eae 52/09 EARTHQUAKE ENGINEERING RESEARCH CENTER
FLUIDSTRUCTURE INTERACTIONS:
ADDED MASS COMPUTATIONS
FOR INCOMPRESSIBLE FLUID
by
JAMES SHAWHAN KUO.
‘COLLEGE OF ENGINEERING.
 UNIVERSITY OF CALIFORNIA « Berkeley, CaliforniaFLUIDSTRUCTURE INTERACTIONS:
ADDED MASS COMPUTATIONS FOR INCOMPRESSIBLE FLUID
by
James ShawHan Kuo
Report to the National Science Foundation
Report No. UCB/EERCB2/09
Earthquake Engineering Research Center
College of Engineering
university of California
Berkeley, California
August 1982ABSTRACT
This report consists of Part I of the dissertation submitted by
‘the author to the Graduate Division of the University of California,
Berkeley, in partial satisfaction of the requirenents for the dagree
of Doctor af Philosophy in Engineering.
In this report, the damreservoir interaction effects considering
incompressible fluid are presented. The hydrodynanic effect represented by’
an addednass matrix 1s evelusted by two basically different. procedures~
2 Generalized Nestergaard Formila and the Galerkin Finite Elenent Method.
Pressure solutions acting on gravity dane, cylindrical arch dams and
general arch dams are compared for the different procedures. Rigorous
wmode shape and frequency correlations are carried out, and based on the
results of the correlation studies most efficient procedure 4s suggested,
which 1s shown to be adequate for engineering purposes.a
ACKNOMLEDGENENTS
ony people have assisted ne during the course of my graduate studies.
1 am most thankful to Professor Ray H. Clough, my thesis advisor and the
Chairman of my dissertation comntttee.His inspiration, patience ané guidance
are deeply appreciated, and I will always renenber his fine example.
T am also grateful to Professors E. L. iiTson and C.D. Hote, menbers
of ay dissertation comittee, for their valuable suggestions, encosragenent
and frfencly guidance.
T wish to express ny gratitude to Dr. John F. Hall for many vélvable
discussions during the first part of my research, and to Professor
R. L. Taylor for making his computer progran FEAP available.
This research has been part of the U.S.R.0.C. Cooperative Research
Progran. Financial support fron the National Science Foundation ts
gratefully acknowledged. Conputer faciTities were provided by the Lavrence
Berkeley Laboratory, University of California, Berkeley. I also wart to
thank Toni Avery for her excellent typing.
1 am greatly indebted to Ors. BI11 & Lydia Yap and Don & Nancy Mangold,
‘tio Christian couples, who provided care concern and encouragenent
throughout my graduate studies.
Above all, T wish to express ny gratitude to py family for their
unfailing support, understanding,encouragenent and love.TABLE OF CONTENTS
Page
MSTA. ee ee eee 4
ACKNORLEDOEHENTS. ws... eee 8
TABLE OF cONTENTS . ee
UST OF FIOWRES foe ee eee wt
Ve RODUETION eee eee aa
VT Ofectives eee veces 1
1.2 State of the Wt eee 2
VB Sep eee eee 4
GENERALIZED MESTERGAARD FORWWLA eee 8
2.1 Review of the Classical Westergaard Formula ne 5 
2.2 Generalized Westergeard formla, os. seeee 6
2.3 Teibutery Area Lanping Process « 8 :
3. GALEREIN FIOITE ELEMENT HETIOD. ee sok
S.1 Galerkin Method for Nave Equation... . « fice
4.2 Finite Clenent Formulations. ©. 0.22... 8
3.4 Consistent Lumping Process ©... eee 9
COMPUTER THPLEMENTATIONS AND RUMERICAL SOLUTIONS... 2... 2
4.1 Computer toplenentations ©... vee ee eee 2
4.1.1 Generalized Westergaard Formula Procedure... 22
4.1.2 Galerkin Finite Element Procedures... 8
4.1.2.1 Formation of matrix gg... cee ee 2h
4.1.2.2 Formation of matrix hy... ee eee 26
4.2 tydrogynante Pressure Solutions and Their Comparisons... 27
4.201 Gravity Dane. eee 28
4.2.2 Cytinarieat hreh Dams eeRese eee areas ee an RE
Se aaa sf
5.2 Correlation of Frequencies and Mode Shapes... ... 2. 34 
a ee ce 
re 5
ee ee as
Beg nC oorraTnecunencee ‘
Base Earthquake (DBE) Excitation. . . a
pe a ees er
ee ene E
APPENDIX B: Finite Element Interpolation Functions and Their f
eee ee :
ee aca ee aes
acceler ea eee ee a
er ypauseaeacue cornneeeenne iFigure
i
12
13
LIST oF Froures
Pictorial addednass according to Westergaard. ... .
Galerkin discretization of the reservoir... 2...
Lunping hydrodynamic pressures into eq
nodal forces .
ivalent hydrodynamic
Normal direction cosines of curvitinear surface... . .
Finite elements: 2D and 30 in natural coordinates. . .
Gravity dam with vertical upstream face... 2.2...
Arch dam with cylindrical upstream face... 22... oe
Arch dan with cylindrical upstream face and different
abutment angles
Arch dam with general geometry upstream face... .
Foundation rock and dam body of Techi dam model
(x2projection)
Foundation rock and dam body of Techi dam mode]
{yeprojection at crown section cut)... ee ee
Finite elenent mesh of dam bo
face projected on xzplane . .
Finite element mesh of dam boc
face projected on xzplane . .
dy (Techi dam) upstream
dy (Techi dam) downstream
Effect of inconpressible reservoir on the fundanental
Frequency of vibration of the danreservoir systems... .
Comparison of effects of incompressible reservoir on the
fundamental frequency of vibration of Techi damtreservoir
system.
Frequencies of first four modes of Techi dam according
to different reservoir model for various water levels. . .
Frequency correlations of experiments with nunerical
analyses...
Frequency spectra correlations of various water levels
(echt dam).
55
56
57
58
59
a
63
6
Cl
cy
1
n
n
73
%
%
”Figure
120
ra
122
123
128
125(a)
125(b)
125(e)
126
vit
Mole, shape (radia) correlations for SO raservotr
ier nee eee eee ers Bose
Node shape (radial) correlations for 85% reservoir
depth ee ee eee eee an
Cantilever stress and arch stress in Techi dam due to
static Toads
Cantilever stress and arch stress envelopes for upstream
face of Techi dam due to static Toads and design base
esrthquake eee eta eran tas eee
Cantilever stress and arch stress envelopes for dounstrear
face of Tech! dan due to static Toads and design base
earthquake. ee
5x danping elastic response spectra for Operation Base
Earthquake and Design Base Earthquake, based on Newnark's
basic spectra with amplification factors for acceleration
and velocity taken to be 2.6 and 1.9, respectively...
Relation between ground acceleration and probability
of exceedance in fifty years, P, and return period T . .
Relation between ground velocity and probability of
exceedance in fifty years, P, and return period T. . . .
Variable water level... eee eer rte
78
a1
90
98
106
107
108
1081. INTRODUCTION
1.1 objectives
hydrodynamic effects induced by the impounded water may have signi
Ficant influence on the response of a dam subjected to earthquake excitation.
Current technique is well capable of analyzing » Tinear damreservoir
interaction system, taking into account the hydrodynamic effects. (47,
913,16,18,19). But the responses of damreservoir systens to most Design
Base Earthquake (D8E)” are Itkely to be nonlinear, so that we can no Tonger
employ the frequency domain solution technique to deat with hydrodynamic
effects as in many of the works being done up to date. (47,16,18,19).
Time domain solution is left as the only alternative. While to economically
‘include the water compressibility of infinite reservoir® in time domain
solution is still under research, the hydrodynamic effects due to an
incompressible water reservoir can be readily taken into account in the
‘time domain solution of a danreservoir interaction system. The easiest
way to deal with the hydrodynamic effects of an imcompressible water
reservoir is by employing the “addedmass" concept (1). It is the objective
of this work to investigate and select most reasonable and economical
method that can count for the hydrodynamic effects of incompressitle
water reservoir in the form of addedmass. For the general geomezry of
concrete dams, the Generalized Westergaard Formula and the Galerkin Finite
Element Method are among the candidates. Rigorous analyses of node shapes
and frequencies are compared with the field experimental work, ané the
results serve as the major indication of the validity of the method.
DBE is an earthquake intensity corresponding to a return period of 100
years (also see Section 5.3)
* Infinite reservoir has been used in general for easier analytical solution
similating outbound energy radiation condition for a very large reservoir.1.2 State of the Art
The Finite Element Method with the aid of high speed digital conputer
has enabled us to analyze a11 types of complex civil engineering structures.
But one of the difficulties remained in today's structural analysis
techniques 1s to evaluate the effects of various kinds of loadings arise
from the environment where the structures are Tocated. The hydrodynanic
loading effects upon the daus are few anong then, Since the early part of
the twenties, the influence of hydrodynamic effects on the responses of the
dans have Tong been an interested topic, especially in the event cf earthquake.
In 1933, professor H. M. Westergeard (1) firstly established a
rational standard procedure to take into account the hydrodynamic loadings
‘on gravity dans during earthquakes. Although the case he studied was
Vinited to rigid dans with vertical upstrean face, and infinitely Tong
reservotrs, ignoring surface waves and considering only sna11 displacements
of Flufd particles, this work was regarded as a milestone. Especially
the concept of addednass, hich he introduced for the inconpressible
water reservoir, greatly simplified the enslysis procedure of tie response
of a dam considering hydrodynamic effects during earthquakes. Srahtz and
Heilbron (2) followed up with a discussion on the effects of a finite
reservoir, compressibility of the water and flexibility of the dans. In
1952, Zangar (3) furthered Hestergaard's works by using an electric analog
he investigated the effects of a sloping upstrean face and provided
results on addednass representations of hydrodynamic effects for a
broader class of dans that can be ideal ized as 2dimenstonal monoliths.
Zienkiewicz and Nath (13) Tater used the sane technique to apply Zangar's
work to 3dimensional arch dans.
‘3
Lately, Chopra has carried out a series of investigations (47) on
various aspects of hydrodynamic effects in the earthquake response of
gravity dams; in the more recent work he included also effects of the
Foundation modelled as an elastic halfspace (24). Following pretty: much
the same path, Porter (18) extended the work of Perumalswani (25) to
formulate explicit mathematical solutfons for the fluid domain reteined by
‘an arch dam considering the vesponses to a1 components of ground notion.
The reservoir considered was defined by a cylindrical dam face of constant
radius, a horizontal floor, and vertical radial banks enclosing a central
angle of 90°. Recently, Hall (19) has developed a numerical sehen to deal
with arbitrary geonetries of reservoir of arch dams. Effects of water
compressibility, flexibility of the dam, energy radiation in infinite
reservoir and vertical ground motion contributions (14) are thorouchly
treated by his procedure, At the same tine, in modelling an infinite
reservoir, Saini etal.(16) used an infinite elenent and obtained similar
results as Chopra; Nath (17) employed a conformal mapping technique and
‘obtained economical and reasonably good accuracy. However, ali these works
are restricted to solutions in frequency domain.
Priscu et al. (8) used a finite difference method to solve for arch
damreservoir system responses in the tine domain, considering compressible
water reservoir. Contrary to Chopra, he concluded that the water compress
ibility could change significantly the seismic response of a slender dam
(e.g. arch dan). In the particular case he studied, the dam displacements
could reduce up to 50% if water compressibility 1s not neglected. This
discrepancy in findings concerning the effects of water compressibility implies
the need for further research.
Nore recently, Muller (26) attenpted an approximation method in the time
domain, taking into account the water compressibility of the reservoir by a4
"second added mass" concept; the idea is good, but it stil] falls
short fn its ability to handle a large reservoir
1.3 Scope
In Chapter 2, the sinpliest representation of hydrodynamic effect,
‘that is, the addednass derived from Westergoard's classical solution,
is reviewed and generalized, considering arbitrary geometry and orfentation
of the upstream face of arch dams. Also an appropriate lumping process
is described
‘The Galerkin Finite Elenent Discretization of the wave equation that
governs the pressure behavior in an imconpressible fluid davain is presented
‘in Chapter 3. A consistent lumping process for this procedure that maintains
symmetry of the resulting addednass is also presented.
Chapter 4 describes conputer implenentations of the preceding concepts,
and also presents numerical solutions for pressures given by the various
schenes and conpares their results. The range of applicability of each
rethod is indicated.
Jn Chapter 5, numerical solutions of the mode shapes and frequencies
obtained by each method are correlated with results of field measurenents
on Techi Arch Dan; variable water level is considered. From these
correlation studies, a most reasonable and economical method 1s suggested.
Finally, stress responses of Techi Arch Dam due to static Toadings, the
Design Base Earthquake and hydrodynamic effects calculated by the suggested
method are presented.
Final conclusions and renarks concerning the needs in further research
‘on the tine domain solutions of infinite compressible water reservoir are
discussed in Chapter 6.5
2. GENERALIZED MESTERGAARD FORPULA
2.1 Review of the Classical estergaard Formula
In Westergaard's classical work (1), dealing with water pressures on
dams during earthquakes, he did not try to consider every possible effects
rather, as a good engineer will do, he made reasonable assumptions for the
case he studied, and was able to obtain reasonable solutions for engineering
use.
The assumptions he made are the following:
(1) dam was idealized as a 2dinensional rigid monolith with vertical
upstream faces
(2) the reservoir extends ta infinity in the upstrean directfon;
(3) displacements of fluid particles are small;
(4) surface waves are ignored;
(5) only horizontal ground motion in the upstreamdownstream direction
is considered.
Acrarding ta these ascimptions, he posed an initial boundary value problem,
and obtained pressure solutions on the upstrean face of the dam. for the
purpose of practical engineering use, he approximated the pressure solution
(For an incompressible reservoir) with a parabola, which he felt t> be
better than a quadrant of an ellipse. Later, he observed that "the pressures
are the same as if a certain body of water were forced to move bacc and forth
with the dam while the remainder of the reservoir is left inactive’. The
amount of the water included was determined by equating the inertia forces
of this body of water to the pressures that actually were exerted upon the
face of the dam under the sane motion of the dam:
Thus, Westergaard suggested (Fig, I1(a}), that the dynamic pressure
could be expressed as::
Pon «Fg wen
where
horizontal ground acceleration, in units of g (gravitational
acceleration)
unit weight of water
Yi * horizontal ground acceleration
unit mass of water
depth of reservoir above the base of the dam
distance from the base of the dan
hydrodynamic pressure at height Z from the base of the ém,
applied normally to the dam face.
Equation (2.1) indicates that the hydrodynamic pressure exerted normally on
the upstrean face of the dam, et hetght 7 above the base of the den, due to
around acesTeration (that 4s, the total acceleration of dam face at
height Z, because the dam ts rigtd), 1s equtvalent to the inertia force of
a prismatic boay ot water of mt crosssection and tength 2 MACHZTs
attached firmly to the face of the dam, and moving with the dom Bick and
forth in the direction normal to the face of the dam (that is, horizontally)
without fetetfon,
Tis body of water attached to the dan face and moving with the dan,
is the ‘addedsmass" applied hy the reservotr to the dan, 2 concept First
Ineroudced by Westergaard, that has greatly staplitied the dynanic response
analysis of dans with hydrodynamic effects.
2.2 Generalized Westergaard Formula (9,10,27)
Employing the concept of “addedmass" as mentioned in Secticn 2.1 above,
we now generalize it by applying the following assumption:The hydrodynanie pressure exerted on any point of the upstream
ace of a dan, due to the total acceleration # normal to the din
face at that point, 1s equa) to the inertia force produced by @
Be gear
where Z is the height of that point above the base of the dan, that
ee
back and forth withthe dam in the normal direction without friction.
(Fig. 11(b)).
According to thts definition, the "addedmass* 1s generalized to be
applicable to the general geometry of the upstren face of flexible atch
dans, because it depends only on the tot normal acceleration at Tecal
points.
You in the Finite elenent analysis of the response of the dan, if we
have dtscretized the dan body tnto finite eTenents, then, at a certain node
"1" on the upsrean face of the don, the hydrodynamic pressure ts:
Be uty (2)
where
pj * hydrodynamic pressure at node "i", compression as positive
total normal acceleration at node "i
t= Westerguard pressure coefictent F oAQT
ness density of the water
p=
depth of water at the vertical section that includes node “i
2; = height of node "i above the base of the dam
But the total normal acceleration FE, can be represented in tems of
cartesian coordinate conpoents of the ground acceleration and
xt Faye Fon
of acceleration components at node i relative to the base of the dan #4 Fy;
and Pye
at node i, we have:
Making use of direction cosines with respect to the normal direction(2.3)
where
att HE HET totan acoeteration of degrees of freedom at node #
Ay 7 Ay Ay Argo normal direction cosines at node *
a (3 x 3) displacement transformation matrix of which the entry
8 yy Stands for the acceleration of node "i" in Jdirection
(J,k = 1,2,3 representing x,y,zdirection, respectively) due to
& unit ground acceleration in kdirection while the din is
undergoing rigi¢ body notion.
Substituting £q. (2.3) into Eq. (2.2) leads to the hydrodynamic pressure at
node { expressed in terns of ground accelerations and relative accelerations
at nade i:
P,
ah Hee
(648; Fy) (2.4)
Hydrodynamic pressures at any point on the face of the dam can be
found in a similar way. Sut in the finite elenent solution procedure, these
‘external pressures must be integrated over the appropriate surface of the dam
to obtain the nodal Toads. In this Tunping process, the hydrodynamic nodal
Forces are expressed in terms of nodal accelerations, by Eqs. (2.2) and (2.3).
thus, the coefficient in this expression will be the equivalent addednass.
2.3 Tributary Area Lumping Process
‘The easiest way to Tunp hydrodynamic pressures into equiva’ent hydro~
dynamic nodal forces, 1s to multiply by the tributary area associated with anode 13 thus
Fagg 7PM (2.5)
where
‘ni = equivalent normal hydrodynamic nodal force, outward normal from
‘the dam face as positive
P; = hydrodynamfc pressure at node i, compression as positive
A; = tributary area associated with node 1.
Note here, that the hydrodynamic pressure was assuned to be constant over
the tributary area, and to have the magnitude as at node 4. Also, since
the hydrodynamic pressures act normal to the dam face, so 1s the equivalent
hydrodynamic nodal force, in the average sense, also normal to the din
face. Hence, the 3 components of the equivatent hydrodynamic forces at node
4 tm Rectangular Cartesian Coordinate (RCC) frame can be found as before.
Prenultiplying F,y by normal direction cosines at node 4, thus leads to the
cartesian coordinate values
7
at 3 (2.6)
where
4 oT
Fe Fy FG
Oy
Substituting Eqs. (2.4)and (2.5) into Eq. (2.6), leads to:
(2.7)
(2.8)10
Mag, 1% te Addedemass matrix associated with node f and folTovirg the
direct stiffness assenbly procedure, the equivalent hydrodynamic nodal
force equations for the dam became:
or,
sammy "C3I) ~ es am) C2M3) “Hpayn) 29)
were m = total nunber of nodes of the dan on its upstresn face.
Hyg in Eq. (2.9) fs the addedmass coefficient matrix for the dam
resulting from the hydrodynamic pressures upon the upstream face of the
dam. It 45 uncoupted between nodal points. Also, notice that the sane
vesterguard pressure coefficfent {5 used, regardless whether the total
nodal acceleratfons cane from the vertical or horizontal component of
groune acceterattons.
The equivalent hydrodynamic nodal force F from Eq. (2.9) is an
adgitionat Toading vector to be incorporated into the right hand side of the
equation of motion of the dam:
Maan)! (mat) * Stnan) Yonxt) * Stmxn) Yemaa) * 2¢mxt) (2.10)
weve
mass, damping and stiffness matrices respectively of the
dan structures
4, v= velocity and displacement vectors respectively of the entire
dam, including internal degrees of freedon of the dam
structure;
total acceleration vector of the dan structuresu
°
e { =  where F is obtained from Eq. (2.9)3
Esme),
n= total number of degrees of freedom of the entire dam structure.
Alternatively, if we write Eq. (2.9) as follows:
0 0 0 ¥
th . ‘ ] tal var a
eT
total acceleration of internal degrees of freedom of the dan;
a +
# = E(aaaay + Bigs from Ea, (2.9).
‘Then, Eq. (2.10) can be rewritten as:
(M+) eee
+kye0 (2.12)

E12
3. GALERKIN FINITE ELEMENT METHOD
3.1 Galerkin Hethod for Nave Equation (11,12,27
In this formilation of reservoir interaction, the hydrodynamic
pressures In the reservotr are assuned to be governed by the pressure wave
equation (Fig. I2(a)):
Wolaanzst) » Uy Blasys2st) aa)
were
plsy.2.t) = pressure distributions in the reservoirs
c
k
WRTp is the sonic wave velocity;
bulk modulus of the fluids
p= mass density of the fluid.
In order to find the hydrodynamic pressures acting on the face of a dam,
Eq. (3.1) mist be solved with appropriate boundary conditions. Since our
interest 1s in finding addedmss representations of the hydrodynamic effects,
after we have found the hydrodynamic pressures due to accelerations at face
of the dam, they must be lumped into equivalent hydrodynamic nodal forces.
Thus, the hydrodynamic forces are related to accelerations at the nodal
degrees of freedom on the face of the dam leading to the addedmass
coefficient matrix. For this purpose, the boundary conditions to be inposed
fon the reservoir boundaries (Fig. I2(b)) are as follows:
(1) at damreservotr interface: $B =  if. 3
y t
(2) at Moor or reservoir: B= 0, or vt = 0:
a Oo
(3) at upstream end of reservoir: $2= 0, or vt = 0
s 's
(4) at free surface of reservoir: p
+ oF surface waves are neglected ;
2)
(5) at canyon watts: 22
any
or, of
s
where ng is the outvard normal direction from the reservoir surface, andb
UE, te the total nom acceleration of the flutd at the bounartes of
the reservoir. The acceleration 1s positive when fluté soves oubard from
ee ee
Surface waves, then, condition (4) becomes a Bs if we want te consider
‘the flexibility of the floor and canyon walls, conditions (2) and (5)
vecone 32«  oot, where the acceteration Vt is detined as was rentioned
stoves 1f'we want to investigate the effects due to relative motion at
upstream end of the reservoir, condition (3) becomes $2  fy , ete.
's
Now we proceed to seek the hydrodynamic pressure solution of Eq. (3.1)
with the boundary conditions of Eq. (3.2). But because the geonetry of the
reservoir generally 1s irregular, it will be extremely difficult to find a
closed form solution; therefore, we seck 2 numerical solution based on the
Galerkin Finite Element Method.
‘The Galerkin Method is a weighted residual method; its residual is
weighted in such 2 way that the approximate numerical solution wilT be
orthogonal to the error of the nunerical solution; and thus, in the energy
norm, the nunerical solution minimizes the residual caused by the error.
Letting B be the approxinate nunerical solution of Eq. (3.1) with
boundary conditions Eq. (3.2), then the residual of Eq. (3.1) due ts the
error in approximate solution p is,
o
per 3.3)
ae
where R is @ residual of very smal magnitude.
The Galerkin Method is expressed as,
re Nt z wv (3.4)
here Nis a row vector of weighting functions. Applying Green's
‘Theorem (or integration by parts) to Eq. (3.4), we have,W
g at Bean Leste nl bev  ssn) pao (3.5)
mye foe" oa
a:
The first term of Eq. (3.5) only exists along the boundaries; applying the
boundary condition £9. (3.2) to it, we have,
lB Tot
YBa ogs ah ot an (3.6)
which no arly exists along damreservir interface, because ot vanishes along
all other boundaries according to Eq. (3.2). To relate this to the mtion of dam
face, we observe that at the sane point on the damreservoir interface,the fluid
acceleration vt can be expressed in terms of the acceleration of the dam
",
face. ower, ft mst be noticed that # ts positive wen outard
:
normal from the dam face (Fig. 12(a)), while ie ‘is positive when
outward normal from the reservoir (Fig. 12(b)). thus,
{in terms of the normal direction cosines and three ROC components;
that 1s,
(3.8)
where
's
FUE UT total acceleration of the dan face im RCC components,
Substituting Eqs. (3.6), (3.7) and (3.8) into (3.5), we have,
oop ot a
EM
i18
For conpressible water, C = 4720 ft/sec, but for incompressible water,
C+, thus Eq. (3.9) becomes,
fgg WB av = ost NT Ut an (3.10)
lies Secreted
‘This is the Galerkin weak form of pressure wave equation; for an incompressible
reservoir, it relates liquid pressures to the accelerations of the face of
the dan,
3.2 Finite elenent Formulations
According to Galerkin Finite Element Method, after the Galerkin form
of the field differential equation has been obtained, we discretize the
donain by Finite Elenent Method, using as weighting functions the baste
‘interpolation functions of the elenents.
Examine again Eq. (3.10),
nwt wp evs oss nt a it an (3.10)
eave Ms ad
Uihere V is the fluid domain, 5 is the damreservoir interface. We now
discretize the fluid donain into 30 finfte elenents, and the interface
correspondingly into 20 finite elenents(the interface ts 20 in natural coor
dinates, but 3D in RCC space). For @ point within the clement ¢ in the fluid
domain, its appropriate hydrodynamic pressures (can be expressed as follows,
according to the Finite Element Method:
pled = yled pfed an)
where
ul) = row vector of interpolation functions associated with todes of
the elenent e,
(©) = column vector of nodal pressures of the elenent e.16
Then, substituting Eq. (3.11) into the left hand side of Eq. (3.10), we
have for element e:
1 1
sap onlMe oplal®) « rrp aul) oyledgyledg(ed
ye) ye”
or, o
sr syle). gpleayle) « (3.12)
yer
were
tea) * ara al) (3.13)
KD = nunber of nodes of 3D fluid elenent e.
For 20 interface elements, the accelerations iit can be approxinated in a
similar way. For 2D interface elenent i, we haves
c :
art. 6G) a
=a te
os
8 seo) “matin of Interpolation funtion: asocinted wich mol
eareos of freadon of interface etenent + ,
t
107 yey) * tums vector oF totat pada accelerations of £D
interface element 1 in RCC components ,
ND = nunber of nodes of 20 interface element i.
Substituting Eq. (3.14) into the right hand side of Eq. (3.10), we have for
element #1
CTC) GE™ gal (074) lye sink
ADE AG) GE gD ne” if an
may Bey Pay te Pgs
Bag AGOT AG) GCE Q(D gl (3.15)
s(t) >here
alt) eT of Daal)
“S(wOx3KD) (4) “$ e9)
ay
"
i = the part of the 30 fluid elenent interpolation function
~$(1x4D)
that reduce to its 2D interface boundary only, vhich 4s
corresponding ta the 20 interface element 4.
Now assembling Eq. (3.12) for all 30 fluid elenents in the reservoir, and
assenbling £9. (3.15) for all 20 interface elements on the damreservoir
interface, we have the discretized form of Eq, (3.10):
x gle) ple) 5 enie) oe (a7)
Ser Sirs Be 9 a a
(NERXNER) (NERXNES)] )(NERXT) (WERXNLL)
Ser es ) Be (P10 hy mf G1)
(NESXNER) (NESxNES)] / (NESx1) (NESXNLL)] (NLLXI)
where
= nodal pressures of fluid elenents that are not on an interface
nor on a free surfaces
nodal pressures of fluid elenents that are on the danreservoir
interface, but not at a free surface;
Gor, Seg? Sop Sag * Submatrices of g partitioned according to p, and
Pst
4 = assenbled matrix of Eq. (3.13) over the entire fluid domains
hh, = assenbled matrix of Eq. (3.16) over the danreservoir interfaces
EE = nodal total accelerations of dam face, including those nodes
at free surfaces
NER
nunber of nodes of fluid elenents that are not on an interface
nor on a free surfacesry
NES = number of nodes of fluid elenents that are on an interface,
but not on free surfaces
NERINES = NEQ, total nurber of nodes in the reservoir, excluding al?
‘ree surface nodes;
NLL = total nunber of nodal acceleration RCC conponents on the
interface including those at the free surface.
‘The pressures on the free surface vanish according to boundary condition
(4) of €q. (3.2), therefore, they do not enter into the assenbling process
of matrix g.
Since we are only interested in finding the hydrodynamic pressures
acting upon the interface, that is the vector p., we don't have to solve
the entire system of equations of £4. (3.18); rather it is convenient and
more economical to do a static condensation on Eq. (3.18) first, to condense
out NER equations that are associated with p,. Since there fs
nothing on the right hand side associated with p,. we need only to do static
condansation operations on the 1aft hand side of fa. (3.18), that ts, on
matrix g. Thus we hoves
e
Bs Oh, Fe xe)
were
% Ses “Sen Ser Ors (3.20)
(Nesotes)
is a symetric matrix, and,
so git hg &t (2.21)
After pz, that 4s the hydrodynamic pressures acting upon the danreservoir
interface, have been found from Eq. (3.21), the next step 1s to Tum the
hydrodynamic pressures into equivalent nodal hydrodynamic forces, ard thus
‘to obtain the addednass coefficient matrix. This operation is equivalent19
to that described in Chapter 2.
3.3 Consistent Lumping Process
The consistent lumping process making use of the virtual displacement
method is the most appropriate procedure for converting the hydroiynantc
pressures of &9.(3.21) into equivalent nodal forces (Fig. 13(c)); that is,
by introducing virtual displacenent field into the domain and equating to
zero the virtual work. Let's now consider a 2D interface elenen: #, which
corresponds to a 3+D fluid clement e; in other words, element i overlaps
the upstrean surface of concrete elenent ¢ of the dan at danreservoir
‘interface. From Eq. (3.21) we can obtain pressure values at discrete nodal
Points that is, for element 1, we can obtain its nodal pressures.
Then, the pressure distributions over the domain of elenent { can be
expressed as,
pet alt) 9) (3.22)
where
BLT) ~ pressure at any point on Interface etenent 1 5
a part of fluid elenent © interpolation functions that are
veduced to its 2D interface boundary only, that 1s,
nonvanishing only on the surface corresponding to interface
element 1 5
pli) = nodai pressures of interface element i.
°5(NDx1)
Now we introduce a virtual displacenent field, oul") at any point, into the
domain of interface elenent 1 by introducing nodal displacenents st nodal
points of element 1, ér(t), and, as in Eq. (3.14), with the interpolation
functions of, ve have,
ay) 2 gf) gpl) (3.23)20
Then, equating the virtual work done by the hydrodynamic pressures to
‘the virtual work done by their equivalent hydrodynamic nodal forces, we
have,
1
el TBD 5s gy yt) 9D gd (3.24)
. sli) Ms
where
Foe) = equivalent hydrodynamic nodal forces of 2D interface
elenent i
trl!) « norma virtua? dtsplacenent at the potnt snare Fl!)
applies.
Notice here that the negative sign is due to the fact that wis positive
for compression and cy is positive when the interface moving outward normal
from the dam fcesterefoesresulting Hn negative virtual work. Furthermore,
as in Eq. (38), we can express srl!) in tems of sult), tnat is,
:
en
sel) ng gfe (7 F (3.25)
where \ is the normal direction cosines at the point where su") tocates.
Substituting Eqs. (3.22), (3.23) and (3.25) into Eq. (3.24), we have
el PANS Sg Ty DT TCAD gd 90)
3)
or, :
2. tat i
fa Hf of (3.26)
iw ed in
where n{') is defined in Eq. (3.16).
Assenbling Eq. (3.26) over the entive damreservoir interface, we
have,
z
F 8 p (3.27)
SHLD) “Sy Lates) “5 (eS)a
Then, introducing Eq. (3.21) into Eq. (3.27), we find,
or,
(3.28)
where
(3.29)
is the symetric eddedonass coefficient matrix for the dan, resuiting from
the hydrodynante prasuras acting upon the upstrean face of the can.
Notice that the addedmass coefficient matrix of Eq. (3.29) whict came from
the Galerkin Finite Element discretization of the wave equation in the
reservofr using a consistent lumping process, is in generat a full matrix,
coupled not only Between nada points but also anong nodet degrees of
freedom that are perpendicular to each other.
Expressing total acceterations # in tems of relative and ground
accelerations, we can rewrite £q. (3.28) in the form similar to Eq. (2.9),
shu,
=e t  a
Fomtaa) ~~ Moseuucauey Se ~ scurry 206t29) f¢350)
(3.30)
As before, the equivalent hydrodynamic nodal force vector F of Eq. (3.30)
is ready to be incorporated into the right hand side of the equation of
motion of the dan as an additional effective loading vector.2
4. COMPUTER IMPLEMENTATIONS AND NUMERICAL SOLUTIONS
4.1 Computer _Inplementations
In order to illustrate the efficiency and validity of the schene
for addednass coefficient matrix computation presented in Chapter 2 and
3, a Fortran program RSVOIR (28) was developed. It serves as a general
purpose incompressible fluid addednass preprocessor for arch dans of
general geometry. The addedmass coefficient matrix can be computed from
this preprocessor and then assenbled with the concrete mass matrix of the
dam at appropriate locations, so that the response of the dam including
incompressible hydrodynamic effects can be obtained. The computational
procedures (20,21,22) for obtaining the addedmass matrix with each
schene are detailed in the following sections.
4.1.1 Generalized Westergaard Formula Procedure
From Eq. (2.8) we have the addedmass coefficient matrix according
to the Generalized Westergeard formula. Tor the computation of added
mass coefficients associated with node 1, three pieces of infomation are
necessar
(a) Westergaard Pressure Coefficient a, = § ov,
(b) tributary ares A;
(c) normal direction cosines 2, = ©,
Ly ae
‘The First item, a;, may be calculated readily when the location of node
1 1s known. The second, A,, is a collection of the arez contributions
from every interface elenent associated with node 1, that is (Fig. 13(b)).
aah) @ay
k
fencer te eteenea ante yact23
AME) = Al) pxoca (4.2)
where A(K) = area of interface elenent k
ND(k) = number of nodes of interface k
The norm direction cosines, 2; (Fig. I4), can be found from the
properties of the interface finite elements assoctated with node 4
(Appendix A). Due to the finite element discretization, the normal direction
cosines of the same node may differ from elenent to element. Therefore,
‘the nodal mass {s obtained through elenent by elenent assenbly:
tay EA te
vere 2{8) « normal divection cosines of interface elenent ky at rede f.
by element, globally. Therefore, the local assembly of Eq. (4.3) is
ee ee
carried out for every interface elenent on the danreservoir interface.
4.1.2 Galerkin Finite Elenent Procedure
The addedmass matrix of Eq. (3.29) is not obtained from the assenbly
process; rather, it coves from the solution at the global level. fut
its ingredient matrices, g, and h,, are the result of assenbling elenental
contributfons. In order to find addednass matrix of Eq. (3.29), it is
necessary to form the matrices g. and h, first.4
4.1.2.1 Formation of matrix g,
‘As shown by Eq. (3.20), matrix g, is a product of static conden
sation operating on matrix g, which in turn is the global assenbled
form of the elenent coefficient matrices gl) of Eq. (3.13).
For 30 fluid element e, the integral of Eq. (3.13) 1s computed
numericatly, thus, recall
1
ane) aye) ay ae) ae anf
(2) 6 sry CO y(e)
Se oe meer)
(2) ayle) aul) aye) aye? aye)
anle) anf®) —an(®) aule) —aule) an
9 oe ee
aay
Enploying Gaussian Quadrature to numerically integrate Eq. (4.4), we
have,
fe)
8) = Em ty My Dry ssysty)l9fedL (4.5)
BM Me
where
(rpSgity) = Sausstan Quadrature Integration points in natural
coordinates (r,5,t)
Mcnily * Gaussian Quadra weighting fimctfons
anfe) aule) —ayle) ule) —anfe) ane)
i i 1
Be i a ae at
clement ©, evaluated at (ry sSqity)>
HDC aSpety) =
ja{2)] = determinant of Jacobian matrix of elenent e,a(°),
evaluated at (rjsSpst,)» where
pe a Teele
af) =] yfe) gle ged yO), zl
ne) xf2d gg) ye) gle) gfe)28
fn wnten ayle)—gyled—gyfed
LO ALE), gee en a respectively, and
xl), le), 22) « x, y, 2 components, respectively, of
coordinates of all the nodes pertaining to element e.
Finally,
on(e) on(e)
a a
fe) . (e)
pe ger © 9 ME
wwf mo
ale eo
ae) aE
Now, the nunerical calculations (Ea. (6.5) Jare carried out for alt fluid
elements in the reservoir, and the results are assenbled according to the
direct stiffness method to forn the global matrix g. But so far, we
have not mentioned what kind of 30 fluid elenent should be used and how
Dig should be the domain of reservoir included in the discretization.
Jn general, 13 to 20 node 30 elenents (Fig. 16(D) and Appendix
8) allowing quadratic variation in upstreandownstream direction are
appropriate to be used as fluid elements, because they provide possible
exponential decay of pressure solutions in the upstream direction. As
for the size of reservoir donain to be included, it is clear that it's
impossible to model an infinite reservoirs therefore, the reservo'r
‘extent in the upstrean direction should be found by studies on the
convergence of hydrodynamic pressures white gradually increasing reservoir
domain. It has been found that for an incompressible reservoir, the
hydrodynantc pressures converge adequately when the reservoir donain
extends in the upstream direction three tines the height of the dan.
Now, after g has been formed, using appropriate 13 to 20 node2%
3D fluid elements and an adequate reservoir domain, we statically
condense out those nodal degrees of freedom in the reservoir that are
not on the danreservoir interface. The nurber of equations eliminated
is NER as defined in Section 3.2, resulting in matrix g, of dinension
(NES x NES), where number NES is also defined in Section 3.2
4.1.2.2 Formation of Matrix hy
The matrix he in Eq. (3.28) fs the assenbled form of Hi") in
Eq. (3.16). The integration of Eq. (3.16) is only appropriate to be
carried out numerically as ahove. Therefore, Tet us recall?
nt egy DT GD
(A) galt 3.16)
Bony ee
‘and apply Gaussian Quadrature integration to the integral, to obtain
' (D
Ah) = 2 2 uy My alr arg) (4.6)
where
gM = weighting functions of Gaussian Quadrature
(rgeSq) = Saussian Quadrature integration points in natural
coordinates (r,s)
1
afang) = nb AC of? evatuated at the Integration goints
(rqpSq)
AW) eal al oan, normal direction cosines at (rqs5q) of
element { (Fig. 14 and Appendix A)
2 yt tye
ay = pal] = OLMS DS LD evatunted at (rgesq) (Ref. 29)a
aster nf") in €q. (4.6) has been numerically evaluated, it is assenbled
‘nto the global matrix h,. These calculations and assenbly operations
are carried out for all 20 interface elenents on the danreservoir
interface. Thus far we have not considered what kind of 2D element
should be used, but 6 to & nodes 20 elenents (Fig. I5(a) and Apoendix
8) seen to be appropriate
Now, after the matrices , and h, have been fomed as above, We
can proceed to carry out the computations for the global addednass
matrix My. of Eq. (3.28). Because inversion of a matrix 1s very
inefficient, the conputation of Eq. (3.29) 4s carrted out as fottews:
(a) Solve 9, 0 = hg, for 9 an
using any suitable equation sotvers
() Form the matrix preduct i = » ig
thus, the global addednass matrix Hof Eq. (3.29) fs obtained.
4.2 Hydrodynamic Pressure Solutions and Their Comparisons
Although the addednass coefficient matrix is the most convenient
vay to account for the hydrodynamic effects, it is hard to tell whether
{2 schene 1s good or not just by looking at the addedmass matrix
coefficients. In order to evaluate an addednass matrix schene, ue have
to examine it with respect to the hydrodynamic effects it produces.
If we recall the procedures to formulate the addedmass matrix,
it is clear that the hydrodynamic pressures are the quantities of interest.
The definition of addednass coefficients can be termed as nodal resisting
forces caused by unit nodal accelerations acting into the reservoir. Yet,
tthe nodal accelerations actually cause distributed hydrodynamic pressures
to act on the dan face, and the nodal forces are obtained through
processes of lunping the pressures. Therefore, the hydrodynamic effects28
represented by the addednass coefficient matrix can be evalucted by the
studies on the pertinent hydrodynamic pressures.
Now, we can impose any pattern of accelerations on the upstream
face of the dam and obtain, from Eq. (2.4) or Eq. (3.21), the hydro=
dynamic pressures distributed over the dam face. The simplest pattern
of accelerations that can be applied is unit uniform motions in the
Uupstreandovmstream direction. Physically, this is just the rigi¢
body motfon of the dam with unit accelerations acting in the upstream
downstream direction, while the reservoir Floor and canyon walls are
fixed.
Figures 16 through I 9 show the hydrodynamic pressure distri
butions over the upstream face of dams of various geometries subjected
to unit unfform acceleration in the upstreandownstrean direction.
‘The cases studied included a gravity dam with vertical upstrean face,
cylindrical arch dans, and a general doubly curved arch dam. Notice
that in all cases, the ydimensfons of successive fluid elements have
a ratio of 1.25 in the upstream direction.
4.2.1 Gravity Dans
Figure 16(a) shows the reservoir of @ gravity dam with yertical
upstream face, discretized into 16node 30 fluid elenents and 8node
20 interface elements. Since the geouetry of the reservoir and the
excitations of the interface boundary do not vary with x, this is a 20
problem and the pressure solutions are independent of x. Actually, this
4s exactly the case Westergaard (1) studied, and the exact solution is
available. As shown in Fig. 16(b), the hydrodynamic pressure solutions
from finite elenent method converges sufficiently when L/H = 3. Also,
‘the figure indicates that both Westergaard approximate solution and theFinite elenent solution with L/H = 3 are good approximations to the exact
solution.
Due to the discretization error, it is seen that the finite
elevent solution converges to a value lover than the exact result.
Tt can be shown that 20node 3D fluid elenents will yield a better
solution because they are more flexible, but the additional computational
cost my not be justified.
In this case study, the Finite elenent method is demonstrated
to be a reasonably accurate and efficient method to evaluate hydrodynamic
effects; the Westergaard approximate solution is expected to be good for
this case only, because it was derived from the closed form solution for
this case.
4.2.2 Gylindrica?_ Arch Dams
As a bridge from the study of gravity dans to the case of general
arch dans, several cylindrical arch dans were studied. Fig. I7(a) shows
‘the reservoir of a cylindrical arch dam with vertical upstream fice. The
reservoir has parallel vertical side walls and horizontal bottom Fig.
17(b) shows the hydrodynamic pressure distributions at the crom section
due to unit upstream acceleration. The finite elenent solution converges
sufficiently when L/H = 3, and the Nestergaard approximate solution is
virtually fdentical to that for the gravity dan in Fig. I6(b). Fig.
17(e) shows that similar results are found halfway between the crown
and abutment, although the finite element results are increased slightly
while the modified Westergaard results are decreased. In Fig. I7(d),
‘the hydrodynamic pressure distribution at the vertical section next to
‘abutment shows significant changes: increases for finite elenent and
decreases for Westergzard. The latter results are unacceptable; the30
Finite element results are effectively converged at L/H The
underestimations of hydrodynamic pressures by the Westergaard epproxi
mate solutions can be attributed to the fact that it only recognizes
the water depth and normal direction cosines of the dam face; it is not
anare of the existence of the bank which forms sharp angle with the dan
thus restricting the lateral direction flow of the reservoir water,
If we now enlarge the angle between the bank and the dan, by
introducing diverging reservoir walls as shown in Fig. I8(a), it can be
seen from Fig. 18(d) that the Westergaard approximate solution 1s
again close to finite elenent solution, It should be noticed that in
Figs. I7(d) and 18(4), the Westergaard epproxinate solutions do not
change, but the finite element solution varies due to the boundary
restrictions imposed by the bank. In general, the Westergaard
approximate solution overestimates hydrodynamic pressures if the abut
rent angle between bank and dam {s reasonably wide as shown in Figs
T8(b) and (c). These Figures also compare the effects of the reservoir
bank flare angles for the finite element solutions.
4.2.3 General Arch Dams
In Fig. 19(a) we present the reservoir of a general arch dan,
where the reservoir has constant section (prismatic form) in the up
stream direction. Because of the discretization approximation in the
Finite element method, the normal direction cosines for nodes at the
face of the dam, may not be calculated accurately especially for the
corner nodes. A general conclusion of this comparison as depicted in
Fig. 19 {3 that the Westergaard approximate solution is too conservative
and overestimates hydrodynamic effects. Its chief advantage 1s that it
is the Teast expensive means to represent hydrodymanic effects. However,a
by actually carrying out computations of hydrodynamic pressures by
both methods on Techi Arch Dan, an existing nonsynmetric arch dam
(Fig. 110), we found that it was not very expensive to use Galerkin
Finite Element Hethod. Results obtained by the two methods, shown in
Fig, 19(d), demonstrate the significant overestimation given by the
Mestergaard approach.
From the cases studied, it is evident that the Generalized
Westergaard approxination can be used for crude preliminary analysis
purposes taking advantages of its relative economy. The Galerkin
Finite Element Method with its competitive Tow cost should be usec to
represent incompressible hydrodynamic effects for final design studies.
The validity of this method will be strengthened by the correlation
studies presented in the following chapter.32
5. NUMERICAL CORRELATIONS WITH EXPERIMENTS ON TECHI ARCH OAM
5.1. Properties of Techi Arch Dam
The Techi Dan, located in the middie part of Taiwan in a moderately
active seismic zone, was completed in Septenber 1974. It is 160 min
height, with @ crest 290 m long at an elevation of 1411 m above sea.
The thickness of the dam is 4.5 m at the crest and 20m at the base. It
is a double curvature arch dam with total concrete volume of 420,000 m?.
A perinetral joint surface has been provided between the dam body and the
pulvino block; the dam body has 22 vertical cantilever monoliths with a
contraction joint between each monolith (Fig. 110).
‘The mechanical properties of the dan and its foundation are as
follows
Young's Modutus
Potsson Ratio
Thermal Coefficient
Mass Density
Compressive Strength
Tensile Strength
Young's Modulus
Poisson Ratio
Mass Density
Compressive Strength
Tensile Strength
Dam Body
E 5.6774 x 10° psi
Fgynamte *
v = 0.21 (0.19 ~ 0.23)
a= 5.6 x 106
150 inet?
e
jc 5965.15 pst
re 500 psi
Foundation
E = 8.516 x 10° psi
v= 0.21
p= 162 16/ft?
og: 2:1 x 108 pst
042 930 psi
In 1979, several sets of dynamic experiments were carried out on
‘the dam, including measurenents of both ambient vibrations and forced3
vibrations. Numerical modeling of the dynanic behavior of Techt Dam
was carried out on the CDC 7600 machine at Lawrence Berkeley Laboratory.
A certain volume of massless foundation rock was included (Fig. 111,
112) in the model, and both the dan body and foundation rock were
discretized by 30 finite elements. Thick shell elements, transition
elements and 3D shell elements as described in the conputer program
ADAP (23) were used for the dam body whereas 8node brick elenents were
used for the foundation rock. (Figures I11 through I14). Various
addednass matrix representations of hydrodynantc effects were comuted
from the preprocessor RSVOIR (28), and assenbled with the concrete mass
matrix computed by the program ADAP. The addednass matrices tha: were
considered included the Generalized Mestergaard approximation solution,
a consistent addedmass matrix fron Galerkin Finite Element Method, and
a diagonalized addednass matrix obtained by diagonalizing the consistent
addedmass matrix. (22). Also, various different water levels were
considered in the study (Appendix C).
The vibration Frequencies and mode shapes were calculated by the
ADAP program for each case including foundation and hydrodynamic effects.
Also, stresses in the dam due to the static Toads, namely hydrostatic
and gravity, were evaluated. Because there was no information on
temperature changes, no thermal load was included. Finally, a response
spectrum dynamic analysis was carried out. The first 8 modes of the
case considering the diagonalized addednass matrix from Galerkin
Finite Element Method were used in the dynamic response analysis, and
stresses envelopes were obtained for the damfoundationreservoir system
subjected to hydrostatic load, gravity load, plus the Design Base
Earthquake excitations. Mode shape and frequency correlations with theexperimental results are shown in Section 5.2, and the stress responses
including earthquake effects are presented in Section 5.3,
5.2 Correlation of Frequencies and Node Shapes
Because of heavy rainfall that occurred during the field measure
nent program, the vibration measurements ware taken at differert water
Tevels. Therefore, munerical solutions were also calculated for various
water levels using different methods of approxination; by curverfitting
we can interpolate frequency values at intermediate water levels.
5.2.1 Frequency Correlations
Various types of frequency correlations are presented in Figures
116 through 119. Fig. I18 shows the variation of fundamental frequency
of Techi Dan with changes of reservotr height, calculated with the
modified Westergaard reservoir approxination and with the finite eTenent
nrodel. Also shown is corresponding information on frequency changes in
a typical gravity dam. This Figure shows that the decrease in
Fundanental frequency of an arch dan is more rapid than that of @ gravity
dam wen the water level in the reservoir increases. This shows that the
Addednass which represents the hydrodynamic effects has a greater
‘influence on the fundanental frequency of the arch dan than on that of the
gravity dan, This 1s obvious because the arch dans are in general nore
Flexible and have Tess conerete mass. The figure also shows that when
the water level is close to the crest, the addedaass has similar influence
on the fundamental frequency for both gravity dam and arch dam. However:
‘this cannot be taken as a general rule, because it depends on tre geonetry
of the reservoir and of the dan being considered. If the reservoir cross
section is of a wide Vshape, it is expected that the influence of the%
addednass upon the fundamental frequency will be much greater on arch
dams. Tt 48 also interesting to see that the addedmass cause almost
‘the sane rate of change of fundanental frequency of the arch dan,
whether 1 was computed from General ized Mestergeard approximate sclutfon
or from the Galerkin Finite Elenent Method, but the significant over
estinate of reservoir effects given by the Hestergaard model is obvious.
‘The correlations between the analytical procedures and also with the
experinental results are presented in a different form in Fig. I16,
where the changes due to water level are based on the full reservoir
condition. The similar slopes among all 3 curves, i.e., the rates of
‘increase in the fundamental frequency per unit decrease in water death
show that the analytical procedures for taking account of hydrodynanic
effects are fairly accurate. The sharp drop on the curve of the
Finite elenent solution shows that the hydrodynamic effect has negligible
‘influence on the fundanental frequency of the dam when the water level
4s below 40z of the height of the dan. The increase in fundanental
frequency of the dam without water over that of the dam with full
reservoir, 15 approxinately 38% when the addednass is computed by
Finite element method. Expressed in the context of Fig. I16, if the
fundamental frequency of the dan without water 4s 100%, the fundamerta?
Frequency of the dam with full reservoir is about 742, In general, a
full reservoir will reduce the fundamental frequency of an arch dam by
20% to 308.
Figure 117 shows comparisons between results of the two numerical
representation of addednass for the frequencies of first 4 modes of
vibration of the reservoirdam system when the water level varies.
This Figure again shows that the rate of change of Frequencies with water%
level is similar for both numerical representations of addedmass.
But obviously the addedmass obtained by Generalized Westergaard solution
is greatly overestimated conpared to that computed by Galerkin Finite
Element Nethod. The difference between different schenes 1s greater than
between different water levels for the same schene. 
The correlation with forced vibration experimental results shown in i
Fig. 18, demonstrates that the hydrodynante effects evaTuated by
Finite Elenent Method in the form of eiagonalized addedmass netrix ofves
better agreenent than the Mestergaard procedure. Although the finite
clement analysis underestinates the addednass effect for the fundamental
node, it overestimates it for the second mode and gives quite cood
agreement for the third and fourth modes. Therefore, as a whole, it may be E
concluded that hydrodynamic effects are well evaluated by Finite Element,
Nethod.
A #inal comparison of all experimental and analytical results for E
‘the various water levels that have been studied fs presented fr Fig. 119.
On this comparison the question has to be raised as to how to identity
‘the ode nunber corresponding to a given Frequency. Secause of the
possibility of missing mages in the experinents, as wiTl be seen in the
following section, we have to be sure that the frequency correlatfons
actually apply to the sase mode nunber. We know that in solution of an
eigenproblen, any errors in the computational procedure affects eigen
vectors nore than eigenvalues; that 1s, the computed eigenvalues tend
to be mate accurate than the computed eigenvectors. However, because
the frequency errors also may be due to inaccurate material properties,
in this study it was necessary to identify the mode nunber by similarity
in mode shape; i.e., if the associated frequencies are different, it
1s presumed to be because the material properties are not modelled
IEa
accurately. In the following section, mode nunbers are identified by
wmode shapes, considering both radial and tangential components; in this
way sone missing modes {n the experiments were discovered. On this
basis the correlations of frequencies associated with the sane node
number are proven to be valid.
5.2.2 Mode Shape Correlations
The correlations of mode shapes are illustrated in Figures 120 and
121. Fig. 120 shows the correlatfons between experiments (36) and numer
ical solutions using the Galerkin Finite Element Hethod for the case when
water level is at 90% of dam hefght. Fig. 121 presents the corelations
for the case when water level is at 85% of the dam height, excep: that the
analysis in this case used the Generalized Westergaard reservoir model.
As mentioned above, the correlations should be done for the same
mode number as identified by sinilar node shapes (because some tangential
components were not measured, only radial components are comparee in the
Figures). Some different modes, as shown by Figures 120(b) and (e), say
have similar radial components of the node shapes; but they are indeed
different modes because displacement shapes on vertical sections are
different. This is far more evident for higher modes. In Figures 120 and
121 we only show the node shapes of the crest, but for the higher modes it
may be necessary also to show the mode shapes of selected sections below
the crest in order to differentiate mode nunbers.
In Fig. 120, we see that the correlations of mode shapes ar very
Good for the first several modes, but some modes were missed in
‘the experimental results. The measured Frequencies associated with
‘those missing modes are listed on the Figures, although the node shapes
Were not measured. The measured and calculated frequencies for the8
similar mode shapes are within Si, indicating the hydrodynamic effects
represented by the addednass matrix computed by Galerkin Finite Elenent
Method are reasonable, while also considering the uncertainties in the
concrete material modelling. It is interesting to see that the 3rd
mode shape may not be easy to obtain in the forced vibration test because
its crest nodes are so close to those of the 2nd mode. In general,
because the behavior of a structure is normally dominated by its Tower
modes, Fig. 120 shows that the hydrodynanic effects represented by the
diagonal ized addednass matrix conputed fron the Galerkin Finite Element
Method are certainly adequate for engineering purposes. Although they ave
not shown, it is worthwhile to mention that the rode shapes given by the
consistent addednass matrix are almost identical to those according to
the diagonalized addednass matrix for both radial and tangential con
ponents, except for mode numbers above the Sth.
Fig. 121 shows that the correlations of mode shapes for 5% water
depth are good up to 4th mode with the 3rd mde missing from excerinental
results. This indicates that the addedmass matrix conputed according to
the Generalized Hestergaard Formula has good relative distribution on the
dan face. However, frequencies of similar mde shapes have errers of up
to 205 thus, the addedmass matrix according to the General izec liester~
gaard Formila averestinates the hydrodynamic effects in magnituce. One
may notice from Figures I21(d) and (e) that the crest mode shares are
similar for modes 4 and 5 of the numerical solution in both radial and
tangential components. If only the crest made shapes were coupered in
this case they would appear to represent the same mode number, Gut they
are truly different mades because the vertical crown section shepes (not shown)
are completely different from each other, Nhen we examine correlations of30
node shapes and their associated frequencies of mode 9, in Fig. I21(1)
we can only say that this is a beautiful note in the melody played by
Generalized Hestergaard Formula.
Fron the correlation studies on frequencies and mde shapes present
in this section, ft may be concluded that the hydrodynamic effects
represented by the diagonalized addednass matrix from the Galerkin Finite
Element Method can be considered as a good approximation of the true
behavior. Nevertheless, the Generalized Westergaard approxinate solution
stil] can be useful for crude preliminary studies.
5.3 Stress Response Representations
Stresses calculated in a structure subjected to specified
Joadings provide the basis for the design of the structure. According
‘to the assumed mechanical properties of the structural material, the
structure is designed so that when subjected to the design loads, it won't
develop excessive stresses that will lead to damage. Because of the
potential disaster associated with the failure of a dam, it is very
inportant to analyze the stresses in the dam accurately. when it is
subjected to the maximum expected loadings. For arch dans, tho critical
stresses may be represented conveniently in terms of normal components
‘in the horizontal and vertical directions, usually called arch and
cantilever stresses
Three intensity loads of earthquakes often are used in the design
of a dam:
(1) Maximum Credible Earthquake (HCE): This is the maximum possible
earthquake that might occur at the site of the dan. When subjected
to the NCE, the dam may suffer damages, but must retain the
reservoir,40
(2) Design Base Earthquake (DBE!
This is an earthquake intensity
corresponding to a return period of 100 years, (Fig. 125(b) and (c)),
tthe expected life of the structure. hen subjected to the DBE,
‘the dam should sustain only repairable damages, and its equipment
should be able to operate normally.
(3) Operation Base Earthquake (08E): This is an earthquake intensity
corresponding to a return period of 25 years. (Fig. 125(b! and (c))+
it is very likely to occur during the life of the structure. hen
subjected to the OBE, the dam should not sustain any damage.
The Design Base Earthquake was used for this study. Figures 122
to I24 present the stress responses of Tech Dam when subjected to
‘various types of loadings. In the figures, SIGXx denotes horizortal normal
stress (arch stress) while SIGYY denotes vertical nomal stress
(cantilever stress). ALI cases presented in Figures 122 to 12 are
discussed in the following sections.
5.3.1 Stress Response to Static Loadings
Because no tenperature change data was available, the only static
Joads considered were hydrostatic and gravity. Figs.122 shows the
static stress results in the form of contour plots; a1l the tension
zones are shaded. Four separate plots are presented (Figs.122(e) (b),
(c), and (4)), showing arch and cantilever stresses on the upstream
and downstream faces. Obviously, all the tensile stresses, either
cantilever or arch, due to the static loads, are well below the tensile
strength of the concrete (which is assumed to be 500 psi here, see
Section 5.1)
The compressive stresses are also well below the material strength.
The static compressive cantilever stresses shown on the downstream face.a
at the foot of the dan in Fig. 122(d) while it 1s subjected to static
Joads are beneficial because they compensate for dynamic tensile stresses
which may be quite high in this region.
5.3.2 Stress Response to Combined Static Loads and Design Base
Earthquake (OBE). Excitation
The response spectra for the DBE are shown in Fig. J25(a). From
Fig. 119, the fundanental frequency of Teché Arch Dam, with 90%
reservoir depth, is approximately 2.7 Hz according ta finite elenent
solution with diagonalized addedmass matrix, it corresponds to approx
‘imately 2 g of the pseudoacceleration intensity on the response svectra
for the DBE in Fig. 125(a). If we consider pseudoacceleration intensity of
0.2 g and above as significant (10% of the intensity of fundamental node) ,
‘then the DBE has important intensity associated with excitation
frequencies up to 7 Hz, which wiTl excite the first 8 modes of the Techi
Dam considering hydrodynamic effects (see Fig. 119). Thus, the first
© modes were included in the response spactrum dynamic analyzic. Secause
‘the dynante stresses resulting from the response spectrum dynamic
analysis are in absolute value, we present the stress response due to
conbined static and dynamic loadings in terms of maximum and minimus
stress envelopes.
The stress envelopes have minimum values obtained by subtracting
response spectrum dynanic stresses from corresponding static stresses,
‘and the maximum values are evaluated by adding response spectrum dynamic
Stresses to the corresponding static stresses. Fig. I23 illustrates the
stress envelope contours on the upstream face while Fig. 124 presents
the stress envelope contours on the downstream face, AS before, SIGKX
indicates arch stresses and S1G¥¥ represents cantilever stresses:
and here, minimun stresses show the largest possible compressive stressesa
whereas maxinun stresses show the largest possible tensile stresses.
Fron Figures 123 (a},(b),(e) and (f), it fs evident that tre
macinum arch and cantilever compressive stresses on the upstream face
due either to upstreandowstrean excitations or crosscanyon excitations,
are all well below the conpressive strength of the concrete, which is
about 5000 psi.
Similarly, Figures 123(d) and (h) show that the maxtmun tensile
cantilever stresses are well below the tensile strength of the material;
hence, no cantilever cracking will occur. However, in Figures I23(c)
‘and (g), it is clear that the tensile arch stresses indicated near the
rest ave beyond the tensile strength of the material; thus cracks are
‘expected to be formed there. But it must be renenbered that vertical
contraction joints were built into the structures; thus the tensile
arch stresses will merely open these Joints. Therefore, no cracking
4s expected from these indicated tensite stresses in the crest region
‘and the dam will be able to withstand the earthquake without significant
Gamage.
Figures 124(a) to (h) present. the corresponding stress results
for the downstream face. Figs. 124(c) and (g) indicate that tensile
arch stresses on the dovnstrean face also exceed the tensile strength
of the material, and again it may be assumed that the contraction joints
\iT1 open to release the tensile arch stresses. Figs. 124(d) and (h)
show that cantilever cracking also is unlikely on the downstream face,
and Figs. 124(a), (b), (e) and (F) show that the conpressive stresses
fare wel] within the compressive strength of the concrete.
Thus, the discussions in this section have shown that Techt Oam
wil1 not have significant damage when it is subjected to hydrostatic3
‘oad and gravity Toad combined with the Design Sase Earthquake
excitations, taking account of incompressible hydrodynantc effects and
foundation flexibility. However, one other aspect of the dynantc
behavior should be considered: the nonlinear response mechantsn
associated with the opening of the contraction Joints due to the action
of dynante tensile arch stresses. It is evident that such Joint opening
on one face of the arch dam wiTl be acconpanied by the modification of
the arch stresses on the other face (and vice versa), and also the
possibiTity of changes in the state of stresses in the cantilever direction.
This type of nonlinear response mechanism is the subject of the second
part of this thesis.6. CONCLUSIONS AND RENARKS.
The hydrodynapic effects represented by an addedness matrix
associated with incompressible fluid reservoir of arch dans are reported
here. Two basically different conputational procedures, nanely
Generalized Mestergaard Formula and Galerkin Finite Elenent Method, are
described in detail, and pressure solutions obtained with each are
compared. Rigorous vibration frequency and mode shape analyses are
carried out, and based on their comparisons with eld measurenents,
a best suitable standard procedure is proposed, i.e., Galerkin Finite
Element Method with diagonal ized addednass matrix
safety evaluations of Techi Dan, that subjected to either static
oads alone or to conbined static and dynamic earthquake Toads, are
discussed. [t 4s shown that Techi sm won't sustain major darages
ue to these loading conditions, but that minor darages might occur near
the crest spitlway. This condition may need further study that includes
Joint opening nonlinear response, it nevertheless should not prohibit
‘the normal operation of the gates.
It was found that hydrodynanic effects of an incompressible liquid
reservoir were represented adequately by e reservoir model that extends
‘in upstream direction 3 tines the height of the dan. This greatly
reduces the cost of the finite elenent analysis of the reservoir inter
action, but the corresponding conclusion may not apply to compressible
water. Inclusion of the water compressibility greatly complicates the
reservoir analysis, and {t 1s not known at present whether or not
the results neglecting water compressibility conservative. Further
veserach {5 needed to verify the significance of the influence of water
compressibility on the rea? time response of an arch dem, especially when
superposition procedure 4s not valid.6
However, the results of the analyses and of the correlations with
field measurements contained in this report have shown that the hydro
dynanic effects represented by addedmass of incompressible water should
bbe satisfactory for engineering purpose in the analysis and design of
an arch dom
In view of nontinear dynamic response analysis of arch dans,
diagonalization of the full addeéonass matrix was deened necessary to 
reduce the computational cost. This diagonalization has neverthetess
destroyed the coupling effectsof addednass, whether these coupling effects
are important or not require further research.46
APPENDIX A: COMPUTATIONS OF NORMAL DIRECTION cOSINES
The normal direction cosines at any point on a curvilinear surface,
as on the 2D interface elenents, can be found (23) from the intrinsic
property of finite elenent interpolation functions which use the natural
coordinates as curvilinear coordinates.
Figures 14(b) and (c) show 2 kinds of possible 20 interface
element (20 in natural coordinates, but 3D in RCC space), where points
pare regular points and point q is 2 degenerate corner point. The
‘computations of normal direction cosines for points p are different from
that for point 9.
Normal Direction Cosines for a Regular Point p
Fron basic finite elenent property, we have,
rs) (aay
where
Xj = 2% Yy B+ the coordinates of node i of tha atamant
The unit normal vector at point p, n, can be found as,
(A.2)
where
2
3: Fhe (r,s)X;. vector tangent to rcurve at point >
PB any eis met target to secure at pont»
or, in finite elenent formation,a
i i k
Cp Tay FEM Mle EM alma, 2m rele, .)
EM sles EM sls Fay clelay
Thus,
; nr
By = THT (yy Pt yp 3+ ayy (aay
a re
exe
and
5 2) «2, v8, «eB
Bx B= O82 ky (as)
A.2: Normal oirection Cosines for a Degenerate Corner Potnt
2%
Because 0 at the degenerate corner point q (Fig. I4(b)), we
cannot find ng as above.
4
Tnstead,the unit normal vector n, can be found most conveniently
as follows:
By 7 EMC) (.6)
(a.7)
and
{A.8)
‘The sequence of the crossproduct 1s expressed in Eq. (A.?) according to
‘the convention that the connectivity of the elenent is defined in the
counterclockwise direction (Appendix 8). In finite elenent formula:ion,48
Eq. (A.7) becomes
i i é
EM gl TOs), ENG gMOoshyg FM) c(1.0.8)25  (8.9)
BM TOF, FM 6(1.0,8)94
and,
(A.10)
(AT)43
APPENDIX Bs FINITE ELEMENT INTERPOLATION FUNCTIONS
‘AND THEIR DERIVATIVES (21,22)
Variable node finite elements are used in the analysis described
in this reports the convention of their connectivities is shown in
Fig. 15.
Interface Elenents (Fig. 15(a)
For interface elements, 20 in natural coordinates and 3D in
RCC space, their interpolation functions and the derivatives of their
‘interpolation functfons with respect to the natural coordinates are as
foVows
1
1725+ 5)
raat a) 4
558
where ty + 0 $F node k fs not factuded
5 * Gl rar, 6(5.5,)
88.0) = } (1488). B= £1, 8 = rs
6(8,0,)
and derivatives
Sg =O pltorIGl508,) + Gran )Geg(545,)
1
Bg (88) = J Bs 8 = 41
(38) = 28 5 8, = 0
For degenerate elenents, if several nodes are degenerated into one node,
their associated interpolation functions also have to be degenerated
into one function, and similarly the derivatives of the interpolation50
functions.
B.2 30 Fluid Elements (Fig. 15(b
For 30 fluid elenents, the suggested interpolation func:
ns and
the derivatives of the interpolation functions with respect to natural
coordinates are as follows:
1
Nyt > 7 Gy # Bg * 7)
1
2 (ier * Sian? Siete) Ta
pag, eee
Ng = %  2 (S13 * Se * G7)
5° 2 Cpa * Syeat Sora) 5 68
Ny 7 k= 920
tq 7 Orff node m 15 not included
Gq = SlrorQ)6U5,5_)6( tt)
G8) =F (1+ 88) oy ET, Be SAE
6(e08,) = 182 Bq 20
and derivatives
Gasp 7 SogltsrgiOlSs5q)6(t, ty) + Glr, rylGy (5. 5q )ECEaty) +
G(r, Fg) B(5, 5q)5 g(t t)51 I
For degenerate elements, if several nodes are degenerated into one
node, their associated interpolation functions also have to be degenerated
into one function, and similarly for the derivatives of the internolation
functions.52
APPENDIX C: VARIABLE WATER LEVELS
When the dam is discretized in such a way that its element bound=
aries on the upstrean interface do not match the boundaries of the
fluid elements at the interface, the addednass matrix found ty the
rnethods presented in Chapter 3 cannot be assenbled directly with the
concrete mass matrix of the dam, This is because different nodal points
or degrees of freedom apply to the water and the concrete elenents.
This is most likely to occur when the dam fs discretized so that the
water surface fs Tocated between the horizontal boundaries of the
dan elenent:
This problem presents no difficulty for the addednass matrix
forned by Generalized Westergaard Formla. But in using Galerkin
Finite Element Method, the difficulty arises in the integration of
Eq. (3.16):
t
gy Ha af) al 3.6)
4
1A lho x a0)
sane donin, rather the domain of 1") is fncluded inthe donsin of gl!)
(Fig. 126). The integration only can be carried out for in the domain
(4) ; (4), pecause 1") are discontinue
of ul") and cannot be in the domain of 9"), pecause 1") are discontin
ous functions in the domain of gf!) (they vanish above the water Tevel,
see Fig. 126). Therefore, Eq. (3.16) can be written, in this case,
as follows:
t
nfs 4 ars) a6) (ras) of eEGrs), nbesy) al) (6.1)
s
or, in the form of quadrature integrations:58
t
i Gi 4 (i)
afd Bem PCr; 8 9 Mp) AM Eley s Iantry s DIT 51]
(C.2)
AN] the terms in &q. (C.2) are defined similarly to those in Eq. (4.6).
The task 45 tn eveluating vues of of"? at Antegration pots
(ryosy)e wintTe al") are functions of (Eyn). Naturally, we have to Find
coordinates (5,n) at points where (r;,5,) locates, and they are only
related through the RCC coordinates of the nodes that associate with each
clenent respectively.
Therefore, Firstly we denote the RCC coordinates of fluid eTenent
nodes as x, y. 2 and RCC coordinates of concrete elements as X.Y, Z
Then, we can have the expression:
(c.3)
where
xcoordinate of node n of the fluid element at the interface
X = vector of Xcoordinates of nodal points of concrete elenents
at interface
15 = vector of shape functions associated with nodes of the concrete
elements.
X, and X are known, and if we position x, properly, that is, at the
Vocations where & values are 
+0) #1, then, Eq. (C.3) will reduce to
2 simple form of a quadratic equation with n as unknown. Then if we
solve n, for the corresponding xy, knowing 1 < nq <1, we can find the
coordinates (Eyn) for all nodes of the fluid elenent on the interface.
Furthermore, the integration points (1,454) has an xcoordinate, x,, givenby,
Nelrye5Q) (c.a)
lihere x = vector of xcoordinates of nodal points of the fluid element
at interface (1.€., %'5)
relating Eq. (C4) to &q. (C23), we have,
ak (c.5)
vihere y= collection of 4, tn Eq. (C.3) with known (
Documente similare cu EERC8209 FluidStructure Interactions Added Mass Computations for Incompressible Fluid
Mai multe de la BMWrider1453
Populare în Components Of A Chemical Plant
Mult mai mult decât documente.
Descoperiți tot ce are Scribd de oferit, inclusiv cărți și cărți audio de la editori majori.
Anulați oricând.