You are on page 1of 17

Engineering Structures 153 (2017) 443–459

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Spatial distribution of Winkler spring stiffness for rectangular mat


foundation analysis
D. Loukidis ⇑, G.-P. Tamiolakis
Department of Civil & Environmental Engineering, University of Cyprus, Nicosia, Cyprus

a r t i c l e i n f o a b s t r a c t

Article history: Structural design of mat foundations of buildings is often done by performing static analysis of a slab
Received 11 January 2017 resting on vertical uncoupled Winkler springs. It is already well established that the simplifying assump-
Revised 7 September 2017 tion of a uniform modulus of subgrade reaction throughout the mat foundation leads to inaccurate results
Accepted 2 October 2017
that significantly underestimate the bending moments in the mat. This paper examines the spatial vari-
Available online 1 November 2017
ation of the Winkler spring stiffness constants that is necessary for the mat-on-springs analysis to pro-
duce the same slab deflections and bending moment diagrams as finite element analysis that treats
Keywords:
the soil as continuum. For this purpose, three-dimensional parametric analyses of slabs resting on elastic
Mat foundations
Finite element analysis
soil are performed using the finite element method for various values of soil elastic properties, slab geo-
Plate bending metrical characteristics and column load configurations. The finite element analysis results were used for
Winkler springs back-calculating analytically the equivalent Winkler spring constants at each node of the mat. Based on
Soil-stricture interaction the numerical results, equations describing the spatial distribution of spring stiffness are proposed. The
performance of the proposed equations is compared against existing spring stiffness spatial distribution
approaches used in practice.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction because it can be easily applied in most commercial structural


analysis computer programs.
The mat is the type of shallow foundation commonly used in Many studies have pointed out the shortcomings of the original
the case of buildings sitting on soft and weak soils and/or transfer- basic Winkler spring approach, which assumes that the modulus of
ring large loads to the ground. It is also often used in special pur- subgrade reaction has the same value everywhere under the mat,
pose structures such as storage tanks, silos and water towers. A and have proposed alternative methods of various degrees of com-
mat foundation is in essence a stiff slab that acts as a single foun- plexity (e.g. [23,16,36,26]). One of these alternatives is the pseudo-
dation element that covers the entire plan area of the structure. As coupled approach, in which the mat still rests on vertical springs,
any shallow foundation, a mat foundation has to be checked but with spring constants that vary across the mat depending on
against bearing capacity failure and excessive total (average) set- the location of a given spring. The pseudo-coupled approach is
tlement. It has to be also structurally designed to adequately resist meant to improve the accuracy of the original Winkler spring
differential settlements and withstand the bending and shear method while retaining its simplicity.
forces that develop inside it. The goal of the present study is to establish the spatial distribu-
A variety of methods are available for the structural analysis of tion of spring stiffness coefficients under a mat foundation that
mat foundations, ranging from simple static analogues to elaborate helps attaining the best possible accuracy in terms of deflections,
three-dimensional finite element analysis. The most common bending moments and shear force diagrams. This is achieved by
method in current design practice is the Winkler spring approach first performing a series of 3-dimensional finite element analyses
[40], in which the soil is represented in the analysis as linear ver- of rectangular mats resting on a linear elastic continuum repre-
tical springs supporting the mat. Despite being a rather rough ide- senting the foundation soil. The resulting nodal displacements of
alization of reality and the emergence of more accurate methods, the mat are then used as input in the back-calculation of the equiv-
the Winkler spring approach still constitutes the state-of-practice alent spring stiffness coefficient values at each node of a mat hav-
ing the same characteristics but this time resting on Winkler
⇑ Corresponding author. springs. A parametric study is performed in order to investigate
E-mail addresses: loukidis@ucy.ac.cy (D. Loukidis), geomitam@gmail.com the influence of the various parameters of the mat problem on
(G.-P. Tamiolakis).

https://doi.org/10.1016/j.engstruct.2017.10.001
0141-0296/Ó 2017 Elsevier Ltd. All rights reserved.
444 D. Loukidis, G.-P. Tamiolakis / Engineering Structures 153 (2017) 443–459

the spring stiffness distribution. The back-calculated spring con- foundation (e.g., width and length of the mat). Hence, ks is project
stants are also compared to those resulting from an advanced iter- specific and is usually estimated by performing calculations of the
ative pseudo-coupled approach based on the Boussinesq solution. overall settlement of the building assuming that the foundation is
Finally, equations are proposed for calculating the equivalent perfectly rigid and using well established methods for footing set-
spring stiffness values at any point of a rectangular mat foundation tlement calculation [8]. The settlement of the rigid foundation (uni-
and their performance is compared to that of existing approaches form settlement) is very close to the average settlement of the
used in practice. It should be noted that the present analysis actual flexible foundation.
assumes that the soil behaves linear-elastically (an assumption
commonly made in current mat design practice). The results of this
2.2. Coupling and the pseudo-coupled approach
study and the proposed equations could be used in the future as a
basis for the development of more rigorous spring models that take
In the static analogue of the Winkler spring method (Fig. 1), the
into account the nonlinear character of soil behavior.
springs are ‘‘uncoupled”, meaning that they respond indepen-
dently of each other and the compression of one spring does not
2. Knowledge background influence the behavior of the neighboring spring. In reality, the
compression of the soil in one location under the mat will generate
2.1. Basic Winkler model vertical shear stresses and strains that are transmitted laterally and
thus influence the deformation at the locations of neighboring
The Winkler spring method [40] assumes that the mat founda- springs. This coupling phenomenon is especially intense closer to
tion sits on vertical linear springs representing the deformable (lin- the edges of the mat foundation. Ignoring the coupling effects con-
ear elastic) soil (Fig. 1). The stiffness coefficient of a Winkler spring stitutes a source of errors that leads to substantial underestimation
Ks is expressed as the product of the area As of the portion of the of mat differential settlements and maximum bending moments
slab influenced by the spring (the tributary area) and the parame- [20]. The shortcoming of the original (uncoupled) spring method
ter known as modulus of subgrade reaction ks, which is defined as: is clearly illustrated by the simple example of a perfectly flexible
rectangular foundation (zero shear and bending stiffness) on an
q
ks ¼ ð1Þ elastic half-space loaded by a uniform vertical distributed load.
w
In this case, the uncoupled spring approach would predict uniform
where q is the foundation pressure exerted to the soil and w is the settlement throughout the mat, while the rigorous solution based
resulting settlement. The estimation of ks presents inherent difficul- on elasticity theory [35] predicts that the settlement at the center
ties because it does not constitute a fundamental material property of the loaded area should be twice the settlement at the corners.
of the soil (such as the Young’s modulus), but a parameter defined A number of more accurate soil - mat foundation interaction
based on an arbitrary one-dimensional pressure-settlement con- models have been developed that take into account coupling
cept. As such, it is a variable that depends on both on the elastic effects [23,24,25,16,17,19,9,10,22] and predict the correct shape
properties of the soil and the geometrical characteristics of the of the loaded soil surface. Apart from springs, this models employ

mat foundation discretized in 2-


dimensional plate elements
column loads

tributary area Αi
node i

spring i

spring constant :

K si = ks × Α i
Fig. 1. Computational model of the Winkler spring analysis approach.
D. Loukidis, G.-P. Tamiolakis / Engineering Structures 153 (2017) 443–459 445

other mechanical components, such as shear and flexural layers the settlements of the contact surface caused by the pressure of
and membranes, thus effectively departing from the one- each discrete area are calculated using Boussinesq [6] and Wester-
dimensional nature of the Winkler model and introducing the gaard [39] theories (in the case of uniform and layered soil profiles,
effects of shearing. Recently, soil non-linearity was introduced in respectively) and the principle of superposition. Then, new esti-
coupled soil springs under a piled raft [22]. Despite the enhanced mates of ks are calculated at each nodal point by dividing the cor-
rigorousness, these components inevitably increase the degree of responding contact pressure with the calculated settlement at that
complexity and the difficulty of implementing the coupled point, and the static analysis of the mat is repeated. In each com-
foundation-soil interaction models in ordinary structural analysis putational iteration, the settlements are compared to the previ-
software. Moreover, they often require extending the analysis ously calculated mat deflections at nodal points (compatibility
domain beyond the planar dimensions of the mat. check) and the iterative procedure is terminated once sufficient
A simpler and more approximate alternative to the coupled convergence between the two is achieved.
models is the ‘‘pseudo-coupled” approach [2,36,26,7,8], in which Coduto [8] proposed a simpler version of the pseudo-coupled
the static analogue is the same as in the classical Winkler spring method, which does not require a trial analysis or iterations.
method (Fig. 1), but with ks being different from point to point According to this version, the mat foundation is divided in N
across the mat, taking larger values near the edge of the foundation (two or more) concentric zones, with the central zone having half
compared to its center. For example, for the static analogue of the width and the length of the mat. Each zone i is assigned a dif-
Fig. 1 to be able to predict the correct variation of settlement across ferent value of ks,i in such a way that ks increases from the inner to
a perfectly flexible (zero rigidity) foundation subjected to a uni- the outer zones and the outmost zone has twice as much ks as the
form distributed load, the ks value assigned at the corners of the central zone (i.e. ksN = 2ks1), with the additional condition that the
foundation needs to be twice as much the value assigned at the weighted (based on the zone areas) average of all ks,i (ks,ave) is
center. equal to the ks obtained from Eq. (1) using settlement w calcula-
To date, the suitable spatial distribution of ks has not been tions that treat the slab as a large rigid footing (kr). Applying this
firmly established and the existing variations of the pseudo- concept to a mat divided in three zones as shown in Fig. 2, with
coupled method differ significantly between each other. The most areas A1 (central), A2 (intermediate), A3 (outer), and assuming that
basic form of pseudo-coupling is to simply use for the springs con- for the intermediate zone ks2 = 1.5ks1, we obtain
nected to the edge of the foundation twice as much ks as for the
A1 þ A2 þ A3
springs at rest of the foundation [7]. A more elaborate version pro- ks1 ¼ ks;ave  ð2Þ
posed by ACI [2] and Bowles [7] dictates that ks should decrease A1 þ 1:5A2 þ 2A3
progressively from the edge to the center of the mat. This is
achieved by first performing a trial analysis with ks being the same 2.3. Continuum finite element analysis
throughout the slab to obtain an estimate of the contact pressure
distribution. Then, based on the contact pressure, the average value The increase in computer power in the last two decades allowed
Drv,ave of the profile of vertical stress increase caused by the foun- the use of the finite element method in mat foundation design in
dation down to a depth of 4B (B: mat width) is calculated at an increasing rate. The finite element method can automatically
selected points along the central axis of the slab using the New- consider coupling and soil-foundation interaction effects by simu-
mark [29] method. The new ks values are estimated by multiplying lating the soil, not as springs or other mechanical equivalents, but
the ks value operative at the edge of the foundation with the ratio as a volume of three dimensional continuum elements. Another
of the Drv,ave at the edge point to that at the point under advantage of continuum finite element analysis (cFEA) is that it
consideration. can easily consider soil stress-strain non-linearity (e.g.,
Banavalkar and Ulrich [3] and Ulrich [36] developed an [37,38,13,4]).
advanced iterative version of the pseudo-coupled approach, called Despite being the computational method that can yield the
the discrete area method. As in ACI [2], the problem is first ana- most realistic results to date (provided that sufficiently fine
lyzed assuming uniform ks distribution to obtain an estimate of meshes are used), its penetration in routine mat foundation engi-
the mat deflections and the contact pressure distribution. Based neering is still rather limited because many structural analysis
on the resulting contours of contact pressure, the mat-soil contact and design software still do not include continuum elements.
area divided in small rectangular areas in which the contact pres- Nonetheless, cFEA results can be used as a reference in examining
sure could be considered to be practically constant. Subsequently, the validity and performance of the mat-on-springs model (Fig. 1).

Fig. 2. Division of mat into zones with different modulus of subgrade reaction values for pseudo-coupled mat analysis [8].
446 D. Loukidis, G.-P. Tamiolakis / Engineering Structures 153 (2017) 443–459

Liao [26] performed 2-dimensional plane strain finite element forming the mat were squared-shaped with edge length equal to
analysis of an elongated foundation slab (strip) resting on linear 0.5 m. These choices with respect to shape, size and order of ele-
elastic 2D continuum finite elements with the goal of establishing ments are in line with common foundation engineering practice
the proper shape of the distribution of the equivalent modulus of and structural analysis software. Moreover, use of 1st order ele-
subgrade reaction across the mat. Moreover, Liao [26] examined ments avoids the complexities of combining the Winkler spring
the effects of the elastic properties of soil and mat, the mat geo- approach with higher order elements, which, aside from corner
metrical characteristics, and the loading configuration. The equiv- nodes, include mid-edge and possibly interior nodes. The meshes
alent ks values were extracted from the FEA output by dividing were constructed in such a way that the locations of the nodes of
the contact pressure at the nodes of the beam elements by the cor- the shell elements coincide with those of the upper faces of the
responding nodal displacements. Liao [26] observed that for rela- underlying soil elements. The foundation nodes were fully tied to
tively stiff mats, the form of loading (positions and relative the corresponding soil nodes, hence there is no possibility of slip-
magnitude of column loads) does not influence significantly the page between the mat and the soil. For optimal numerical accu-
spatial distribution of ks. Hence, based on the numerical results, racy, care was taken during meshing so that the solid elements
Liao [26] compiled tables that give ks values as a function of the immediately underneath and around the mat are cubical (lowest
distance from the central longitudinal axis of the strip mat, the possible aspect ratio).
Poisson’s ratio and the thickness of the deformable soil layer. The Both soil and mat were assumed to behave linearly elastic. The
results show that ks takes an almost constant value in a central part soil profile was assumed to be uniform, with constant Young’s
of the mat that has width approximately equal to 70% of overall modulus Es and Poisson’s ratio vs. All degrees of freedom at the lat-
width of the mat and increases non-linearly towards the edges. eral and bottom boundaries of the model were fully restrained. The
width and length of the soil domain were set to be ten times the
width and length of the mat, respectively. This ensured that the
3. Methodology lateral boundaries of the model had negligible influence on the
mat settlement. The mat was loaded by concentrated nodal vertical
3.1. Continuum finite element analysis forces representing the column loads. The superstructure was not
included in the model and, thus, has no contribution to the stiff-
Three dimensional finite element analysis is utilized in this ness of the mat foundation.
study to establish accurate estimates of the mat deflections. The To validate the FE model configuration described above, two
analyses were performed using the commercial finite element soft- basic elastic settlement example problems were analyzed. First,
ware ABAQUS [1]. The computational model consisted of a block of an analysis was performed with the soil surface being loaded by
8-noded (1st order) hexahedral elements (C3D8) representing the a square uniform pressure distribution in the absence of mat. For
soil. The mat foundation was placed at the central region of the this problem, there is an exact closed-form solution by Steinbren-
upper surface of the soil block (Fig. 3) and was modeled using 4- ner [33]. The finite element analysis was performed for uniform
noded (1st order) shell elements. Both full integration (S4) and pressure 100kPa loading a square area of width B=10m and soil
reduced integration with hourglass control (S4R5) shell element with Es = 10,000 kPa and ms = 0.3. The distance between the free
formulations were tried and it was found that the choice is practi- surface and the bottom boundary was 100 m (H/B=10). The Stein-
cally inconsequential to the analysis results, with discrepancies in brenner solution yields settlement 0.046 m at the corner of the
terms of nodal displacements not exceeding 0.06%. The elements loaded area and 0.092 m at the center of the loaded area. The finite
element predictions were 0.045 m corner settlement and 0.096 m
center settlement, corresponding to numerical error of 2.2% and
4.3%, respectively. The second example is the same as the first
one with the difference that an extremely rigid square mat (thick-
ness 3 m, Emat = 32 GPa, B = 10 m) is placed between the dis-
tributed load and the soil. This mat can be considered as
perfectly rigid since the difference between the corner displace-
ment and the center displacement was only 1‰ of the average set-
tlement, which was equal to 0.0695 m. This settlement value
compares favorably with the semi-analytical prediction of
0.0746 m (6.8% error) following Fraser and Wardle [14] for a rigid
square foundation.
In Fig. 4, the response of the Abaqus S4 (full integration) and
S4R5 (reduced integration) elements in a problem of a mat resting
on Winkler springs is compared to that of the more traditional
Reissner-Mindlin plate elements, fully integrated and selectively
integrated with hourglass control. Hourglassing refers to the
unwanted spurious oscillations pattern often observed in the nodal
displacement profiles in the case of reduced or selective integra-
tion elements. Herein, selective integration with hourglass control
is done according to Belytschko et al. [5]. Following Belytschko
et al. [5], full Gauss quadrature (e.g. 2  2 quadrature rule in the
case of 4-noded elements) is used for the bending component of
the element stiffness matrix, while the shearing component of
the element stiffness matrix is the weighted average of two evalu-
ations, one using a single (reduced) point quadrature (with weight
factor 1  e) and one using a 2  2 (full) quadrature rule (with
weight factor e). For the boundary value problem studied herein,
Fig. 3. Typical mesh used in the continuum finite element analysis. we found that a factor e = 0.2 provides the best element perfor-
D. Loukidis, G.-P. Tamiolakis / Engineering Structures 153 (2017) 443–459 447

the associated membrane stiffness were omitted since they play


negligible role in the boundary value problem at hand.
The problem of a mat on Winkler springs is governed by the fol-
lowing system of n linear equations:

½K tot ½U ¼ ½F ) ð½K mat  þ ½K springs Þ½U ¼ ½F ð3Þ

where [Ktot] is the total stiffness matrix of the static analogue


(Fig. 1), [Kmat] is the stiffness matrix of the assembly of plate ele-
ments composing the mat, [Ksprings] is the diagonal stiffness matrix
of the contribution of the Winkler springs, [U] is the degrees of free-
dom matrix (vertical nodal displacements wi and rotations ui):
2 3
w1
6u 7
6 27
6 7
6 u3 7
6 7
6w 7
½U ¼ 6
6
47
7 ð4Þ
6 u5 7
6 7
6 7
6 u6 7
4 5
..
.

and [F] is the nodal actions matrix:


2 3 2 3
F1 F1
6 7
6 M2 7 6 07
6 7 6 7
6 M3 7 6 7
6 7 6 07
6 7 6 7
½F ¼ 6 F 7 6 F4 7
6 4 7¼6 7 ð5Þ
6 M5 7 6 7
6 7 6 07
6 7 6 7
6 M6 7 4 0 7
6
5
4 5
.. ..
. .

Given that in this study it is assumed that the foundation


receives from the superstructure only vertical forces Fi, the nodal
moments Mi are all equal to zero. Each of the n/3 equations of
the above system that pertain to vertical force equilibrium can
be written as

K mat i;1 w1 u2 þ K mat i;3 u3 þ    þ K mat i;n2 wn2


þ K mat i;2
Fig. 4. Comparison of behavior of various types of elements: (a) vertical displace- þ K mat i;n1 un1 þ K mat i;n un þ K si wi ¼ F i ð6aÞ
ments along the edge and (b) along the centerline (y = 0) of a mat on Winkler
springs.
and each of the 2n/3 equations pertaining to moment equilibrium
can be written as

K mat j;1 w1 þ K mat j;2 u2 þ K mat j;3 u3 þ    þ K mat j;n2 wn2


mance in terms of hourglass control. The mat foundation dimen-
sions and load configuration are shown in Fig. 4. The modulus of þ K mat j;n1 un1 þ K mat j;n un ¼ 0 ð6bÞ
subgrade reaction ks was set equal 1000 kN/m3 throughout the Knowing the wi and ui from the cFEA, Eq. (6a) can be directly
slab. It can be observed that the S4 and S4R5 elements produce solved with the respect to the unknown spring constant Ksi. Alter-
almost identical results, which are in close agreement with those natively, the rotations ui could be treated as unknowns (2n/3) and
obtained using Mindlin plate elements with selective integration. the system of Eqs. (6a) and (6b) (n equations) be solved with a total
Contrarily, full integration Mindlin elements exhibit comparatively of n unknowns, i.e. the n/3 spring constants Ksi plus the 2n/3 rota-
overly stiff behavior. Selective integration is known to have supe- tions ui. Either approach produces exactly the same results. This is
rior performance over full integration [11,21]. because the equations of the form of Eq. (6b) can be seen as an
independent system of 2n/3 equations and 2n/3 unknowns that
3.2. Determination of equivalent spring constants uniquely relate the nodal rotations to the nodal displacements,
and this relation is influenced only by the type of element used
The mat deformations resulting from the continuum finite ele- for the discretization of the mat. So, for a given set of wi values,
ment analyses (cFEA) are used here as input in the back-calculation the system of equations (6b) will produce the same ui values as
of the equivalent spring coefficient Ks values at each node of the those developing in the cFEA.
mat. The back-calculation was done using an algorithm [34] writ- Analysis of a mat-on-springs problem using the Ksi values
ten in MATLAB [27], where the discretization of the mat with plate extracted using this back-calculation methodology was found to
elements was made to be the same as in the cFEA (4-noded planar result in nodal vertical displacements and rotations that are equal
elements with size 0.5 m  0.5 m). The plate elements had exactly down to the 9th significant digit to those produced originally by
the same formulation as the Abaqus S4R5 shell elements with the the corresponding cFEA solution, with this negligible discrepancy
difference that the in-plane (horizontal) displacement DOFs and being due to truncation error [34].
448 D. Loukidis, G.-P. Tamiolakis / Engineering Structures 153 (2017) 443–459

3.3. Boussinesq-based iterative solution corresponding vertical displacement at the center of the tributary
area is calculated using the Steinbrenner [33] solution:
For comparison purposes, computations were also done using
F si ð1  v 2s Þdi
an iterative solution procedure based on the discrete area method DwB;ii ¼ 1:12 ð8Þ
of Ulrich [36], which is the most advanced among the available Ai Es
pseudo-coupled approaches. This iterative method is slightly mod- where di is the width of the tributary area of node i.
ified herein to enhance automation and facilitate its algorithmic The spring reactions could have been applied to the soil as uni-
implementation. The final product of a mat-on-springs analysis form pressures at the corresponding tributary areas for all nodes,
using the Ulrich [36] method is not only the mat displacements, i.e. not only for i = j but also for i – j. This alternative approach
but also the corresponding distribution of spring constants com- would have been completely equivalent to the discrete area
patible with the Boussinesq theory. The computational steps of method of Ulrich [36], with the discrete areas being the tributary
the method are as follows: areas. Nonetheless, it can be proven mathematically (using Eqs.
(7) and (8)) that, for mat discretization in elements of 0.5 m 
(1) Initially, the mat rests on Winkler springs having stiffness 0.5 m in size, the maximum possible difference in DwB,ij between
that is consistent with a uniform modulus of subgrade reac- the two approaches is only 4%.
tion. This initial uniform value can be set equal to that cor- Adding the contributions of all reaction forces Fsj exerted on the
responding to a perfectly rigid foundation. It should be elastic half-space, the vertical displacement of the surface of the
noted that the computational procedure always converges elastic half-space at the location of any given node i is
to results that are independent of the initial guess of modu-
lus of subgrade reaction, which affects only the number of X
N total

iterations needed to achieve convergence. The initial static


wB;i ¼ DwB;ij ð9Þ
j¼1
solution of the mat-on-springs problem yields the first guess
of nodal vertical displacements w and spring reactions Fs. where Ntotal is total number of nodes in the mat.
(2) The obtained spring reaction at any given node j, Fsj (Fig. 5) is (3) The displacements wB,i of the surface of the elastic half-space
used as loading for the calculation of the vertical displace- are compared to the mat displacements wi from the mat-on-
ments wB at the surface of an elastic half-space based on spring calculations and used for obtaining new guesses for
the Boussinesq theory. The reaction Fsj, applied as an equal the Winkler spring constants at each mat node using the fol-
and opposite force on the surface of the elastic half-space, lowing equation:
causes a vertical displacement DwB,ij at the surface of the
ðnewÞ wi ðprev iousÞ
half-space at the location of a node i given by the following K si ¼ K ð10Þ
wB;i si
equation [6]:

The spring constant values obtained from Eq. (10) are used in a
F sj ð1  v 2s Þ new solution of the mat-on-springs problem.
DwB;ij ¼ ð7Þ
pEs rij Steps 2 and 3 are repeated until wi and wB,i converge to each
other down to a small tolerated relative difference. In this study,
convergence is assumed to have been achieved once the ratio
where rij is the distance between points (nodes) i and j. The Boussi- |wi  wB,i|/|wB,i| becomes less than 104 for all nodes. The analyses
nesq theory predicts that the displacement exactly at the point of based on the iterative method described above were performed
application of the force (i.e. i = j) is infinite. Hence, for i = j, Fs is using an algorithm written in MATLAB.
not applied as a concentrated load but as an equivalent uniform
pressure over the tributary area Ai of spring i. In this case, the
4. Numerical results

Continuum finite element analyses (cFEA) were performed for


the load configurations (i.e. number, location, and relative magni-
tude of column loads) shown in Fig. 6. Most of the configurations
reflect the fact that the corner and edge columns usually carry
smaller loads than the interior columns. Configurations C1, C6
and C8 have loads that are proportioned according to the column
tributary areas, while C2 and C4 deviate from this rule. In the case
of C3, C5, C7 and C10, the resultant of the column forces does not
pass through the center of the slab, i.e. the foundation is loaded
eccentrically, inducing overall rotation of the foundation. For C7,
in particular, there is load eccentricity in both x- and y-axes. The
column spacing S ranged from 4 m to 8 m, resulting in mat dimen-
sions in the 10 m to 20 m range (typical for single and multi storey
residential buildings) and the thickness of the mat d from 0.5 m to
1 m. It should be noted that, unlike the mat settlements, moments
and shear forces, the back-calculated spring stiffness Ks coefficients
are independent of the value of the load Q shown in Fig. 6, given
that both soil and mat materials are assumed to be linear elastic.
Column loads in indeterminate structures, such as typical frame
buildings, depend on the differential settlements. A convex
deformed shape of the slab, such as the one shown in Fig. 4, would
lead to an increase of the central column loads and a reduction of
Fig. 5. Schematic of the Boussinesq-based iterative pseudo-coupled method. the peripheral ones. The opposite will happen in the case of a
D. Loukidis, G.-P. Tamiolakis / Engineering Structures 153 (2017) 443–459 449

Fig. 6. Loading configurations considered in the parametric study.

concave deformed shape. Nonetheless, the large thickness of the The Young modulus of the mat Em was set equal to 32 GPa and
mats constructed in reality (d = 0.5 m for single storey dwellings the Poisson’s ratio equal to 0.2 in all analyses. In most of the cases
to 3 m for high rise buildings) and considered in this study examined, the Young modulus of the soil Es was 10 MPa, a value
(d = 0.5–1.0 m), combined with the relatively small stiffness of that is consistent with soft clay or loose sand, and the Poisson’s
soils, results in mat differential settlements between consecutive ratio was equal to 0.3. Values of Es equal to 1 MPa and 100 MPa
columns that rarely exceed 1.5 cm, even for tall multi-storey build- and vs equal to 0.15 and 0.45 were also considered. To quantify
ings (e.g., [36,18]). Consequently, any redistribution of column how much flexible is the foundation compared to the soil, Meyer-
loads is expected to be less than 50% of the prescribed column hof [28] introduced the relative stiffness factor given by the follow-
loads [30,32] and the actual column load distribution for case C1, ing expression:
for example, will be between C1 and C2 or C4. This is one of the
reasons the parametric study includes loading combinations C2 Em d
3

and C4, which deviate from the column tributary area concept Rs ¼ ð11Þ
12Es B3
(C1, C6, C8). In any case, the analysis results presented is subse-
quent sections indicate that the equivalent spring stiffness For the combinations of problem parameters examined herein,
distribution is practically insensitive to the exact form of the Rs ranges from 0.011 to 1.1. A foundation is characterized as rigid
column load distribution. if Rs > 0.5.
450 D. Loukidis, G.-P. Tamiolakis / Engineering Structures 153 (2017) 443–459

4.1. Vertical displacement profiles is a function of the foundation aspect ratio B/L and the thickness
H of the compressible soil underneath the foundation (Fig. 3). Based
Fig. 7 shows examples of vertical displacement profiles along on numerical results by Fraser and Wardle [14] and Chow [12], who
the x central axis of the mat (y = 0). The vertical displacements employed a computational method that combines the Boussinesq
are normalized with respect to the settlement wr should the foun- analytical solution with finite element analysis, the factor Cf can
dation were of infinite rigidity [15]: be adequately approximated by the following formula:
 0:45
qave Bð1  v 2s Þ 0:85 BL
wr ¼ C f ð12Þ Cf ¼    1þexpð5v 3s Þ ð13Þ
Es 1 þ 0:1 2 þ BL HB
where qave is the average bearing pressure, which is equal to the
A general observation from (Fig. 7) is that the vertical displace-
sum of all the column loads divided by the mat area. The factor Cf ments of the mat are comparable to wr (w/wr around 0.95). Hence,
wr could play effectively the role of a reference displacement. By
extension, the spring stiffness corresponding to a perfectly rigid
foundation (Kr) could be used as a basis for establishing the varia-
tion of Ks across the mat, as shown later in this paper.
The displacement profiles are either predominantly concave or
convex depending mainly on the loading configuration and the
length of the mat (Fig. 7b). The larger the number of loads across
the mat (configurations C6 and C8) and the larger the mat size,
the more likely is for the displacement profile to be concave
despite having column loads at the mat edges. Fig. 7c contains
one case were the loading is non-symmetric (C3), causing the
mat to develop overall rotation.

4.2. Back-calculated Winkler spring stiffness

Fig. 8 shows the spatial distribution of back-calculated spring


stiffness Ks from four continuum finite element analyses with dif-
ferent mat geometry and loading configurations. The common
characteristic among the distributions is that Ks has relatively
small values in a rather wide central region of the mat, indepen-
dently of the mat aspect ratio (L/B=1 or L/B=2) and whether the
loading is concentrated in the middle (loading C9) or more evenly
distributed across the mat (loading C6). Closer to the mat edges
(excluding the corner regions), the Ks rises sharply to values 2.5–
3.5 times the Ks at the mat center. This is in agreement with the
findings of Liao [26] based on 2D plane strain cFEA (strip mat).
For similar relative stiffness ratio and Poisson’s ratio values, the
Liao [26] results give back-calculated spring constant Ks at the edge
of the mat approximately 2.75 times (5.5 times in terms of modu-
lus of subgrade reaction ks) larger than that at the center. Accord-
ing to the present cFEA, at the mat corners, Ks attains its maximum
values, which are 3–4.5 times larger than those operative near the
center.
For the reminder, the Ks distributions will be presented normal-
ized with respect to a reference spring stiffness Kr. The Kr is chosen
to be the spring constant corresponding to a perfectly rigid mat
foundation (zero flexibility), which is calculated according the fol-
lowing formula derived based on Eq. (12):
Es Ainner
K r ¼ kr Ainner ¼ ð14Þ
C f ð1  m2s ÞB
where kr is the modulus of subgrade reaction for a perfectly rigid
mat and Ainner is the tributary area of the spring neglecting the pres-
ence of the boundaries of the mat. This means that, for springs
attached to the edge and corners of the mat, the tributary area is
calculated as if they were inner springs. The choice of using Ainner
instead of the actual tributary area for the springs at the mat bound-
ary is made in order to get smooth Ks distributions that could be
easily fitted by an analytical expression, as it will be shown in the
next section.
In Fig. 9, the Ks distributions back-calculated from cFEA are
compared to those from the Boussinesq-based iterative procedure
for a square mat 10 m wide and 0.5 m thick, on soil with Es = 10
Fig. 7. Examples of profiles of settlement along mat central axis (y = 0) from cFEA. MPa and vs = 0.3, for three different loading configurations
D. Loukidis, G.-P. Tamiolakis / Engineering Structures 153 (2017) 443–459 451

Fig. 8. Distribution of back-calculated spring stiffness Ks from analyses with soil properties Es = 10 MPa, vs = 0.3 and mat thickness d = 0.75 m.

(C1, C2 and C3). It can be seen that the distributions from the two configuration C3 is equal to 1/12 of the mat width. We see that
methods are qualitatively similar, with the basic characteristics the Ks distribution also turns out to be asymmetric, with Ks taking
being the large central region with small Ks and the sharp rise of the largest values at the edge that sinks the most into the soil
Ks near the mat edges, especially at the corners. The same trends (x = 5 m) due to the overall rotation of the slab. Moreover, the con-
can be observed in the Ks profiles (along the centerline and the trast in edge spring stiffness in the case of eccentric loading
edge of the mat) shown in Fig. 10 for a square mat 16 m wide (Fig. 9c) is much more striking than in the symmetric loading case
and 0.75 m thick loaded with 9 column loads (loading C1) and 25 of Fig. 9b.
column loads (loading C6). Both methods agree on that the Ks is
around 0.5Kr in the central region of the mat. However, according 5. Proposed equations for Ks distribution
to the Boussinesq-based iterative method, the increase of Ks
towards the corners of the mat is distinctively sharper compared In this section, two different equations approximating the Ks
to that calculated based on cFEA. For example, the corner Ks values distribution will be proposed for potential use in mat static analy-
from the iterative method in Fig. 9a and b are 43% and 39% larger sis. The first equation is a 6th degree polynomial that fits relatively
than those from cFEA, respectively. In addition, the Ks predicted well the ‘‘cup” shape of the back-calculated Ks distributions for the
by the iterative method in the central plateau region is slightly range of problem parameters examined in the present study, while
lower than that from cFEA (Fig. 10). the second equation, although much simpler and very roughly
In the case of load configuration C2 (Fig. 9b), the loading is sym- approximating the Ks distribution, is found to yield equally accu-
metric with respect to both x and y-axes, but the edges parallel to rate predictions with respect to mat bending moments.
the x-axis are loaded with larger loads and, as a result, sink into the The Ks distributions back-calculated from cFEA can be approxi-
soil more than the other two edges. This results in a more intense mately fitted by the following 6th degree polynomial:
coupling effect and, consequently, in larger Winkler spring con- ( " 6  6 #
stants along the edges parallel to the x-axis than to the y-axis. x þ 0:1ex y þ 0:1ey
K s ¼ K r ð0:55 þ C H1 Þ 1 þ 2C H2 þ
The case shown in Fig. 9c (loading configuration C3) illustrates L=2 B=2
the effect of an non-symmetric distribution of column loads lead-     

ex x ey y
ing to total load eccentricity (i.e. resultant of the column loads does þ4 þ ð15Þ
L L=2 B B=2
not pass through the center of mat). The load eccentricity in load
452 D. Loukidis, G.-P. Tamiolakis / Engineering Structures 153 (2017) 443–459

connuum FEA Boussinesq-based

(a)
connuum FEA Boussinesq-based

(b)

connuum FEA Boussinesq-based

(c)
Fig. 9. Comparison of Ks distribution back-calculated from continuum finite element analysis (cFEA) with that from the Boussinesq-based iterative solution for a mat with L =
B = 10 m, S = 5 m, d = 0.5 m on soil with Es = 10 MPa, vs = 0.3 and different loading configurations: (a) C1, (b) C2, and (c) C3.

with the x-axis oriented parallel to the longer dimension of the mat where xQi and yQi are the lever arms of column load Qi with respect
(length L), where ex and ey are the eccentricities of the resultant of to the mat center lines. The factors CH1 and CH2 introduce the influ-
column loads along the x- and y-axes, respectively. The eccentrici- ence of the presence of an undeformable substratum at depth H
ties are calculated using the formulas below the mat foundation and are given by the following
P expression:
Q xQ i
ex ¼ P i ð16aÞ  
Qi H
C H1 ¼ 0:45 exp 2:2 ð17aÞ
P B
Q iy
ey ¼ P Qi ð16bÞ  
Qi B
C H2 ¼ exp 0:4 ð17bÞ
H
D. Loukidis, G.-P. Tamiolakis / Engineering Structures 153 (2017) 443–459 453

Fig. 10. Comparison of Ks profiles from cFEA and Boussinesq-based iterative


solution for a mat with L = B = 16 m along centerline and along edge.

The coefficients of the 6th degree polynomial were determined


using non-linear regression over the cFEA back-calculated Ks val-
ues. The regression revealed that the coefficients of the 2nd to
5th degree terms were approximately equal to zero and thus could
be omitted from the expression. The 1st degree terms in Eq. (15)
are operative only if mat is loaded eccentrically (ex – 0 and/or ey
– 0) and their role is to effectively rotate the Ks distribution, as
observed in Fig. 9c. The coefficient of the 6th degree term is set
to decrease with the ratio of mat width B to the thickness of the
deformable soil H. This is because the analyses show only a mild
increase of Ks towards the mat edges in cases of a mat resting on
a relatively thin compressible layer. The quality of the fit of Eq.
(15) can be seen in Figs. 11–17, which are presented in the next
section.
Implementation in foundation engineering practice of Eq. (15),
which requires calculation and assignment of a different Ks value
at each mat node, may be cumbersome. Hence, an alternative
and simpler equation is also proposed, according to which all inner
nodes are connected to springs with stiffness 65% of Kr (provided
that H  B), while much higher values are assigned to the springs
of the edge nodes depending on the number of edge nodes (nodes
along the mat perimeter) Nedge relative to the total number of
nodes Ntotal in the mat: Fig. 11. Effect of soil Young’s modulus on the spring stiffness distribution.
(
0:65K r cH1 for inner nodes
Ks ¼ ð18Þ
0:65 þ 0:35 NNtotal K r cH2 for edge nodes for cases of H  B (i.e., cH1 = cH2 = 1.0). The latter condition stems
edge
from the fact that, for most of the cases examined with H  B,
The multipliers cH1 and cH2 are factors that introduce the effect the average of the cFEA back-calculated Ks is 0.9–1.0 times the
of the deformable soil thickness H and are given by the following Kr. The coefficients appearing in Eqs. (18) and (19) were chosen
equations, provided that B/H < 5: after establishing, using the steepest descent method, the optimal
uniform Ks value for the inner nodes and the optimal uniform Ks
B value for the edge nodes that lead to the best possible match
cH1 ¼ 1 þ 0:08 ð19aÞ
H between the deflections from the mat-on-springs analysis and
those obtained from cFEA.
B Eq. (18) does not take into account the effect of load eccentricity
cH2 ¼ 1  0:12 ð19bÞ
H and, thus, does not predict the associated rotation of the Ks distri-
The basic idea behind Eq. (18) is to have an enhanced edge bution. Nonetheless, it will be shown in the next section that using
spring stiffness, as suggested by Bowles [7], while the total stiff- this simple Ks distribution results in adequately accurate mat
P bending moments predictions.
ness of all springs combined, K s;i , is always equal to Kr  Ntotal,
454 D. Loukidis, G.-P. Tamiolakis / Engineering Structures 153 (2017) 443–459

Fig. 12. Effect of soil Poisson ratio on the spring stiffness distribution.
Fig. 14. Profiles of Ks for various loading configurations without eccentricity.

Fig. 13. Effect of mat thickness on the spring stiffness distribution.


Fig. 15. Effect of thickness of deformable soil on the spring stiffness distribution.

6. Influence of problem variables on Ks distribution Nonetheless, it will be demonstrated later that this shortcoming
does not affect the bending moments obtained from mat static
The effect of the soil Young’s modulus Es on the normalized analysis employing Eq. (15).
spring stiffness distribution can be seen in Fig. 11, which presents The Poisson’s ratio m and the thickness of the mat foundation d
results from analyses of the same mat problem, i.e. mat 10 m wide, have practically no influence on the Ks/Kr (Figs. 12 and 13), in
0.75 m thick, with loading configuration C1 (Fig. 11a) and C8 accordance with the findings of Liao [26] for strip (2D) mats. The
(Fig. 11b), but with different Es values ranging from 1 MPa to loading configuration of the mat has a slightly more pronounced
100 MPa. We see that, in all three cases, Ks at the mat corners turns effect (Fig. 14), with the curves of Ks/Kr corresponding to the mat
to be 2.5–3 times larger than Kr, while Ks/Kr falls in the 0.5–0.6 edge being more spread apart compared to the previous figures.
range in the central region. The main qualitative difference Generally, a large number of column loads acting on the edge
between the distributions shown in Fig. 11 is that for Es = 100 (e.g., cases C6 and C8) results in roughly 35% larger Ks/Kr values
MPa, which for the given problem corresponds to a relative stiff- at the mat edge compared to the case of a mat loaded only by a
ness factor 0.011, the Ks distribution shows small local peaks at central load (C9). Nonetheless, it was judged that the effect of load-
the points of application of column loads along both the edge ing configuration on Ks is not strong enough, from a practical per-
and the centerline. Eq. (15) cannot capture these local peaks. spective, to be taken into account in Eq. (15).
D. Loukidis, G.-P. Tamiolakis / Engineering Structures 153 (2017) 443–459 455

Fig. 17. Profiles of Ks for loading configuration with eccentricity in both axes.

100 MPa). Analyses using the Boussinesq-based iterative approach


with mat thickness values well outside this range show that for d <
0.25 m (Rs < 0.004), Ks distributions start to deviate significantly
from those shown in the figures of this section. On the contrary,
increasing d to 3 m (or even more) leads to no change in the Ks dis-
tribution. Hence, it has to be noted that the proposed Ks distribu-
tions (Eqs. (15) and (18)) may not be applicable for very thin
mats and/or very stiff soils/soft rocks (for which E > 100 MPa).

7. Effect of Ks distribution on moment and shear force diagrams

One of the main purposes of the static analysis of a mat founda-


tion in practice is the determination of the required steel reinforce-
ment based on moment and shear force diagrams. Fig. 18
demonstrates the influence of the assumed spring stiffness distri-
bution under a mat on the calculated bending moments (per unit
mat width), mx, along the mat central longitudinal axis (y = 0) in
a mat-on-springs static analysis. The moment is normalized with
respect to the sum of the column loads RQ times the column spac-
ing S. The moment diagrams shown in each plot correspond to the
spring stiffness distributions established herein, namely the Ks dis-
tributions established using the cFEA (Abaqus) results and the
Fig. 16. Profiles of Ks for loading configurations with eccentricity in one axis.
Boussinesq-based iterative approach of Ulrich [36], along with
those predicted by the proposed equations, Eqs. (15) and (18). In
addition, Fig. 18 plots results obtained using the Ks distributions
The Ks/Kr distribution depends heavily on the presence of an commonly assumed in practice, namely (1) uniform modulus of
undeformable substratum at depth H, provided that H/B is less subgrade reaction ks throughout the mat (i.e., the original, uncou-
than 1 (Fig. 15). As the thickness of the deformable foundation soil pled Winkler approach), (2) multiplying by a factor of 2 the ks for
layer decreases, the inner Ks/Kr values increase, while the Ks/Kr at the springs lying at the edge of the mat [7], (3) considering concen-
the edge decreases, apparently converging progressively towards tric zones of different ks as shown in Fig. 2 [8]. For the calculations
unity. The dashed lines (Eq. (15) predictions) follow the observed using the method of Fig. 2, a size factor rB for the intermediate zone
trends thanks to the soil thickness factors CH1 and CH2. The perfor- (zone B) equal to 1/8 is assumed herein, resulting, according to Eq.
mance of Eq. (15) in predicting the equivalent Ks distribution in the (2) and Fig. 2, to ks1 = 0.63kr, ks2 = 0.94kr and ks3 = 1.25kr. It should
case of eccentrically loaded mats can be seen in Fig. 16 (single axis be noted that, in the case of the distributions used in practice (uni-
eccentricity) and Fig. 17 (double axis eccentricity). Eq. (15) cap- form ks, doubling ks at the edge, and the Fig. 2 ks distribution), the
tures adequately the rotation of the Ks/Kr distribution, with devia- concept of tributary areas applies as discussed in the Knowledge
tions becoming relatively large in the case of loading C5 (Fig. 16b), Background section and shown in Fig. 1. In the case of an eccentri-
in which case the load eccentricity is equal to 1/6 of the mat width. cally loaded mat (Fig. 18g and h), the plots show also the bending
The results of the parametric study suggest that there is no need moment curves obtained using Eq. (15) but with the terms related
for the Ks distribution to be a function of the relative stiffness factor to load eccentricity omitted (i.e., inputting ex = ey = 0). The
Rs. The parametric runs were performed for mat thickness values Boussinesq-based solution, as employed herein, assumes an infi-
encountered in practice (d  0.5 m) and not very stiff soils (Es  nitely extending soil layer under the mat. Hence, this method
456 D. Loukidis, G.-P. Tamiolakis / Engineering Structures 153 (2017) 443–459

Fig. 18. Diagrams of bending moment along mat central axis (y = 0) for various spring stiffness distributions.
D. Loukidis, G.-P. Tamiolakis / Engineering Structures 153 (2017) 443–459 457

was not applied in the analysis of Fig. 18f, in which the thickness of
the deformable soil layer H is very small (¼ of the mat width B).
The agreement in moment predictions using the Ks from the
proposed equations, cFEA and the Boussinesq-based iterative solu-
tion [36] is very good in all plots of Fig. 18. Eq. (18) performs
equally well as Eq. (15) in predicting bending moments, even in
the cases of eccentrically loaded mats (Fig. 18g, h), despite the sim-
plicity of Eq. (18) and the fact that it does not take into account the
effects of load eccentricity on the shape of the Ks distribution. In
fact, analysis with Eq. (15) yields almost the same bending
moment diagram, even when omiting the eccentricity related
terms of the equation (Fig. 18g, h). This suggests that the
eccentricity-induced rotation of the Ks distribution, which was
observed in Figs. 9c, 16 and 17, does not practically influence the
inner stress state of the mats. However, neglecting the eccentricity
related terms in Eq. (15) or using Eq. (18) leads to a significant
overestimation of the overall rotation of the mat (Fig. 19) by
>80%. Only consideration of the complete Eq. (15) ensures an accu-
rate prediction of the mat rotation due to eccentric loading that
Fig. 19. Profile of settlement along mat central axis (y = 0) from analysis with matches adequately the results based on cFEA and the
eccentric loading for various spring stiffness distributions. Boussinesq-based solution.

Fig. 20. Examples of shear force diagrams along mat central axis (y = 0) for various spring stiffness distributions: (a) central loading, (b) eccentric loading.
458 D. Loukidis, G.-P. Tamiolakis / Engineering Structures 153 (2017) 443–459

Generally, the approaches used in current practice underesti- moment diagrams that are close to those based on contin-
mate substantially the positive peak bending moments (bottom uum finite element analysis.
fiber in tension) and overestimate peak negative moments (top (4) The spring stiffness distributions commonly assumed in
fiber in tension). The uniform ks approach exhibits the worst per- practice, namely the uniform modulus of subgrade reaction
formance, with errors that exceed 70% in certain cases ks approach and the approach of doubling the ks at the edge
(Fig. 18a, b), an issue that has already been raised in the technical nodes, may lead to an underestimation of the peak positive
literature [7,8,19]. For the cases shown in Fig. 18, the uniform ks bending moments (bottom fiber in tension) of the order of
distribution approach, the approach of doubling the ks at the edge 40% and 30%, respectively, and to an overestimation of peak
nodes, and the ks zoning of Fig. 2 result in an underestimation of negative bending moments (top fiber in tension) of the order
the peak positive bending moments by 38%, 29% and 16% on aver- of 70% and 50%, respectively.
age and 71%, 59% and 31% maximum, respectively, while they (5) The shear force diagrams are largely unaffected by the
result in an overestimation of peak negative bending moments assumed spring stiffness distribution.
by 72%, 49% and 44% on average and 106%, 69% and 71% maximum,
respectively. Nonetheless, it is interesting to note that there are Finally, it must be stressed that the conclusions of this study
cases in which all approaches perform well. The most striking rely on the assumption that the soil is a linear elastic material.
one is that of a mat loaded only by a central load (loading C9, Hence, the present findings should be further checked in the future
Fig. 18e), where the moment at the critical cross-section (point of against results from analyses in which the soil is modeled as a non-
load application) turns to be practically independent of the linear material. This is because the part of the foundation soil that
assumed Ks distribution (relative differences with respect to peak lies near the edges of the mat is subjected to more shearing than
moment do not exceed 6%). In addition, the differences among all soil near the mat center, where stress and deformation conditions
approaches appear comparatively small when the thickness H of are more oedometric, leading to a redistribution of contact pres-
the deformable soil layer is small (Fig. 18f, H/B = 0.25). sures due the non-linear nature of the soil shearing behavior
Fig. 20 plots the vertical shear force Vx (per unit mat width) [37,13]. The shear stresses close to the mat edges may be near or
along the mat central longitudinal axis (y = 0) for two examples at the shear failure envelope, even under serviceability loads in
(centrally and eccentrically loaded mat). It can be observed that, the case of weak soils (e.g., loose sands, normally consolidated
contrarily to the bending moments, the vertical shear forces seem clays), despite that the bearing pressures of mat foundations are
to be quite insensitive to the choice of the distribution of Winkler typically small, only about 12 kPa per supported storey [31], and
spring constants. This can be explained by the fact that the values that the operative factor of safety (FS) against bearing capacity fail-
of the extrema of the diagrams, which coincide with the points of ure in the case of mat supported buildings is usually very large
application of column loads, are predominantly and tightly con- (much larger than the requirement for FS = 2.0 or 3.0). Nonetheless,
trolled by the values of the column loads according to static equi- in the case of soils with a strongly cohesive characters (clayey
librium principles, given also the fact that the spring reactions of soils), the ground tends to remain away from failure for a substan-
the immediately underlying springs are comparatively small. The tial range of magnitude of superstructure loading. Trial cFEA anal-
relative differences in Vx between the analyses using different Ks yses of a rectangular mat on soil modeled as an elastic-perfectly
distributions become large near midspan cross-sections, which plastic material following the Tresca failure criterion (e.g., clay
are nevertheless non-critical with respect to shear or punching under undrained conditions) using Plaxis 3D show that, while
failure. the soil near the mat corners starts entering yield state when the
operative FS values become smaller than 7.5, the operative FS
needs to become less than 4.0 for the soil along the mat edges to
8. Conclusions commence yielding. In the case of cohesionless soils (sand and
gravers) or slightly cohesive soils, the soil along the perimeter of
Finite element analysis using three-dimensional continuum ele- the mat is expected to be at yield state for markedly smaller load-
ments (cFEA) was used for establishing the appropriate values of ing magnitudes. Viladkar et al. [38] performed cFEA of a soil-mat-
Winkler spring constants Ks for use in static analysis of rectangular structure interaction problem considering a hyperbolic (non-
mat foundations using the pseudo-coupled approach. Based on the linear) stress-strain law for cohesive soil. It was observed that
FE results, equations approximating the spring constants spatial the inclusion of the soil non-linearity in the problem solution
distribution are proposed. According to the findings of the present resulted in only ±6% change in the maximum bending moments
study, the following conclusions can be drawn: compared to a purely linear elastic solution. Thus, the use of the
proposed equations is likely more reliable for the case of cohesive
(1) The cFEA back-calculated Ks values in the central 60% of the soils than cohesionless soils.
mat are practically constant and equal approximately to
0.55 times the spring constant corresponding to a perfectly
rigid foundation Kr. Closer to the edges of the mat, Ks increases References
sharply, attaining values that are roughly 1.5–3 times larger
[1] ABAQUS. Analysis user’s manual, Simulia, Dassault systems; 2011.
than Kr, with the larger values being operative at the mat cor- [2] ACI Committee 336. Suggested analysis and design procedures for combined
ners. The average value of Ks across the mat is very close to Kr. footings and mats (ACI 336.2R-88). American Concrete Institute (ACI); 1988. p.
(2) The form of the Ks distribution is rather insensitive to the 336.2R-1-336.2R-21.
[3] Banavalkar PV, Ulrich Jr., EJ. Structural and geotechnical features of
relative stiffness of the mat and the Poisson’s ratio of the
Republicbank center. In: Proceedings, council on tall buildings and urban
soil. The distribution of the column loads also appears to habitat, Singapore; 1984. p. 363–70.
have only a small effect on the shape of the Ks distribution, [4] Basile F. Non-linear analysis of vertically loaded piled rafts. Comput Geotech
2015;63:73–82.
provided that the mat loading does not induce significant
[5] Belytschko T, Tsay CS, Liu WK. A stabilization matrix for the bilinear Mindlin
overall rotation of the mat (case of eccentric resultant of col- plate element. Comput Methods Appl Mech Eng 1981;29:313–27.
umn vertical loads). If the column load resultant is eccentric, [6] Boussinesq J. Application des potentiels a l’équilibre et du mouvement des
the Ks distribution appears also rotated. solides élastiques. Gauthier-Villars; 1885.
[7] Bowles JE. Foundation analysis and design. 5th ed. McGraw-Hill; 1996.
(3) The Boussinesq-theory-based iterative approach of Ulrich [8] Coduto DP. Foundation design – principles and practices. 2nd ed. Prentice Hall;
[36] (discrete area method) yields Ks values and bending 2001.
D. Loukidis, G.-P. Tamiolakis / Engineering Structures 153 (2017) 443–459 459

[9] Colasanti RJ, Horvath JS. Practical subgrade model for improved soil-structure [26] Liao SSC. Estimating the coefficient of subgrade reaction for plain strain
interaction analysis: software implementation. Pract Period Struct Des Constr condition. Proc Inst Civil Eng: Geotech Eng 1995;113(3):166–81.
2010;15(4):278–86. [27] Matlab. User’s guide, mathworks; 2012.
[10] Horvath JS, Colasanti RJ. Practical subgrade model for improved soil-structure [28] Meyerhof GG. Some recent foundation research and its application to design.
interaction analysis: model development. Int J Geomech 2011;11(1):59–64. Struct Eng 1953;31(6):151–67.
[11] Cook RD, Malkus DS, Plesha ME, Witt RJ. Concepts and applications of finite [29] Newmark NM. Simplified computation of vertical pressures in elastic
element analysis. Wiley; 2001. foundations. Engineering Experiment Station. University of Illinois at
[12] Chow YK. Vertical deformation of rigid foundations of arbitrary shape on Urbana-Champaign; 1935.
layered soil media. Int. J. Numer. Anal. Method. Geomech. 1987;11(1):1–15. [30] Roy R, Dutta SC. Differential settlement among isolated footings of buildings
[13] Edgers L, Saayei M, Alonge JL. Modeling the effects of soil-structure interaction frames: the problem, its estimation and possible measures. Int J Appl Mech
on a tall building bearing on a mat foundation. Civil Eng Pract 2005:51. Eng 2001;6(1):165–86.
[14] Fraser RA, Wardle LJ. Numerical analysis of rectangular rafts on layered [31] Salgado R. The engineering of foundations. International ed. McGraw-Hill;
foundations. Géotechnique 1976;26(4):613–30. 2008.
[15] Holtz RD. Stress distribution and settlement of shallow foundations. In: Fand [32] Savaris G, Hallak PH, Maia PCA. Influence of foundation settlements in load
H-Y, editor. Foundation engineering handbook. Springer; 1991. p. 166–222. redistribution on columns in a monitoring construction-Case Study. Rev
[16] Horvath JS. New subgrade model applied to mat foundations. J Geotech Eng IBRACON Estrut Mater 2010;3(3):346–56.
1983;109(12):1567–87. [33] Steinbrenner W. Tafeln zur Setzungsberechnung. Die Strasse 1934;1:121–4.
[17] Horvath JS. Subgrade models for soil-structure interaction analysis. In: [34] Tamiolakis G-P. Study of the spatial distribution of the Winkler spring stiffness
Foundation engineering: current principles and practices. New York: for the static analysis of mat foundations MSc Thesis. Nicosia
American Society of Civil Engineers; 1989. p. 599–612. (Cyprus): University of Cyprus; 2012 [in Greek].
[18] Horvath JS. Subgrade modeling for mat foundations: a review and critique of [35] Timoshenko SP, Goodier JN. Theory of elasticity. McGraw Hill International
analytical methods. Special Publication SP-152, ACI; 1995. p. 117–60. Book; 1982.
[19] Horvath JS. Soil-structure interaction research project – basic ssi concepts and [36] Ulrich Jr EJ. Subgrade reaction in mat foundation design. Concr Int 1991;13
applications overview. Report no CGT-2002-2, Center for Geotechnology, (4):41–50.
Manhattan College; 2002. [37] Viladkar MN, Godbole PN, Noorzaei J. Soil-structure interaction in plane
[20] Horvilleur JF, Patel VB. Mat foundation design - a soil-structure interaction frames using coupled finite-infinite elements. Comput Struct 1991;39
problem. In: Ulrich EJ, editor. Design and performance of mat foundations; (5):535–46.
state-of-the-art review. Detroit: ACI; 1995. p. 51–94. [38] Viladkar MN, Noorzaei J, Godbole PN. Interactive analysis of a space frame-
[21] Hughes TJR. The finite element method: linear static and dynamic finite raft-soil system considering soil nonlinearity. Comput Struct 1994;51
element analysis. Dover Publications; 2000. (4):343–56.
[22] Jeong S, Cho J. Proposed nonlinear 3-D analytical method for piled raft [39] Westergaard HM. A problem of elasticity suggested by a problem in soil
foundations. Comput Geotech 2014;59:112–26. mechanics: soft material reinforced by numerous strong horizontal sheets. In:
[23] Kerr AD. A study of a new foundation model. Acta Mech 1965;1:135–47. Contributions to the mechanics of solids, Stephen Timoshenko 60th
[24] Kerr AD, Rhines WJ. A further study of elastic foundation models. Report no S- anniversary volume. New York: MacMillan; 1938. p. 260–77.
67-1. New York University, Bronx, USA; 1967. [40] Winkler E. Die Lehre von Elastizität und Festigkeit. Prague: H. Dominicus;
[25] Levinson M, Bharatha S. Elastic foundation models—a new approach. In: Proc 1867.
4th symposium on eng appl solid mech; 1978.

You might also like