Nuclear Engineering and Design 48 (1978) 377-392 North-Holland Publishing Company

377

THE SPRING METHOD FOR EMBEDDED FOUNDATIONS * Eduardo KAUSEL Stone and Webster Engineering Corporation, 245 Summer Street, Boston, Massachusetts 02107, USA Robert V. WHITMAN Department of Ovil Engineering, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA Joseph P. MORRAY and Farid ELSABEE Stone and Webster Engineering Corporation, 245 Summer Street, Boston, Massachusetts 02107, USA Received 1 February 1978 The paper presents simplified rules to account for embedment and soil layering in the soil-structure interaction problem, to be used in dynamic analyses. The relationship between the spring method, and a direct solution (in which both soil and structure are modeled with finite elements and linear members) is first presented. It is shown that for consistency of the results obtained with the two solution methods, the spring method should be performed in three steps. The first two steps require, in general, finite element methods, which would make the procedure unattractive. It is shown, however, that good approximations can be obtained, on the basis of one-dimensional wave propagation theory for the solution of step 1, and correction factors modifying for embedment the corresponding springs of a surface footing on a layered stratum, for the solution of step 2. Use of these rules not only provides remarkable agreement with the results obtained from a full finite element analysis, but results in substantial savings of computer execution and storage requirements. This frees the engineer to perform extensive studies, varying the input properties over a wide range to account for uncertainties, in particular with respect to the soil properties.

1. Introduction In recent years a significant effort has been directed towards formulating engineering solutions to the problems of vibration of foundations and seismic response of buildings. The aim has been to develop practical rules and formulae to be used in dynamic analyses o f structural systems, accounting for s o i l - s t r u c t u r e interaction [10,28]. This problem is o f special interest in the seismic analysis and design o f massive buildings such as nuclear containment structures. The subgrade rigidity, or the lack of it, is often accounted for in analyses by a set o f springs and dashpots, representing the soil stiffnesses. The spring constants used are derived for the greater part from theo* Expanded version of Paper K2/6 presented at the 4th International Conference on Structural Mechanics in Reactor Technology, San Francisco, California, 15-19 August 1977.

retical solutions for surface footings on ideally elastic, homogeneous, isotropic halfspaces. While these solutions apply to idealized cases o f rigid plate foundations underlain by elastic halfspaces, they found wide application in machine foundation design. Building foundations, however, are usually buried to some extent beneath the surface of the ground. This embedment has in may cases considerable effect on the dynamic response of the structure, both in terms of relative frequency contents and amplitude o f the resulting motions. Because of the complicated boundary conditions that must be satisfied in a theoretical formulation, rigorous analytical solutions for embedded foundations are non-existent at present. It is for this reason that numerical (finite element), experimental and approximate analytical techniques are currently being used to provide a solution to the problem at hand for these complicated geometries. An awareness of the effects associated with embed-

378

E. Kausel et al. / Spring method for embedded foundations

ment [2,13,17,22,23,29,30], coupled with the availability of numerical solutions and the lack of rigorous solutions, has been the basis in recent times for discrediting the spring method as a tool for analysis, particularly in the nuclear power industry. This is unfortunate, since the spring method constitutes a technique extremely attractive to the engineer for its reliability, economy and simplicity. The detraction of the spring method has been argued by some researchers on the basis of comparisons between the classical halfspace spring method, and the more involved finite element solutions. Many of these comparisons are not meaningful, since they are based on inappropriate values for both the 'spring constant' and the 'support motion'. In fact, the spring method and finite element solutions can be shown to be mathematically equivalent; if they are classified as different, it is because of inconsistencies in their implementation. It is the purpose of this paper to show the relationship between a more general spring method and the solutions provided by direct finite element procedures. At the same time, practical rules for use in dynamic analyses will be presented and applied to particular examples.

reference system. The solution of this equation is equivalent to the solution of the two matrix equations [13] M I [ ) 1 +C};" 1 + K Y 1 = 0 and
MY2 + C Y 2 + K Y 2 = -M2~rl , (2)

(3)

2. The basic superposition theorem With reference to fig. 1, we assume that the general equations of motion of a structure-foundation system are given by the matrix equation
MU+CY + K Y = O,

(1)

where M, C, K are the system mass, damping and stiffness matrices; U and Y are the absolute and relative displacement with respect to some general ground
M S L S STRUCTURE A SE S

\\\\\\\\\\\\\
COMPLETE SOLUTION KINEMATIC INTERACTION DYNAMIC INTERACTION

Fig. 1. Superposition theorem.

where U1 = Y1 + Ug, U = U1 + Y2, Y = Y1 + Y2, and M = M 1 + M 2. M 1 excludes the mass of the structure, while 342 excludes the mass of the soil. Ug is some generalized ground motion vector. The equivalence of (2) with (1) is demonstrated by simple addition. In eq. (2) the response of the massless structure is found first and will be referred to as the kinematic interaction. The results of this step are then used in eq. (3), which shall define the inertial interaction, and which is solved by application of fictitious inertia forces applied to the structure alone. In the solution of the second step, it is irrelevant whether the soil is modeled with finite elements, or equivalently, with a (far-coupled) matrix of stiffness functions modeling the subgrade, and defined at the soil-structure interface. These stiffness functions can be regarded as resulting from a dynamic condensation of all the degrees of freedom in the soil (a frequency domain solution is implied). For the particular situation where the combination foundation-structure is very rigid, it becomes legitimate to replace the matrix of stiffness functions by the overall vertical, torsional, rocking and swaying stiffness functions, i.e. by frequency dependent 'springs' and 'dashpots'. It follows also that the solution of the kinematic interaction phase is completely defined by the rotations and translations of the massless structure, which moves as a rigid body. Hence, one can replace the massless structure in eq. (2) by a rigid massless foundation, subjected to the same ground excitation as the original system. Also, a more careful examination of eq. (3) will show that the solution Y2 can be regarded as a vector of displacements relative to a fictitious support, while the rigid body translations and rotations of the massless foundations in eq. (2) are the equivalent support motion. Provided that the assumption of rigid foundation is pertinent, it is therefore valid to break the solution into three steps [13] (see also fig. 2). (1) Determination of the motion of the massless

E. Kausel et al. / Spring method for embedded foundations

KINEMATIC INTERACTION SUBGRADE IMPEDANCES

379

_ (kxx kx, ? K-lk,..,l

AFV
+

,\\\\\\\\\\\\\
YG

~\\\\\\\\\\\\\

TOTAL SOLUTION

Fig. 2. The three-step solution.

rigid foundation, when subjected to the same input motion as the total solution. This is the solution of eq. (2). For an embedded foundation it will yield, in general, both translations and rotations. (2) Determination of the frequency dependent subgrade stiffnesses for the relevant degrees of freedom. This step yields the so-called soil 'springs'. (3) Computation of the response of the real structure supported on frequency dependent soil springs, and subjected at the base of these springs to the motion computed in (1). Notice that the only approximation involved in this approach concerns the deformability of the structural foundation. If this foundation were rigid, the solution of this procedure would be identical to that of the direct (or one-pass) approach (assuming, of course, consistent definitions of the motion and the same numerical procedures). The superposition principle is valid only for a linear system. While the modulus and damping of the soil are strain dependent, studies [ 16] have shown that most of the non-linearity occurs as a result of the earthquake motion, and not as a result of soft-structure interaction. Thus, the soil properties consistent with the levels of strain in the free field (i.e. before the structure has been built) may be used in steps 1 and 2 without further modification to account for the additional strains imposed by the structure. The first two steps require, in general, finite element methods, and thus it might appear that the three-

step method has no advantage as compared to considering both kinematic and inertial interaction together in a single step. However, reasonable approximations can be obtained on the basis of one-dimensional wave propagation theory for the solution of step 1, and correction factors modifying the corresponding springs of a surface footing on a layered stratum to take into account embedment effects, for the solution of step 2. Use of these rules not only provides remarkable agreement with the results obtained from a full finite element analysis, but results in substantial savings of computer execution and storage requirements. This frees the engineer to perform extensive studies, varying the input properties over a wide range to account for uncertainties, in particular with respect to the soil properties. Also, deviations from axial symmetry may be introduced into the model of the structure in step 3 (which means that a torsional 'spring' and 'dashpot' must be evaluated in step 2) and the effect of changes in the mass and stiffness of the structure be evaluated without having to rerun and entire analysis. 3. The substructure t h e o r e m A useful and illustrative alternative formulation of the superposition theorem can be obtained using the substructure technique. This approach leads to a set of equations relating the free field solution with fictitious forces applied at the foundation of the structure [4,9].

380
STRUCTURE

E. Kausel et al. / Spring method Jbr embedded foundations (c) subgrade, free field solution

s?,,

i~];;

Kff Kfg

Kfr] [Ut*t [P~]

Lb
O ~r g r

/gg K:ift;if=lO f.
Krg

(0}

2~

L~
b) FREE FIELD PROBLEM

O)

SOIL STRUCTURE INTERACTION PROBLEM

Fig. 3. Substructure theorem.

/ ss

Consider a finite element discretization of the soil-structure system as shown in fig. 3a. The soil and the structure have been separated, and equilibrium enforced by application of the internal forces Pb, Pf. The model is subjected to an arbitrary excitation (in the form of specified displacements) along the boundary. Fig. 3b depicts the discretization of the 'free field' problem, that is, the situation that would exist if no structure had been built. The 'structure' in this case represents the excavated portion of soil. For a solution in the frequency domain, the matrix equations relating forces and displacements are of the form ( - ~ 2 M + i~2C + K) U = P, where M is the mass matrix, C is the damping matrix, and K the stiffness matrix. P and U are the force and displacement vectors, while ~2 is the driving frequency. For the sake of simplicity, the frequency dependent complex submatrices of the dynamic stiffness matrix K d = K + i~C - ~2M will be denoted by subscripted K. The force-displacement relationship for the various substructures shown in fig. 3a,b are then: (a) structure:

The subindices above refer to the following: s, for the nodes of the structure, excluding the soil-structure interface; b, for the nodes of the structure along the interface; f, for the nodes of the soil along the same interface; g, for the nodes of the soil, excluding the interface and the boundaries; and r, for the nodes along the boundary. The asterisk refers to the free field solution. Notice that both tile 'free field' problem and the soil-structure interaction problem are subjected to the same excitation Ur. However, in general Pr 4: Pr: unless tile boundary is far away from the structure. Subtracting (6) from (5) leads to Kff Kfg K f r ] [ g f - - U ~ )
Pf-P~t

KI: KggK~itUg U;t-(0

Condensing the matrix equation above then gives
X ( Uf u ; ) = & - p~ , (s )

where X = X(~) is the subgrade impedance matrix. On the other hand, equilibrium and compatibility require that Pf = P b , P~ = --Pt~, Uf = Ub, U~ = UI~ so that
X(Ub - Ub) = --Pb -- P~ = Pb -- Pb

(9)

Substitution of (9) into (4) then yields

Kbs Kbb + X

XU~ + P~ '

/ {0/
= Pb

{Kbs KbbJ Ub

(4)

(b) subgrade, including soil-structure interaction

Kff Kfg Kfrltef I IPfl

Krf Krg Krr)

ggfgggKgr~ :}=i~;

(5)

which states that the solution to the soil-structure interaction problem can be obtained (for the structure) by application of fictitious forces Pb = P~ + XU~ at the foundation-soil interface. P~ and U~ are once again the 'free field' forces and displacements along this interface, and can easily be computed for a number of cases, e.g. layered soil systems subjected to seismic waves propagating at any arbitrary angle [ 11, 27, 31 ]. However, the subgrade impedance matrix X is not easily obtainable, except for the particular case of

E. Kausel et al. / Spring method for embedded foundations

381

zero embedment [3,8,32]. For ideally rigid embedded foundations, the components of the displacement vector Ub can be expressed in terms of the rigid body displacements and rotations of the foundation, defined at some point (for instance, the centroid of the base):
Ub = TUo,

(1 1)

where Tis the rigid body transformation matrix, and Uo contains the rigid body displacements and rotations. Nodal forces Pb and foundation forces P0 are similarly related:
Po = TTpb = TT(p~ + X U ~ ) .

(12)

For a plane condition, U0 = (u, o, ~b}T ,

Po = (H, V, 114}T .

Combining eqs. (10), (11) and (12) finally gives 0 [ TTKbs TT(Kbb + X) (13) Inspection of eq. (13) reveals that
K o = TTXT

(14)

is the subgrade stiffness matrix containing the (generally coupled, frequency dependent) soil 'springs'. Also, TT(p~ + XU~) is a vector containing the overall forces and moments of the fictitious forces P~ + XU~. Unfortunately, it is not possible to express TTXU~ in terms of the subgrade stiffness matrix K o in the general case. However, for foundations with no embedment, and under the additional assumption of vertically propagating waves, such a relation is indeed possible. 4. Dynamic response of embedded foundations The following statements can now be made, concerning the dynamic response of embedded foundation systems in layered, viscoelastic halfspaces, in view of eqs. (2), (3) and (13). (i) Kinematic interaction (wave passage) effects are always present in embedded foundations, whether the seismic waves propagate vertically or not. An assessment of these effects can be made either directly (solving eq. (2)), or using eq. (13). The former is as expensive as the direct solution, of eq. (1), and therefore, not attractive. The latter procedure requires knowl-

edge of the subgrade impedance matrix X, which generally is not readily available. However, i f X is known, and the foundation is rigid, the response of the system can easily be obtained with eq. (13) by application of the fictitious forces Po, or alternatively, through a 'support' motion which follows from Ko 1 TT(p~ + XUf*). If the foundation is not rigid, no such equivalent support motion exists, and eq. (1) or (10) is to be used. (ii) For surface foundations (no embedment), kinematic interaction effects vanish only if the seismic waves propagate vertically. For an arbitrary seismic environment, wave passage effects can efficiently be handled with the aid of eq. (13), as severa ! algorithms are available for the determination of the impedance matrix X [3,8,32]. Again, if the foundation is not rigid (for instance, a dam), no equivalent support motion can be defined for arbitrary spatial variations of the seismic motion. (iii) Whenever wave passage effects are present, their determination requires knowledge of the subgrade impedance matrix X, or alternatively, a solution of eq. (2), which is as elaborate as the direct solution of eq. (1). The availability of the subgrade stiffness matrix Ko is not sufficient for this purpose. Note, however, that the subgrade impedance matrix X and the subgrade stiffness matrix K o are independent of the seismic environment (except for possible non-linear effects, which are not being considered here). Therefore, parametric studies can easily be conducted once any of these matrices is available. (iv) Many comparisons performed in the past between direct finite element and 'lumped spring' solutions for soil-structure interaction were inconsistent and gave discrepant results because (a) the spring 'constants' used did not account for embedment, and (b) the support motion used did not account for kinematic interaction; the free field motion was applied instead directly under the subgrade springs.

5. Approximate solution for circular, embedded foundations The solutions presented in the following sections have been obtained with a three-dimensional axisymmetric finite element formulation. A fundamental feature of the program used is the exact representation of the model boundary (fig. 4) which separates the

382

E. Kausel et al. / Spring method for embedded foundations

Fig. 4. Finite element idealization of massless foundation. finite element region from the semi-infinite continuum (the free field). This consistent energy transmitting boundary was developed for the plane strain case by Waas and Lysmer [20,34], and was extended to the three-dimensional case by the first author [12,14]. In essence, it can be regarded as a virtual extension of the finite element mesh to infinity, and can be placed without loss of accuracy immediately next to the foundation.

6. Approximations to the kinematic interaction

Both the direct solution and the first phase of the spring method start normally with the specification of
MASSLESS FOUNDATION

a uniform motion along the bottom boundary. They assume, therefore, that the motion is known at bedrock. Furthermore, since the motion is uniform along the rock-soil interface, the validity of one-dimensional amplification theory is implied as a fundamental assumption. In the following section it will be assumed that the motion which the ground experiences before any structure (or hypothetical massless foundation) has been built, can be described by means of onedimensional wave propagation theory. The control motion, i.e. the specified earthquake record, will be assumed to take place at the free surface in the 'free field'. Motions at other points in the free field can be obtained by the so-called 'deconvolution' process which makes use of one-dimensional wave propagation theory. Since one-dimensional wave amplification theory is well understood [26,27], a description of this theory will be omitted here. During a seismic excitation, an embedded, massless, circular, rigid foundation will experience two planesymmetric displacement modes. The first one is a translational motion, which contributes the major portion of the excitation to the structure in the third step of the analysis; it is caused by the overall translation of the subgrade. The second type is a rotational motion, caused mainly by the shearing stresses developed along the sidewall-soil interface. These shearing stresses result from the differential horizontal translation of the soil in the embedment region (fig. 5) producing a 'pseudorotation' of the soil. The foundation, on the other hand, being rigid, cannot defbrm in this manner, and as a result it will rotate. This rotation is mainly resisted by the subgrade stiffness acting on the mat and sidewalls. The third author investigated the kinematic interaction problem in a parametric study [21] using a suite of embedment ratios E/R and stratum ratios H/R, covering a range of values typically found in nuclear
F R E E FIELO UA= I (HARMONIC1

I' / f / /

/' / / = uA - U B E

,~\~~.\\\\\\\~\\\\\\\,

~1 t

/L-.L. . . . . . . . . s--t i l/ i / F-~ ..... -# -~, / s~ / // / / /' / // / / / / /

~a

BEDRO CK

/I i/

//I/z

Fig. 5. Kinematic interaction problem.

E. Kausd et al. /Spring method for embedded foundations

reactor design. He found that the translation of the massless foundation is similar to the translation of the free field at the foundation level, except that the Fourier spectrum of the former is smoother: it does not show the pronounced absence of frequency components at the resonant shear beam frequencies of the soil column in the embedment region (for uniform soil properties above the foundation level, these would be given by fr = (1 + 2n) Cs/4E, where Cs is the shear wave velocity, E is the depth of embedment, and n = 0, 1, 2 ..... ). On the other hand, the rotation experienced by the foundation can be related to the pseudorotation in the free field, i.e. to the differential displacement between the free surface and the foundation level, divided by the embedment depth. In the following sections a summary of the parametric study will be presented, and rules to approximate the kinematic interaction will be proposed. More details can be found in ref. [21]. With reference to fig. 5, a unit harmonic displacement was specified at the free surface, and deconvolved to bedrock. Using the finite element program, frequency dependent transfer functions u, for the displacement and rotation of the massless foundation relative to the motion at the free surface were determined. Similarly, the frequency dependent transfer

383

functions for the displacement in the free field at the elevation of the foundation, and for the pseudo-rotation of the free field ~bB(~)=

uA--UB_I--u B
E E

'

(lS)

were computed. Typical results are shown in fig. 6. The transfer functions u and Un, @and @B were then compared for a range of embedment and stratum depth values, and rules to approximate these functions were suggested [21 ]. A simplified version of these rules is given below. Let F(I2) be the Fourier transform of the acceleration at the free surface in the free field (in most cases, the design earthquake). The translation and rotation of the massless, rigid foundation are then given approximately by /J = IFT

IF(a)L c \(2T]~l F o~f nf J ' s

t F ( ~ ) [0.453],

iff<,O.7fn,
(16a) if f > 0.7 fn;

= IFT ~F(~) [0.257/R] (~b"is positive clockwise)

2fn/ iff>fn,
(16b)

10
08

t.~.~. N

\ %,
~\

/" /.i i

/"

'~,

\ \

//

/'

/'I"--'~'-~,~.-FREE FIELD, FOUNDATION "~ LEVEL

\.\
'\

06 04 O2
'~ i

/__ ~
"~ . i ' / / /

/ ... ..
~ ~

~\
.

./
~'

~.\
-"
EXACT

.,/

TRANSLATION

17

81

,~

f(H~)

(R@)A
06

~
I ~ ' -' . . . . . . . . . . . .
I 1 I I I I 1 I

-(~--i--;D:7o0 SEC ";/ FT,

~l ...... 1 ~:'~" ..... 2
I IO
l I I
. . . .

04

02

IL

11

12

13

14

15

f(Hz)

Fig. 6. Motion of massless foundation, transfer functions (absolute value).

384

E. Kausel et al. / Spring method Jbr embedded foundations

-g

07

g
t.u o

06

05

04

OI

02

03

04

05

06 (SEC}

07

PER(OD

Fig. 7. Motion of massless foundation, response spectra, 1% oscillator damping.

where IFT stands for inverse Fourier transformation; fn is the fundamental shear beam frequency of the embedment region (for uniform soil properties in the embedment region, this value is given by fn = Cs/4E with Cs being the shear wave velocity, and E the depth of embedment). The expression in square brackets describes an approximation to the transfer functions for the translation and rotation of the massless foundation. For surface footings, E = 0, fn = 0% and cos 0 = 1 ; it follows that no kinematic interaction takes place for this case. The procedure described yields satisfactory results for a wide range of embedment ratios, see for instance fig. 7 which shows the response spectra for the case depicted in fig. 6. It should be noted, however, that the rotational component is sensitive to the lateral soil conditions, and particularly to the flexibility of the lateral walls. For flexible sidewalls the actual rotation is significantly smaller, and in the extreme case of no sidewalls, the rotation even changes sign.

7. Approximations for the stiffness functions

As with the kinematic interaction, the values of the subgrade stiffness function (impedance functions) depend only on the geometric configuration of the foundation and on the properties of the founding soil. These functions can be evaluated using analytical, experimental or numerical methods.

Before turning to numerical methods, the engineer should consider whether the desired functions might be available in the literature. Analytic solutions to many problems have been obtained; many of these are available in useful form in the book by Richart et al. [24], Erden [6] provides a very extensive bibliography of other available solutions as well as a description of the conditions assumed for each. Luco and Westmann [ 18] investigated the vertical, horizontal, torsional, and rocking modes of excitation of a circular plate on an elastic halfspace for a wide range of frequencies. Veletsos and Wei [33] presented numerical results for compliance and stiffness functions for the rocking and sliding modes of a circular disk on an elastic halfspace. Luco [19] extended his previous formulation to solve the case of circular foundations on layered halfspaces, and presented results for a single layer underlain by elastic rock. Gazetas [7] developed an analytical solution for strip footings and rectangular foundations on layered halfspaces or layered strata, for the rocking, sliding, vertical, and torsional modes. All these solutions can be mathematically exact, although sometimes small approximations are introduced (smooth versus rough footing). While these solutions apply only to surface foundations, studies with the finite element method [15] seem to indicate that the frequency variation of these functions is not very different for embedded foundations. The main effect of embedment seems to be an overall increase in the stiffness values over the entire frequency range. This would allow use of the finite element techniques just for the static case, with considerable savings. It would also seem that good approximate relationships can be obtained to account for the increase in stiffness in terms of two embedment ratios: depth of embedment to depth of stratum, and depth of embedment to radius of foundation. Experimental results are less easily obtained than theoretical and analytic ones, so the literature is less extensive. Erden [6] describes results from a broad experimental program, and also gives a comprehensive bibliography of other experimental work. The results presented in this paper are based on parametric studies performed by the fourth writer to determine approximate expressions for stiffness functions of circular, embedded foundations[5]. For each particular geometry, the static values were evaluated with two or three meshes (a fine, a standard, and a

E. Kausel et al. / Spring method for embedded foundations

385

coarse mesh), and the results were corrected for mesh size error in a manner similar to that described in ref. [ 15]. The study was limited to the coupled horizontal translation and rotation (rocking-swaying) of a rigid, circular embedded foundation in a homogeneous stratum. For this particular case, the force displacement relationship can be written

K ~ ' = 3(1 - v) [1

611/
(19)

In these formulae G = the shear modulus of the soil underneath the mat; R = the radius of the foundation; E = the depth of embedment; H = the depth to bedrock; and v = Poisson's ratio. Also, kl 1, k22 halfspace solution (i.e. ref. [33]) =(0.65/3~/(1 - ( 1 - 213) ~2), 11 for ~ = ao/aol ~

tKc~x K~)
where H = the horizontal force;M = the rocking moment; and u, q~ = the corresponding displacements (rotations). The elements Kxx, Kx~, Ke~ of the stiffness matrix depend on the frequency of excitation f2 of the forces (moments). Since these forces and the resulting displacements are generally not in phase with each other, these elements are complex functions of frequency. Each stiffness function is of the form Ko(1 + 2i13)(k + iaoc), where Ko = the static stiffness; /3 = a measure of the internal damping in the soil (of a hysteri~tic nature); i = x / - 1 ; and ao = the dimensionless frequency = f2R/Cs. (f2 is the circular frequency of the motion and excitation, R the radius of the foundation slab, and Cs a reference shear wave velocity.) k, c = frequency dependent coefficients normalized with respect to the static stiffness. The coefficient c is related to the energy loss by radiation. Using the program described earlier, the dynamic stiffness functions were computed for a range of embedment and stratum depth ratios, and written as described above as:

[Halfspace solution for ao > aol l

1, = (rr/2)(R/H),

=/0.5013~/(1 - (1 - 2/3) ~2), for ~ = ao/ao2 <~1, c22 tHalfspace solution for ao > ao2 = aol
c12 = Cll ,

Cp/Cs,

k12 = k l l ,

(20)
where/3 = the internal (hysteretic) damping in the soil; Cp, Cs = the dilatational wave and shear wave velocities in the subgrade, and aol, ao2 = the (non-dimensional) fundamental shear beam and dilatation frequencies of the stratum (defined above for uniform soil properties). As can be seen, the frequency variation of the stiffness functions has been approximated by that of the halfspace, except for c 1 and c2 in the low frequency range. Fig. 8a and b show a comparison of the 'true' stiffness coefficients with the approximation given above for E/R = 1, H/R = 3, v = ~,/3 = 0.05, while fig. 9a and b give a close-up of the cl and c 2 functions in the low frequency range, i.e. below the fundamental resonant frequencies of the stratum. On the other hand, fig. 10 shows a similar low frequency close-up for a surface footing with H/R = 2, v = 1 and a range of hysteretic damping ratios, while fig. 11 presents corresponding results for various stratum depths. Finally, fig. 12 shows a comparison of the 'true' coupling term versus the approximation suggested. Except for the stiffness coefficient k l l , which displays a somewhat wavy nature, the suggested approximation for the stiffness and damping coefficients provide reasonable substitutes for the true functions. It can be observed that the radiation damping coefficients cl, c2 are larger for the embedded case than for the

Kxx =Kx(kll +iaoc11) (1 Kx~ = Ke~(k12 + iaoc12)

+ 2i3), (18)

(1 + 2i/3),

K = K~;b(k22 + iaoc22 ) (1 + 2i/3). Analysis of the results obtained then provided the following approximations to the static values Kx, Ke~, K ~ , and to the dynamic stiffness coefficients kl 1, k12, k22, Cll, Cl2, c22 (modified from ref. [51]): 8GR /' + I R ~ ( I + ~ ) ( 2+ 5 E IE /~xx = --C-~[1 21-1] 2 o 2E l~xo = KxxR (5 -d0.03) ~

4H]'

(~ is positive clockwise)

386

E. Kausel et al. / Spring method for embedded foundations

3
..............
kll'
1.0

!-'J
t,,,tA,,,,,,,..
k]zl
c) b)
G)

c)

.........

. . . . . . . . . . . .

Io

05

05

~
cl
ROCKING ,7/2

b)
a)

"772
0I
Cll I0

SWAYING
tT ao
04

27- Qo

02

0.3

0.5

'

011

'

0!2

'

0'3

0!4

'

015 fo

y2o
05

~
~TI2
O1

~
o~

_
-

_ ...........
o)

b)
c)
ff a<~

77

ao

02

03

04

05

fo

02

03

-o'~

'

o'5

70

Fig. 8. Dynamic

stiffness coefficients,

u = ~, ~3

0.05.

011,

~11 015

0 I0

005

)-//
S/c)
005

1
i
olo
ROCKING 010

,5 swAY,,G i
010

]~

SWAYING

I-

oo~F' o',o JI~o

C22

L ~..o.,o//' ,//
/ ,/"

/I

,Y

!/

,7

rl I

i .,! ...... 4......

~I

'

I-

I/
005

~-O,o/'2
010

,,'7,~
015 fo /

cz2~

I I

ROCKING
o IO

illl//ll///
/1 ../,.I. // /, Z / /#~ // -//

l" //"++

I
010

) !

O0
005

"~Z
Ol5

I I
I
I i

05

/ //

)o

005

010

015

020

o Z5

fo

Fig. 9. Damping coefficients in low frequency range. Geometry as shown in fig. 8.

Fig. 10. Damping coefficients in low frequency range vs. internal damping H/R = 2, v = 1.

E. Kausel et al. / Spring method for embedded foundations

387

"I
01C

L ~=S

H:4

/
/
/

H=2

/ /
!

SWAYING O05
0115

020

~fo

/" /

//

//

01

//" 03oo~.
- - -

ROCKING
/

I *oo z

,-"7' //

.,,

ratio h = Kx~/Kxx is essentially a real quantity independent of frequency. Physically, it represents the height of the center of stiffness, i.e. the location above the mat at which a force must be specified in order to produce only a translation. The approximation suggested for the coupling term implies that the height of the center of stiffness is approximately one-third of the embedment depth. The jump in the damping coefficients c 1 , c2 in the neighborhood of the stratum resonant frequencies follows from the fact that, in a lossless medium, no energy can be radiated away below these critical frequencies [ 1 ]. Further studies and improvement in the approximation of these functions in the low frequency range are desirable, although they are not important whenever the fundamental rocking-swaying frequency is greater than the shear-beam frequency of the stratum. Useful approximations to the halfspace stiffness and damping coefficients can be obtained as follows,
025 -fO

OIO

I 015

020

Fig. 11. Damping coefficients in low frequency range vs. depth of stratum, u = 3, 3 = 0.05.

v:~12
J " ~:0 45

kn:IST~FFNESS COEFFICIENT

~ ~
I/3

surface footing; therefore, the suggested procedure should give conservative results for an embedded structure. On the other hand, the lack of agreement of k] ] is not important in practical situations since (1) rocking is typically more important in soil-structure interaction problems, and (2) at intermediate and high frequencies the response is controlled by damping (13, c 11, c22 ) and the structural inertia (which is proportional to the mass of the structure and the square of the frequency). On the other hand, fig. 12 shows that the
1C ~ N
h= Kx4'

v: ........................... DAMPING O E F F I C I E N T C
c,:O 576

~ :O 45 - 0 5

SWAYING 2 ] ~7
05

21rf
~0 fo

ROCKING ~ STIFFNESSCOEFFICIENT
k=2: 1-O2o0

Kxx
IC

v:O

DAMPING COEFF CIENT

v : 045 0

REAL PART

", ',,

,2,

~,7

,:

,'~L\

, 6, 2rr
I0

= 7

800 fo

,-

0.5

v= ~ 0 ~

......

,:o ...... 2S

oo
Fig. 13. Halfspace stiffness and damping coefficients vs. approximation (halfspace values aTterVeletsos and Wei [33]).

Fig. 12. Cross coupling term, case (b) in fig. 8.

388

E. Kausel et al. / Spring method jbr embedded/bundations

"Fable 1 Radius of cylinder Height of center of mass Total mass Mass m o m e n t o f i n e r t i a with respect to center of mass Mass m o m e n t o f i n e r t i a w i t h r e s p e c t to c e n t e r o f base R = 70 ft Yo = 8 0 ft m = 3086 Kslug I 0 = 1 5 . 1 5 1 0 6 K s l u g ft 2

see fig. 13: kl1 = 1,

k22 = 1 -

0.2 a o

(a o ~< 2.5), (u~<l/3) / (21)

{0.5 1 -0.2a o Cli = 0.576,

c= =

(u>~0.45))a>~2"5'

I = 3 4 . 9 0 1 0 6 K s l u g ft 2

0.3

a2/(1 + a2o).

Additional research is needed to extend the range of application of the approximate formulae. In particular, it is desirable to develop similar formulae for the vertical and torsional degree of freedom of embedded, circular foundations.

8. Soil-structure interaction problem

Once the input motion and the base impedances are known, the last step is reduced to a simple dynamic analysis of a multidegree-of-freedom system. The 'stiffness' and 'damping' terms can be added directly to the corresponding terms for the structure in a solution in the frequency domain. A time domain solution, and a modal solution in particular, offer a simpler physical interpretation of the results. They require, however, frequency independent stiffness and damping coefficients, and for the latter, in addition, the existence of normal modes. Several approximations are then introduced. (i) The soil springs and dashpots are assumed frequency independent. The constant values can be (a) the static values, or (b) the values at a frequency corresponding to the fundamental frequency of the combined soil-structure system. This last option requires a trial-and-error procedure. (ii) The existence of normal modes is assumed. To compute the values of effective damping in each mode, weighted averages of the internal damping in the different components are combined. Several methods have been suggested to obtain weighted modal dampings; see for instance Ro6sset et al. [25].

a nuclear containment structure was analyzed, using the proposed rules, and compared to the 'exact' solution. To keep matters simple, the structure was modeled as a rigid cylinder, with the properties as given in table 1 (see also fig. 14). The structure is embedded 70 ft in a homogeneous stratum of depth H = 140 ft, Poisson's ratio u = 1, specific weight 7 = 0.12 kcf, and having a shear wave velocity Cs = 700 ft/s. The situation is similar to that shown in fig. 6 if the massless foundation is replaced by the idealized structure. The control motion (see also fig. 7) is an artificial earthquake record rich in all frequencies, specified at the free surface in the free field, and having a peak acceleration of 0.125 g. The shear modulus of the soil for this case is G = 1830 ksf, while the shear beam and dilatational frequencies of the stratum are 1.25 and 2.50 Hz, respectively. For

160

i
i
7(?'

/.

i,ao

t'
v : !/3 ~-: o 121~cf

k! r
!

,o

i
9. Example of application
To illustrate the usefulness of the approximate procedures suggested in this paper, an idealized model of

c~:7oo,~

7s

I
f

Fig. 14. I d e a l i z e d m o d e l u s e d in e x a m p l e .

050

050

040

040 o,

~ 030
II~, 0 2 0 W

~
I-<

050

W 020

U
010
o) APPROXIMATE MOTION, TRUE STIFFNESSES
i I ~ I i I , i , i _

010

b) TRUE MOTION, APPROXIMATE STIFFNESSES

i i i i i i [ I i I

%
050

O 2O

O 40

060

080

100 r

020 o 5oA

PERIOD (SEC.)

0.40 0.60 080 PERIOD (SEC)

1.00v

04C

040

05C

z 050 9

n~ .~ 0 2 0

<~

n.. 020 I,d c) APPROXIMATE MOTION, APPROXIMATE STIFFNESSES

, ,

010

O.lO

d)TRUE MOTION (TRANSLATIONAL COMPONENT ONLY), TRUE STIFFNESSES

[ I

. . . . . . . .
0.40 060 0.80 PERIOD (SEC) I O0 ~

0 ~0

00

020

I i I i t 0.40 060 080 PERIOD (SEC.)

i . ICO-

Fig. 15. R e s p o n s e s p e c t r u m a t m a t - s o i l i n t e r f a c e , 1% o s c i l l a t o r d a m p i n g .

200

200' a) APPROXIMATE MOTION, TRUE STIFFNESSES

~ i:n

b)TRUE E

MOTION, STIFFNESSES

o,
z

0 100

g
100

020 2 O0

040 060 080 PERIOD (SEC.)

~ O0

20

040 0 60 080 PERIOD (SEC.)

1.00~

0o
c) APPROXIMATE MOTION, d)TRUE MOTION ( T R A N S L A T I O N A L COMPONENT ONLY),TRUE STIFFNESSES

ob v

IOC

_.1 w

0 0

I 0.20

I 040

I 060

Ii

I,

0.80

1.00

0.20

PERIOD (SEC.)

0.60 0 80 PERIOD (SEC.)

0,40

lO0

Fig. 16. R e s p o n s e s p e c t r u m o f t o p o f s t r u c t u r e , 1% o s c i l l a t o r d a m p i n g .

390

E. Kausel et al. / Spring method ]or embedded foundations the simplified support motion is an excellent substitute for the more difficult to determine true motion. Figs. 15b, c and 16b, c, on the other hand, reveal that the actual embedded foundation has a somewhat larger value of radiation damping than the one predicted by the approximate formulae. This was expected, because the radiation damping coefficients Cl, c2 of an embedded foundation are larger than those of the corresponding surface foundation. While further research could narrow the differences between true and approximate stiffnesses, the procedure gives satisfactory results, particularly considering all sources of uncertainty affecting the problem. Additional studies made with deeper strata, in which the resonant shear beam and dilatational frequencies of the stratum were smaller than the fundamental rocking-swaying frequency of the soil-structure system, revealed an even better agreement between true and approximate solutions. On the other hand, analyses made using the classical halfspace springs and limiting the damping to 10% critical, and/or subjecting the system to the control motion directly under the springs, gave results in gross disagreement with the direct (or true) solution.

these conditions, use of eqs. (19) gives the following static stiffness matrix for the foundation, relative to the center of the base: K-[ 2.079 53.85 / 11040.00]" 106"

-t53.85

The 'exact' static stiffness matrix (not shown) is almost identically the same. The coupled rocking-swaying frequencies follow then from the characteristic equation det {K - 4nZf2M) = 0, where M is the mass matrix

30 6
my o I )-(246.92

469 /
34900. )

Solution of the secular equation yields a fundamental frequency f l = 2.78 Hz. If this value is used to recompute the stiffness matrix to account for frequency dependence, a new number for the natural frequency can be calculated. Repeating the process until convergence is established gives a final value f = 2.37 Hz when using the approximate stiffness coefficients given by eqs. (20). This natural frequency is below the resonant dilatational frequency of the stratum (2.5 Hz), so that low radiation damping values can be expected. Figs. 15 and 16 show a comparison between the 'true' solution (using the true stiffness functions and the true support motion) and the solution obtained using (a) the approximate motion and the true stiffnesses, (b) the true motion and the approximate stiffnesses, (c) the approximate motion and the approximate stiffnesses, and (d) the translational component of the true motion only, and the true stiffnesses. In all cases the analyses were performed in the frequency domain, determining time histories by convolution of the transfer functions and the Fourier transforms of the input motions. The 'true motion' (a translation and rotation) and the 'true stiffnesses' were determined with the finite element program mentioned earlier. As a check, the structure was also modeled with finite elements and the response computed directly. The results so obtained were identical to the 'true solution' referred above, thus demonstrating the validity of the three-step solution. Inspection of figs. 15 and 16 reveals that the approximate procedure suggested, while not exactly duplicating the 'exact' solution, provides remarkable agreement for the response spectra at the structural elevations shown. In particular, figs. 15a and 16a show that

10. Conclusions It was the purpose of this paper to illustrate the connection between direct finite element procedures for the evaluation of soil-structure interaction effects in an embedded structure and a more general spring method that accounts for embedment in the foundation stiffnesses and the support motion. At the same time, approximate rules were proposed for embedded, circular foundations, which greatly simplify the use of the spring method while providing results in good agreement with the more involved 'exact' solution. A number of comparative studies indicate that it is more important to reproduce correctly the static stiffnesses than their complete frequency variation. It is also worth noticing that the increase in stiffness due to embedment is very sensitive to the properties of the lateral soft, which may be disturbed, and to the degree of adhesion of the lateral walls to the soil. Considering, in addition, the uncertainties in the soil properties and its non-linear behavior, it is clear that engineering judgement is needed in the selection of the most appropriate model, and that parametric studies,

E. Kausd et al. / Spring method for embedded foundations

391

varying the assumed conditions within reasonable limits, are advisable. This, of course, is equally true whether the analysis is carried out in a single step (onepass method) or in three steps, as suggested here. l h e spring method has the advantage of being less time-consuming when approximations are used for the kinematic interaction problem. It allows, therefore, more parametric studies, and the accuracy of each step is subject to better control. Of particular importance is the possibility of this method to make use of symmetry or cylindrical conditions if the foundation meets these requirements, even if the structure does not (which is a frequent situation). The coupling between the corresponding terms will come in naturally in the third step. From a practical standpoint, the procedure has an additional advantage when the design motion is specified by a broad band response spectrum not tailored to the soil conditions at the site. If the direct approach is applied to such a case, deamplification of certain frequency components with depth resulting from the use of one single wave pattern (i.e. vertically propagating waves) could lead to unconservative estimates for the motions of the structure. Under such conditions it may be better to regard the design motion as an 'average' motion in the vicinity of the structure, and to use it directly as input to step 3. Alternatively, the approximate procedure could be generalized to account for non-vertically incident body waves and surface waves which can introduce torsional components of motion even in perfectly axisymmetric structural systems.

C = damping matrix Cs = shear wave velocity Cp = dilatational wave velocity E = depth of embedment f = frequency F = Fourier transform of input earthquake g = acceleration of gravity G = shear modulus h = height of center of stiffness H = horizontal force; depth of stratum i = imaginary unit (X/-1) I = mass moment of inertia k = stiffness coefficient K = stiffness matrix m = mass M = mass matrix M = moment (positive clockwise) P = force vector R = radius of foundation T = rigid body transformation matrix u = horizontal displacement U = absolute displacement vector o = vertical displacement V = vertical force X = impedance matrix Y = relative displacement vector /3 = damping ratio = Poisson's ratio = rotation (positive clockwise) ~2 = circular frequency = 21rf

Acknowledgements.
This paper is based in part on the graduate theses of J. Morray and F. Elsabee at the Massachusetts Institute of Technology, and on research and calculations performed in the Engineering Mechanics Division of Stone and Webster Engineering Corporation. Special thanks are due to Prof. J.M. Ro6sset for his useful criticism and to Mr. C.F. Reeves for his encouragement and support of the Engineer Resident Program which made this study possible.

Superscripts
T = transpose = derivative with respect to time

References
[ 1] R.N. Arnold, G.N. Bycroft and G.B. Warburton, J. Appl. Mech. 77 (1955) 391. [2] J. Bielak, Earthquake Eng. Struct. Dyn. 3 (3) (1975) 259. [3] A.K. Chopra, P. Charabarti and G. Dasgupta, Report No. EERC 75-22, Earthquake Engineering Research Center, University of California, Berkeley, Aug. (1975). [4] A.K. Chopra and J.A. Gutierrez, Report No. EERC 73-13 Earthquake Engineering Research Center, University of California (1973). [5] F. Elsabee, Thesis presented to the Massachusetts Insti-

Notation
ao = dimensionless frequency (D,R/Cs) c = damping coefficient

392

E. Kausel et al. / Spring method ]or embedded foundations tute of Technology, at Cambridge, MA, in 1975, in partial fulfillment of the requirements for the degree of Master of Science. S.M. Erden, Thesis submitted to the Graduate School of the University of Massachusetts, at Amherst, Mass., in 1974, in partial fulfillment of the requirements for the degree of Doctor of Philosophy. G.C. Gazetas, Thesis presented to the Massachusetts Institute of Technology at Cambridge, Mass., in 1975, in partial fulfillment of the requirements for the degree of Master of Science. G.C. Gazetas and J.M. Ro6sset, Proc. of the National Structural Engineering Conference, ASCE, held August 1976, Madison, Wisconsin, Vol. 1, pp. 115-131. J.A. Gutierrez, Report No. EERC 76-9, Earthquake Engineering Research Center, University of California, Berkeley, April 1976. J.R. Hall, Jr. and J.F. Kissenpfenning, Proc. ELCALAP Seminar, Berlin, Sept. (1975) Paper U2/2. T.J. Jones and J.M. Ro~sset, MIT Research Report 70-3, Massachusetts Institute of Technology, Cambridge, Mass., Jan. 1970. E. Kausel, MIT Research Report R74-11, Soils publication no. 336, Struotures publication no. 384, Massachusetts Institute of Technology, Cambridge, Mass., Jan 1974. E. Kausel and J.M. Ro~sset, Electric Power and the Civil Engineer, ASCE, Proc. of the Power Division Specialty Conference held in Boulder, Colorado, Aug. 1974. E. Kausel, J.M. Ro~sset and G. Waas, J. Eng. Mech. Div., ASCE 101 (EM5), Proc. Paper 11652, Oct. (1975) 679. E. Kausel and J.M. Ro6sset, J. Eng. Mech. Div., ASCE 101 (EM6), Proc. Paper 11800, Dec. (1975) 771. E. Kausel, J.M. Ro~sset and J.T. Christian, J. Geotech. Eng. Div., ASCE 102 (GT11), Proc. Paper 12579, Nov. (1976) 1159. R.J. Krizek, D.C. Gupta and R.A. Parmlee, J. Soil Mech. Found. Div., ASCE 98 (SM12) (1972) 1347. J.E. Luco and R. Westmann, J. Eng. Mech. Div., ASCE 97 (EM1) (1971) 1381. [19] J.E. Luco, Nucl. Eng. Des. 31 (2) (1974) 204. [20] J. Lysmer and G. Waas, J. Eng. Mech. Div., ASCE 98 (EM1), Proc. Paper 8716, Feb. (1972) 85. [21] J.P. Morray, Thesis presented to the Massachusetts Institute of Technology, at Cambridge, Mass. in 1975, in partial fulfillment of the requirements for the degree of Master of Science. [22] M. Novak, ASCE National Structural Meeting, April 1973, San Francisco, California. Meeting preprint No. 2029. [23] M. Novak and Y.O. Beredugo, J. Soil Mech. Found. Div., ASCE 98 (SM12) (1972) 1291. [24] F.E. Richart, Jr., J.R. tiall, Jr. and R.D. Woods, Vibration of Soils and I:oundations (Prentice Hall, New Jersey, 1970). [25] J.M. Ro~sset, R.V. Whitman and R. Dobry, 1. Struct. Div., ASCE 99 (ST3) (1973) 389. [26] J.M. Ro6sset and R.V. Whitman, MIT Research Report R69-15, Soils publication no. 231, Massachusetts Institute of Technology, Cambridge, Mass., March 1969. [271 J.M. Ro~sset, Lecture notes for the Course on Soil Dynamics for Earthquake Design, International Center for Computer Aided Design (ICCAD), Lecture Series No. 276, Santa Margherita, Italy, 21-23 Jan. 1976. [28] M.A. Sarrazin, MIT Research Report R70-59, Massachusetts Institute of Technology, Cambridge, Mass., Sept. 1970. [ 2 9 ] H. Tajimi, Proc. 4th World Conference in Earthquake Engineering, Santiago, Chile (1969). [30] S.A. Thau, A. Umek and R. Rostamian, ASCE National Structural Engineering Meeting, Meeting Preprint 2200, Cincinatti, Ohio, 22-26 April, 1974. [31] W.T. Thomson, J. Appl. Phys. 21 (1950). [32] A.K. Vaish and A.K. Chopra, J. Eng. Mech. Div., ASCE 100 (EM6) (1974) 1101. [33] A. Veletsos and Y. Wet, J. Soil Mech. Found. Div., ASCE 97 (SM9) (1971) 1227. [34] G. Waas, Thesis presented to the University of California, at Berkeley, California, in 1972, in partial fulfillment of the requirements for the degree of Doctor of Philosophy.

[6]

[ 7]

[8]

]9]

[10] [11]

[12]

[ 13]

[14] [15] [16]

[17] [18]