) [] []
Inte (wcgporder) 2 2
Init it) {0;0} {0;0}
Sub ings
Sub 1, 3
Sha hape) shlag(1,'T') shlag(1,'P') shlag(2,'u') s
Inte (gporder) 2
Con (cporder) 1
wea 0
dwe ak) 0
con tr) (T(13+Ttimefunction*(8765*FB
Diffusion
175
176
4.4. Application Mode: PDE, Coefficient Form
Application mode type: PDE, Coefficient Form
Application mode name: H (hydraulic model)
4.4.1. Application Mode Properties
Property Value
Default element type Lagrange  Quadratic
Wave extension Off
Weak constraints Off
4.4.2. Variables
Dependent variables: P, P_t2
Independent variables: x, y, z
Shape functions: shlag(1,'P'), shlag(2,'P')
Interior boundaries not active
4.4.3. Settings
Point Settings Edge Settings
Point 140 Edge 161
weak term (weak) 0 weak term (weak) 0
dweak term (dweak) 0 dweak term (dweak) 0
constr term (constr) '0' constr term (constr) '0'
weakconstr 1 weakconstr 1
Shape functions (wcshape) [] Shape func
Initial value (wcinit) {0;0} Integration or
Initial value
Boundary Settings
Boundary 1, 25 34
Type Neumann
boundary
condition
Dirichlet bound
condition
weak term (weak) 0 0 0 0 0
dweak term (dweak) 0 0 0 0 0
constr term (constr) 0 0 0 0 0
(q) 0 0 0 0 0
(h) 1 1 1 1 1
(g) 0 0 0 0 0
(r) 0 ((HboundB1*(y100)
+HboundB2*(y+100))
/200zp)*row*g
(HboundB1
zp)*row*g
(HboundB2
zp)*row*g
((HboundB1*(y100)
+HboundB2*(y+100))
/200zp)*row*g*(1
Htimefunction)
weakconstr 1 1 1 1 1
Shape functions
(wcshape)
[] [] [] [] []
Integration order
(wcgporder)
2 2 2 2 2
Initial value (wcinit) {0;0} {0;0} {0;0} {0;0} {0;0}
ary Dirichlet
boundary
condition
Dirichlet
boundary
condition
Dirichlet boundary
condition
tions (wcshape) []
der (wcgporder) 2
(wcinit) {0;0}
2 14 613, 1519
Subdomain Settings
Subdomain 1, 3
Shape functions (shape) shlag(1,'T') shlag(1,'P') shlag(2,'u') shlag(2,'v') shlag(2,'w')
Integration order (gporder) 2
Constraint order (cporder) 1
weak term (weak) 0
dweak term (dweak) u_test*(b1*ux_time+b6*uy_time+b5*uz_time)+
v_test*(b6*vx_time+b2*vy_time+b4*vz_time)+
w_test*(b5*wx_time+b4*wy_time+b3*wz_time)
constr term (constr) 0
Diffusion coefficient (c) {{'k1/muw','k6/muw','k5/muw';
'k9/muw','k2/muw','k4/muw';
'k8/muw','k7/muw','k3/muw'}}
Absorption coefficient (a) 0
Source term (f) (thetaeq*betaTw)*T_time
Mass coefficient (da) 1/G
Conservative flux convection coeff. (al) {{0;0;0}}
Convection coefficient (be) {{0;0;0}}
Conservative flux source term (ga) {{'row*g*k5/muw';'row*g*k4/muw';'row*g*k3/muw'}}
weakconstr 1
Subdomain initial value 1, 3
P ((HboundB1*(y100)+HboundB2*(y+100))/200zp)*row*g
4.5. Application Mode: PDE, Coefficient Form
Application mode type: PDE, Coefficient Form
Application mode name: M (mechanic model)
4.5.1. Application Mode Properties
Property Value
Default element type Lagrange  Quadratic
Wave extension Off
Weak constraints Off
4.5.2. Variables
Dependent variables: u, v, w, u_t2, v_t2, w_t2
Independent variables: x, y, z
Shape functions: shlag(2,'u'), shlag(2,'v'), shlag(2,'w')
Interior boundaries not active
4.5.3. Settings
Point Settings Edge Settings
Point 140 Edge 161
weak term (weak) {0;0;0} weak term (weak) {0;0;0}
dweak term (dweak) {0;0;0} dweak term (dweak) {0;0;0}
constr term (constr) {'0';'0';'0'} constr term (constr) {'0';'0';'0'}
weakconstr 1 weakconstr 1
Shape functions (wcshape) [] Shape functions (wcshape) []
Initial value (wcinit) {0;0;0;0;0;0}Integration order (wcgporder) 2
Initial value (wcinit) {0;0;0;0;0;0}
177
Boundary Settings
Boundary 613, 1519 4 1, 25 2, 14 3
Type Neumann
boundary
condition
Neumann boundary
condition
Dirichlet
boundary
condition
Dirichlet
boundary
condition
Dirichlet
boundary
condition
weak term (weak) {0;0;0} {0;0;0} {0;0;0} {0;0;0} {0;0;0}
dweak term (dweak) {0;0;0} {0;0;0} {0;0;0} {0;0;0} {0;0;0}
constr term (constr) {'0';'0';'0'} {'0';'0';'0'} {'0';'0';'0'} {'0';'0';'0'} {'0';'0';'0'}
(q) {0,0,0;0,0,0;
0,0,0}
{0,0,0;0,0,0;0,0,0} {0,0,0;0,0,0;
0,0,0}
{0,0,0;0,0,0;
0,0,0}
{0,0,0;0,0,0;
0,0,0}
(h) {1,0,0;0,1,0;
0,0,1}
{1,0,0;0,1,0;0,0,1} {1,0,0;0,0,0;
0,0,0}
{0,0,0;0,1,0;
0,0,0}
{1,0,0;0,1,0;
0,0,1}
(g) {0;0;0} {0;0;'(40035)*roeq*g*
Mtimefunction'}
{0;0;0} {0;0;0} {0;0;0}
(r) {0;0;0} {0;0;0} {0;0;0} {0;0;0} {0;0;0}
weakconstr 1 1 1 1 1
Shape functions
(wcshape)
[] [] [] [] []
Integration order
(wcgporder)
2 2 2 2 2
Initial value (wcinit) {0;0;0;0;0;0} {0;0;0;0;0;0} {0;0;0;0;0;0}{0;0;0;0;0;0} {0;0;0;0;0;0}
Subdomain Settings
Subdomain 1, 3
Shape functions (shape) shlag(1,'T') shlag(1,'P') shlag(2,'u') shlag(2,'v') shlag(2,'w')
Integration order (gporder) 2 2 2
Constraint order (cporder) 1 1 1
weak term (weak) {0;0;0}
dweak term (dweak) {0;0;0}
constr term (constr) {'0';'0';'0'}
Diffusion coefficient (c) {{'t11';'t16';'t66';'t15';'t56';'t55'},{'t16','t12','t14';'t66','
t26','t46';'t56','t25','t45'},{'t15','t14','t13';'t56','t46','
t36';'t55','t45','t35'};{'t16','t66','t56';'t12','t26','t25';'
t14','t46','t45'},{'t66';'t26';'t22';'t46';'t24';'t44'},{'t56','
t46','t36';'t25','t24','t23';'t45','t44','t34'};{'t15','t56','
t55';'t14','t46','t45';'t13','t36','t35'},{'t56','t25','t45';'
t46','t24','t44';'t36','t23','t34'},{'t55';'t45';'t44';'t35';'
t34';'t33'}}
Absorption coefficient (a) {0,0,0;0,0,0;0,0,0}
Source term (f) {'(b1*Px+b6*Py+b5*Pz)+(betaTs*(t11+t12+t13)*Tx+betaTs*
(t16+t26+t36)*Ty+betaTs*(t15+t25+t35)*Tz)';'(b6*Px+b2*Py
+b4*Pz)+(betaTs*(t16+t26+t36)*Tx+betaTs*(t12+t22+t23)*T
y+betaTs*(t14+t24+t34)*Tz)';'roeq*g+(b5*Px+b4*Py+b3*Pz)
+(betaTs*(t15+t25+t35)*Tx+betaTs*(t14+t24+t34)*Ty+betaT
s*(t13+t23+t33)*Tz)'}
Mass coefficient (da) {0,0,0;0,0,0;0,0,0}
Conservative flux convection coeff. (al) {{0;0;0},{0;0;0},{0;0;0};{0;0;0},{0;0;0},{0;0;0};{0;0;0},
{0;0;0},{0;0;0}}
Convection coefficient (be) {{0;0;0},{0;0;0},{0;0;0};{0;0;0},{0;0;0},{0;0;0};{0;0;0},
{0;0;0},{0;0;0}}
Conservative flux source term (ga) {{0;0;0};{0;0;0};{0;0;0}}
weakconstr 1
178
Subdomain initial value 1, 3
u 0
v 0
w 0
5. SOLVER SETTINGS
Solve using a script: off
Auto select solver on
Solver Time dependent
Solution form weak
Symmetric off
Adaption off
5.1. Direct (UMFPACK)
Solver type: Linear system solver
Parameter Value
Pivot threshold 0.1
Memory allocation factor 0.7
5.2. Time Stepping
Parameter Value
Times Linspace(0,1e8,100)
Relative tolerance 0.01
Absolute tolerance 0.0010
Times to store in output Tsteps
Time steps taken by solver Free
Manual tuning of step size Off
Initial time step 0.0010
Maximum time step 1.0
Maximum BDF order 5
Singular mass matrix Maybe
Consistent initialization of DAE systems 2
Error estimation strategy 0
Allow complex numbers Off
5.3. Advanced
Parameter Value
Constraint handling method Eliminate
Nullspace function Auto
Assembly block size 5000
Use Hermitian transpose On
Use complex functions with real input Off
Type of scaling Auto
Manual scaling
Row equilibration On
Manual control of reassembly Off
Load constant On
Constraint constant On
179
Mass constant On
Jacobian constant On
Constraint Jacobian constant On
6. POSTPROCESSING
(see Chapter 6 of the thesis)
7. VARIABLES
7.1. Subdomain
Name Description Expression
absTx_T grad(T) sqrt(Tx^2+Ty^2+Tz^2)
absPx_H grad(P) sqrt(Px^2+Py^2+Pz^2)
absux_M grad(u) sqrt(ux^2+uy^2+uz^2)
absvx_M grad(v) sqrt(vx^2+vy^2+vz^2)
abswx_M grad(w) sqrt(wx^2+wy^2+wz^2)
180
APPENDIX XIII: Full article of the reference [22] (Preprint)
This APPENDIX contains a facsimil of the article labelled in the list of
references as [22], which corresponds to the time series analysis performed in the
Chapter 3 of the thesis. The complete reference of this article is again given here:
Caamn, I., Elorza, F.J ., Mangin, A., Martn, P.L. & Rodrguez, R. Wavelets
and Statistical Techniques for Data Analysis in a MockUp HighLevel Waste Storage
Experiment. International Journal on Wavelets, Multiresolution & Image Processing
2(4): pp. 351370. December 2004.
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
APPENDIX XIV: Full article of the reference [3] (Preprint)
This APPENDIX contains a facsimil of the article labelled in the list of
references as [3], which corresponds to the dualcontinuum model described in
APPENDIX VI as an alternative for the hydraulic part of the THM model of the
Chapter 6 of the thesis. The complete reference of this article is again given here:
Ababou, R, Caamn, I., Elorza, F.J . ThermoHydroMechanical Simulation of
a 3D Fractured Porous Rock: Preliminary Study of Coupled MatrixFracture
Hydraulics. Proceedings of the COMSOL Multiphysics Conference 2005, pp. 193198.
Ed. J . M: Petit and J . Daluz. France. November 2005.
203
Abstract We present a problem involving the
modeling of coupled flow and elastic strain in a 3D
fractured porous rock, which requires prior
homogenization (upscaling) of the fractured medium
into an equivalent darcian anisotropic continuum. The
governing equations form a system of PDEs (Partial
Differential Equations) and, depending on the case
being considered, this system may involve two
different types of couplings (in a real system, both
couplings (1) and (2) generally take place):
1) Hydraulic coupling in a single (no exchange)
or in a dual matrixfracture continuum (exchange);
2) ThermoHydroMechanical interactions
between fluid flow, pressure, elastic stress, strain, and
temperature (after Ababou et al. 1994 [1]).
We present here a preliminary model and
simulation results with FEMLAB, for the hydraulic
problem with anisotropic heterogeneous coefficients.
The model is based on data collected at an
instrumented granitic site (FEBEX project) for
studying a hypothetical nuclear waste repository at the
Grimsel Test Site in the Swiss Alps.
Keywords: FEMLAB. Numerics. Porous
fractured media. Coupled PDE systems. Darcys law.
Permeability upscaling. Dual continuum. Fluid
exchange. Biot. Hydromechanics. Poroelasticity.
1 Introduction
This article presents a preliminary study of
fractured rock, including fracture network
reconstruction and numerical flow simulations, as a
first step towards a fully coupled ThermoHydro
Mechanical (THM) analysis of a fractured granite
formation located at the Grimsel Test Site (GTS,
Switzerland), where the FEBEX experiment is
located. FEBEX is an experiment to test the THM
behavior of a crystalline highlevel waste repository.
The aim of the preliminary simulations presented
below is to reproduce the hydraulic behavior of the
fractured medium using either single or dual
continuum approaches to the fractured porous rock.
The macroscale continuum equations and coefficients
are obtained by upscaling from the local Darcy
equation (matrix) and Poiseuilletype equation
(fractures) up to block scale, where each
homogenized block contains ideally many fractures.
But, to obtain the upscaled equations requires
knowledge of the morphology of the 3D fracture
network. The overall procedure can be summarized as
follows.
First, the 3D network is obtained by a statistical
reconstruction method (or inversion method)
based on various fracture statistics and on
detailed observations of fracture traces on tunnel
drifts and boreholes.
Secondly, the domain of interest is partitioned
into subdomains, in which the fractured rock is
represented as a set of singlefractured blocks.
The tensorial upscaled coefficients are computed
at the scale of the subdomains based on
superposition approximations.
Thirdly, the corresponding system of continuum
PDEs are solved numerically for initial
boundary conditions, with a numerical mesh
finer than block scale, using (here) 3D finite
element PDE solvers in FEMLAB[3].
In this preliminary paper, only the hydraulic
upscaling will be applied. A set of 3D numerical
experiments with either single or dual continuum
equations will be presented. For this reason, the
upscaling of hydraulic coefficients is briefly
presented. The hydraulic simulations are performed
using the FEMLAB multiphysics software.
Although the full THM model is not implemented
here, the principles of coupled hydromechanics are
still briefly explained. The THM model yields a
tensorial nonorthotropic (rank 4) PDE system to be
solved with FEMLABmultiphysics.
2 Characterization of the 3D fractured
medium (network)
2.1 Experimental site: geology, fracture data
GEOLOGY, TUNNEL, BOREHOLES.
The GTS is located in the southern part of the
Central Aar Massif, around 400m below the surface.
The rocks in this area are mostly granitic, and are
Rachid ABABOU, Israel CAAMN, Fco. Javier ELORZA
ThermoHydroMechanical simulation of a 3D fractured porous
rock: preliminary study of coupled matrixfracture hydraulics.
204
1
0,8
0,6
0,4
0,2
0
0,2
0,4
0,6
0,8
1
1 0,
8
0,
6
0,
4
0,
2
0 0,2 0,4 0,6 0,8 1
S1+S2
K4
K2+L
S3
K1+K3
S4
Non clas.
3
1
2
4
intruded by lamprophyres and aplites. The FEBEX
experiment is located in the northern part of the GTS
Laboratory tunnel, where a marked water discharge
was encountered. Two exploratory boreholes where
made: FEBEX95001 and 95002.
FRACTURE NETWORK STRUCTURE (DATA)
A general lithologic and structural cartography of
the gallery was developed (Figure 1), where five
different zones were distinguished along the main axis
of the gallery:
Figure 2: Fracture orientation data coming from
boreholes FEBEX95001 and FEBEX95002 : pole
diagram with 5 families.
OPTIMIZATION/RECONSTRUCTION ALGORITHM
An optimization procedure, based on simulated
annealing, was used to adjust fracture size
distribution so as to minimize (via an appropriate
error norm criterion) the discrepancy between the
frequency histograms of observed and simulated
tunnel traces. The criterion also takes into account the
observed vs. simulated discrepancy of the number of
fracture intersections with the exploratory boreholes
and the tunnel. The main features of the fracture
system where imposed in order to preserve the
models geometric and hydraulic consistency given
other sources of information (major geologic
structures, observed tunnel seepages, etc). Here, we
only show a preliminary example of the optimally
reconstructed 3D fracture network. The optimization
algorithm will be detailed in another paper.
Figure 3 shows the reconstructed 3D fracture
network, which has N =18 272 disc fractures. The
3D generation domain consists of a block of
70mx100mx70m (a volume of 490 000 m3) centered
on the FEBEX gallery. The negative X axis points
towards the Geographic North (SouthNorth axis).
Figure 3: Reconstructed fractured network.
X
Y
Z
Figure 1: Map of traces on the wall of the FEBEX
gallery, divided into five different zones [8].
For example, the 2nd tunnel zone (x=14.025.5 m)
has high fracture density: there are some breccified
zones (breccias) with more than 10 fractures/m; the
granite is highly altered in this zone, and water flow is
approximately 70 l/day overall. A pole plot of the
fracture orientations and their partition into 5 sets
(families) was established according to the GTS
geologic studies [6, 9]. The distribution of fractures
along the two exploratory boreholes is available [not
show here for lack of space].
2.3. Statistical network generation : inverse
problem, optimization, reconstruction
A synthetic fractured medium was generated from
the field data, i.e., reconstructed statistically based on
field data. First, the following fracture parameters and
statistics were defined (and if possible pre evaluated)
by using the abovedescribed geologic information:
 Fracture positions. Homogeneous Poisson
process for the (x,y,z) coordinates of fracture centers.
 Fracture orientations. Four different families of
fractures were defined according to both
morphological (stereonet) and genetic criteria.
Uniform distributions for the direction and the plunge
were used. Figure 2 shows the pole diagram.
 Fracture densities. One of the densities used in
the calibration process is the density 21 of tunnel
traces (trace length / intersecting plane area).
 Fracture aperture. Data on fracture aperture are
only qualitative. Distinguishing filled, open and
wet fractures, leads to assigning a priori apertures of
1E8m, 1E5m and 1E2m respectively.
 Fracture size. A power law distribution was used
for fracture diameters or radii. The parameters of this
distribution (R
MIN
, R
MAX
, b exponent) were optimized
so that the synthetic network fits the geologic data.
205
206
3 Hydraulic model for fractured porous
media: upscaling, equations
3.1. Single and dual continuum flow equations
SINGLE MEDIUM : MATRIXFRACTURE CONTINUUM
FLOW EQUATION.
The governing equation for the upscaled single
continuum is of the same form as the classical Darcy
equation for incompressible single phase flow (where
the upscaled permeability must be considered a 2nd
rank tensor, as will be seen):
P
t
P
. K
1
. C
DUAL CONTINUUM MATRIXFRACTURE FLOW SYSTEM.
The governing equations for the upscaled dual
continuum model are a coupled system of two
equations (generalizing the dual porosity model
initially developed by [2]):
( )
( )
f m f ff
f
f
f m m mm
m
m
P P P
t
P
P P P
t
P
. K
. K
1
. C
1
. C
f f
m m
where :
P
m
is pressure in the matrix medium [Pa]
P
f
is pressure in the fractures medium [Pa]
K
mm
=K
M
is the equivalent upscaled permeability
tensor of the matrix medium [m2]
K
ff
=K
F
is the equivalent upscaled permeability
tensor of the fractures medium [m2]
is the dynamic viscosity of the fluid [Pa.s] ( is
the ratio of a stress / grad V)
is the exchange coefficient between matrix
and fractures [Dimensionless]
m
is the volumetric fraction of the matrix
medium per unit volume of space [m3/m3]
f
is the volumetric fraction of the fractures
medium per unit volume of space [m3/m3]
m
is the intrinsic porosity of the matrix [m3/m3]
f
is the intrinsic porosity of the fractures
medium [m3/m3]
C
m
is the specific elastic storage coefficient of the
matrix continuum [Pa1]
C
f
is the specific elastic storage coefficient of the
fractures continuum [Pa1]
Note that we have distinguished here volume
fractions from porosities, and we have introduced a
fracture porosity
f
, even though
f
=1 usually for a
clean rock joint. However we also consider the case of
depositfilled fractures, faults, disturbed zones, and
other geologic bodies such as lamprophyres: in all
such cases we may specify
f
<1.
The pressures P [Pa] in the matrix and fractures
are phaseaveraged over the upscaled unit volume or
block. The intrinsic permeabilities K [m2] are the
equivalent upscaled permeabilities for each
continuum, defined over the same scales as pressure.
They are taken here to be positive second rank tensors
(as will be seen).
The coefficients C [Pa
1
] are specific storage
coefficients, or capacities. They express the
capacity to store or drain a m
3
of fluid per m
3
of
medium due to a unit variation of pressure. Since the
medium is saturated (single phase flow) this capacity
is due solely to compressibility effects. Thus, it is
assumed that the fluid and the two continuous media
(matrix, fractures) all react like elastic isotropic
continua to changes of pressure. It is possible to
rewrite the above equations using a new set of elastic
capacities c [s]:
m m
C
m m
c = ;
f f
C
f f
c =
where lower case capacities c are in units of time
[s], while upper case capacities C are in inverse
pressure units [Pa
1
]. The system can now be
expressed in terms of (c
m
,c
f
) as follows:
( ) ( )
( ) ( )
+ =
f m f ff
f
f
f m m mm
m
m
P P P
t
P
c
P P P
t
P
c
. K
. K
.
.
PURE EXCHANGE DUAL CONTINUUM EQUATIONS.
To obtain a pure exchange process with negligible
pressure gradient terms, let us assume that the matrix
is only weakly conductive compared to the fracture
network, that is: K
M
<<K
F
or, perhaps more
accurately:
m
K
M
<<
f
K
F
. This allows us to neglect
the pressure gradient terms ("transport" terms) in the
matrix flow equation (Pm/t) above. In addition, we
assume that, since the fracture medium permeability
is comparatively high, the pressure gradient in the
"fracture continuum" is negligible in each upscaling
domain. This allows us to neglect the pressure
gradient terms in the fractures medium (equation
Pf/t). If we also neglect, for simplicity, the spatial
fluctuations of (, , C) in each medium, we can
show that the pressure difference is governed by a
purely kinetic equation (pure exchange) :
P
t
P
~
C
~
where
2
~ f m
P P
P
=
.
This equation was used by Kfoury, Ababou et al
(2004) to analyze the second upscaling problem,
where the local (at the scale of unit cell) is
randomly heterogeneous, and the upscaling of is
sought on a larger scale comprising many unit cells.
3.2. Single medium hydraulic upscaling
In this section, we develop the macroscale Darcy
flux equations at the scale of a homogenization sub
domain, possibly containing many fractures or
fracture blocks (as explained in the following). The
upscaling approach being used will only be outlined
here. It is essentially based on a flux superposition
principle similar to that used in the thermohydro
mechanical upscaling equations developed in [1] and
[9] for the case of a fractured rock with impervious
porous matrix. The fact that matrix permeability is
no longer neglected here, constitutes a generalization
of the abovecited works.
The region of interest is partitioned into sub
domains, in which the flow equation must be
homogenized. Most importantly, in each subdomain,
the 3D medium is idealized as an assembly of blocks,
each block being constituted of the porous matrix
traversed by a single planar fracture. Let us now
express some intermediate steps and give the final
result.
First, we express as follows the microscale
tensorial conductivities characterizing the local head
loss law within each medium (single fracture and its
surrounding matrix) composing a block:
Isotropic porous matrix (M):
Anisotropic fracture, or fault, or coarse medium (F):
where:
is conductivity parallel to fracture,
e.g. Poiseuille law (for a real fracture)
is conductivity transverse to
fracture, e.g. quasiinfinite (real fracture)
I K
M
M
M
M
M
K
K
K
K
=
=
0 0
0 0
0 0
F
F
F
F
K
K
K
0 0
0 0
0 0


K

F
K
F
K
Secondly, the exact equivalent conductivity for the
singlefracture block of Figure 4 is calculated,
according to the low order upscaling approach
evoked earlier. The result is:
( ) ( )
H j i A j i ij ij
K n n K n n + =
where K
A
is the arithmetic mean of (K
F
, K
M
), and K
H
is their harmonic mean. Both are weighted by the
corresponding volumetric fractions of F and M at
block scale [ and (1) respectively]. Also, ni is the
ithcomponent of the unit vector normal to fracture.
Finally, we define the geometry of individual
fracture blocks. One approach is to let each block be a
prismatic volume having same crosssection as the
corresponding planar fracture. Thus, if the fracture is
a disc, the block is a circular cylinder; if the fracture is
rectangular, the block is a parallelepipedic box. The
heights (b
k
) of the blocks (k=1,2) can be chosen
constant (b
k
=b) within the homogenization domain,
such that the cumulated volume of all blocks (V
k
)
equals the volume of the domain (V ):
b
S
V
b
F
N
m
m
k
= =
1
b/2
b/2
a
x
z
l
l
I
F
A
C
B
z
H
J
z
=
x
H
J
x
=
C
B
A
Figure 4: Basic building block of a fractured porous
medium, showing Boundary Conditions with
imposed piecewise linear hydraulic head H (total
pressure P).
Finally, a flux superposition approach is applied
over all the individual blocks. Several alternative
approaches are considered, including: (i) volume
weighted average of blockscale Darcy flux density
vector q [m/s]; and (ii) direct sum of blockscale
fluxes Q [m3/s]. The latter approach just adds up the
individual block contributions to the global flux:
=
k
k
Q Q
which requires a prior estimate of the individual
block flux Q
k
. But only the flux density q
k
is known,
since the earlier blockupscaling step yields a block
scale Darcy equation in terms of flux density.
Therefore, to obtain the block flux Q
k
requires a prior
estimate of the effective crosssectional flux area of
the block. Dividing the flux Q
k
by the total cross
sectional area (A
T
) of the homogenization domain,
yields finally Darcys law in terms of Darcy velocity
(flux density). This yields finally the equivalent
conductivity tensor :
( )
( ) ( )
( )
( )
( )
( )
( )
+ +
=
k
F
k
M
k k
j
k
i ij
T
k
ij
K K n n K ... 1

( ) ( )
( )
( )
( )
F
k
M
k
k
j
k
i
K
K
n n
1
1
This K
ij
tensor gives the mean response of the
fractured porous medium to a given global gradient.
3.3. Dual medium model : exchange
coefficient (and permeabilities)
The upscaling approach developed for a single
matrix/fracture medium, can also be used for
upscaling, separately, the matrix and fracture system
permeabilities in the dual medium equations:
UPSCALED PERMEABILITY OF THE 1RST MEDIUM THE
POROUS MATRIX. Assuming the porous matrix
composing the intact rock to be homogeneous and
isotropic, the upscaled matrix permeability tensor is
simply the local scalar permeability, that is:
=
=
V
S
S
V
F
N
m
m
k
k
1
.
M M
K K =
207
UPSCALED PERMEABILITY OF THE 2ND MEDIUM THE
FRACTURE SYSTEM. To upscale separately the fracture
network, constituting the 2nd medium, we may reuse
the previous matrixfracture upscaling, simply by
inserting K
M
=0 there. The result is a fracture
permeability tensor of the form:
( )
( )
( ) ( )
( )
( )
[ ]
=
k
F
k k
j
k
i ij
T
k
ij F
K n n K

UPSCALED MATRIX/FRACTURE EXCHANGE THE
SCALAR EXCHANGE COEFFICIENT . The exchange
coefficient needs to be evaluated at subdomain scale
(homogenization scale). The upscaled proposed in
this paper is an extension from the original Warren
Root formula [10]. More recent studies [5] confirm, to
some degree, the WarrenRoot intuition, at least for a
regular or periodic array of fractured blocks.
Following in part [5], we propose to evaluate the
dimensionless coefficient at subdomain scale as:
( )
2
2
K
=
;
( )
M F M
K K K K
+ =
1
1 1
2
1
and the characteristic length scale is the effective
interspacing distance between fracture planes. We
evaluate it from the specific area density 32 defined
in the FRACMAN code [4], following also [1]:
32
1
.
4 Implementation of coupled hydraulics:
the FEMLAB model
The domain of investigation is a fractured rock
block of 70mx100mx70m that contains the FEBEX
tunnel. The Xaxis points to the North, and the
FEBEX tunnel has an orientation of N270E. The
origin of the domain is located at the center of the
fractured block, and coincides with the center of the
tunnel. A regional flow parallel to +X is considered.
Three measurement points, or probes, are defined
for output purposes: (i) the lamprophyre dyke sample
point; (ii) the tunnel sample line (right wall of tunnel);
(iii) the flow sample line (along Xaxis). Figure 5 and
Table 1 show the main features of the domain and the
mesh used in FEMLABs 3D finite element model.
Figure 5: The 70mx100mx70m region of interest
showing the models measuring points (probes).
#of
elements
#of
nodes
#of boundary
elements
#of edge
elements
8315 24798 1294 110
Table 1: The 3D Finite Element Mesh.
5 Flow simulation results and discussion
Several flow experiments have been modeled.
Each of them focuses on a different aspect of the
hydraulic model. The experiments are divided in:
Pure exchange experiments: The initial
boundary conditions are such that there is no net
flow at the scale of the domain. Only matrix
fracture exchange takes place, controlled by
coefficient estimated from the synthetic
fractured network (see final equation of section
3.3). Two different pure exchange experiments
were carried out: (I.1) homogeneous ; (I.2)
heterogeneous on a 3x3x3 partition.
Flowthrough experiments: In this set of
experiments, the boundary conditions are such
that there exists a mean flux, and a non zero
pressure gradient, across the entire domain
(flowthrough experiments). Several numerical
experiments were conducted, some coinciding
with the full dual medium model, others reducing
to the single medium without exchange (all with
partitions of 3x3x3): .
(II.1) Single medium / no exchange;
(II.2.1) Dual medium / low flow regime;
(II.2.2) Dual medium, negligible exchange;
(II.2.3) Dual medium with both flow+exchange.
Figure 6a. shows the spatial distribution of the
estimated for the fractured medium. It is
represented as spheres of radius proportional to ,
one for each of the subdomains in the 3x3x3 partition.
Figure 6b. shows ellipsoids of K
eq
for each partition.
This value of K
eq
has been computed for a matrix
conductivity K
M
=0, and is the one used for the dual
medium model, whereas in the single medium mode
we compute K
eq
using a value for K
M
=1e12m
2
. In all
the experiments, the final simulation time was
sufficiently large to reach steady state. In most cases,
this final time was around T=1E+11 s 3000years. In
this paper we only show the results of the experiment
(II.2.1).
a. Exchange coefficient
.
b. Equivalent hydraulic
conductivity K
eq
.
Figure 6: 3D distribution of the model parameters
represented in the centers of the partition blocks.
Right wall
sampleline
Flow direction
sampleline
Lamprophyre
samplepoint
208
209
EXPERIMENT II.2.1: DUALMEDIUM HYDRAULICS WITH
LOW FLOW REGIME
In this experiment the pressure P
f
(t) in the fracture
medium decreases quickly during the early stages of
the experiment. However, steady state flow is reached
later than in the single medium experiment. This is
because the upscaled permeability of the single
medium combines matrix and fracture, whereas it
only considers the fracture medium permeability in
this experiment.
The evolution can be divided in two stages: in the
first stage (early times) we get similar results than
those found for the pure exchange experiment (I.2)
(Figure 7), and in the second stage (late times) we
obtain the same behavior as that observed for the
single medium flowthrough experiment (II.1)
(Figure 8).
In this case, after the exchange equilibrium is
reached, both fracture and matrix media behave
exactly the same in the flow transitory. This means
that, in the latest stages, this model behaves
completely equivalent to the single medium model.
Figure 7: Distribution of P
f
at time T=1e+6s
(12 days) at measuring points (probes), for the
dual medium flowthrough experiment with low
flow regime.
Figure 8: Distribution of P
f
at time T=1e+11s
(3000 years) at measuring points (probes) for the
dual medium flowthrough experiment with low
flow regime.
6. Conclusions and future work
We presented here a preliminary model and
simulation results with FEMLAB, for the hydraulic
problem with anisotropic heterogeneous coefficients.
The model is based on data collected at an
instrumented granitic site (FEBEX project) for
studying a hypothetical nuclear waste repository at
the Grimsel Test Site in the Swiss Alps. This
approach allows us to quantify the influence of a 3D
fracture network on the hydraulics of fractured rocks.
In this paper, thermohydromechanical (THM)
coupling is not implemented. However, the aim of
this work in the future is to fully implement hydro
mechanical coupling (and thermal stresses as well).
This will be done, in a first step, by considering the
fractured rock as a single equivalent poroelastic
continuum. The reader is referred to the THM
upscaling theory developed in [1] and also reported in
[9], for an impervious matrix. This theory will be
extended to the case of a permeable porous matrix
(Km0).
7. Acknowledgements
This work has been partially supported by the
FEBEX project under contract number FIKWCT
20000016 with the European Commission and
ENRESA.
References
1. Ababou R., A. Millard, E. Treille, M. Durin, and F.
Plas : Continuum Modeling of Coupled Thermo
HydroMechanical Processes in Fractured Rock.
Comput. Methods in Water Resources, Kluwer
Academic Publishers, A. Peters et al. eds. Vol.1, Ch.6,
pp.651658 (1994).
2. Barenblatt G. , Zelthov I., Kochina I. Basic concepts
in the theory of seepage of homogeneous liquids in
fissured rocks. J . Appl. Math. 24 (1960), 213240.
3. COMSOL, 2004: FEMLAB Users Guide 3.1.
COMSOL AB. October 2004. 581 pp.
4. Dershowtiz W. et al. (1992): FRACMAN v.2.3 :
Interactive discrete feature data analysis, geometric
modelling, and exploration simulation User Doc.
Redmond WA: Golder Assoc. 1992.
5. Dykhuisen, R.C. (1990). A new coupling term for
dualporosity models. Water Resour. Res., 1990,
26(2):351356.
6. Keusen, H.R., Ganguin, J ., Schuler, P., Buletti, M.
(1989): Grimsel Test Site: Geology. NAGRA Tech.
Report. NTR 8714E. (Ref [6])
7. Kfoury M., R. Ababou, B. Noetinger and M. Quintard,
2004. Matrixfracture exchange in fractured porous
media: stochastic upscaling. ComptesRendus
Acadmie Sciences (Paris). C.R. Mecanique, 2004,
Vol.332, pp.679686.
8. Pardillo J ., Campos R., Guimer J . (1997):
Caracterizacin geolgica de la zona de ensayo
FEBEX(GrimselSuiza). CIEMATIMAM201.
9. Stietel A., Millard A., Treille E., Vuillod E., Thoraval
A., Ababou R.: Continuum Representation of Coupled
HydroMechanical Processes of Fractured Media:
Homogenisation and Parameter Identification. In :
Devts. Geotech. Engg.: Coupled ThermoHydro
Mecha. Processes (DECOVALEX). O.Stephansson,
L.J ing, CF.Tsang eds. 79:135164, Elsevier (1996).
10. Warren J.E. and P.J . Root (1963). The behavior
of naturally fractured reservoirs. Soc. Petrol.
Eng. J ., 3(5), 245255, 1963.
210
APPENDIX XV: Full article of the reference [23] (Preprint)
This APPENDIX contains a facsimil of the article labelled in the list of
references as [23], which corresponds to the simulation of the threedimensional
fractured network described in Chapter 4 of the thesis. The complete reference of this
article is again given here:
Caamn, I., Elorza, F. J ., Ababou, R. 3D Fracture Networks: Optimal
Identification and Reconstruction. Accepted for publication in the proceedings of the
XIth IAMG Annual Congress. Belgium. September 2006.
211
International Association of Mathematical Geology
XI
th
International Congress,
Universit de Lige  Belgium
3D Fracture Networks: Optimal Identification and
Reconstruction.
I. Caamn
1
, F. J. Elorza
1
, R. Ababou
2
1 Departamento de Matemtica Aplicada y Mtodos Informticos, E.T.S.I. Minas,
Universidad Politcnica de Madrid, Spain
2 Institut de Mcanique des Fluides de Toulouse, Institut National Polytechnique de
Toulouse, France
Corresponding author: icanamon@dmami.upm.es
ABSTRACT : In this paper we generate a 3D fractured medium that reproduces the nonuniform map of
fracture traces left on the wall of a cylindrical gallery. A specific algorithm calculating the analytic intersection
of the fractures with the gallery has been developed, and appropriate statistical measures to compare both real
and simulated media have been defined. In addition, other geological data such as orientations and fracture
densities have been used, and size distribution parameters have been determined in the optimization procedure to
best fit the tracemap in a real case study. Satisfactory concordance between measured and generated fractured
media maps of traces is reached.
KEYWORDS : 3D fractured medium, Montecarlo, cylindrical wall tracemap, inverse problem,
reconstruction, nonuniform tracemap.
1. Introduction
Works on fractured media reconstruction generally take trace data from flat or pseudoflat
walls coming from the vertical parts of arc galleries (Gillespie 1993, La Pointe 2002), but
there are a lot of cylindrical galleries whose tracemaps cannot be used by considering there
walls as flat. We could approximate the cylinders as a set of elongated planes parallel to the
axis of the gallery, but then long traces get cut through the several planes, so that tracelength
is underestimated and reconstruction cannot be realistic in 3D.
In this paper we describe a methodology to use the traces left on cylindrical galleries without
plane approximations. To do that, analytical solution of the intersection in 3D of a disk
shaped fracture with a cylinder has been obtained. Then, some statistics have been defined to
characterize the tracemap to be reproduced. Finally, a Montecarlo optimization procedure
based on a variation of Simulated Annealing (Metropolis 1953, Goffe 1994) has been used to
find the parameters of the fracture size distribution that best fits the tracemap. Additionally, a
specific algorithm to reproduce the nonhomogeneous trace density along the tracemap has
been defined.
The granitic medium reconstructed in this paper is located at the Grimsel Test Site
(Switzerland) in the southern part of the Central Aar Massif, around 400m below the surface.
The cylindrical gallery from which fracture trace data have been collected corresponds to the
FEBEX experiment, and has been excavated in the northern part of the Laboratory tunnel of
the GTS.
212
International Association of Mathematical Geology
XI
th
International Congress,
Universit de Lige  Belgium
2. Fractured network data
Different discontinuity systems have been described within the Central Aar Masif: ductile
systems (S1, S2 and S3), brittle systems (S4/K4, K2/L=lamprophyre direction, K1, K3 and
S5) and tensile systems (ZK1 and ZK2). For a detailed description of the fracture systems
present in the GTS see (Steck 1968, Keusen et al. 1989). According to this classification,
genetically and morphologically similar fractured systems have been grouped and four
fracture families have been identified. The statistical orientation distributions of these families
have been obtained with the data coming from two exploratory boreholes near the FEBEX
gallery (Pardillo et al. 1997) (see Figure 1).
1
0,8
0,6
0,4
0,2
0
0,2
0,4
0,6
0,8
1
1 0,
8
0,
6
0,
4
0,
2
0 0,2 0,4 0,6 0,8 1
S1+S2
K4
K2+L
S3
K1+K3
S4
Non clas.
3
1
2
4
1
2
Fig. 1. Pole diagram of the fractures intersecting boreholes FEBEX95001 and FEBEX95002 and
regrouping into four fracture families.
The general cartography of fracture traces in the FEBEX gallery (Pardillo et al 1997) is
presented in Figure 2, where the most important lithologic and structural features can be
distinguished. Five different zones with different structural characteristics and trace density
can be distinguished along the main axis of the gallery (see Figure 2).
Tracemap of the FEBEX gallery has been digitalized and two types of measures have been
obtained for identification purposes:
 Linear density
21
: defined as the ratio between total trace length and the wall surface. It has
been calculated for the five different zones.
 Histograms: of the trace length and the 3D trace chord in the whole gallery, the last obtained
by rebuilding the cylindrical gallery from the tracemaps and getting the 3D coordinates of the
traces.
Fig. 2. Development of the cylindrical tracemap in the FEBEX gallery with the five different zones of
fracture intersection density. Zone 4 includes a lamprophyre dyke highly transmissive.
Zone 1 Zone 2 Zone 3 Zone 4 Zone 5
213
International Association of Mathematical Geology
XI
th
International Congress,
Universit de Lige  Belgium
214
3. Methodology for the reconstruction of the fractured medium
The exact analytical solution of the intersection of a disk fracture with a cylindrical tunnel
comes out from solving the following system of equations:
2 2
,
2 2 2
t
t
t
t t t
L
x
L
R z y = + (1)
(2)
where the first equation represents the cylinder and the second one the fracture disks, both of
them in their respective local coordinated systems. R
t
is the tunnel radius, L
t
is the tunnel
length, R
i
is the ith fracture disk radius, and N
disks
is the number of fracture disks. The
solution of this system of equations yields to the analytical equation defining the trace. We
obtain the extremes of the trace by solving the second equation as equality.
From all the fracture traces appearing in the FEBEX gallery tracemap, seventeen features
have a complete or nearcomplete trace. These fractures are big, and some of them are
opened, and therefore relevant for creating preferential pathways for the fluid flux through the
rock. It is important to take into account this features for reconstruction purposes, in order to
preserve some geometric and hydraulic consistency in the model. According to previous
geological studies, these features are large regional fractures, so a radius R=100, sufficiently
large for the fractures to traverse the whole domain, has been assigned to them. To obtain
their direction and plunge from the fracture traces, the following geometric relations have
been used:
disks i i i i
N i z R z y ,..., 2 , 1 0 ,
2 2 2
= = +
+ =
m
D
arctan
2
(3)
( )
sin
arctan arctan
l
D
L
D
(4)
where D is the diameter of the FEBEX gallery, m is the arc distance between the intersections
of the trace with a horizontal plane passing through the gallery axis, and l is the arc distance
between the intersections of the trace with a vertical plane passing through the gallery axis.
A homogeneous Poisson process is used to define the coordinates ( )
fc fc fc
z y x , ,
ly distributed random
wever, the nonuniform pattern
roduced, so
. Ba
of each
fracture center. Accordingly, these three coordinates are uniform
variables within the bounds of the rectangular box domain. Ho
of the traces in the gallery wall showed in Figure 2 has also been rep that the
distribution of centres in the proximities of the gallery must also be nonuniform sically,
fracture density
21
is recalculated during the generation of fractures, until it reaches the
experimental density on each zone of the FEBEX gallery. Each fracture lying in a zone where
the maximum
21
has already been reached will then be moved towards another different
zone.
A Montecarlo algorithm has been implemented to reconstruct the synthetic fractured medium.
An optimization procedure based on simulated annealing has been used to adjust fracture size
distribution so as to minimize the discrepancy between the statistical characteristics of the
International Association of Mathematical Geology
XI
th
International Congress,
Universit de Lige  Belgium
synthetic fractured medium and those of the real fractured medium, according to the geologic
data available. In this work, the power law or Pareto distribution has been used to
characterize the size of the fracture network, and only the R
min
and b parameters have been
optimized (R
max
has been set to 100m).
4. Results and discussion
The optimization procedure yields to the determination of the parameters of the power law
size distribution that best fit the geologic data: R
min
=0.1985m, R
max
=100m (fixed), and
b=3,3048. Table 1 lists the main characteristics of the fractured network in terms of number
of fractures and intersections with the different measured objects and with the corresponding
simulated objects. Figure 3 shows the tracemap of the FEBEX gallery resulting from the
optimized fractured medium, and Figures 4 and 5 present two 3D views of the fractured
medium within the generation domain, which consists of a block of 70x200x70m
3
centered in
the FEBEX gallery. The simulated 3D fracture network has N =2906474 disc fractures.
Fracture density
21
#
traces
#
intersect.
FBX01
#
intersect.
FBX02
zone1 zone2 zone3 zone4 zone5 global
Measured 614 155 410 1.493 1.154 2.096 0.796 2.083 3.077
Generated 800 144 234 1.498 1.201 2.012 0.824 2.084 3.084
Tab. 1. Number of intersections with the gallery and the exploratory boreholes and fracture density
in the five gallery zones of the real and simulated fractured media.
Fig. 3. Development of the cylindrical tracemap obtained as intersection of the 3D simulated fractured
medium with the FEBEX gallery.
Figure 4: Disks of the simulated fractured
medium.
Figure 5: Simulated fractured medium inside
the domain.
A good agreement has been obtained between the observed features, i.e. number of traces in
tunnel and boreholes, fracture density and tracemap, and the ones produced by the simulated
fractured medium. The nonuniformity of the tracemap has been also coherently reproduced.
N
x
y
z
y
z
N
215
International Association of Mathematical Geology
XI
th
International Congress,
Universit de Lige  Belgium
216
Although satisfactory results have been obtained, here are some comments for future work:
 The nonhomogeneous fractured network simulated near the gallery could be assumed
to be the same in the overall domain, so the statistical functions producing the
fractured medium would be extended everywhere. To do that, nonhomogeneous
Poisson processes should be used (Stoyan et al. 1987), and some kind of density
measure must be defined for every point in the 3D space. The reduced second moment
function (Hanisch & Stoyan 1983) of the tracemap could be used for this purpose.
 The objective function has a stochastic nature, because it depends on the generation of
a fractured medium obtained with statistical distributions. This means that, for the
same set of parameters R
min
, R
max
, and b, we can obtain different values of the
objective function for each realization. Therefore, an average of the objective
functions of several realizations should be used to obtain a more reliable value given a
set of parameters. The number of realizations to use in that average has to be
determined according to the confidence interval needed in the objective function.
 For a hydrological validation of the fractured medium, hydraulic and transport tests
available in the site should be used, with hydraulically conditioned fractured networks.
5. Conclusions
We have developed a methodology to simulate a 3D fractured network that fits optimally a
cylindrical tunnel tracemap. The optimization procedure searches for the best parameters of
the size distribution to minimize the discrepancies between measured and simulated trace
length and 3D trace chord histograms and number of intersections with the tunnel and two
exploratory boreholes. A good agreement between both fractured mediums has been found in
the results. This methodology provides a good starting point for the use of cylindrical
tracemaps to simulate geological 3D fractured networks, and completes the classical use of
flat wall tracemaps that is more extensively developed in the literature.
Acknowledgments: The authors of this article would like to thank the FEBEX II project (EC contract
FIKWCT200000016) to have provided them the opportunity to accomplish this study and to have
provided with the experimental data used on it.
REFERENCES
Gillespie, P.A., Howard, C.B., Walsh, J .J ., & Watterson, J ., 1993. Measurement and characterization
of spatial distributions of fractures. Tectonophysics (226): 112141.
Goffe, Ferrier and Rogers, 1994. Global Optimization of Statistical Functions with Simulated
Annealing. Journal of Econometrics (60, 1/2): 65100.
Hanisch, K.H., Stoyan, D., 1983. Remarks on statistical inference and prediction for a hardcore
clustering model. Math. Operationsf. Statist., ser. statist. (14): 559567.
Keusen, H.R., Ganguin, J ., Schuler, P., Buletti, M., 1989. Grimsel Test Site: Geology, Nagra Tech.
Report, NTR 8714E.
La Pointe, P.R., 2002. Derivation of parent fracture population statistics from trace length
measurements of fractal fracture populations. Int. Jour. Of Rock Mech. & Mining Sci. (39): 381388.
Metropolis, N., Rosenbluth, A., Rosenbluth, M., Teller A., and Teller, E. 1953. Equation of state
calculations by fast computing machines. Jour. of Chem. Phys. (21): 10871092.
Pardillo, J ., Campos, R., Guimer, J . , 1997. Caracterizacin geolgica de la zona de ensayo FEBEX
(GrimselSuiza). CIEMATIMAM201.
Steck, A., 1968. Die alpidischen Strukturen in den zentralen AareGraniten des westlichen
Aarmassivs. Eclogae geol. Helv. (61): 1948.
Stoyan, D., Kendall, W.S., Mecke, J ., 1987. Stochastic Geometry and its Applications. J .Wiley&Sons.
217
218