Escolar Documentos
Profissional Documentos
Cultura Documentos
BSc, MSc
August 2009
This work has been primarily motivated by the lack of sophisticated monolithic tools
for modelling nonlinear soil–structure interaction problems, while recognising the
existence of advanced tools for nonlinear analysis of structure and soil in isolation.
Although coupled modelling of soil-structure interaction problems may be achieved
using a monolithic treatment, the partitioned treatment has been advocated as
offering major benefits in the context of coupled modelling of nonlinear soil-
structure interaction. Accordingly, the aim of this work has been to develop
advanced numerical methods for nonlinear coupling of soil-structure interaction
problems, where the partitioned approach is adopted as a framework for coupling
field-specific tools with minimal intrusion into codes.
In this respect, the partitioned approach for soil-structure interaction analysis has
been fully investigated in this work. Various coupling techniques are developed in
the context of soil-structure interaction analysis, and their computational
characteristics are discussed. Novel formulations for coupling soil-structure systems,
based on relaxation coupling methods and also utilizing the tangent stiffness matrix
of the partitioned sub-domains at the interface, are proposed, and their relative
performance is evaluated. The proposed approaches are believed to possess superior
convergence characteristics in comparison with existing coupling methods, rendering
these methods more general procedures for modelling soil-structure interaction
problems using coupled field-specific tools.
2
Abstract
high potential are demonstrated. Indeed, it is shown that the partitioned treatment is a
feasible and realistic approach for coupled modelling of nonlinear soil-structure
interaction problems, providing an integrated interdisciplinary computational
approach which combines the advanced features of both structural and geotechnical
modelling for a variety of challenging soil-structure interaction problems.
3
Acknowledgements
First and foremost I would like to express my deep and sincere gratitude to my
supervisors, Prof. Bassam A. Izzuddin and Dr. Lidija Zdravkovic. I owe a great deal
of appreciation for this thesis to Prof. Bassam A. Izzuddin for his insightful
supervision, invaluable guidance and unfailing support from very early stages of this
research, without which this thesis would not have been possible. I thank him for his
suggestions, refinements and constructive comments on full drafts. Likewise, I am
deeply grateful to Dr. Lidija Zdravkovic for her continuous encouragement and
support, substantive guidance, valuable comments on full drafts, and her overall
contribution to the final outcome.
Being part of the research groups in the Systems & Mechanics section and the Soil
Mechanics section at Imperial College has been a great experience. Thanks are due
to all academic staff and students for creating such an enjoyable and unique
environment. Special thanks are due to Mrs. Fionnuala Donavan the Postgraduate
Administrator, whose kindness will always be remembered. Many thanks are owed
to my friends and colleagues for the remarkable moments I have shared with them.
I am very grateful and deeply indebted to my parents for their inseparable love and
support throughout my life. Their support has been instrumental in all my
achievements thus far. I would like to thank my brother for being supportive and
caring.
Words fail me to express my deep appreciation to my wife Ahoura for her dedication
and persistence confidence in me. Without her love and understanding I would not
have been able to complete this thesis.
4
Table of Contents
Abstract 2
Acknowledgements 4
List of Tables 9
List of Figures 11
Notation 19
Chapter 1
1 Introduction 21
1.1 Soil-Structure Interaction 21
1.2 Coupled Systems 23
1.3 Treatment of Coupled Soil-Structure Interaction 24
1.4 Partitioned vs. Monolithic Approaches 26
1.5 Aims and Scope of Research 27
1.6 Layout of Thesis 29
Chapter 2
2 Literature Review 32
2.1 Background 32
2.2 Modelling of Soil-Structure Interaction 34
2.2.1 Field Elimination 35
2.2.2 Finite Element Method (FEM) 40
2.2.3 Integrated Modelling 42
2.2.4 Partitioned Analysis 46
2.2.5 Staggered Approach 48
2.2.6 Iterative Coupling 52
2.3 Concluding Remarks 60
Chapter 3
3 Staggered Approach 63
5
Table of Content
3.1 Introduction 63
3.2 Dynamic Analysis Formulation 65
3.3 Partitioning 67
3.4 Staggered Coupling Procedure 70
3.5 Predictors 71
3.6 Stability Analysis by Amplification Method 72
3.7 Stability Analysis of Staggered Coupling Scheme for a Test System 76
3.8 Accuracy 84
3.9 Example 84
3.10 Conclusion 91
Chapter 4
4 Iterative Coupling 92
4.1 Introduction 92
4.2 Iterative Coupling 96
4.2.1 Sequential Dirichlet-Neumann Iterative Coupling 97
4.2.2 Sequential Neumann-Dirichlet Iterative Coupling 98
4.2.3 Parallel Dirichlet-Neumann Iterative Coupling 100
4.2.4 Parallel Neumann-Dirichlet Iterative Coupling 102
4.2.5 Parallel Dirichlet-Dirichlet Iterative Coupling 103
4.2.6 Parallel Neumann-Neumann Iterative Coupling 105
4.3 Treatment of Interactive Boundary Conditions 106
4.4 Convergence of Iterative Coupling 107
4.4.1 Convergence of Sequential D-N with Trivial Update 107
4.4.2 Example 112
4.5 Simulation Environment 117
4.6 ADAPTIC 117
4.7 ICFEP 119
4.8 INTERFACE 121
4.9 Simulation Environment Architecture 122
4.10 Data Communication 125
4.11 Concluding Remarks 127
6
Table of Content
Chapter 5
5 Interface Relaxation 129
5.1 Introduction 129
5.2 Constant Relaxation 131
5.2.1 General Convergence Analysis 134
5.2.2 Convergence Studies 142
5.2.2.1 Example 1: Dynamic FEM-FEM Coupling 142
5.2.2.2 Example 2: Static FEM-FEM Coupling 154
5.3 Adaptive Relaxation 159
5.3.1 Convergence Analysis 159
5.3.2 Convergence Studies 163
5.3.2.1 SDOF Test System 163
5.3.2.2 Example 1: Dynamic FEM-FEM Coupling 164
5.3.2.3 Example 2: Static FEM-FEM Coupling 170
5.4 Soil-Structure Interaction Analysis 172
5.5 Concluding Remarks 184
Chapter 6
6 Reduced Order Method 187
6.1 Introduction 187
6.2 Condensed Interface Tangent Stiffness 188
6.2.1 Numerical Example 193
6.2.2 Analogy between Interface Relaxation and Condensed Interface
Stiffness Approaches 198
6.3 Approximation of the Condensed Tangent stiffness 201
6.3.1 Condensed Interface Secant Stiffness Matrix 202
6.3.1.1 Example 1: Static FEM-FEM Coupling 204
6.3.1.2 Example 2: Dynamic FEM-FEM Coupling 208
6.3.1.3 Discussion on Nonlinear Analysis 213
6.3.2 Reduced Order Method 214
7
Table of Content
8
List of Tables
Table 5.1: Range of suitable and optimal relaxation parameter for different
m1
145
m2
Table 5.2: Range of suitable and optimal relaxation parameter for different
ET
147
EB
Table 5.3: Range of suitable and optimal relaxation parameter for different
ET / EB 155
Table 5.4: Number of coupling iterations with adaptive and optimum relaxation
for 1000 time-steps ( t 0.01s ) with a tolerance of 10-4 m 165
Table 5.5: Number of required coupling iterations with adaptive and optimum
relaxation 171
9
List of Tables
Table 6.2: Required coupling iterations for different coupling schemes 206
Table 6.4: Required coupling iterations for different coupling schemes 229
Table 7.7: Loading scenario in different incremental stages of the analysis 290
Table 7.8: Loading scenario in different incremental stages of the analysis 293
10
List of Figures
Figure 2.4: Field elimination model for SSI analysis of lateral vibration of an
elevated water tank 39
Figure 2.6: a) Schematic illustration of the pile and soil model b) Schematic diagram
of the structure model 44
11
List of Figures
12
List of Figures
Figure 5.1: Variation of error reduction factor against relaxation parameter 133
Figure 5.8: Error reduction for different relaxation schemes (Time = 3.6s) 148
Figure 5.9: Error reduction for different relaxation schemes (Time = 3.5s) 149
Figure 5.10: Error reduction for different relaxation schemes (Time = 4.18s) 149
Figure 5.17: Rotation at the tip of the cantilever (Tolerance 105 ) 153
Figure 5.20: Error reduction for different relaxation schemes (model M4) 157
13
List of Figures
Figure 5.31: Error reduction for adaptive and constant relaxation schemes 171
Figure 5.38: Error reduction for different relaxation schemes for the first load
increment 178
Figure 5.39: Error reduction for different relaxation schemes for the fifth load
increment 178
Figure 5.40: Error reduction for different relaxation coupling schemes for the last
load increment 179
Figure 5.44: Vector plot of displacements in the soil partitioned sub-domain 182
14
List of Figures
Figure 5.46: Deformed shape and bending moment (kN.m) of the frame (scale=5) 183
Figure 6.3: Error reduction of different coupling schemes for model A1 206
Figure 6.4: Error reduction of different coupling schemes for model A4 207
Figure 6.5: Error reduction of different coupling schemes for model A6 207
Figure 6.10: Comparison between different coupling schemes for Model C1 212
Figure 6.11: Comparison between different coupling schemes for Model C5 213
Figure 6.18: Error reduction of different schemes (Time=2.74s) for Model C1 236
Figure 6.19: Error reduction of different schemes (Time=3.63s) for Model C1 236
Figure 6.20: Error reduction of different schemes (Time=4.92s) for Model C1 237
Figure 6.21: Error reduction of different schemes (Time=4.42s) for Model C1 237
Figure 6.22: Error reduction of different schemes (Time=1.19s) for Model C1 238
15
List of Figures
Figure 6.23: Error reduction of different scheme (Time=2.74s) for Model C1 238
Figure 7.11: Deformed shape (scale=5.0) and bending moment (kN-m) in final load
step for a) non-interactive case b) interactive case 263
Figure 7.12: Contours of stress level in soil sub-domain (at final increment) 263
16
List of Figures
Figure 7.18: Bending moment (kN-m) at the base of C1 for different load-steps 271
Figure 7.19: Bending moment at the middle of beam B1 for different load-steps 271
Figure 7.20: Deformed shape (scale=5.0) and bending moment (kN-m) in final load
step for a) linear Winkler foundation b) nonlinear partitioned analysis 272
Figure 7.24: Plane frame resting on soil subject to ground excavation 275
Figure 7.27: Vertical settlement of the left footing for different load cases 279
Figure 7.28: Vertical settlement of the middle footing for different load cases 279
Figure 7.29: Vertical settlement of the right footing for different load cases 280
Figure 7.32: Vertical displacement of ground surface for different excavation depths
(Case 6) 282
Figure 7.33: Horizontal displacement of the excavation wall for different excavation
depths (Case 6) 282
Figure 7.34: Vertical settlement of the left footing for different Le (Case 6) 283
Figure 7.35: Vertical settlement of the right footing for different Le (Case 6) 284
Figure 7.36: Vertical settlement of the right footing for different Le (Case 6) 284
17
List of Figures
Figure 7.38: Contour plots of stress levels and plasticity induced in soil sub-domain
in increment 6 (Case 6) 286
Figure 7.40: Contour plots of stress levels and plasticity induced in soil sub-domain
in increment 12 (Case 6) 287
Figure 7.41: Deformed shape (scale=5) and bending moment (kN-m) of structure for
(a) 1st, (b) 6rd , (c) 7th and (d) 12th increment 288
Figure 7.44: Deflection (scale: 5) and bending moment (kN-m/m) distribution of the
retaining wall for (a) 1st, (b) 4th, (c) 7th, (d) 8th and (e) 9th increment 292
Figure 7.47: Deflection (scale: 5) and bending moment (kN-m/m) distribution of the
retaining wall for (a) 1st, (b) 2nd , (c) 3rd , (d) 4th and (e) 5th increment 295
18
Notation
All symbols used in this thesis are defined where they first appear. The reader is
cautioned that some symbols denote more than one quantity; in such cases the
meaning should be clear when read in the context.
19
Notation
U X
X : displacement vector of the non-interface degrees of freedom in sub-
domain ΩX
F X
X
: force vector of the non-interface degrees of freedom in sub-domain ΩX
U i
X : displacement vector of the interface degrees of freedom in ΩX
F i
X : force vector of the interface degrees of freedom in sub-domain ΩX
20
Chapter 1
Introduction
There are numerous problems in Civil Engineering construction that require the use
of realistic models for the structure, the supporting soil and the soil-structure
interface. Examples include the assessment of various structures under earthquake
loading, the analysis of offshore jackup structures under extreme wave loading and
the evaluation of the structural damage due to excavations to name but a few. This
interaction can sometimes modify the stresses and deflections of the whole structural
system significantly. In fact, the structure with its loading conditions imposes
stresses and forces on the ground, which in turn deforms and as a consequence
transmits back additional forces and deformation to the structure. This process
continues until full equilibrium of the whole soil-structure system is satisfied, or
until both the soil and the structure fail in the case of excessive loading and
deformations of the system.
Numerical analysis, typically using the finite element or finite difference method, is
currently the most advanced tool available to facilitate soil and structure analysis.
Chapter 1 Introduction
With the development of these numerical tools, there are advanced techniques
employed for addressing the behaviour of structure and soil. Although there are
common techniques for structural and geotechnical modelling, the distinct demands
of the two fields meant that modelling has evolved differently in each domain, thus
leading to two modelling disciplines with distinctive high-level features.
The above distinct challenges in the two fields have been reflected in the
development of discipline-oriented computational tools, which offer sophisticated
nonlinear modelling for their respective domain but, at best, a crude approximation
of the other, leading to poor representation of soil-structure interaction. This usually
means that structural analysis simplifies soil behaviour, while geotechnical analysis
simplifies structural behaviour using field elimination techniques. It is therefore a
real challenge to achieve the same amount of sophistication in modelling both the
soil and the structure in a single soil-structure interaction analysis.
22
Chapter 1 Introduction
One of the generic classes of coupled systems in the field of Civil Engineering
relates to the modelling of soil-structure interaction problems, where due to the
interaction between the physically distinct sub-domains, soil and structure, at the
interface, neither domain can be solved separately from the other without undue
simplifications, especially in the nonlinear range of response.
23
Chapter 1 Introduction
In principle there are three possible approaches for dealing with soil-structure
interaction problems, namely: i) field elimination, ii) direct/monolithic/
simultaneous, and iii) domain decomposition/partitioned analysis (Rugonyi & Bathe,
2001).
In the field elimination treatment, one or more sub-domains of the coupled problem
are eliminated using a simple reduction technique, and the remaining sub-domain(s)
are considered under appropriate boundary conditions representing the eliminated
sub-domain(s). An example of implementing the field elimination technique in
modelling soil-structure interaction problems is illustrated in Figure 1.1, where the
soil sub-domain is replaced by spring type boundary conditions in order to model the
interaction effects.
Field elimination techniques are mostly restricted to simple problems that permit
efficient decoupling. In the context of nonlinear soil-structure interaction problems,
the simplification made in the modelling brings with it a loss of generality and
accuracy, due to the complexity of numerical modelling of this phenomenon.
24
Chapter 1 Introduction
In contrast with the field elimination treatment, the monolithic and partitioning
treatments as discussed in the following are general in nature. In the monolithic
treatment of soil-structure interaction, the whole problem is modelled as a single
computational entity and the solution of the complete system of equations of the
coupled problem is attenuated in one analysis scheme. Alternatively, in partitioned
analysis as shown in Figure 1.2, the soil-structure coupled system is physically
partitioned into soil and structure sub-domains. These partitioned sub-domains are
then computationally treated as isolated entities, and the response of each sub-
domain is calculated using already developed soil and structural solvers. In this
procedure, the interaction effects are viewed as force/displacement effects at the
interface of soil-structure system which are communicated between the individually
modelled soil and structure sub-domains using prediction, substitution and
synchronization techniques.
25
Chapter 1 Introduction
26
Chapter 1 Introduction
However, despite the significant potential benefits of the partitioned approach, this
method should be formulated and implemented with great care, since the stability
and convergence characteristics of coupling algorithms are typically conditional and
problem dependent. Moreover, gain in computational efficiency using the partitioned
approach over monolithic treatment is not guaranteed.
This work aims at developing an advanced generic numerical method for nonlinear
coupling of static and dynamic soil-structure interaction problems. In this respect,
this work is primarily motivated by the lack of sophisticated monolithic tools for
modelling nonlinear soil–structure interaction. Recognising the existence of
advanced tools for nonlinear analysis of structure and soil, in isolation, the
partitioned approach is adopted as a framework for coupling field-specific tools with
minimal intrusion. Accordingly, in adopting the partitioned treatment, the focus is on
providing an advanced capability for modelling nonlinear soil–structure interaction
with existing field-specific codes rather than on achieving superior computational
performance in comparison with the monolithic treatment.
27
Chapter 1 Introduction
28
Chapter 1 Introduction
Finally, the developed tool is used in a number of case studies involving nonlinear
soil-structure interaction in which nonlinearity arises in both the structure and the
soil, thus leading to important conclusions regarding the adequacy and applicability
of the proposed coupling approaches for such problems, as well as the prospects for
further enhancements.
29
Chapter 1 Introduction
30
Chapter 1 Introduction
In this respect, various novel reduced order method coupling algorithms are
developed and their performance is examined. The applicability of the presented
coupling techniques is demonstrated for nonlinear soil-structure interaction analysis
via an example investigating the interactive behaviour of a plane frame and a
supporting soil system.
Chapter 7 (Case Studies): The applicability of the partitioned treatment and its
great potential towards performing realistic nonlinear soil-structure interaction
analysis are illustrated through various case studies in this chapter.
31
Chapter 2
Literature Review
2.1 Background
For example, consider the frame of Figure 2.1 supported on soil. As the loading on
the frame is symmetric, it is clear that there is a higher concentration of load over the
middle column and its support. This results in the soil below the middle footing
sustaining greater vertical settlements than the adjacent footings. In turn, framing
Chapter 2 Literature Review
action causes redistribution of the loads to the end columns due to the generated
differential settlements. These differential settlements among various parts of the
structure alter both the axial forces and the moments in the structural members
considerably, where the amount of redistribution of loads depends upon the rigidity
of the structure and the load-settlement characteristics of soil (Dutta & Roy, 2002).
This considerable influence of the structural rigidity on the interactive response has
been qualitatively explained in the literature long back (Taylor, 1964).
Indeed, such interaction between structure and soil at their common interface leads
to an actual response for both physical sub-domains, which is considerably different
from what is obtained for each sub-domain in isolation. In view of the
interdependence between the responses of the two sub-domains, the frame structure,
its foundations and the soil on which it rests constitute together a coupled system.
The effect of soil–structure interaction under both static and dynamic loading has
attracted significant research interest over many years due to its important role in the
analysis of real civil engineering problems. In this respect, research by
Noorzaei et al. (1993), Allam et al. (1991), Roy & Dutta (2001) and Potts &
Zdravkovic (2001) on static soil-structure interaction analysis, and also research by
33
Chapter 2 Literature Review
Stewart et al. (1999) and Dutta et al. (2004) on dynamic soil-structure interaction
analysis, demonstrates the significant effect of soil-structure interaction and its
importance in predicting the overall coupled response. There are also numerous
reported case histories in the literature which demonstrate the significant influence of
considering soil-structure interaction effects in civil engineering design and practice,
such as the destruction of buildings in Carcas earthquake in 1967 (Dowrick, 1977),
the enormous damage to structures in the Hanshin and Awaji areas during Kobe
earthquake in 1999 (Inaba et al., 2000), the case of leaning tower of Pisa (Burland &
Potts, 1994), and the reported damage to several building structures due to
differential settlements (Charles & Skinner, 2004). All of these researches deliver an
important message that considering soil-structure interaction effects can be essential
for a multitude of real civil engineering construction problems.
The primary expected results from a coupled soil-structure interaction (SSI) analysis
are the stresses and displacements of the structure and the soil sub-domains. SSI
analysis tools can be used in design to calculate the stresses and deformations of the
structure and soil sub-domain to compare with the allowable values and, if
necessary, to modify the system configuration so as to meet specific design criteria
of serviceability, safety and economy. In addition, SSI tools can be used for the
assessment of existing coupled systems, enabling the prediction of potential damage
and the proposal of strengthening and repair measures.
Towards this end, modelling the material behaviour of both structure and soil,
modelling the geometric nonlinearity of structure under extreme events, modelling of
special boundary conditions such as excavation in soil, etc., are amongst the most
central issues that should be considered in a reliable SSI analysis.
Works by Jardine et al. (1986), Wrana (1993), Noorzaei et al. (1993), Noorzaei et al.
(1995a), Inaba et al. (2000), Estorff & Firuzian (2000), Krabbenhoft et al. (2005)
and Bourne-Webb et al. (2007), concentrating on the nonlinear modelling of soil-
34
Chapter 2 Literature Review
There exist various models, algorithms and implementations for modelling SSI
problems. These all depend on the availability and limitations of computational tools
to carry out the SSI analysis, the physical effects that are to be captured, the desired
degree of accuracy, etc. In the following some of these modelling techniques are
presented and their advantages and disadvantages are briefly discussed.
For a long time, field elimination techniques have been favoured over fully coupled
analysis of soil-structure interaction problems, largely due to their computational
simplicity. In this respect, numerous idealisation and elimination techniques
representing either soil or structure, depending on their relative significance for the
problem under consideration, have been employed and evaluated. In most cases,
however, the lack of accuracy of field elimination approaches is apparent.
One of the most common field elimination techniques for idealization of the soil sub-
domain is Winkler model (Winkler, 1867), where the soil medium is removed and
the soil-structure interface is modelled by a system of independent and closely
spaced linear springs as shown in Figure 2.2. Clearly, in this type of modelling the
35
Chapter 2 Literature Review
deformation at the interface is only confined to the loaded region, which is simply
incorrect.
Numerous studies for SSI analysis have been carried out using the Winkler
hypothesis, including studies by Terzaghi (1955), Vesic (1961), Brown et al. (1977)
and Bowles (1996). Despite the simplicity and low computational cost of the
Winkler idealization, the fundamental problem is the determination of the stiffness of
the associated elastic springs replacing the soil sub-domain. As a coupled problem,
the value of the sub-grade reaction is not only dependent on the sub-grade but also
on the parameters of the loaded area as well. However, the sub-grade reaction is the
only parameter in Winkler idealization, thus great care is required in determination
of the sub-grade parameter (using plate load test, consolidation test, triaxial test or
CBR tests (Dutta & Roy, 2002)).
As expected with field elimination techniques, there are serious limitations with
Winkler type idealizations. The major obvious pitfall of such an approach is that it
provides no or very little information regarding the stress and deformation state
within the soil mass. Dutta and Roy (2002) summarises some other primary
limitations of such an idealization:
‘…the basic limitation of Winkler hypothesis lies in the fact that this model
cannot account for the dispersion of the load over a gradually increasing influence
area with increase in depth. Moreover, it considers linear stress–strain behaviour of
soil. The most serious demerit of Winkler model is the one pertaining to the
independence of the springs. So the effect of the externally applied load gets
localized to the sub-grade only to the point of its application. This implies no
cohesive bond exists among the particles comprising soil medium.’ (Dutta & Roy,
2002: p.1582)
Beside the Winklerian approach, there is also a conceptual approach for physical
idealization of the infinite soil sub-domain using the theory of continuum mechanics
(Harr, 1966). The continuum idealization of the soil sub-domain, originally
introduced by Odhe and the research work of Boussinesq (Bowles, 1996), is used to
36
Chapter 2 Literature Review
Such approaches in principle overcome the inadequacy of the Winkler spring model
to transmit lateral shear stresses, where a load acting on the soil surface cannot
produce any settlement except over its specific loading area. In addition, compared
to the Winklerian approach, the continuum approaches provide much more
information on the stress and deformation state within the soil mass. Furthermore,
these approaches are more realistic as they are based on explicit data of geotechnical
investigations, in contrast to the Winkler method, where the sub-grade reaction
modulus does not represent a soil property.
Nevertheless, there are major drawbacks associated with the elastic continuum
approach such as its inaccuracy in reactions calculated at the edges of the foundation.
Furthermore, although solutions for practical problems idealizing the soil sub-
domain as elastic continuum are available for some limited cases, this more
sophisticated method has never gained popularity among the designers, mainly
because of it leading to various mathematical intricacies and its inherent inability for
a direct analytical implementation for design purposes. Adding to these, the inability
of evaluating complex material behaviours, such as the nonlinear and elasto plastic
behaviour of soil, severely limits the application of this model in practice (Dutta &
Roy, 2002).
37
Chapter 2 Literature Review
that of Kurian et al. (2001), where the springs are intermeshed so that the
interconnection is automatically achieved. Similarly, different types of continuum
models for foundation modelling have been proposed (Harr et al., 1969; Nogami &
Lain, 1987; Vallabhan & Das, 1991). There are also models that take into account
the elasto-plastic behaviour (Zeevrat, 1972) of foundation soil as illustrated in Figure
2.3a or the visco-elastic behaviour (Noda et al., 2000) of soil as presented by Figure
2.3b. A brief summary of such foundation models could be found in the review paper
of Dutta & Roy (2002).
The search for physically close and mathematically simple models in soil-structure
interaction problems has not been limited to field elimination techniques
representing the soil sub-domain. There are several field elimination techniques
which tend to simplify the structural sub-domain both computationally and
38
Chapter 2 Literature Review
mathematically. The work of Standing et al. (1998) falls into this category, where the
movements of the Treasury building in London arising from the construction of twin
tunnels are studied using a finite element model for the soil, while the building
structure is simply modelled as an equivalent raft. Also in the work by Dutta et al.
(2004) on dynamic soil-structure interaction modelling of elevated tanks, as shown
in Figure 2.4, it can be seen that both the structure and soil sub-domains are
simplified.
Figure 2.4: Field elimination model for SSI analysis of lateral vibration of an
elevated water tank (Dutta et al., 2004: p.828)
Field elimination techniques cannot deal accurately with geometric and material
nonlinearity in the replaced sub-domain, hence modelling the nonlinear response of
both soil and structure becomes complex for which more sophisticated modelling
approaches would be required. With the increasing availability of powerful
computers and the wider applicability of numerical methods compared to analytical
approaches, the use of the finite element method has become a common means for
modelling such complex interactive behaviour.
39
Chapter 2 Literature Review
The finite element method is a special form of matrix analysis, where the whole
continuum is discretized into a finite number of elements connected at different
nodal points. The general principles and use of the finite element method are well
documented (e.g. Desai & Abel, 1987; Zienkiewicz et al., 2005).
In structural analysis, nonlinear modelling using finite element analysis has evolved
to address complex issues of various structural forms (Izzuddin, 1991; Zienkiewicz
et al., 2005; Zienkiewicz & Taylor, 2005). In fact, due to the universal nature of the
finite element method in modelling real-life complex conditions including geometric
and material nonlinearity in the response (Izzuddin, 1991), this numerical technique
has been widely used in design and assessment of complex structures (Bull, 1988;
Smith & Coull, 1991; Rombach, 2004).
On the other hand, utilising numerical methods in geotechnical engineering has also
provided geotechnical engineers with an extremely powerful analysis and design tool
(Potts & Zdravkovic, 1999; Potts & Zdravkovic, 2001; Potts, 2003). According to
Potts & Zdravkovic (1999), the numerical methods in modelling the soil medium are
far superior to conventional analytical methods, which tend to relax one or more of
the basic solution requirements (see Table 2.1).
40
Chapter 2 Literature Review
41
Chapter 2 Literature Review
In view of the above, the finite element method currently stands as the most
powerful and versatile tool for solving complex soil-structure interaction problems.
A finite element procedure for the general problem of soil-structure interaction
involving nonlinearities due to material behaviour, geometric changes and interface
behaviour is presented by Desai et al. (1982). A three-dimensional visco-elastic
finite element formulation has also been proposed by Viladkar et al. (1993) for
studying the interactive behaviour of a space frame with the supporting soil sub-
domain. The interactive behaviour of a plane frame-footing-soil system with elastic-
perfectly plastic soil has been investigated by Viladkar et al. (1991) and Noorzaei et
al. (1995a). The influence of soil strain-hardening on the elasto-plastic soil-structure
interaction of framed structures has also been undertaken by Noorzaei et al. (1995b).
Several studies have been carried out to model the discontinuous behaviour that may
occur at the interface of soil and structure, including the work of Beer (1985) and
Viladkar et al. (1994) who developed interface elements to model this discontinuity.
All of the above mentioned research has two major conceptual characteristics in
common. Firstly, the finite element method is used as a tool to obtain the complex
response of soil-structure interaction, and secondly the soil and structure sub-domain
are modelled simultaneously in a single computational model.
Although numerical analysis, and particularly finite element analysis, is currently the
most advanced approach for modelling soil-structure interaction, the application of
finite element modelling to soil-structure interaction problems is often limited by the
availability of a simulation environment offering advanced modelling capabilities for
both soil and structure sub-domains. Despite the existence of common techniques for
nonlinear structural and geotechnical analysis, the distinct demands of the two fields
in terms of material modelling and solution procedures have led to differently
evolved modelling techniques in each field. In fact, this has led to development of
two modelling disciplines with distinctive high-level features. In turn, this has been
reflected in the development of discipline-oriented computational tools, which offer
42
Chapter 2 Literature Review
This usually means that structural analysis simplifies soil behaviour, while
geotechnical analysis simplifies structural behaviour using field elimination
techniques. For example, in the work of Vlahos et al. (2006) on soil-structure
interaction of spudcan footings on clay soil subjected to cyclic loading, the structural
model is connected to a hybrid footing model (developed based on available
experimental data) representing the soil medium, as illustrated in Figure 2.5.
Figure 2.5: Idealization of jack-up models for SSI (Vlahos et al., 2006: p.217)
43
Chapter 2 Literature Review
et al. (2004) on the assessment of the influence of soil flexibility on the dynamic
behaviour of building frames (Figure 2.8).
Figure 2.6: a) Schematic illustration of the pile and soil model b) Schematic diagram
of the structure model (Jin et al., 2005: p.1324)
44
Chapter 2 Literature Review
On the other hand, Tian & Li (2008) undertook work on the dynamic response of
building structures subject to ground shock from a tunnel explosion, where the multi-
story building structure is simplified and represented by a shear lumped mass model
as depicted in Figure 2.9.
Figure 2.9: Idealization of the coupled soil-structure interaction system (Tian & Li,
2008: p.1171)
45
Chapter 2 Literature Review
46
Chapter 2 Literature Review
The above research groups were focused on different applications and pursued
different types of problem-decomposition methodology. For Instance, Belytschko &
Mullen (1978) studied node by node partitioning and sub-cycling. Hughes & Liu
47
Chapter 2 Literature Review
48
Chapter 2 Literature Review
substitution stage typically stands for the substitution of the reaction forces from the
structural model into the soil model at the same interface.
This approach is approximate in nature due to the fact that the predicted interface
displacements, obtained using displacements, velocities and accelerations from
previous steps, are invariably different from the displacements evaluated following
force substitution into the soil model, thus violating the compatibility conditions at
the interface.
The aforementioned advantages make the proposed approach appealing for the
solution of large coupled systems. Further to this, O’Brien & Rizos (2005)
considered the application of the staggered approach for the simulation of high speed
train induced vibrations. The Boundary Element Method is used to model the soil-tie
system, while Finite Element Method, along with Newmark’s integration, is used for
the modelling of the rail system. The two methods are coupled at the tie-rail interface
and the solution is obtained following a staggered, time marching scheme
(Figure 2.14).
49
Chapter 2 Literature Review
Figure 2.13: Schematic representation of the staggered coupling approach (Rizos &
Wang, 2002: p.878)
Besides the inapplicability of staggered approach to static analysis, there are major
issues in relation to the staggered approach that need to be addressed in coupled
dynamic analysis, namely stability and accuracy. A staggered solution procedure
should not degrade the numerical stability of any individual system nor the overall
50
Chapter 2 Literature Review
coupled problem. The accuracy degradation of the solution procedure is also another
concern in a staggered coupling procedure. In general, for nonlinear mechanical
problems the stability and accuracy of a staggered solution procedure are intertwined
and should be studied concurrently.
Considerable research has been undertaken on the stability and accuracy of the
staggered approach for different coupled problems (Farhat & Park, 1991; Huang &
Zienkiewicz, 1998). For Instance, in the work of Rizos & Wang (2002) on staggered
coupling of soil-structure interaction, it is concluded that in the absence of elastic
(restoring) forces in the structure (e.g. rigid massive foundations on elastic soils), the
FEM solver in their staggered scheme becomes unstable; therefore, a stabilisation
should be augmented into the FEM equations with an equivalent stiffness of the soil
region under the foundation.
51
Chapter 2 Literature Review
To address the above shortcomings, coupling algorithms have been developed that
are stable and accurate for a wider range of time-step size, which is mainly achieved
by introducing corrective iterations into the staggered approach, hence the name
iterative sub-structuring/coupling methods (Quarteroni & Valli, 1999). In addition to
the enhancement of stability and accuracy, iterative coupling approaches facilitate
parallel computing through problem partitioning which can lead to much greater
computational efficiency. In addition, iterative coupling approaches can be applied to
both static and dynamic problems.
The general procedure of iterative coupling is illustrated in Figure 2.15, which refers
to a soil-structure interaction coupled system decomposed into the soil and structure
sub-domains (T, B). The governing equations of the partitioned sub-domains are
solved independently at each load increment (or time step in the case of dynamic
analysis), using predicted boundary conditions (either force or displacement) at the
interface. These predicted boundary conditions are then successively updated using
52
Chapter 2 Literature Review
Unlike the staggered approach, the compatibility and equilibrium conditions at the
interface of partitioned sub-domains are enforced in the iterative coupling approach,
provided convergence is achieved, thus the stability and accuracy concerns
associated with the staggered approach are no longer relevant.
Despite the significant potential benefits of iterative coupling, a major issue relates
to whether convergence to equilibrium and compatibility at the interface can always
be achieved through successive iterative substitutions. In this respect, the utilised
technique in successive update of boundary conditions at the interface of the
partitioned coupled system is the most critical algorithmic stage. In fact, it is the
update technique that dictates the convergence behaviour of the algorithm rather than
the time/load step size.
53
Chapter 2 Literature Review
domains is carried out simultaneously during every coupling iteration. Unlike the
parallel coupling, in sequential coupling the partitioned sub-domains are solved one
after the other at each iterative stage. In addition to the parallel or sequential nature
of iterative coupling procedures, these algorithms also differ in relation to the
treatment of prescribed Dirichlet (Displacement) and Neumann (Force) effects at the
interface of the partitioned sub-domains. Elleithy & Tanaka (2003) presented and
categorized a range of different iterative coupling algorithms according to the
aforementioned characteristics for elasto-static BEM-BEM and FEM-BEM coupling
(for example, Sequential Dirichlet–Neumann coupling algorithm, Parallel Neumann–
Neumann coupling algorithm, Parallel Dirichlet-Neumann coupling algorithms, etc.).
Such algorithms will be described in detail in a more general form in the context of
static and dynamic soil-structure interaction FEM-FEM coupling in Chapter 4.
The application of iterative coupling algorithms in partitioned analysis has not been
limited to BEM-FEM coupling or fluid-structure interaction. In fact, these
algorithms are general in nature and with some modification could be applied to any
desired type of discretization technique in partitioned analysis of multi-physics
interaction problems. For instance, Collenz et al. (2004) utilised an FEM-FEM
coupling method based on a sequential approach for their work on modelling the
interactive nonlinear behaviour of the micro-beams under electrostatic loading.
54
Chapter 2 Literature Review
Figure 2.16: Coupled dam-reservoir-soil system (Hagen & Estorff, 2005a: p.10)
55
Chapter 2 Literature Review
Figure 2.17: Flowchart of iterative coupling method by Elliethy et al. (2004: p.852)
Although the superiority of the iterative coupling partitioned approach has been
recognised more recently for coupled modelling of soil-structure interaction, there
remain significant technical challenges related to algorithmic and computational
issues, particularly with reference to convergence.
In all of the abovementioned iterative coupling algorithms, the update technique used
in the successive iterations is the constant interface relaxation scheme. As shown
later in this work, convergence to compatibility and equilibrium at the interface of
the partitioned sub-domains could be guaranteed through the choice of a suitable
value for the relaxation parameter, though the number of required iterations may be
unrealistically large.
Despite the relative simplicity of iterative relaxation schemes, there are several
issues that need to be carefully considered: i) determination of the range of suitable
relaxation parameters for the specific problem under consideration in order to
achieve convergence, and ii) selection of the optimum relaxation parameter in order
to achieve maximum computational efficiency.
56
Chapter 2 Literature Review
These results have also been confirmed by the research of Estorff & Hagen (2005)
on dynamic analysis of coupled BEM-FEM problems, where the significant
influence of different problem parameters on the choice of suitable relaxation
parameter is demonstrated. It has also been shown that, depending on the
characteristics of the problem under consideration, there is a range of applicable
relaxation parameters outside of which convergence to compatibility and equilibrium
could not be guaranteed. Moreover, there exists an optimum relaxation parameter in
the convergent range that holds the highest convergence rate and computational
efficiency. In this regard, there is a lack in the literature with regard to establishing
the general mathematical convergence behaviour with relation to the relaxation
parameter. Moreover, the determination of the optimum constant relaxation
parameter would typically require a process of trial and error for every coupling case
under consideration, which is prohibitive for real large scale soil-structure interaction
problems.
Building on the above, as shown later in this work, the adaptive coupling technique
in the context of nonlinear soil-structure interaction is proposed. In this respect, the
57
Chapter 2 Literature Review
need for establishing the general convergence behaviour of the scheme for both
dynamic and static FEM-FEM coupled problems is demonstrated mathematically
and illustrated through various case studies in Chapter 5. In this regard, it is shown
that the performance of interface relaxation in iterative coupling of soil-structure
interaction problems is enhanced significantly through the use of an adaptive
relaxation, using error minimization techniques.
In general, due to the lack of general convergence studies on the application of such
methods and the effects of nonlinearity on the convergence behaviour, specifically in
the context of soil-structure interaction, there remain significant issues to be covered
in relation to the convergence characteristics and performance of constant and
adaptive relaxation methods.
Furthermore, all the presented adaptive relaxation schemes have been employed
within the sequential Dirichlet-Neumann form of coupling algorithm, thus an
adaptive method for evaluating the relaxation parameter in the parallel forms of
coupling algorithms is still largely absent from current research on iterative coupled
modelling via interface relaxation approach.
Albeit, it is possible to extend the existing iterative coupling methods to enhance the
computational efficiency of both adaptive and constant relaxation coupling
algorithms, while overcoming the problematic issues regarding the trial and error
process embedded in evaluation of the constant relaxation parameter.
58
Chapter 2 Literature Review
In the view of above and considering the need for developing iterative coupling
methods with unconditional convergence characteristics and high convergence rate, a
new approach in coupled nonlinear soil-structure interaction analysis is proposed
later in this thesis. This approach is based on utilizing the condensed interface
tangent stiffness matrices of the structure and soil models, depending on the variant
coupling algorithms under consideration (Chapter 6).
Although the condensed tangent stiffness matrix can be determined with current
nonlinear field modelling tools, its implementation for nonlinear soil-structure
interaction coupling using discipline oriented softwares would normally necessitate
direct access to the source codes. In this respect, it is proposed in this work that the
condensed tangent interface stiffness matrices of the partitioned soil and structural
sub-domains may be reasonably approximated by constructing reduced order models
of the structure and soil sub-domains. The benefit of such an approach is that it does
not require the explicit assembly of the stiffness matrices. Indeed, with minimal
intrusion to the source codes of the solvers, a potentially efficient coupling technique
for coupled modelling of soil-structure interaction will be achieved. This builds on a
previous approach presented by Vierendeels (2006) and Vierendeels et al. (2007),
who utilized a procedure for constructing the reduced order model of partitioned sub-
domains throughout the coupling iterations for implicit coupling of fluid–structure
59
Chapter 2 Literature Review
In this regard, there is a pitfall associated with the traditional reduced order method.
For instance, in FEM-FEM iterative coupling problems, where the number of
coupling iterations exceeds the number of interface degrees of freedoms, due to
existence of linearly dependent displacement/force modes, the reduced order method
approximation for the condensed interface tangent stiffness matrix will be poor. This
shortcoming is discussed and addressed comprehensively in Chapter 6, where a
significant modification in reduced order method formulations is proposed to
overcome this shortcoming. Moreover, an advanced technique for approximating the
reduced order method (mixed reduced order method) which possesses a very high
convergence rate, especially in nonlinear problems, is further proposed.
It is shown that, with the current discipline oriented structural and soil solvers,
achieving the same amount of sophistication in numerical modelling of both soil and
structure in a single soil-structure interaction analysis is still a challenge. Since
sophisticated discipline-oriented structural and geotechnical solvers are readily
available, and continue to be used for either application, the partitioned treatment of
coupled soil-structure interaction problems is introduced as an alternative to the
monolithic approach for tackling interaction problems.
60
Chapter 2 Literature Review
The background and origin of the partitioned approach are presented. The staggered
coupling procedure, including its previous application to several transient dynamic
coupled problems, is discussed, and its main disadvantages are highlighted. These
disadvantages will be further demonstrated in Chapter 3, where the staggered
solution procedure for coupled modelling of FEM-FEM problems is
comprehensively examined. To address these shortcomings, iterative coupling
procedures have been introduced as an alternative to the staggered approach. Not
only do these iterative methods address the stability and accuracy issues of the
staggered approach, but they are applicable to both static and dynamic analysis.
It is also pointed out that further enhancement of iterative coupling beyond the use of
relaxation schemes may be achieved through the use of the condensed tangent
stiffness matrices at the soil–structure interface. As a special case of such an
approach, the condensed tangent stiffness matrix may be approximated via reduced
order models of the partitioned sub-domains, where various highly efficient reduced
order methods will be presented in Chapter 6. This builds on previous work by
Vierendeels (2006) and Vierendeels et al. (2007) concerned with implicit coupling of
fluid–structure interaction problems, though major modifications are proposed in this
research for coupled modelling of nonlinear soil-structure interaction.
61
Chapter 2 Literature Review
62
Chapter 3
Staggered Approach
3.1 Introduction
One of the earliest approaches proposed for coupling partitioned mechanical systems
is the staggered approach (Felippa & Park, 1980), which is particularly suited to
transient dynamic analysis. In the staggered solution approach, the equations for
each sub-domain are solved once at each time-step and predicted values of the
coupling boundary conditions are used to obtain the response of the individually
modelled sub-domains.
The two most important issues of the staggered approach that should be addressed in
any scheme are the stability and the accuracy of the algorithms. The stability of
staggered algorithms depends on the suitable choice of the predictor operator, the
selected time integration scheme in each domain, the employed time-step for the
time integration schemes and the utilized discretization techniques at the interface
and through each field.
Once satisfactory stability is achieved, the next concern is accuracy. The predicted
interface displacements, obtained using velocities/accelerations at the beginning of
the step, are invariably different from the displacements evaluated following force
64
Chapter 3 Staggered Approach
substitution into the soil model, thus leading to compatibility defaults at the
interface. The accuracy of the staggered approach may be improved by reducing the
time-step and using an optimal predictor, though stability requirements may dictate
step sizes that are too small rendering this approach prohibitively expensive for
large-scale soil-structure interaction problems.
One of the most used techniques in assessing the dynamic response of mechanical
systems is the direct method, where the dynamic equilibrium equations are integrated
directly at the overall structural level, typically using a step-by-step time integration
scheme. This type of analysis is the most general and is readily applicable to
nonlinear analysis.
65
Chapter 3 Staggered Approach
The dynamic equilibrium equations for a discrete linear structural system can be
expressed as:
matrix, U is the acceleration vector, U is the velocity vector, U is the
1
U n1 U n t U n t 2
Un Un 1 (3.3)
2
66
Chapter 3 Staggered Approach
The above equations, allow the three unknowns U n 1 , U n 1 and Un 1 to be
determined as:
F Uˆ C Uˆ
1
1
2
U n 1 M C K n 1 M (3.5)
t t n n
where:
3.3 Partitioning
Assume that the above coupled system is composed of ΩT and ΩB sub-domains with
the interface (ΩΓ= ΩT∩ ΩB), as shown in Figure 3.2 and formulated by:
67
Chapter 3 Staggered Approach
In the above, vectors UXX , U XX , U XX and FXX correspond to the displacement
vectors and external loads, respectively, for the non-interface degrees of freedom in
sub-domain ΩX, while vectors U , U , U and F correspond
i
i
i
i
to
Assuming that the above coupled system is partitioned into two independently
modelled sub-domains (ΩT, ΩB) and discretised by FEM using a step-by-step time
68
Chapter 3 Staggered Approach
integration method, such as the Newmark method, the governing discrete equations
of each independently modelled partitioned sub-domain can be formulated as
follows:
In the above, vectors U Xi and FXi are the displacements and external loads,
K11X K12X
X = stiffness matrix of sub-domain ΩX,
K 21 K 22X
M 11X M 12X
X = mass matrix of sub-domain ΩX, and
M 21 M 22X
C11X C12X
X = damping matrix of sub-domain ΩX.
C21 C22X
Clearly, the time marching Equations (3.12) and (3.13) cannot be solved
independently. This is due to existence of unknown displacement (i.e. U Ti , U Bi )
and force vectors (i.e. FTi , FBi ) at the interface. However, by employing a
69
Chapter 3 Staggered Approach
Consider the partitioning strategy presented in the previous section, where ΩT refers
to the partitioned structure sub-domain, while ΩB corresponds to the partitioned soil
sub-domain. The general algorithmic framework of staggered coupling in the context
of soil-structure interaction analysis using FEM can be described by the following
procedure, in which n denotes the current time-step number.
For n = 1, 2,...
F F
i
T n
i
B n 0 (3.14)
STEP 4: Based on the interface forces, FBi , and the external loading applied to the
n
Equation (3.13).
U
i
T n 1
U Bi , U Bi , UBi , t
n n n
(3.15)
70
Chapter 3 Staggered Approach
while the compatibility of the initial condition for the structure domain and soil
domain is enforced:
U U
i
T n
i
B n (3.16)
U U
i
T n
i
B n (3.17)
The above staggered coupling scheme couples the response of the partitioned sub-
domains by enforcing equilibrium condition at the interface level, while the
compatibility default is minimized by choosing a suitable predictor function and
small time step in order to level the response of the coupled sub-domains.
3.5 Predictors
where iP , iP and iP are numeric coefficients of the predictor. Since for second-
U n i 1
terms are not considered. Note also that Equation (3.18) involves m past
71
Chapter 3 Staggered Approach
General Form:
U predicted n 1 P U n
Trivial Predictors
Trivial:
U predicted n1 U n
General Form:
U predicted n1 P U n t P U n
First Order Predictors
Central Difference Method:
U predicted n1 U n t U n
General Form:
U predicted n1 P U n t P U t 2 P U
n n
Second Order Predictors Central Difference Method:
t 2
U predicted U n t U U n
n 1 n 2
Some examples of the single-step predictors that can be used here for second-order
dynamic systems are presented in Table 3.1.
The term “stable” informally means resistant to change. For technical use, the term
has to be defined more precisely in terms of the mathematical model, but the same
connotation applies. In mathematics, stability theory is typically concerned with
whether a given function is sensitive to a small perturbation (Felippa & Park, 2004).
U (t T ) U (t ) (3.19)
Stability of the system requires that, in studying the behaviour of the system after
application of an arbitrary perturbation in the initial displacement or velocity, the
motion remains within small prescribed limits of the unperturbed motion.
72
Chapter 3 Staggered Approach
One of the widely used methods in stability analysis is the amplification method,
which is also called von Neumann stability analysis (Zienkiewicz & Taylor, 1991;
Felippa & Park, 2004). The approach is based on decomposition of motion into
normal modes and checking the growth or decay of perturbations from one step to
the next, and it can be implemented using standard linear algebra procedures.
in which [A] is known as the amplification matrix. Since methods such as Newmark
represent approximations that are used to derive equations of type (3.20), error is
introduced into the solution U n 1 at each time step. Since the solution U n 1 at time
tn 1 depends on the solution U n at time tn the error can grow with time. It can be
clearly observed that any error presented in the solution will of course be subjected
to amplification by precisely the same factor:
U k A U n
k n
(3.21)
Knowing that the general modal solution of any recurrence algorithm can be written
as (Zienkiewicz & Taylor, 1991):
U n1 U n (3.22)
and by substituting Equation (3.22) into Equation (3.20), it can be observed that is
given by the eigenvalues of the amplification matrix as:
73
Chapter 3 Staggered Approach
A I U n
0 (3.23)
absolute value, 1 , all initially small errors will increase without limit and the
solution will be unstable. Therefore the stability of the recurrence algorithm requires
that 1 (Zienkiewicz & Taylor, 1991). In the case of complex eigenvalues, the
The calculation of the eigenvalues in such problems is not trivial. Therefore, two
general procedures are employed here to make the task of stability checks more
practical.
The first procedure is named z transformation (Zienkiewicz & Taylor, 1991). The
determinant equation A I 0 provides the characteristic polynomial yielding
1 z
(3.24)
1 z
where z and are in general complex numbers, it is easy to show that the
74
Chapter 3 Staggered Approach
I zI
R zR
R i I z z R iz I
C1 0 ,
C1 C3
0
C0 C2
C1 C3 C5
C0 C2 C4 0
0 C1 C3
and so on up to,
C1 C3 C5 ...
C0 C2 C4 ...
0 C1 C3 ...
0
0 0 C2 ...
... ... ... ...
0 Cn
75
Chapter 3 Staggered Approach
By utilizing the above tools, the stability of the proposed staggered algorithms can
be assessed by implementing the general procedure described in the next section.
Consider a coupled system which is partitioned into the sub-domains (ΩB) and (ΩT),
with each sub-domain discretised in the time domain using Newmark method, and
assume that both domains are condensed to a single degree of freedom at the
interface, neglecting the effects of damping for simplicity. The governing
equilibrium equations for the partitioned sub-domains of the test system are given
by:
76
Chapter 3 Staggered Approach
where FTi and FBi are the interface forces applied to the structure and soil sub-
domains, respectively.
Following the staggered approach algorithmic steps presented in Section 3.4, the
response of the partitioned sub-domains of the above test system can be coupled as
discussed in the following, yielding to the construction of the amplification matrix:
U U
i
T n
i
B n (3.27)
U U
i
T n
i
B n (3.28)
U U
i
T n
i
B n (3.29)
Choosing the second-order central difference predictor from Table 3.1, the
prescribed displacement in partitioned sub-domain ΩT at time t tn1 is:
t 2
UT n1 UT n t U T n UT n (3.30)
2
STEP 2 - Solving the ΩT sub-domain for FTi using Newmark method (see
n 1
Section 3.2):
MT
t
2
KT U Ti FTi M T Uˆ Ti 0
n 1 n 1
F
i
M T t 2 KT U Ti
n 1
t 2 M T Uˆ Ti
(3.31)
T n 1
t 2
where:
77
Chapter 3 Staggered Approach
F
i
F
i
M T t 2 KT U Ti
n 1
t 2 M T Uˆ Ti
(3.33)
B n 1 T n 1
t 2
MB
t
2
K B U Bi FBi M B Uˆ Bi 0
n 1 n 1
U
i
t 2 FBi n 1 M B Uˆ Bi (3.34)
M B K B t 2
B n 1
where:
U
i
B n 1
t
U Bi Uˆ Bi
n 1 (3.36)
U
i
B n 1
1
U i Uˆ Bi
t 2 B n 1 (3.37)
where:
78
Chapter 3 Staggered Approach
U
i
T n 1
U Bi
n 1
(3.39)
U
i
T n 1
UBi
n 1
(3.40)
Using a central difference predictor from Table 3.1, the prescribed displacements for
ΩT at time t tn 2 is:
t 2
UT n 2 U B n1 t U B n1 U B n1 (3.41)
2
M B KT t 2 i
U
i
B n 1
M B K B t 2
U B n
M KT t 2 t
B
M B K B t 2 U
i
B n (3.42)
1 M B 2 M T 2 M T KT t t i
2 2
2 M B K B t 2
U B n
The velocity at the interface of sub-domain ΩB at time tn+1 can also be calculated by
expanding Equation (3.36), as:
KT K B t
U
i
B n 1
M B K B t 2 U
i
B n
M B K B t 2 K B KT t 2 i
+
M B K B t 2
U B n (3.43)
1 KT t K Bt 2 M T 2 M B 2 K B t t i
2 2 2
-
2 M B K B t 2
U B n
79
Chapter 3 Staggered Approach
Similarly the acceleration at the interface of sub-domain ΩB at time tn+1 can also be
expressed in terms of displacement, velocity and acceleration of the previous time
step (tn ) as:
KT K B
U
i
B n 1
M B K B t 2
U Bi n
K T K B t
M B K B t 2 U
i
B n (3.44)
1 KT K B t 2 M T 2 K B t i
2 2
2 M B K B t 2
U B n
Now, the amplification matrix of the presented test system can be constructed in the
form of U n 1 A U n , by knowing Equations (3.42), (3.43) and (3.44), with
A being the amplification matrix of the proposed staggered coupling scheme. The
amplification matrix can be derived as:
1 K B 2 4 1 t 2 KT (2 2 1)t 2 2 M T 4 M B 2
3
2 M B K B t 2
1 K B 2 2 1 t 2 KT 2 4 1 t 2 4M T 2 M B
(3.46)
2 M B K B t 2
M T KT t 2
0
M B K B t 2
As mentioned before it is sufficient and necessary for stability that the modulus of all
the eigenvalues is less than or equal to one ( i.e. i 1 ).
80
Chapter 3 Staggered Approach
1
M B K B t 2
K B 4 2 t 2 KT 4 2 t 2 4M B 4M T z 3
1
2 B
K 4 1 t 2 KT 4 1 t 2 4 M T 4 M B z 2
M B K B t
(3.47)
1
2 T
2 K K B t 2 z
M B K B t
1
M B K B t 2
K T K B t 2 0
Hurwitz conditions:
C0 0
K 4 2 K 4 2 t
B T
2
4( M B M T )
0
(3.48)
M B K B t 2
C1 0
K 4 1 K 4 1 t
T B
2
4 M B MT
0
(3.49)
M B K B t 2
C1 C3
0
C0 C2
K 2 1 K 2 1 t
T B
2
M B 2 1 M T 2 1
0 (3.50)
M B K B t
2 2
81
Chapter 3 Staggered Approach
C1 C3 0
C0 C2 0 0
0 C1 C3
K 2 1 K 2 1 t
T B
2
M B 2 1 M T 2 1
0
(3.51)
M K B t 2
3
B
Consider the above stability conditions under the assumption of very small time
steps (i.e. t 0 ):
lim(C1 ) 0 M B M T 0 (3.53)
t 0
C1 C3 0
C1 C3
lim
t 0 C C4
0 lim C0
t 0
C2 0 0 M B 2 1 M T 2 1 0 (3.54)
0
0 C1 C3
The Newmark method is unconditionally stable in linear analysis for all modes
regardless of the choice of t, if:
1
2 Newmark method unconditionally stable (3.55)
2
82
Chapter 3 Staggered Approach
The fact that the stability in the presented staggered procedure is conditional and is
guaranteed provided that M B M T 0 as t 0 , can be further demonstrated by
If [A] is a square n×n matrix with real or complex entries and if ( 1 , 2 ,, n ) the
eigenvalues of [S] are listed according to their multiplicities, then:
i 1 (i 1,..., n) (3.57)
Considering Equations (3.56) and (3.57), it is clear that a necessary (but not
sufficient) condition for the stability of Equation (3.45) would be:
det A 1 . 1 n 1 (3.58)
Using Equation (3.45), it can be shown that the above necessary stability condition
simplifies to:
M T KT t 2
det A 1 (3.59)
M B K B t 2
which confirms the conditional stability of the scheme and shows that conditional
stability may be achieved provided that M B M T 0 as t 0 . Moreover, it is
83
Chapter 3 Staggered Approach
3.8 Accuracy
3.9 Example
84
Chapter 3 Staggered Approach
85
Chapter 3 Staggered Approach
expensive as the stability requirements dictate very small time steps M B M T .
This would be clearer when compared to the monolithic treatment which is stable for
t 0.01 s .
86
Chapter 3 Staggered Approach
87
Chapter 3 Staggered Approach
The stability/accuracy of the staggered approach with the specified time step size
( t .0005 s ) is clearly not satisfactory, hence another analysis was carried out
reducing the time step to ( t .0001 s ). Although this is an extremely small time
step that makes the staggered approach extremely computational expensive, the
results of staggered analysis with t .0001 s confirms that by reducing the time
step size a full coupling of the partitioned sub-domains can be achieved. The fact
that the stability and accuracy are achievable by the staggered coupling approach,
provided that small time steps are employed, is confirmed by comparing the results
against those of the monolithic treatment.
Figures 3.10 and 3.11 show the variation with time of the rotation and horizontal
displacement, respectively, at the interface for both the staggered and monolithic
approaches. The variation with time of the horizontal displacement and rotation at
the free end of the cantilever is also depicted in Figures 3.12 and 3.13, respectively.
88
Chapter 3 Staggered Approach
89
Chapter 3 Staggered Approach
90
Chapter 3 Staggered Approach
3.10 Conclusion
91
Chapter 4
Iterative Coupling
4.1 Introduction
the aforementioned treatment of the stability and accuracy issues of the staggered
approach, facilitate parallel computing through problem partitioning which could
lead to much greater computational efficiency. The other advantage of iterative
coupling algorithms is that, unlike the staggered approach, they could be readily
employed in both static and dynamic analysis of soil-structure interaction problems.
Structure sub-domain:
93
Chapter 4 Iterative Coupling
Soil sub-domain:
In the above, M X , C X and K X are the mass, damping and the stiffness
correspond to displacement, velocity, acceleration and external load vectors for the
non-interface degrees of freedom in sub-domain X , while U Xi , U Xi , UXi and
domains is just for illustration, where the proposed method is in fact also applicable
to different desired types of discretisation techniques. Moreover, the static analysis
of the presented coupled system can be considered as a special case of the presented
dynamic analysis.
Although Equations (4.1) and (4.2) of the partitioned soil and structure sub-domains
cannot be solved independent of each other, by applying an iterative coupling
scheme and coupling the response of the partitioned soil and structure sub-domains
at the interface level, the partitioned sub-domains can be analysed separately. The
proposed solution scheme couples the response of the soil and structure sub-domains
by enforcing explicitly compatibility and equilibrium conditions at the interface.
94
Chapter 4 Iterative Coupling
Figure 4.2 shows a soil-structure interaction coupled system, decomposed into the
soil and structure sub-domains treated by iterative coupling algorithms. The
governing equations of the partitioned sub-domains are solved independently at each
time step (or load increment in the case of static analysis), using predicted boundary
conditions (either force or displacement) at the interface. These predicted boundary
conditions are then successively updated using corrective iterations, until
convergence to equilibrium and compatibility is achieved at the interface and within
the partitioned sub-domains. This enables the coupling procedure to have an
effectively similar overall accuracy and stability to the monolithic treatment.
Compatibility condition:
U U
i
B n
i
T n
0 (4.3)
95
Chapter 4 Iterative Coupling
Equilibrium condition:
F F
i
B n
i
T n 0 (4.4)
Although in the above the assumption is that the soil and the structure always remain
in contact at the interface, the treatment of separation and slip can be a simple
extension through the use of interface elements that may be considered to be either
part of one of the sub-domains or even part of the interface model.
In the following, the various types of iterative coupling algorithms (Elleithy &
Tanaka, 2003) are adapted and categorized in the context of soil-structure interaction
analysis. Throughout this discussion, subscript n and superscript I denote the
time/load increment and the iteration number, respectively.
96
Chapter 4 Iterative Coupling
STEP 1: At the start of each increment, the structure domain is loaded by the
external forces, FTT , while the displacements (Dirichlet data) at the interface
n
U U
i I
T n
STEP 2: The structural solver computes the response of the structure using Equation
F
i I
T n .
STEP 3: The corresponding interface forces at the soil domain can be calculated by
applying equilibrium:
F F
i I
B n
i I
T n 0
STEP 4: Based on these forces, FBi , and the external loading applied to the soil
I
sub-domain, F
, the soil solver computes the response of the soil domain for
B
B n
U and U .
i I
B n
B I
B n
By comparing the initial prescribed interface displacement and the obtained interface
displacements in STEP 4, the following two scenarios can occur:
STEP 5: If convergence to compatibility has not been achieved, the new estimation
of the interface displacements according to the compatibility condition of Equation
97
Chapter 4 Iterative Coupling
(4.3) is applied to the structure domain, and iteration continues (I=I+1) from STEP 2
until convergence to compatibility is achieved.
U
I 0
i
T n
U
I 0
i
T n 1
U i I 1
F
I
i
T n B n
STEP 1: At the start of each increment, the structure domain is loaded by the
external forces, FTT , while the forces (Neumann data) at the interface nodes,
n
F
i I
T n , are prescribed in accordance with the initial conditions:
F F
i I
T n
98
Chapter 4 Iterative Coupling
STEP 2: The structural solver computes the response of the structure using Equation
U .
i I
T n
U U
i I I
B n
i
T 0
n
n n
By comparing the initial prescribed interface forces and the obtained ones in STEP 4
the following two scenarios can occur:
STEP 5: If convergence to equilibrium has not been achieved, the new estimation of
the interface forces according to the equilibrium condition of Equation (4.4) is
applied to the structure domain, and iteration continues (I=I+1) from STEP 2 until
convergence to equilibrium is achieved.
99
Chapter 4 Iterative Coupling
F
I 0
i
T n
F
I 0
i
T n 1
F
I 1
U
I
i i
T n B n
STEP 1: At the start of each increment, the structure sub-domain is loaded by the
external forces, FTT , while the displacements (Dirichlet data) at the interface
n
U U
i I
T n
Concurrently, the soil sub-domain is loaded by the external forces, FBB , while the
n
forces (Neumann data) at the interface nodes, FBi , are prescribed in accordance
I
F F
i I
B n
100
Chapter 4 Iterative Coupling
STEP 2: The structural solver computes the response of the structure using Equation
(4.1), for non-interface displacements, U TT , and interaction forces at the interface,
I
F
i I
T n . Simultaneously, the soil solver computes the response of the structure using
By comparing the initial prescribed interface displacements and forces and the
obtained interface displacements and forces in STEP 2, the following two scenarios
can occur:
STEP 3: If convergence to compatibility and equilibrium has not been achieved, the
new estimation of the interface displacements and forces according to the
compatibility and equilibrium conditions of Equations (4.3) and (4.4) is applied to
the structure and soil sub-domains respectively, and iteration continues (I=I+1) from
STEP 2 until convergence to compatibility and equilibrium is achieved.
U
i I 0
T n
U
i I 0
T n 1
U
i I 1
T n F
i I 1
B n
F
i I 0
B n
F
i I 0
B n 1
101
Chapter 4 Iterative Coupling
STEP 1: At the start of each increment, the structure sub-domain is loaded by the
external forces, FTT , while the forces (Neumann data) at the interface nodes,
n
F i I
T n , are prescribed in accordance with the initial conditions:
F F
i I
T n
Concurrently, the soil sub-domain is loaded by the external forces, FBB , while the
n
U U
i I
B n
STEP 2: The structural solver computes the response of the structure using Equation
(4.1), for non-interface displacements, U TT , and interaction displacements at the
I
interface, U
I
i
T . Simultaneously, the soil solver computes the response of the
n
By comparing the initial prescribed interface displacements and forces and the
obtained interface displacements and forces in STEP 2, the following two scenarios
can occur:
STEP 3: If convergence to compatibility and equilibrium has not been achieved, the
new estimation of the interface displacements and forces according to the
compatibility and equilibrium conditions of Equations (4.3) and (4.4) is applied to
102
Chapter 4 Iterative Coupling
the structure and soil sub-domains respectively, and iteration continues (I=I+1) from
STEP 2 until convergence to compatibility and equilibrium is achieved.
F i I 0
T n
F
i I 0
T n 1
F U
i I 1
I 1
i
T n B n
U i I 0
B n
U
i I 0
B n 1
STEP 1: At the start of each increment, the structure sub-domain and the soil sub-
domain are loaded by the external forces, FTT and FBB , while the displacements
n n
103
Chapter 4 Iterative Coupling
U
i I
B n are prescribed in accordance with the initial conditions (preferably enforcing
compatibility):
U U Bi U
i I 0 I 0
T n n
STEP 2: The structural and soil solver compute the response of the structure and soil
concurrently using Equations (4.1) and (4.2) for non-interface displacements, U TT
I
and U BB , and interaction forces at the interface, FTi and FBi , simultaneously.
I I I
n n n
By comparing the obtained forces at the interface of soil and structure sub-domains
in STEP 2, the following two scenarios can occur:
U i I 0
I 0
U
i I 0
I 0
T n U Bi T n1 U Bi
n n 1
U U
I 1
i i I 1
T n B n
U
I 0 I 0
U
I 0 I 0
i
UTi
i
T n1 U Bi
B n n n 1
104
Chapter 4 Iterative Coupling
STEP 1: At the start of each increment, the structure sub-domain and the soil sub-
domain are loaded by the external forces, FTT and FBB , while the forces
n n
(Neumann data) at the interface nodes of the partitioned sub-domains FTi and
I
F
i I
B n are prescribed in accordance with the initial conditions (preferably enforcing
equilibrium):
F
i I 0
FBi F
I 0
T n n
STEP 2: The structural and soil solvers compute the response of the structure and
soil concurrently using Equations (4.1) and (4.2) for non-interface displacements,
simultaneously.
By comparing the obtained displacements at the interface of soil and structure sub-
domains in STEP 2, the following two scenarios can occur:
105
Chapter 4 Iterative Coupling
The presented sequential and parallel coupling algorithms can be used in partitioned
treatment of the soil-structure interaction problems. However, the algorithms in
which the partitioned structure sub-domain is considered as subjected to Dirichlet
(displacement) boundary conditions at the interface, while the partitioned soil sub-
domain is subjected to a Neumann (force) or Dirichlet (displacement) boundary
condition at the same interface, are more suitable in the context of static soil-
structure interaction analysis. This is purely due to the fact that in static problems
only essential Dirichlet boundary conditions can be imposed on the interface of the
structure sub-domain, where applying the natural Neumann boundary conditions at
this interface results in singularity of the equilibrium equations for the structural sub-
domain. In fact, treating the structure interface with Neumann boundary conditions
in such a problem requires some additional conditions to make the structure sub-
domain solvable.
The above requirement narrows the choice of coupling algorithms in a general soil-
structure interaction analysis to Dirichlet-Neumann or Dirichlet-Dirichlet family of
106
Chapter 4 Iterative Coupling
Despite the significant potential benefits of iterative coupling methods, a major issue
relates to whether convergence to equilibrium and compatibility at the interface can
always be enforced through successive iterations. Thus, besides choosing a suitable
iterative scheme it is important to address the issues related to the convergence
behaviour of the selected iterative scheme.
The objective of this section is to elaborate the basic procedure used for convergence
analysis of sequential D-N iterative coupling algorithm in the case of trivial update
of boundary conditions. Consider the following iterative coupling solution form:
cte
New Old
(4.5)
The subsequent iterations for the above equation, in which the solution process is
107
Chapter 4 Iterative Coupling
C
1
(4.6)
C
2 1
(4.7)
C
3 2
(4.8)
and in general:
C
I 1 I
(4.9)
Considering Equations (4.6), (4.7) and (4.8) the solution error, , in different
2
2 1
1
(4.10)
3 3 2
2
2 1
1
(4.11)
Therefore, generally the solution error after I iterations would take the following
form:
I
I 1
1
(4.12)
that is less than one. This can be extended to the matrix form as, for a successive
approximation cte to converge for any initial vector , it is
I 1 I
108
Chapter 4 Iterative Coupling
necessary and sufficient that the moduli of all eigenvalues of the error reduction
matrix, , are less than one.
Using this theorem we can investigate the convergence behaviour of the iterative
coupling algorithms by constructing the matrix during successive iterations. This
treated by the sequential D-N iterative coupling procedure. Assume that each sub-
domain is discretised in time using the Newmark step-by-step integration method
and that both sub-domains are condensed at the interface. The governing equilibrium
equations for the partitioned soil and structure domains of the above mentioned
system are given as:
Structure sub-domain ( T ):
Soil sub-domain ( B ):
Following the steps of the sequential D-N coupling algorithm presented in Section
4.2.1 and considering a trivial update of boundary conditions at the interface in
n n
STEP 1: Set the initial interface displacements, U Ti , for the structure sub-domain.
I
STEP 2-3: Compute the response of the structure forces at the interface and apply
equilibrium:
109
Chapter 4 Iterative Coupling
F
i I
T n 1
1
t 2
M T KT t 2 CT t U Ti M T Uˆ Ti
I
n 1 C Uˆ
n
T
i
T
n
(4.15)
F
i I
FTi
I
B n 1 (4.16)
n 1
M T KT t 2 CT t U Ti M T Uˆ Ti M B Uˆ Bi
I
U Bi n1
I n n n n
M B K B t CB t
2
(4.17)
U U Bi
I 1 I
i
T (4.18)
n 1 n 1
M T KT t 2 CT t i I
U
i I
B n 1
M B K B t 2 CB t
UT n1 cte (4.19)
It can be shown that the compatibility default at the interface after K iterations can
be obtained from the following difference equation:
U
I K
U Bi
I K
n
U Ti
I K
n
K . U Bi
I 0
n
U Ti
n
I 0
(4.20)
with:
110
Chapter 4 Iterative Coupling
Comparing Equations (4.20) and (4.12), the necessary and sufficient condition for
convergence of the presented coupled test system can be written as:
KT
1 (4.23)
KB
Equation (4.22) shows that for the presented sequential D-N iterative coupling
algorithm with trivial update of boundary conditions to converge for any initial
prescribed displacement value, the condensed effective stiffness of the sub-domain
treated by Dirichlet data at the interface should be less than the condensed effective
stiffness of the sub-domain treated by Neumann boundary conditions. Here,
condensation is a process by which some of the degrees of freedom are eliminated
from the overall equilibrium providing a reduced set of equilibrium equations for the
remaining degrees of freedom.
K 11 K 12 U x Fx
K = (4.24)
21 K 22 U i Fi
K 11 U x K 12 U i Fx (4.25)
111
Chapter 4 Iterative Coupling
K 21 U x K 22 U i Fi (4.26)
K * U i = F* (4.27)
where
K * = K 22 - K 21 K 11-1 K 12 (4.28)
F* Fi K 21 K 111 Fx (4.29)
The fact that the convergence behaviour is influenced by the ratio of the condensed
effective stiffness of the partitioned sub-domains can be further established by
constructing the general convergence behaviour of the example presented in the next
section.
4.4.2 Example
Here we examine the convergence behaviour and the effect of mass and stiffness of
coupled domains in a partitioned interaction problem treated by sequential D-N
iterative coupling algorithm using a trivial update of boundary conditions at the
interface. Consider the dynamic problem of the mass-spring system illustrated in
Figure 4.9a, where the problem is partitioned into two sub-domains B and T as
112
Chapter 4 Iterative Coupling
U
i
B
F
B
i
U
i
T
F
i
T
partitioned sub-domain T
The governing equilibrium condition for the partitioned domain T can be written
U T F T Uˆ T
KTeffective Ti Ti M T T 0 (4.30)
U T n 1 FT n 1 ˆ i
U T n
with
M 0
M T 01 M 2
(4.31)
K1 K 2 K2
KT K 2
(4.32)
K2
113
Chapter 4 Iterative Coupling
1
KTeffective KT M (4.33)
t 2 T
given by:
U i Fi Uˆ i
K Beffective BB BB M B B 0 (4.34)
U B n 1 FB Uˆ BB n
with
M 3 0
M B M 4
(4.35)
0
K K3
K B K3 K3 K 4
(4.36)
3
1
K Beffective K B M B (4.37)
t 2
U
i I
U Ti
I
B n 1 cte (4.38)
n 1
where:
K 3 K 4 t 2 M 4 K 2 K1 2 t 4 M 2 K 2 M 1 K 2 M 2 K1 t 2 M 2 M 1
(4.39)
K 1 K 2 t 2 M 1 K 4 K 3 2 t 4 M 3 K3 M 3 K 4 M 4 K 3 t 2 M 4 M 3
114
Chapter 4 Iterative Coupling
The essential and sufficient requirement for the above system to converge using the
sequential Dirichlet-Neumann algorithm as demonstrated before would be:
K 3 K 4 t 2 M 4 K 2 K1 2 t 4 M 2 K 2 M 1 K 2 M 2 K1 t 2 M 2 M 1
1 (4.40)
K 1 K 2 t 2 M 1 K 4 K 3 2 t 4 M 3 K 3 M 3 K 4 M 4 K 3 t 2 M 4 M 3
Considering Equations (4.33) and (4.37), and applying the condensation process of
Equation (4.28) the exact convergence condition of Equation (4.40) can be achieved:
M2 K22
K2
t 2 K K M 1
1 2
t 2
1 (4.42)
M3 K 32
K3
t 2 K K M 4
3 4
t 2
Considering the above convergence condition for very small time steps (i.e. t 0 ):
1 M2 M3
M2
lim (4.43)
t 0 M3
Equation (4.43) shows that if the mass at the interface is assigned such that the mass
of the domain treated by Dirichlet boundary condition is less than that of the sub-
domain treated by Neumann boundary condition, the convergence criteria of the
algorithm is satisfied in the limit as the time step becomes very small. It is important
115
Chapter 4 Iterative Coupling
lim K t
t 0
3
2
0 , which would not be the case if one of the corresponding springs is
lim lim
t 0 K 2 M1 M 2
M3
1 (M1 M 2 ) M 3 (4.44)
In fact due to the infinite stiffness K 2 , the mass participating at the interface level at
holds for the case of K 3 , where in this case M Ti M 1 should be less than
M Bi M 3 M 4 :
lim lim
t 0 K 3 M2
M3 M4
1 M 2 (M 3 M 4 )
Considering the above conditions, it can be concluded that in the sequential D-N
iterative coupling algorithm with trivial update of boundary conditions a conditional
coupling convergence can be achieved. In fact, the convergence can only be
achieved provided that the problem under consideration allows for suitable
partitioning strategies at the interface that satisfy the condition of Equation (4.41).
116
Chapter 4 Iterative Coupling
In order to deal with different coupling methods utilizing robust update techniques
for nonlinear-soil structure interaction problems, as proposed in Chapters 5, 6 and 7,
a novel iterative coupling simulation environment is developed, utilizing discipline-
oriented solvers for nonlinear structural and geotechnical analysis. The developed
simulation environment is used to demonstrate the relative performance
characteristics and merits of various presented algorithms. This tool is also applied in
a number of case studies involving nonlinear soil-structure interaction with
nonlinearity in both structure and soil, thus leading to important conclusions
regarding the adequacy and applicability of various coupling techniques for such
problems as well as the prospects for further enhancements.
The simulation of soil-structure interaction via the partitioned approach is carried out
in this work through the coupling of two powerful FEM codes, ADAPTIC (Izzuddin,
1991) and ICFEP (Potts & Zdravkovic, 1999) that have been developed at Imperial
College London for advanced nonlinear structural and geotechnical analysis,
respectively.
It should be emphasised that although the approach developed in this work is applied
to the coupling of ADAPTIC and ICFEP, the related algorithms are general in nature
and equally applicable to the coupling of other existing nonlinear soil and structural
analysis tools.
4.6 ADAPTIC
117
Chapter 4 Iterative Coupling
of steel and composite frames, slabs, shells and integrated structures. The program
features are described briefly hereafter.
The development of ADAPTIC was initially driven by the needs of the offshore industry
for accurate and efficient, nonlinear analysis of offshore structures subject to extreme
loading conditions. This motivated the development of pioneering adaptive nonlinear
dynamic analysis techniques for framed structures, accounting for geometric and
material nonlinearity. The program has been extensively developed to deal with different
extreme loading, such as earthquake, fire and blast, as well as numerous additional
structural forms, such as reinforced concrete and steel-decked composite slabs, cable and
membrane structures, and curved shells.
In ADAPTIC the loading can be either in the terms of applied forces or prescribed
displacements/accelerations at nodal points. The loads can vary proportionally under
static conditions, or can vary independently in the time or pseudo-time domains. The
latter variation can be utilised for static or dynamic analysis. Different types of analysis
in ADAPTIC are namely:
dynamic analysis,
eigenvalue analysis.
118
Chapter 4 Iterative Coupling
The following show some of the various assumptions that can be made using
ADAPTIC in structural modelling:
elastic modelling,
elasto-plastic modelling,
4.7 ICFEP
ICFEP (Potts & Zdravkovic, 1999) is a powerful finite element program specifically
written for the analysis of geotechnical engineering problems. It has been used for
many numerical analysis research projects at Imperial College and is continually
being developed. ICFEP has successfully been applied to numerous practical
engineering projects. The comprehensive formulation of ICFEP makes it possible to
consider a very wide range of problems. The most common applications are the
analysis of deep basement excavations, embankments, slopes and tunnelling.
119
Chapter 4 Iterative Coupling
120
Chapter 4 Iterative Coupling
Drucker Prager,
Boundary surface models for both clays and sand (e.g. MIT-E3, Papadimitriu
and Bouckovalas, respectively).
4.8 INTERFACE
Coupling of the ADAPTIC and ICFEP is carried out using a coupling program called
INTERFACE which utilises a sequential Dirichlet-Neumann type of iterative
coupling algorithms. The INTERFACE program is written in FORTRAN 95.
ADAPTIC and ICFEP run on separate processors as independent black box solvers,
where the task of communication and synchronization between the two individual
codes is achieved via INTERFACE that implements the iterative coupling methods
utilizing various update techniques. In this respect, the interface program manages
the retrieval, manipulation and passing the necessary data between the two field
programs during coupled analysis (See Figure 4.10).
121
Chapter 4 Iterative Coupling
The client software, here the INTERFACE program, can send data requests to the
connected servers, ADAPTIC and ICFEP, which in turn accept these requests,
process them, and return the requested information to the client. The important
characteristics of the INTERFACE as a client program are: i) initiation and
synchronisation of the requests to ADAPTIC and ICFEP, and ii) waiting for,
receiving and processing the replies from ADAPTIC and ICFEP. On the other hand,
122
Chapter 4 Iterative Coupling
ADAPTIC and ICFEP as server programs never initiate any activity without
receiving a request from the INTERFACE program.
In the developed simulation tool, ADAPTIC and ICFEP run on separate processors
as independent black box solvers and are started separately. After start-up, the two
programs wait until receiving a run request and prescribed interaction data from the
INTERFACE program in a specific sequence which is managed by the INTERFACE
program. Then the programs start to solve their own partitioned sub-domain as black
box solvers, during which the INTERFACE is waiting for the either program to
solve for equilibrium in its respective sub-domain under the prescribed interface
conditions and to return the complementary displacement/force conditions at the
interface boundary.
Using the obtained results, the state of coupling convergence at the interface of the
partitioned sub-domain is checked. If convergence is not achieved at the interface,
the INTERFACE program calculates the new estimates of the prescribed boundary
conditions and initiates the iterative coupling procedure within the same time/load
step in ADAPTIC and ICFEP. If convergence is achieved, the interface sends a
request of updating the initial condition for the next time-load increment to
ADAPTIC and ICFEP and assigns new prescribed initial boundary condition for the
structure partitioned sub-domains, moving the solution to the next time/load
increment. The sequence of interaction between the INTERFACE, ADAPTIC and
ICFEP programs is detailed in Figure 4.11.
123
Chapter 4 Iterative Coupling
124
Chapter 4 Iterative Coupling
For a given problem, the user specifies the input data files for ADAPTIC and ICFEP
corresponding to structure and soil partitioned sub-domains, respectively. These
contain the typical relevant information such as the finite element model, material
models, analysis type, etc. In addition to these, an input file for the INTERFACE
program is also specified which defines the coupling region, where the coupling
interaction occurs (soil-structure interface), for both the structure and soil partitioned
models. Moreover, the required number of time/load increments, the coupling
convergence criteria at the interface and the type of update technique required for the
coupling analysis are specified in the same data file. A sample structure of the
INTERFACE data file is given in Appendix A. After launching all three programs,
while ADAPTIC and ICFEP are waiting for the initiation request and the prescribed
interactive values to be posted by the INTERFACE, the INTERFACE starts
generating a formatted direct access FORTRAN data communication file based on
its own input data file. This file is just for communication purposes between the
programs, and is read-write accessible to all programs, where a sample structure is
given in Appendix B. After the communication data file has been generated by the
INTERFACE, the first calls of the augmented subroutines in ADAPTIC and ICFEP
codes identify the coupling regions taking part in the coupling process as described
in Section 4.9 (see Figure 4.12).
125
Chapter 4 Iterative Coupling
The coupling procedure continues until the required time/load steps are completed.
At this stage, both ADAPTIC and ICFEP produce the required results of their own
solved partitioned sub-domain in separate output files, which can be accessed
directly via their own post processing tools for analysing the results. In addition, the
effect of different coupling parameters on the convergence characteristics, including
the achieved compatibility and equilibrium defaults at the interface and the
convergence rate, can be analysed by considering the output file generated by the
INTERFACE program.
126
Chapter 4 Iterative Coupling
In this chapter, different iterative coupling methods are presented for partitioned
analysis of soil-structure interaction problems, where it is assumed that the overall
domain is divided into physical partitions consisting of soil and structure sub-
domains. The advantage of the iterative coupling schemes in which the partitioned
structure sub-domain is treated by Dirichlet boundary was discussed, and particular
emphasis is therefore given in this work to D-N iterative coupling methods.
The developed tool is used in this research to assess the adequacy and applicability
of various coupling methods in nonlinear soil-structure interaction analysis,
including the new methods proposed in this work. In this context, the developed
approach is believed to offer great potential towards providing an integrated
interdisciplinary computational framework which combines the advanced features of
127
Chapter 4 Iterative Coupling
128
Chapter 5
Interface Relaxation
5.1 Introduction
U 1 I U Ti I U Bi
i I 1 I I
T (5.1)
n n n
In the above, I is a real positive parameter, which is called the relaxation parameter
It is clear from Equation (5.1) that the interface relaxation method possesses the
benefits of being simple to implement and having an undemanding process of
estimating new updates of boundary conditions in corrective iterations.
130
Chapter 5 Interface Relaxation
analogous procedure for evaluating the adaptive relaxation parameter in the context
of soil-structure interaction using FEM-FEM coupling is proposed here, and its
convergence characteristics is investigated. It is shown that, in contrast to the
traditional relaxation scheme, in which the relaxation parameter is typically
evaluated by trial and error, an adaptive relaxation scheme offers improved prospects
for achieving convergence and computational efficiency in complicated large scale
nonlinear problems. The evaluation of such prospects and the comparison of
different relaxation schemes are therefore primary objectives of this chapter,
particularly considering soil-structure interaction problems with nonlinearity in both
structure and soil.
be specified prior to the start of coupling analysis. Since the constant relaxation
parameter should be determined in advance, there are two major issues regarding the
applicability of such a technique, namely: i) determination of the range of suitable
relaxation parameters for the specific problem under consideration in order to
achieve convergence, and ii) selection of the optimum relaxation parameter in order
to achieve maximum computational efficiency.
131
Chapter 5 Interface Relaxation
It can be shown that by introducing a suitable constant relaxation parameter, not only
an unconditionally convergent algorithm may be achieved, but also significant
improvement in the convergence rate may be attained by prescribing an optimum
relaxation parameter.
This can be illustrated by considering the same SDOF soil-structure interaction test
system and following the sequential D-N coupling algorithmic steps, while
employing the constant interface relaxation:
U 1 U Ti U Bi
i I 1 I I
T (5.3)
n n n
U
i I
U Ti cte
I
B n (5.4)
n
with,
U U 1 ( 1) U
i K
B n
i
T
K
n
K i I 0
B n U Ti
I 0
n (5.6)
Now, it can be easily shown that for this relaxation scheme to converge,
1 1 should be less than one:
1 1 1 1 1 1 1 (5.7)
132
Chapter 5 Interface Relaxation
1 1 1 0
2 0 2 (5.8)
1 1 1
1
Figure 5.1 shows the 1 1 value against the relaxation parameter for
different assumed values. Considering Equations (5.8) and (5.7), it can be shown
that for all possible partitioned soil and structure sub-domains stiffness ratios,
The figure also illustrates that there exists an optimum relaxation value that ensures
convergence at the highest rate:
0 Optimum( ) 1 (5.9)
133
Chapter 5 Interface Relaxation
1 1 K BEffective
Optimum( ) (5.10)
1 K Effective K BEffective KTEffective
1 TEffective
KB
where K , U and F represent the global stiffness matrix (effective stiffness
matrix for dynamic analysis), the displacement vector and external force vector
(effective force vector for dynamic analysis) of the coupled soil-structure interaction
system.
form as:
vectors U i
and F i
ext correspond to displacements and external loads
134
Chapter 5 Interface Relaxation
Henceforth for the sake of simplicity, but without loss of generality, it is assumed
that there is no external load applied at the interface of the coupled system (i.e.
F 0 ).
i
ext
In the above, vectors U Xi and FXi correspond to displacements and external loads
The response of the above separately modelled sub-domains can now be coupled by
enforcing compatibility and equilibrium conditions at the interface using a sequential
D-N iterative coupling scheme, as follows:
STEP 1: At the start of each increment, the structure domain is loaded by the
external forces FTT , while the displacements at the interface nodes, UTi , are
n n
U U
i I
T (5.15)
n
135
Chapter 5 Interface Relaxation
STEP 2: The structural solver computes the response of the structure, using
n n
U
T
T
I
n
K11T
-1
F K
T
T I
n
T
12
U Ti
I
n (5.16)
F
i I
U TT K 22 U Ti
I I
T n K 21
T T
(5.17)
n n
It should be noted here that although Equations (5.16) and (5.17) are for a linear
response, the same entities can also be readily obtained for a nonlinear response from
the model of the structural sub-domain.
STEP 3: The corresponding interface forces at the soil sub-domain can be calculated
by applying equilibrium:
F F
i I
B n
i I
T n 0 (5.18)
STEP 4: Based on these forces FBi and the external loading applied to the soil
I
domain F
B I
B n , the soil solver computes the response of the soil domain, using
n n
U B I
K12B FB n
1 B I
B n K11
B
I
B (5.19)
U Bi K 21 K 22B F i I
n B n
Again, it is noted that although Equation (5.19) is for a linear response, the same
entities can also be readily obtained for a nonlinear response from the model of the
soil sub-domain.
136
Chapter 5 Interface Relaxation
STEP 5: If convergence to compatibility has not been achieved, the new estimation
of the displacements will be applied to the structure domain:
I 2
If U U
i I
B n
i
T n 2
, then
U 1 U Ti U Bi
i I 1 I I
T n n n
I I 1 , go to STEP2
If U Bi U Ti , then n n 1 , go to STEP 1
I I
n n
U i I
U Ti cte
I
B n (5.20)
n
where,
1
KCB KCT (5.21)
F F
1 1 1
cte KCB B I -1 I
K 21B K11B B K CB K 21
T
K11T T
T
(5.22)
137
Chapter 5 Interface Relaxation
It is noted that, as evident from Equations (5.21) and (5.22), cte would be constant
K CB represent the condensed stiffness matrix of the soil and structure sub-
K CT K 22
T
K 21
T -1
K11T K12T (5.23)
-1
K CB K 22B K 21B K11B K12B (5.24)
U i 0 U
T n
Iteration no. 0 (5.25)
U Bi U Ti cte
0 0
n n
U i 1 U i 0 1 U i 0
T n B n T n
Iteration no. 1 (5.26)
U Bi U Ti cte
1 1
n n
U i 2 U i 1 1 U i 1
T n B n T n
Iteration no. 2 (5.27)
U Bi U Ti cte
2 2
n n
Accordingly, the compatibility defaults at each of the above iteration stages are
given by:
U U I U cte
i 0 i 0 i 0
B n T T (5.28)
n n
U U I U cte
i 1
B n
i 1
T n
i 1
T n (5.29)
138
Chapter 5 Interface Relaxation
U U I U cte
i 2 i 2 i 2
B n T T (5.30)
n n
U U I U 1 U cte
i 1 i 1 i 0 i 0
B n T n B n T (5.31)
n
U U
i 1
B n
i 1
T n
(5.32)
0 0
I U Bi n U Ti n I U Ti n cte
0
U U
i 1
B n
i 1
T n
(5.33)
0 0
I U Bi n U Ti n U Bi n U Ti n
0 0
Similarly it can be shown that the compatibility default at iteration number 2 takes
the following form:
and in general:
n
i I 1
B n U Ti
I 1
n (5.36)
139
Chapter 5 Interface Relaxation
In view of Equation (5.37), the convergence of the presented coupling algorithm can
now be related to the eigenvalues of matrix I I
K
, where the matrix
For the above successive iteration process to converge for any initially prescribed
value, it is necessary and sufficient that all eigenvalues of the matrix
Assuming that eigenvalues of are positive, which is true for a positive definite
as would be the case in linear analysis, the above condition simplifies to:
1 max 1 1 (5.39)
Considering Equations (5.39) and (5.21), it can be shown that for all possible
partitioned soil and structure sub-domains, there exists a range of relaxation
parameters that guarantee convergence to compatibility of the coupling algorithm:
2
0 2 (5.40)
1 max
140
Chapter 5 Interface Relaxation
Moreover, the best choice of relaxation parameter, opt , as depicted in Figure 5.2, is
Therefore, the optimum relaxation value that ensures the convergence and holds the
best convergence rate can be defined as:
1 1
0 opt 1 (5.42)
1 av 1 max min
2
The above convergence analysis, illustrated in Figure 5.2, clearly indicates that
convergence to compatibility in sequential D-N iterative coupling algorithms can be
ensured by using the interface relaxation update technique, at least if the coupled
system under consideration is relatively linear. Indeed, by predefining the relaxation
parameter in the range given by Equation (5.40), the reduction of compatibility error
at the interface is guaranteed. Moreover, Equation (5.42) demonstrates the existence
of an optimum relaxation parameter in the convergent range for which not only the
141
Chapter 5 Interface Relaxation
The previous theoretical study has shown that the convergence of the constant
relaxation scheme for D-N iterative coupling is highly sensitive to the partitioned
sub-domain parameters, specifically the condensed stiffness of the partitioned sub-
domains at the interface. Here, these findings are demonstrated through some
illustrative FEM-FEM coupling examples.
142
Chapter 5 Interface Relaxation
Figure 5.3, where E 210 109 Nm 2 , 0.01 and y 300 106 Nm2 . The above
5.5, with three degrees of freedom at the interface (one rotational and two
translational).
E
y
y
E
143
Chapter 5 Interface Relaxation
conditions (forces).
Firstly, this problem is analysed to illustrate the trial and error process embedded in
the coupling procedure via constant relaxation scheme, for finding the convergence
range and the optimum relaxation parameter. In this regard, in order to demonstrate
the effect of the condensed effective stiffness matrix on the convergence behaviour,
the proposed problem is analysed for different problem partitioning types. This is
achieved by analysing the same problem with different mass ratios at the interface of
the partitioned sub-domains T and B ( m1 / m2 ). The different analysed models,
including the considered partitioned mass ratios at the interface, range of suitable
relaxation parameters and the optimum relaxation parameter associated with the least
computational cost, are presented in Table 5.1. These values have been obtained by
an initial process of trial and error for each model under consideration for 100 steps
with t 0.01s , where the relaxation parameter is chosen as a real value in the range
of ]0, 2[ according to Equation (5.8).
144
Chapter 5 Interface Relaxation
m1 Range of Suitable
Model Optimum Relaxation
m2 Relaxation parameters
200 kg
S1 0. 1 ]0,1.14] [0.8, 0.9]
1800 kg
400 kg
S2 0.25 ]0,1.13] [0.7, 0.8]
1600 kg
600 kg
S3 0.43 ]0,1.1] [0.6, 0.7]
1400 kg
800 kg
S4 0.6 ]0,1] 0.6
1200 kg
1000 kg
S5 1.0 ]0, 0.9] 0.5
1000 kg
1200 kg
S6 1.5 ]0, 0.7] 0.4
800 kg
1400 kg
S7 2.3 ]0, 0.5] 0.3
600 kg
1600 kg
S8 4.0 ]0, 0.35] 0.2
400 kg
1800 kg
S9 9.0 ]0, 0.15] 0.1
200 kg
m1
Table 5.1: Range of suitable and optimal relaxation parameter for different
m2
Figure 5.6 shows the total number of coupling iterations required for various
constant relaxation parameters and the applicable range of relaxation parameters
ensuring convergence to compatibility. A tolerance of ε = 10-4 m was set for the error
based on the norm of compatibility defaults:
U U
i I I
B n
i
T M (5.43)
n
145
Chapter 5 Interface Relaxation
The results confirm that the convergence characteristics of the D-N iterative scheme
using constant relaxation is very sensitive to the chosen relaxation parameter,
rendering its selection a very difficult task, as evidenced by the significant increase
in number of iterations for model S8, between α0.2 with 126 coupling iterations,
α0.1 with 307 coupling iterations and α0.3 with 331 coupling iterations.
Considering Figure 5.6 more closely, the optimum relaxation parameter for each
model is defined as the value corresponding to the minimum number of iterations
(see Table 5.1). The fact that the optimum relaxation parameter varies and is
sensitive to the partitioned model stiffness ratios is confirmed by the obtained
results.
for model S5 ( i.e. m1 m2 1000kg ). The different analysed models, including the
146
Chapter 5 Interface Relaxation
Range of Suitable
Model Optimum Relaxation
Relaxation parameters
K1 5.0 ]0, 0.27] 0.2
K2 4.0 ]0, 0.45] 0.35
K3 2.0 ]0, 0.75] [0.4, 0.5]
K4 1.0 ]0, 0.9] 0.5
ET
Table 5.2: Range of suitable and optimal relaxation parameter for different
EB
As shown in Figure 5.7, the sensitivity of the convergence rate and value of the
optimum relaxation parameter to the problem characteristics is even more critical for
cases where the condensed effective stiffness at the interface of the sub-domain
treated by Dirichlet boundary condition, is much greater than that of the other
domain treated by Neumann boundary condition. In general, as this ratio increases
147
Chapter 5 Interface Relaxation
the optimum relaxation parameter tends to smaller values, while the range of an
applicable relaxation parameters reduces significantly; moreover, the convergence
rate is also considerably reduced, leading to a significant increase in computational
cost.
The results from the two previous parametric studies clearly indicate that by
employing an optimum relaxation parameter the convergence rate of the coupling
method can be considerably enhanced. This fact is demonstrated in Figures 5.8 , 5.9
and 5.10 for model S7 from Table 5.1, where it is clear that the by using an optimum
relaxation, the prescribed tolerance on compatibility defaults is achieved with far less
iterations than those with non-optimum constant relaxation parameters.
0.018
Relaxation=0.1
0.016
Compatibility error at Interface (m)
Relaxation=0.2
0.014
0.012 Optimum Relaxation=0.3
0.01 Relaxation = 0.45
0.008
Convergence Criterion = 0.00017
0.006
0.004
0.002
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Iteration Number
Figure 5.8: Error reduction for different relaxation schemes (Time = 3.6s)
148
Chapter 5 Interface Relaxation
0.02 Relaxation = 0.1
0.018
Relaxation = 0.2
Compatibility Error at Interface (m)
0.016
Optimum Relaxation = 0.3
0.014
Relaxation =0.4
0.012
Relaxation = 0.5
0.01
0.008 Convergence Criterion = 0.00017m
0.006
0.004
0.002
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Iteration Number
Figure 5.9: Error reduction for different relaxation schemes (Time = 3.5s)
0.006
Relaxation = 0.1
0.005 Relaxation = 0.2
Compatibility Error at Interface (m)
Relaxation = 0.3
0.004
Relaxation = 0.4
Relaxation = 0.5
0.003
Convergence Criterion = 0.00017 m
0.002
0.001
0
0 1 2 3 4 5 6 7 8 9 10
Iteration Number
Figure 5.10: Error reduction for different relaxation schemes (Time = 4.18s)
Despite the shortcomings of the constant relaxation scheme and the difficulties of
determining the optimum relaxation parameter, the superiority of this method over
149
Chapter 5 Interface Relaxation
the staggered approach and its verification against the monolithic treatment are
demonstrated next for Model S7. For iterative coupling, the relaxation parameter is
taken at the optimal value of 0.3, and in all cases the dynamic analysis is undertaken
over 10s with a time step of t 0.01s .
Figures 5.11 and 5.12 show the variation with time of the rotation and horizontal
displacement, respectively, at the interface of the sub-domain T for both the
optimum relaxation and monolithic approaches. Similarly Figures 5.13 and 5.14
show the variation with time of the rotation and horizontal displacement,
respectively, at the interface of the sub-domain B for both the optimum relaxation
and monolithic approaches. The variation with time of the rotation and horizontal
displacement at the free end of the cantilever is also depicted in Figures 5.15 and
5.16, respectively. The graphs confirm that the results obtained from coupled
partitioned analysis match very well with those obtained from the monolithic
treatment within the prescribed compatibility tolerance of 10 4 m .
150
Chapter 5 Interface Relaxation
151
Chapter 5 Interface Relaxation
152
Chapter 5 Interface Relaxation
Figure 5.17: Rotation at the tip of the cantilever reduced tolerance to 105
153
Chapter 5 Interface Relaxation
Considering Figure 5.15 it can be seen, that there are slight differences between the
values obtained by the partitioned treatment and those obtained via monolithic
approach. This slight inaccuracy is due to the prescribed tolerance for convergence
to compatibility. In fact, by prescribing a smaller convergence tolerance a closer
match to the monolithic approach can be achieved, as demonstrated in Figure 5.17
where a smaller tolerance of 10 5 is used.
Here, the static plane strain problem of Figure 5.18a is considered and discretised
using 8-noded quadrilateral elements. The presented system is partitioned into two
sub-domains, namely T and B , where each partitioned domain has 5 interface
nodes (10 DOFs) as shown in Figure 5.18b. The resulting partitioned problem is
treated by the D-N iterative coupling technique. As pointed out before, prescribing
Neumann boundary condition on T will result in singularity of the matrices for
154
Chapter 5 Interface Relaxation
Iterative coupling is undertaken with constant relaxation for different stiffness ratios
of partitioned sub-domains T and B ( ET / EB ). The results obtained from
coupled partitioned analysis match very well with those obtained from the
monolithic treatment within the prescribed compatibility tolerance.
Details of the analysed models, including the elastic modulus ratios, range of suitable
relaxation parameters and the optimum relaxation parameter associated with the least
computational cost, are presented in Table 5.3. These values have been obtained by a
process of trial and error for each model, where the relaxation parameter is chosen as
a real value in the range of ]0, 2[.
Table 5.3: Range of suitable and optimal relaxation parameter for different ET / EB
Figure 5.19 shows the number of coupling iterations required for various constant
relaxation parameters for each of the considered coupled systems, where
convergence is assumed at a tolerance of 1 10 4 L for the compatibility defaults
(see Equation (5.43), with L=1m being the characteristic element size. The results
confirm that the convergence behaviour is significantly influenced by the stiffness
ratios of the partitioned sub-domains.
155
Chapter 5 Interface Relaxation
100
90
Number of Iterations 80
70
60
50
40
30
20
10
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3
Relaxation Parameter
M1 M2 M3 M4 M5 M6
Figure 5.19: Influence of relaxation parameter on convergence properties
Considering Figure 5.19, the optimum relaxation parameter for each model can be
easily defined as the one corresponding to the minimum number of iterations. The
fact that the optimum relaxation parameter varies and depends on the partitioned
model stiffness ratio has been illustrated earlier, and is further confirmed by the
currently obtained results. The sensitivity of the convergence rate and value of the
optimum relaxation parameter to the problem parameters is even more critical for
cases where the domain treated by the Dirichlet boundary condition is relatively
stiffer than the other domain treated by the Neumann boundary condition. In general,
as this stiffness ratio increases, the range of convergent relaxation parameters
significantly reduces, and the optimum relaxation parameter tends to smaller values,
leading to deterioration in the convergence rate and a significant increase in
computational cost. The convergence rates of the different relaxation schemes are
illustrated in Figure 5.20 for model M4, which demonstrate that the optimum
relaxation provided a much faster convergence rate than other non-optimum
relaxation schemes.
156
Chapter 5 Interface Relaxation
Figure 5.20: Error reduction for different relaxation schemes (model M4)
The results show that as the discretisation of the partitioned sub-domains becomes
finer and as a result more interface degrees of freedom are employed, the range of
applicable relaxation parameter ensuring convergence becomes smaller. Moreover,
as evident from Figure 5.22, the optimum relaxation parameter changes noticeably.
For instance, considering the case of Mesh 0.5 the range of applicable relaxation
parameters is ]0, 1.1] and this would significantly be reduced to ]0, 0.6] for Mesh
2.0. In addition, the value of the optimum relaxation parameter for Mesh 0.5 is about
0.8, which is out of the applicable convergence range of Mesh 2.0 for which the
optimum relaxation parameter is about 0.5, highlighting the significant sensitivity of
the constant relaxation scheme to the partitioned sub-domains parameters.
157
Chapter 5 Interface Relaxation
Mesh 0.5
50
Mesh 1.0
Number of Total Iterations
40 Mesh 2.0
30
20
10
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1
Relaxation Number
158
Chapter 5 Interface Relaxation
solvers for large multi-degree of freedom problems. Therefore, the optimum constant
relaxation parameter is usually found by a process of trial and error for every case
under consideration. Convergence conditions clearly indicate that convergence of the
Dirichlet-Neumann iterative coupling method depends on the stiffness ratio of the
partitioned soil and structure sub-domains. In linear analysis, the stiffness matrix
remains constant at all computational steps. However, for nonlinear analysis the
effective stiffness depends on the deformation state. This change of the stiffness of
the partitioned domains will have a significant effect on the convergence
characteristics of the coupling method at the interface level, where the optimum
relaxation parameter will change over the load/time increments. Accordingly,
ensuring convergence and computational efficiency in large scale nonlinear
problems, where the optimum relaxation parameter is to be determined over the full
range of response by trial and error, would be very difficult if not impossible.
Instead, such a scenario requires a dynamic change of relaxation parameter to ensure
optimal convergence and associated computational efficiency. This leads to the
concept of adaptive relaxation, where the relaxation parameter is determined during
coupling iterations, using error minimization techniques. With adaptive relaxation,
the use of trial and error for determining the relaxation parameter is avoided, whilst
leading to significant improvement in the convergence rate of iterative coupling, as
discussed hereafter.
159
Chapter 5 Interface Relaxation
the compatibility and equilibrium defaults at the next iteration by using the
compatibility and equilibrium default history of the previous iterations.
Considering again Equation (5.20) for a linear coupled system with constant [],
while assuming an adaptive change of relaxation parameter, the interface
displacement values at the structure and soil sub-domains at different iteration stages
can be expressed as:
U i 0 U
T n
Iteration no. 0 (5.44)
U Bi U Ti cte
0 0
n n
U i 1 U i 0 1 U i 0
T n 1 B n 1 T n
Iteration no. 1 (5.45)
U Bi U Ti cte
1 1
n n
U i 2 U i 1 1 U i 1
T n 2 B n 2 T n
Iteration no. 2 (5.46)
U Bi U Ti cte
2 2
n n
Accordingly, the compatibility default at each of the above iteration stages is given
by (see Equation (5.36)):
160
Chapter 5 Interface Relaxation
with:
EI U Bi n U Ti n
I I
(5.52)
Considering Equation (5.51), the proposed method defines the adaptive relaxation
parameter at each coupling iteration stage ( I 1 ), based on minimizing the
following:
I I E
2
I 1 I 1 I 1 2 (5.53)
2
(5.54)
Given that:
161
Chapter 5 Interface Relaxation
1 1
EI 1 EI 1 EI EI 1 (5.56)
I I
Substituting Equation (5.56) back into Equation (5.54) leads to the following:
d
I 1 0
d I 1
E 2 EI 1EI
2 (5.58)
I I 1
I 1
EI 1 EI 2
2
d 2 2
EI EI 1 2 0
2
(5.59)
d I 1 I
2
minimisation of I 1 .
Now, defining:
U U U
i I i I i I 1
T T T (5.60)
n n n
U U U
i I
B n
i I
B n
i I 1
B n (5.61)
162
Chapter 5 Interface Relaxation
U U U
I 1
I EI 1
i I i I i
T T T (5.62)
n n n
U U E E
i I i I
B n T I I 1 (5.63)
n
Therefore, Equation (5.58) for the adaptive relaxation parameter can be further
simplified to:
I 2
U U Ti U Bi
i I I
T n 2 n n
I 1 2
(5.64)
U U i I i I
T n B n
2
The automatic choice of relaxation parameter from either Equation (5.64) or (5.58)
can be used for successive evaluation of the interface displacements using the
iterative coupling algorithm for I 0 . For the first iterative stage (i.e. I 0 ), the
relaxation parameter can be chosen as an arbitrary real value to start the coupling
process. Although the choice of the first relaxation parameter does not have
influence on the convergence, a choice which is close to the optimal value of
Equation (5.42) will result in better error reduction for the first coupling iteration,
and hence to a fewer coupling iterations with adaptive relaxation.
The previous theoretical study has shown that the relaxation parameter in D-N
iterative coupling may be adaptively adjusted for every iteration (I≥1), leading to
enhanced convergence and improved efficiency. Here, these findings are
demonstrated through some illustrative FEM-FEM coupling examples.
The coupled test system considered in Section 5.2 for constant relaxation is
investigated here with adaptive relaxation. In this case, since the interface of the
163
Chapter 5 Interface Relaxation
coupled test system consists of a single degree of freedom, Equation (5.64) can be
rewritten in the form of:
I 2
U U Ti U Bi
I I
U
i i I
T n 2 n n T
I 1 n
(5.65)
U U
2 I i I
U U
i I i I i
T B n T n B n
n 2
Expanding Equation (5.65) using Equations (5.61) and (5.4), it can be shown that:
1 1
I 1 (5.66)
U U 1
i I i I 1
B n B n
1
U U
i I
T n
i I 1
T n
In this section, the efficiency of the adaptive relaxation scheme is evaluated through
considering the dynamic FEM-FEM coupling problem of Example 1 in Section
5.2.2.1. The problem is analysed for the different models presented in Tables 5.1 and
5.2, where comparison is made between the results of constant and adaptive
relaxation.
As before, the results obtained from partitioned analysis are in excellent agreement
with the monolithic approach, as expected. A comparison between the number of
164
Chapter 5 Interface Relaxation
Table 5.4: Number of coupling iterations with adaptive and optimum relaxation for
1000 time-steps ( t 0.01s ) with a tolerance of 10-4 m
These results clearly indicate that the adaptive relaxation method is far superior to
the constant relaxation scheme, since not only does it avoid the process of trial and
error for finding the relaxation parameter, but it also enhances the convergence rate
of iterative coupling significantly. This fact is further demonstrated in Figures 5.23-
5.30, for the most critical cases with regard to number of coupling iterations required
for convergence with optimum relaxation, namely Model K1 and Model S8. These
figures present the convergence rate of the adaptive and corresponding optimum
relaxation in different randomly selected time steps of the analysis. The automatic
change of the relaxation parameter value is also illustrated in these figures, where it
165
Chapter 5 Interface Relaxation
166
Chapter 5 Interface Relaxation
167
Chapter 5 Interface Relaxation
168
Chapter 5 Interface Relaxation
169
Chapter 5 Interface Relaxation
Here the static FEM-FEM coupled problem of example 2 in Section 5.2.2.2 is also
analysed using the adaptive relaxation technique with the same prescribed tolerance,
where a comparison between the number of coupling iterations using optimum
relaxation and adaptive relaxation is provided in Table 5.5.
As before, these results confirm that the adaptive relaxation method is far superior to
the constant relaxation scheme. This fact is further demonstrated in Figure 5.31 for
the most critical case with regard to convergence, namely model M1, where it is
clear that adaptive relaxation method achieves the prescribed tolerance for
compatibility defaults with 10 coupling iterations compared to the optimum constant
relaxation with 31 coupling iterations. Although the compatibility error after the first
iteration is significantly increased, this is only due to the initial choice for the
relaxation parameter at I=0 ( 0 1.0 ). In fact, if the constant relaxation approach is
applied with 1.0 , convergence would not be achieved. However, this is not a
170
Chapter 5 Interface Relaxation
serious problem for the adaptive relaxation approach, since the compatibility default
decreases significantly in subsequent iterations.
M1 31 10
M2 20 7
M3 11 6
M4 7 4
M5 6 5
M6 4 4
Table 5.5: Number of required coupling iterations with adaptive and optimum
relaxation
Figure 5.31: Error reduction for adaptive and constant relaxation schemes (Model
M1)
171
Chapter 5 Interface Relaxation
In this section two examples of soil-structure coupling are presented, utilising both
the constant and adaptive relaxation approaches. The first example is a simple linear
soil-structure interaction problem, as shown in Figure 5.32, which is aimed at
verifying the relaxation approaches for iterative coupling against the monolithic
treatment.
The problem consists of linear static plane strain analysis of a concrete cantilever
wall, resting on a flexible soil, loaded at the top with a horizontal force. In
partitioned analysis of the problem, the adaptive relaxation shows reasonable
convergence characteristics and converges within 3 coupling iterations to a
prescribed tolerance of 10-4 m. With the same problem also modelled monolithically,
the obtained results from both monolithic and partitioned analysis are in excellent
agreement, as illustrated in Figure 5.33. This figure shows the horizontal
displacement of the beam obtained for the non-interactive case (rigid base) and for
the interactive case by both the partitioned and monolithic approaches. This further
demonstrates that by enforcing convergence to equilibrium and compatibility in the
172
Chapter 5 Interface Relaxation
sequential D-N iterative coupling algorithm, a strong coupling of the partitioned sub-
domains can be achieved at the interface.
The second example represents a typical urban situation, where due to foundation
settlements the force quantities in the structural members often are revised. The
example considers a steel frame resting on soil and subjected to static loading
(Figure 5.34), where nonlinear elasto-plastic constitutive behaviour of the soil as
well as geometric and material nonlinearity of the structure are taken into account.
The considered frame is taken from a building, which is designed for office purposes
and assumed to be loaded equally on each floor, where plan and elevation views are
shown in Figure 5.35 and 5.36, respectively. The soil-structure interaction analysis is
carried out assuming plane strain conditions in the soil using an effective out-of-
plane width of 1m, where the developed domain decomposition approach is
employed utilising ADAPTIC and ICFEP. The main objective of this study is to
establish the applicability and efficiency of the presented coupling algorithm using
the adaptive relaxation technique, highlighting its merits compared to constant
relaxation scheme.
173
Chapter 5 Interface Relaxation
174
Chapter 5 Interface Relaxation
Table 5.6: Geometric and material properties of the partitioned soil-structure system
175
Chapter 5 Interface Relaxation
To assess the merits of the presented relaxation coupling approaches, both constant
and adaptive relaxation are employed. For the constant relaxation approach, the
considered problem is analysed using different relaxation parameters ( 𝛼 ∈]0,1] ).
Figure 5.37 shows the total number of coupling iterations required for various
constant relaxation parameters and the applicable range of relaxation parameters
ensuring convergence to compatibility over the full range of response consisting of
ten increments. A tolerance ε = 10-4 m was set for the compatibility error of each
coupled degree of freedom at the interface (see Equation (5.43)).
The results confirm that the convergence behaviour of the iterative scheme using
constant relaxation is very sensitive to the chosen relaxation parameter, rendering its
selection a very difficult task, as evidenced by the significant increase in number of
iterations between α=0.55 with 64 coupling iterations and α0.65 with 180
coupling iterations.
176
Chapter 5 Interface Relaxation
Table 5.7 presents the range of constant relaxation parameters which guarantee
convergence, the optimum relaxation parameter and the number of coupling
iterations required for convergence to the prescribed tolerance of 10 4 m for both
adaptive and constant relaxation schemes.
(0,0.65] 0.55 64 46
The convergence rates of the three approaches in the first, fifth and last load
increments are illustrated in Figures 5.38, 5.39 and 5.40 respectively. This
demonstrates that the adaptive relaxation achieves a faster convergence rate than the
relaxation scheme, whether optimal or not.
177
Chapter 5 Interface Relaxation
Figure 5.38: Error reduction for different relaxation schemes for the first load
increment
Figure 5.39: Error reduction for different relaxation schemes for the fifth load
increment
178
Chapter 5 Interface Relaxation
Figure 5.40: Error reduction for different relaxation coupling schemes for the last
load increment
Figure 5.41 shows the number of coupling iterations required in each increment for
the different coupling schemes, which demonstrates that the optimum relaxation
parameter should be determined over the full range of response in nonlinear
problems.
In fact, in nonlinear problems, finding the optimum relaxation parameter by trial and
error is not computationally efficient. This is evidenced by considering the behaviour
of coupling with α0.6, where a relatively good convergence rate is observed in the
first three load increments (even better than the optimum relaxation with α0.55),
but as the coupled system presents more nonlinearity the convergence rate decreases
significantly.
179
Chapter 5 Interface Relaxation
These results demonstrate the superiority of the adaptive relaxation scheme, which
achieves much faster convergence than the constant relaxation scheme, whether
optimal or not. It is, however, worth observing that the adaptive relaxation scheme
on average still requires about 5 iterations per load step (for 10 increments), which is
relatively large in comparison with what would be necessary in a typical monolithic
treatment, thus highlighting the need for further enhancement in iterative coupling
algorithms.
180
Chapter 5 Interface Relaxation
reaches a compatible equilibrium state. In the following, the deformation and stress
states of the coupled problem at the end of load application are briefly described.
The vertical deformation profile of the soil surface, with respect to the distance from
the middle column of the frame, obtained from coupled analysis is given in Figure
5.42, where the three troughs correspond to the locations of the three footings
showing their vertical settlement. Clearly, the generated level of vertical settlement,
of the order of 20cm and vertical differential settlement of the order of 10 cm,
requires the structural analysis model to account for geometric nonlinearity.
The deformed shape of the mesh of the soil partitioned sub-domain is also depicted
in Figure 5.43.
181
Chapter 5 Interface Relaxation
A vector plot of displacements in the soil sub-domain in the vicinity of the structure
foundations is also shown in Figure 5.44, where the relative magnitude of these
vectors reflects the mechanism of ground deformation.
Figure 5.45 shows contours of stress level at the end of analysis under the applied
load on the structure. The stress level is the ratio, at the same mean effective stress,
of the current deviatoric stress to the deviatoric stress at failure. It therefore varies
from 0 to 1, where 1 indicates full plasticity and failure. It is evident from Figure
5.45 that the applied loading conditions have mobilised an extensive plastic zone
182
Chapter 5 Interface Relaxation
underneath the building. This zone is, however, bigger and deeper under the middle
footing, which is in agreement with the previous figures that show most of the
deformation and load concentration under the middle column.
Figure 5.46: Deformed shape and bending moment (kN.m) of the frame (scale=5)
Finally, the deformed shape and the bending moment contours of the partitioned
structure sub-domain are shown in Figure 5.46. It is evident from the deformed
183
Chapter 5 Interface Relaxation
shape of the structure and also the displacement vectors underneath each of the three
footings in Figure 5.44 that they experience rigid tilting (indicated previously in
Figure 5.42) and significant vertical settlements.
This chapter investigates the use of interface relaxation coupling techniques in FEM-
FEM domain decomposition analysis of soil-structure interaction. In this respect, the
overall domain is divided into physical partitions consisting of soil and structure sub-
domains. Coupling of the separately modelled sub-domains is undertaken with
various interface relaxation algorithms based on the sequential iterative D-N sub-
structuring method, which ensures compatibility and equilibrium at the interface
boundaries of the sub-domains. This enhancement of the convergence characteristics
is achieved by employing a relaxation of the interface Dirichlet entities in successive
iterations. Various mathematical and computational characteristics of the coupling
method, including the governing convergence rate and choice of relaxation
parameter, are studied, where it is demonstrated that the convergence behaviour of
the constant relaxation scheme is very sensitive to the stiffness ratio of the
partitioned sub-domains. This renders the selection of an optimum relaxation
parameter very difficult, leading to considerable computational inefficiency,
especially for realistic large-scale nonlinear problems.
184
Chapter 5 Interface Relaxation
It should be emphasised that the considered methods are generally applicable to the
coupling of various computational procedures that are used for nonlinear structural
and geotechnical analysis. In this context, these coupling methods have the potential
to provide an integrated interdisciplinary approach which combines the advanced
features of both structural and geotechnical modelling for a variety of problems in
soil-structure interaction analysis.
Finally, it should be noted that although using the adaptive scheme removes
significant difficulties in the conventional relaxation iterative coupling scheme, there
is a pitfall associated with both constant and adaptive relaxation schemes when the
partitioned sub-domain parameters dictate very small values of the relaxation
parameter for convergence. In such cases, relaxation methods breakdown with poor
convergence rates and significant computational inefficiency.
185
Chapter 5 Interface Relaxation
186
Chapter 6
6.1 Introduction
Although the condensed tangent stiffness may be readily available with some
nonlinear field modelling tools, a more general approximation for the condensed
stiffness matrices is desirable and can indeed be achieved during the course of
coupling iterations, thus avoiding the need for explicit determination and extraction
of the stiffness matrices. In this respect, the condensed tangent stiffness matrix is
approximated via reduced order models, building on a previous approach by
Vierendeels et al. (2007) concerned with fluid–structure interaction problems.
Various significant modifications to this approach are proposed in this chapter,
leading to a versatile and efficient approach for coupled modelling of nonlinear soil-
structure interaction problems. In this regard, a major pitfall associated with the
original method causing divergence is overcome by a new selective
addition/replacement procedure of force and displacement mode vectors. Moreover,
in order to achieve better approximation of the condensed tangent stiffness matrix in
the initial stages of coupling iterations, a mixed reduced order method is proposed,
which achieves a more robust coupling technique than the conventional reduced
order method. In the following such approaches are discussed in detail.
188
Chapter 6 Reduced Order Method
In the above, vectors U XX and FXX correspond to the displacements and external
forces for the non-interface degrees of freedom, while U Xi and FXi correspond to
Assume that an iterative coupling method is employed for coupling of the above
partitioned soil and structure sub-domains, where the compatibility and equilibrium
defaults at the interface of the structure and soil sub-domains for iteration number I
of load/time step n take the form:
U n U Bi n U Ti n
I I I
(6.3)
F n FBi n FTi n
I I I
(6.4)
Consider the new iterative estimation of the interface displacements and forces in the
successive iteration I+1 expressed incrementally in the general form:
U U Ti U Ti
i I 1 I I
T (6.5)
n n n
U
i I 1
U Bi U Bi
I I
B n (6.6)
n n
F i I 1
FTi FTi
I I
T n (6.7)
n n
F i I 1
FBi FBi
I I
B n (6.8)
n n
F i I
KTC U Ti
I
T n (6.9)
n
189
Chapter 6 Reduced Order Method
F i I
K BC U Bi
I
B n (6.10)
n
in which KTC and K BC are the condensed tangent stiffness matrices at the
U i I 1
U Ti
I 1
B n 0 (6.11)
n
F i I 1
FBi
I 1
T n 0 (6.12)
n
Expanding Equation (6.12) using Equations (6.4), (6.7) and (6.8) gives:
F n FTi n FBi n 0
I I I
(6.13)
Similarly expanding Equation (6.11) using Equations (6.3), (6.5) and (6.6) gives:
Now by substituting U Bi from Equation (6.14) in Equation (6.15) the following
I
is obtained:
F n KTC U Ti n K BC
I I
U 0
i
T
I
n U
I
n
(6.16)
Therefore the incremental value of U Ti takes the following form of:
I
190
Chapter 6 Reduced Order Method
U K K
1
U n F n
I I I
i
T
C
T
K BC C
B (6.17)
n
U K K
1
i I
U n F n
I I
B n
C
T
K BC C
T (6.18)
Substituting Equations (6.13), (6.9) and (6.10) in Equation (6.14) the incremental
I
FTi n K BC KTC
1 1 1
K
U
I
n
C
B
1
F n
I
(6.19)
I
FBi n K BC KTC
1 1 1
K
U
I
n
C
T
1
F n
I
(6.20)
The above equations can be used in successive update of boundary conditions of the
different sequential and parallel coupling algorithms presented in Chapter 4 (see
Appendix C). Depending on the utilized coupling algorithm type, the following
equations could provide the new estimation of the iterative Dirichlet/Neumann
boundary conditions ensuring convergence.
U
U Ti KTC K BC K
I 1 I 1
U n F n
i C I I
T B (6.21)
n n
U
U Bi KTC K BC K
1
i I 1
U n F n
I C I I
B n T (6.22)
n
F i I 1
T n
I
n
FTi K BC KTC
1 1 1
K U
I
n
C
B
1
F n
I
(6.23)
191
Chapter 6 Reduced Order Method
F
i I 1
B n
I
n
FBi K BC KTC
1
K
1 1
U
I
n
C
T
1
F n
I
(6.24)
Although, as evidenced by the above equations, this method requires the access to
the stiffness/flexibility matrices of the partitioned sub-domains and poses more
interface related computations compared to interface relaxation schemes, there are
major advantages associated with the above update technique. More specifically,
these advantages are, i) avoiding the need for using predefined relaxation parameters
and trial and error for ensuring optimal convergence in the coupling procedure ii)
offering a high convergence rate compared to relaxation schemes, and iii)
facilitating the development of different coupling algorithms (see Appendix C). In
this respect, as further shown in Appendix C, this approach provides richer and more
general update technique in iterative coupling algorithms compared to both adaptive
and constant relaxation, leading to a versatile and efficient approach for partitioned
modelling of coupled systems.
192
Chapter 6 Reduced Order Method
STEP 2: The structural solver computes the response of the structure for : FTi
I
STEP 3: The corresponding interface forces at the soil domain can be calculated by
applying equilibrium: FTi FBi 0
I I
n n
STEP 4: Based on these forces and the external loading applied to the soil domain,
the soil solver computes the response of the soil domain for: U Bi
I
U
U Ti KTC K BC K
I 1 1
U n
i I C I
T n n B
Here, the coupling procedure and convergence behaviour of the sequential Dirichlet-
Neumann algorithm are examined using the condensed interface stiffness matrix for
updating the Dirichlet boundary conditions. Consider the linear coupled spring
system consisting of eight spring elements associated with different stiffness as
illustrated in Figure 6.1. In order to perform the partitioned analysis procedure, the
coupled system is decomposed into two separately modelled partitioned sub-domains
T and B (Note that nodes 5 and 4 are the interface nodes corresponding to 2
193
Chapter 6 Reduced Order Method
K K i Pi
K
K j Pj
To facilitate the verification of the various coupling schemes presented hereafter, the
coupled problem is initially modelled and solved monolithically. Performing global
structural analysis using a monolithic approach results in the formation and solution
of the following global system of equations:
40 10 0 0 0 2 0
10 40 10 20 0 20
3
0 10 25 0 15 4 0 (6.25)
0 20 0 35 15 5 0
0 0 15 15 40 6 60
194
Chapter 6 Reduced Order Method
67
231
2 268
231
3 541
4 (6.26)
231
5 463
6 231
241
77
Clearly any result obtained by the partitioned analysis of the coupled spring system
should mach the results obtained by the monolithic approach given by Equation
(6.26). In order to perform the partitioned analysis using the condensed interface
stiffness approach, the partitioned sub-domains must be modelled in isolation, and
their corresponding condensed interface stiffness matrices must be obtained.
written in the form of Equation (6.27). In the following PXi and Xi correspond
T T
0 15 4 T P4 T
i i
15
0
15 15 5i P5i (6.27)
T T
15 15 40 60
6
75 45
15 0 15 1 8 8
KTC
15 15 (6.28)
0 15 15 40 45 75
8 8
195
Chapter 6 Reduced Order Method
written as:
40 10 0 0 2 0
10 40 10 20 3 20
i i
0 4 B P4 B
(6.29)
0 10 10
0 20 0 20 5i P5i
B B
In the above PXi and Xi correspond to the forces and displacement of the
B B
22 16
1
10 0 0 10 40 10 0 0 3 3
K BC
(6.30)
0 20 0 20 10 40 10 20 16 28
3 3
Utilizing the obtained condensed interface stiffness matrices of Equations (6.28) and
(6.30) in the update of boundary conditions, the partitioned sub-domains can be
effectively coupled at the interface using various coupling algorithms. In the
following, the presented example is coupled using the sequential Dirichlet-Neumann
coupling algorithm which is the main emphasis of this work. Demonstration of the
applicability of the presented update technique for other types of parallel/sequential
coupling algorithms can be found in Appendix D for the same example.
6.2.1.1 Sequential DirichletNeumann Iterative Coupling
196
Chapter 6 Reduced Order Method
3
i I 0 0 P i I 0
15 0 15 4 T 4 T 6 2
0 i I 0 0 P i I 0 P i I 0 45
15 15 5 T 5 T 4 T
2
(6.31)
15 15 40 6 60 P5i 45
I 0
T
2
45 45
equilibrium (i.e. P4i and P5i
I 0 I 0
), and solving the partitioned sub-
B 2 B 2
domain B gives:
5
2 0 2 6
40 10 0 0 10
10 40 10 20 3 20 3 3
i I 0 i I 0 i I 0
0 4 B P4 B 45 / 2 4 B 67
(6.32)
0 10 10
20 i I 0 P i I 0 45 / 2 i I 0 12
5 B 5 B 5 B 107
0 20 0
24
compatibility at the interface is not achieved at this iterative stage. Therefore, new
estimates of Dirichlet data should be calculated according to Equation (6.21) to
enforce convergence in the next iteration:
1
75 45 22 16 22 16 67
4 T 0 8
i I 1
8 3 3 3 3 12
I 1
.
5i 0 45 75 16
28 16 28 107
T
8 8 3 3 3 3 24 (6.33)
541
231
463
231
197
Chapter 6 Reduced Order Method
Prescribing the above estimate for interface Dirichlet data at the interface of T , at
i I 1 541 241
4 T 231 P4i 77
I 1
15 0 15 T 6
0 463 i I 1 130
15 15 5 P5 T P4 T
i I 1 i I 1
(6.34)
T 231 11
15 15 40
60 P5i 1300
I 1
6
T
77
67
0
2 2 231
40 10 0 0 20 3
268
10 40 10 20 3
i I 1 P i I 1 130 i I 1 231
0 4 B 4 B
11 4 B 541
(6.35)
0 10 10
20 i I 1 i I 1 1300 i I 1 231
5 B P5 B 5 B 463
0 20 0
77
231
B B T
and 5i
I 1
, thus at the first coupling iteration (I=1) convergence to compatibility is
T
achieved. Moreover, the obtained results by the above coupling procedure given by
Equations (6.34) and (6.35) are identical to those obtained with the monolithic
treatment given by Equation (6.26).
and B represented by Equations (6.1) and (6.2). Considering Equation (6.21) and
198
Chapter 6 Reduced Order Method
knowing that the equilibrium condition at the interface is already satisfied for each
F n 0 ),
I
coupling iteration in STEP 6 (i.e. the renewal of the boundary
U
U Ti KTC K BC K
I 1 1
U n
i I C I
T B (6.36)
n n
1
U U Ti I K BC KTC
i I 1 I 1 I
T U n
(6.37)
n n
U β U Bi I β U Ti
i I 1 I I
T (6.38)
n n n
β I + λ
1
(6.39)
and:
1
λ K BC KTC (6.40)
It can be immediately noted that Equation (6.38) resembles the interface relaxation
199
Chapter 6 Reduced Order Method
process of trial and error for each case under consideration. Although the proposed
approach involves more interface-related computations than the relaxation approach,
overall computational efficiency is envisaged due to the enhanced convergence of
iterative coupling.
The convergence of the presented sequential D-N algorithms can now be established
for a linear case by following the algorithmic steps of sequential D-N, while
assuming that the iteratively updated boundary conditions have the following form:
U X U Bi I X U Ti
i I 1 I I
T (6.41)
n n n
Using Equation (6.41) for renewal of the interface Dirichlet data, in coupling of
Equations (6.1) and (6.2), the compatibility default after K successive iterations at
the interface of the decomposed soil-structure system, can be obtained by the
following difference equation:
with:
1
λ K BC KTC (6.43)
1
K BC K 22B K 21B K11B K12B = condensed stiffness matrix of the structure
KTC K 22
T
K 21
T 1
K11T K12T = condensed stiffness matrix of the soil sub-
200
Chapter 6 Reduced Order Method
For the successive iteration process of Equation (6.42) to converge for any initial
value, it can be clearly shown that X should take the form of:
X β I + λ
1
(6.44)
201
Chapter 6 Reduced Order Method
Assume that there are N degrees of freedom at the interface of the coupled problem.
The following N+1 displacement vectors are prescribed at the interface of the sub-
domain treated by Dirichlet boundary conditions ( T ):
U
T
i
T ui1 ,..., uij ,..., uiN , i 0,..., N (6.45)
i
where:
uij
0 i j or i 0 (6.46)
u i j and i 1
F
T
i
T i f i1 ,..., fij ,..., f iN , i 0,..., N (6.47)
202
Chapter 6 Reduced Order Method
the initial secant condensed interface stiffness matrix of the above partitioned sub-
domain corresponding to the chosen incremental displacement εu can be obtained as:
f i1 f 01 f N 1 f 01
u
u
KT
C
, i 1,..., N (6.48)
f iN f 0 N
f NN f 01
u u
Similarly in order to obtain the initial condensed flexibility matrix of the sub-domain
treated by Neumann boundary conditions, the following N+1 force vectors are
prescribed at the interface of the sub-domain ( B ):
F
T
i
B i f i1 ,..., f ij ,..., fiN , i 0,..., N (6.49)
where:
f ij
0 i j or i 0 (6.50)
f i j and i 1
U
T
i
B i ui1,..., uij ,..., uiN , i 0,..., N (6.51)
the initial secant condensed flexibility matrix of the above partitioned sub-domain
corresponding to the chosen incremental displacement f can be obtained as:
203
Chapter 6 Reduced Order Method
6.3.1.1 Example 1: Static FEMFEM Coupling
This example demonstrates the high performance and applicability of using the
initial secant condensed stiffness matrix in coupling linear static FEM-FEM
problems. Consider the plane strain problem of Figure 6.2a, which is discretized
using four noded elements. The presented system is partitioned into two sub-
domains, namely T and B , where each partitioned sub-domain has 3 interface
nodes at its interface (i.e. a total of 6 degrees of freedom), as shown in Figures 6.2b
and 6.2c.
The resulting partitioned problem is treated by the D-N iterative coupling algorithm.
Since, prescribing Neumann boundary conditions on T results in singularity of the
204
Chapter 6 Reduced Order Method
The results obtained from coupled partitioned analysis match very well with those
obtained from the monolithic treatment within the prescribed compatibility tolerance.
The results corresponding to the analysed models, including the considered elastic
modulus ratios, compatibility tolerance and the number of required coupling
iterations for different considered schemes, are presented in Table 6.2.
Partitioned sub-domain T
These results clearly show that using the condensed interface stiffness matrices of
the partitioned sub-domains is a far superior update technique compared to adaptive
relaxation, since it enhances the convergence rate of the coupling method
significantly. In fact, as expected, employing the condensed interface stiffness
matrices for this linear problem leads to convergence after only one iteration. This
fact is further demonstrated in Figures 6.3, 6.4 and 6.5 for models A1, A4 and A6,
respectively.
205
Chapter 6 Reduced Order Method
A1 8.0 6 10 13 16 1
A2 4.0 5 6 10 13 1
A3 2.0 4 6 6 10 1
A4 1.0 3 5 6 6 1
A5 0.5 3 3 5 6 1
A6 0.2 1 3 3 4 1
206
Chapter 6 Reduced Order Method
207
Chapter 6 Reduced Order Method
6.3.1.2 Example 2: Dynamic FEMFEM Coupling
and y 300 106 Nm2 . The above system is partitioned into two sub-domains
namely T and B , as shown in Figures 6.6b and 6.6c, with three degrees of
analysed for duration of 5s with t 0.01s using the initial linear elastic condensed
interface stiffness matrices, as outlined in Section 6.3.1.The obtained results for a
prescribed compatibility tolerance of 10-4m, are compared to adaptive relaxation and
summarised in Table 6.3.
208
Chapter 6 Reduced Order Method
c) Partitioned sub-domain T
209
Chapter 6 Reduced Order Method
Table 6.3: Number of required coupling iterations for 500 time-steps ( t 0.01s )
with a tolerance of 10-4 m
The results obtained from coupled partitioned analysis match very well with those
obtained from the monolithic treatment within the prescribed compatibility tolerance.
Considering Model C1, Figures 6.8 and 6.9 show the variation with time of the
rotation and horizontal displacement, respectively, at the interface of the sub-domain
T for both the partitioned approach with initial stiffness approximation and the
monolithic approaches.
210
Chapter 6 Reduced Order Method
211
Chapter 6 Reduced Order Method
The results show that the update of boundary conditions via the condensed interface
stiffness matrix approach provides a much better convergence rate than via the
adaptive relaxation scheme.
Considering the critical analysed model, C1, it is apparent that the update via
condensed interface stiffness matrix achieves superior convergence rate to adaptive
relaxation by more than 50%. In general, it is evident that the proposed procedure
outperforms the adaptive relaxation scheme in linear FEM-FEM coupling, as
expected. Figures 6.10 and 6.11 show the number of coupling iterations required
over the full 5s duration for Models C1 and C5, where it is clear that the condensed
stiffness approach converges typically after one coupling iteration when the response
is linear elastic.
212
Chapter 6 Reduced Order Method
6.3.1.3 Discussion on Nonlinear Analysis
It has been shown in the previous examples that the condensed interface stiffness
approach in iterative coupling enables convergence to compatibility within one
iteration for linear problems. However, this immediate convergence does not
normally occur for nonlinear problems through the use of the secant matrices given
by Equations (6.48) and (6.52). In nonlinear analysis, the effective stiffness depends
on the deformation state. In such cases the rate of convergence depends on the
change in the effective stiffness. This change of the stiffness of the partitioned sub-
domains can have a significant effect on the convergence characteristics of the
coupling method at the interface level, thus requiring the reformation of the tangent
stiffness matrices during iteration in order to ensure convergence to compatibility
using Equation (6.37), as discussed hereafter.
213
Chapter 6 Reduced Order Method
Although the condensed tangent stiffness matrix at the interface could be determined
with existing nonlinear field modelling tools, this might require significant
modification of such computational tools, the extent and nature of which would vary
between one tool and another. It is therefore proposed that the condensed tangent
stiffness matrix may be reasonably and generally approximated by constructing
reduced order models of the structure and soil sub-domains. The benefit of such an
approach is that it does not require the explicit assembly of the stiffness matrices,
thus providing a general yet potentially efficient coupling technique.
For the partitioned structure sub-domain, the following displacement mode matrix
can be constructed at coupling iteration I 1 :
M I (6.53)
constructed as:
214
Chapter 6 Reduced Order Method
U U .δ
i I
T T (6.55)
n
where:
T
δ 0 ,..., J ,..., I 1 (6.56)
Thus the variation of the interface forces can also be approximated in a similar
manner as:
F F .δ
i I
T n T (6.57)
The coefficients 𝛅 which provide the minimum error in approximating any U Ti
I
1
δ U T . U T U T U Ti n
T T I
(6.58)
Now, the reduced order model of the partitioned structure sub-domain can be
determined from combining Equations (6.57) and (6.58):
F F U . U U U
1
i I T T i I
T n T T T T T (6.59)
n
Comparing the reduced order model of the structure given by Equation (6.59) with
Equation (6.9), it is evident that the condensed tangent stiffness matrix of the
partitioned structure sub-domain can be approximated as:
1
KTC FT U T . U T UT
T T
(6.60)
In a similar manner, the condensed tangent flexibility matrix of the partitioned soil
sub-domain can be obtained by constructing the following force and displacement
mode matrices at coupling iteration I 1 :
215
Chapter 6 Reduced Order Method
(6.61):
F F .η
i I
B n B (6.63)
where:
T
η 0 ,..., J ,..., I -1 (6.64)
U U .η
i I
B n B (6.65)
Accordingly, the reduced order model of the partitioned soil sub-domain can be
determined as:
Comparing the reduced order model of the soil partitioned sub-domain given by
Equation (6.66) with Equation (6.10), it is evident that the condensed tangent
flexibility matrix of the partitioned soil sub-domain can be approximated as:
1 1
K BC U B FB . FB FB
T T
(6.67)
216
Chapter 6 Reduced Order Method
It is worth noting that, in the original reduced order scheme proposed by Vierendeels
et al. (2007) for coupling fluid-structure interaction problems, the prescribed
Neumann data are calculated as a result of solving the reduced order models of the
partitioned sub-domains in consecutive iterations, (except for the first two iterations,
where equilibrium is applied). Here, the reduced order models are solved only once
to obtain the Dirichlet data prescribed to the structure sub-domain, while the
Neumann data prescribed to the soil sub-domain is obtained by enforcing
equilibrium at all iterations.
6.3.2.1 Singularity of U T . U T and FB . FB
T T
Consider both the condensed interface stiffness and flexibility matrices of the
structure and soil partitioned sub-domains as approximated by Equations (6.60) and
(6.67), respectively. These approximations are only valid if and only if the square
FB . FB
T
become singular, Equations (6.60) and (6.67) cannot be used to
invertible if and only if all its rows and columns are independent.
217
Chapter 6 Reduced Order Method
In view of the above, a sufficient condition for non-singularity of Τnn is for the
square matrix Τnn to have a full rank (Horn & Johnson, 1999). Since both U T
freedom and K the current coupling iteration number, it is clear that matrices
Given that:
it is evident that:
Rank Α K K U T . U T Rank U T M K
T
(6.69)
Rank Β K K FB . FB Rank FB M K
T
(6.70)
these matrices as given by Equations (6.53) and (6.61), for U T and FB to have
full column rank, it is essential that in all coupling iterations the newly obtained
displacement and force increments at the interface nodes for T and B are linearly
218
Chapter 6 Reduced Order Method
In coupling iterations for which the coupling iteration number is less than the
number of interface freedoms, there is only the possibility of having linearly
dependent incremental displacement/force vectors. However, in problems where the
required number of coupling iterations for achieving convergence exceeds the
number of interface degrees of freedom, any newly obtained incremental
displacement/force vectors is undoubtedly a linear combination of the previous
corresponding independent vectors. As a result, the construction of the condensed
tangent stiffness matrix at the interface would not be possible in this case.
These facts highlight a major shortcoming of the original reduced order method
proposed by Vierendeels et al. (2007). In the following section, a new approach is
proposed for selective addition and replacement of the incremental
displacement/force vectors in UT and FB , thus overcoming the above
shortcoming and offering a robust reduced order model for approximating the
condensed interface tangent stiffness/flexibility matrices.
As mentioned above, in constructing the reduced order method for approximating the
condensed interface tangent stiffness matrix of the partitioned sub-domains, there is
a need for a procedure to make sure that any newly obtained interface
displacement/force modes added to U T and FB matrices are linearly independent.
In this respect, the procedure proposed in this section ensures that any newly
constructed interface incremental displacement vector, hereafter referred to as
displacement mode, is linearly independent of the previous modes:
U
I 1, I
UT
I 1 T
i
T ζ with ζ 0 , , J , , I 2 (6.71)
Consider the reduced order model for partitioned sub-domain T which is treated by
219
Chapter 6 Reduced Order Method
of sequential D-N coupling method, at the starting coupling iteration (I=0), the
relaxation of the interface boundary conditions is applied. At the first coupling
iteration (I=1), U T and FT matrices would be single column matrices as shown
UT U Ti
I 1 0,1
(6.72)
FT FTi
I 1 0,1
(6.73)
where:
U U Ti U Ti
i I ,J I J
T (6.74)
F i I ,J
FTi FTi
I J
T (6.75)
In the next coupling iteration (I=2), there will be a new set of interface displacement
modes:
U T . U T , U T
T
should be a full column rank matrix. In general at any iteration
220
Chapter 6 Reduced Order Method
previously obtained interface displacement modes given by Equation (6.78) leads to:
U U
i I i J
U
I ,J T T
i
(6.81)
U U
T I J
i i
T T
I 1
ˆ
U could be written in the normalised form of:
T
I 1
Uˆ U Ti ,..., U Ti ,..., U Ti
0, I J ,I I 2, I
(6.82)
T
I 1
U ˆ
U
i I 1, I
.ζ (6.83)
T
T
with
T
ζ 0 ,, J ,, I 2 (6.84)
221
Chapter 6 Reduced Order Method
ζ ω .U Ti
I 1, I
(6.85)
where:
1
ˆ I 1 T ˆ
I 1 ˆ I 1 T
ω UT U
T U T
(6.86)
I 1
UTi ˆ
U .ω .U Ti
I 1, I I 1, I
(6.87)
T
I U i
T
I 1, I
, U Ti
I 1, I
1 (6.88)
ˆ . ω . U i I 1, I
I 1
I 1 U Ti , U T
I 1, I
(6.89)
T
I 1
it can be clearly shown that if U Ti ˆ
I 1, I
is a linear combination of U as
T
assumed in Equation (6.83), the value of I 1 should be equal to I .
independent U T :
I
(6.90)
222
Chapter 6 Reduced Order Method
where the corresponding force modes at the interface of sub-domain T take also
(6.91)
On the other hand, when I 1 I is less than the prescribed tolerance, U Ti
I 1, I
I 1
ˆ . In such a case, the previous U T and FT
I 1 I 1
is a linear combination of U T
can still be used for approximating the condensed stiffness matrix at the interface.
However, a better approach for nonlinear analysis would be to replace one of the
full rank U T .
I
ω .U Ti
T I 1, I
ζ 0 ,, J ,, I 2 (6.92)
UT
I
can be constructed which incorporates the latest incremental mode vectors.
U Ti ,..., U Ti ,..., U Ti ,..., U Ti
I 1, I
UT
I 0, I m,I J ,I
(6.93)
Taken out
and the corresponding force mode matrix at the interface of sub-domain T takes
the form:
223
Chapter 6 Reduced Order Method
FT FT ,..., FT ,..., FT ,..., FT
i 0, I i m,I i J ,I i I 1, I
I
(6.94)
Taken out
F F
i I i J
F i I ,J
B B
(6.98)
F F
B
i I i J
B B
I 1
Fˆ B could be written in the normalised form of:
I 1
Fˆ FBi ,..., FBi ,..., FBi
0, I J ,I I 2, I
(6.99)
B
224
Chapter 6 Reduced Order Method
I 1
FBi Fˆ B
I 1, I
.ζ (6.100)
With
T
ζ 0 ,, J ,, I 2 (6.101)
ζ ω .FBi
I 1, I
(6.102)
where:
1
ˆ I 1 T I 1 ˆ I 1 T
ω FB Fˆ FB (6.103)
B
I 1
F i I 1, I
Fˆ B .ω .FBi
I 1, I
(6.104)
B
I F B
i I 1, I
, FBi
I 1, I
1 (6.105)
I 1 I 1, I
I 1 FBi , Fˆ B .ω .FBi
I 1, I
(6.106)
225
Chapter 6 Reduced Order Method
tolerance, , FBi
I 1, I I 1
cannot be expressed as a linear combination of Fˆ B . In
On the other hand, when I 1 I is less than the prescribed tolerance, FBi
I 1, I
I 1
is a linear combination of Fˆ B . In such a case, the previous FB and U B
I 1 I 1
can still be used for approximating the condensed stiffness matrix at the interface.
However, a better approach for nonlinear analysis would be to replace one of the
rank FB .
I
ω .FBi
T I 1, I
ζ 0 ,, J ,, I 2 (6.109)
226
Chapter 6 Reduced Order Method
FBi ,..., FBi ,..., FBi ,..., FBi
I 1, I
FB
I 0, I m,I J ,I
(6.110)
Taken out
and the corresponding the corresponding displacement mode matrix at the interface
of sub-domain B takes the following form:
U Bi ,..., U Bi ,..., U Bi ,..., U Bi
I 1, I
UB
I 0, I m, I J ,I
(6.111)
Taken out
6.3.2.3 Singularity of Approximated Stiffness/Flexibility Matrices
Considering the presented reduced order method, Equations (6.60) and (6.67) are re-
written as:
U
1
FT M K . U T M K U T K M
T T
KTC T K M (6.112)
M M
1 1
U B M K FB K M . FB M K FB K M
T T
K BC (6.113)
M M
In view of the discussion in section 6.3.2.2, it is assumed that for coupling iterations
K<M both U T M K and FB M K have full column rank (with M and K denoting the
number of interface degrees of freedom and the current coupling iteration number,
respectively).
and considering Equations (6.112) and (6.113), it can be easily shown that:
227
Chapter 6 Reduced Order Method
Rank KTC
M M
K (6.115)
1
Rank K BC
M M
K (6.116)
1
As previously mentioned, the condition for non-singularity of KTC and K BC
requires these square matrices to have a full rank (M). Since K<M, it is clear from
1
Equations (6.115) and (6.116) that both KTC and K BC approximated by reduced
order method are singular matrices in nature. This is not a problematic issue in the
view of updating the boundary conditions in successive iterations using
1
Equation (6.37), since there is no need for inverting KTC and K BC . In fact, the
reason for constructing the flexibility condensed interface matrix of the partitioned
sub-domain B directly by the reduced order method in Section 6.3.2 is due to the
6.3.2.4 Example 1: Static FEMFEM Coupling
Here the static FEM-FEM coupled problem of 6.3.1.1 is also analysed using the
presented reduced order scheme with and without the selective addition and
replacement procedure. A comparison between the number of coupling iterations
using adaptive relaxation and reduced order method is also provided in Table 6.4.
These results clearly show that the reduced order method is far superior to the
adaptive relaxation scheme, since it enhances the convergence rate of the coupling
method significantly. This high convergence rate is further demonstrated in Figures
6.12 to 6.17 for models A1 to A6 respectively.
228
Chapter 6 Reduced Order Method
Furthermore, it is shown in Figures 6.12 and 6.13 that without employing the
proposed selective addition/replacement procedure of section 6.3.2.2, if the number
of required coupling iterations for convergence to a prescribed tolerance exceeds the
number of independent modes, the iterative scheme will diverge. This fact highlights
the significance of employing the selective addition/replacement procedure proposed
in Section 6.3.2.2.
A3 2.0 5 7 10
A4 1.0 5 6 6
A5 0.5 5 5 6
A6 0.2 4 4 4
229
Chapter 6 Reduced Order Method
230
Chapter 6 Reduced Order Method
231
Chapter 6 Reduced Order Method
232
Chapter 6 Reduced Order Method
6.3.2.5 Example 2: Dynamic FEMFEM Coupling
In the first set of methods, named as ‘Adaptive’ and ‘Reduced’ in Table 6.5, the
relaxation parameter is initialised at the start of every load/time step to a constant
value, 1.0 , and thereafter the adaptive relaxation or reduced order method is
applied in the normal way.
On the other hand, the second set of coupling methods, named as ‘Adaptive*’ and
‘Reduced*’ in Table 6.5, continue with the parameter or condensed matrices from
the previous step. For the ‘Adaptive*’ approach, the starting relaxation parameter is
considered to be equal to the last computed adaptive relaxation parameter of the
previous time/load step. Similarly in the case of ‘Reduced*’, the last approximation
of the condensed stiffness/flexibility matrices obtained in previous time/load steps is
used instead of a constant relaxation parameter for the update of boundary conditions
in the first iteration.
The dynamic analysis problem is considered for the different problem partitioning
types. This is achieved by analysing the same system with different mass ratios at the
interface of the partitioned sub-domains T and B ( m1 / m2 ). The different
analysed models are presented in Table 6.5 for a 5s duration of response with
t 0.01s using different coupling techniques.
The presented results in Table 6.5 again show that the reduced order method
possesses a higher convergence rate than adaptive relaxation. Moreover, it is shown
that the performance of ‘Adaptive*’/‘Reduced*’ techniques are far better than their
corresponding ‘Adaptive’/‘Reduced’ counterparts (see Figure 6.23). However, it
should be noted that, unlike the problem under consideration, in problems where
233
Chapter 6 Reduced Order Method
The higher convergence rate of the reduced order method compared to adaptive
relaxation is further demonstrated in Figures 6.18, to 6.22, where the error reduction
of the two schemes, Adaptive*’/‘Reduced*, are compared in different arbitrarily
chosen time steps.
In all the analyses, the procedure introduced in Section 6.3.2.2 for selective
addition/replacement of the displacement and force vectors is utilised. As discussed
earlier, if such a procedure is not employed, any new obtained displacement/force
modes undoubtedly will be a linear combination of the previous independent
displacement/force modes, since the required number of coupling iterations exceeds
the number of independent modes. As a result, the approximation for the condensed
tangent stiffness matrix at the interface will be poor, if numerically possible due to
round-off errors, thus leading to a very low convergence rate and even divergence.
This shortcoming of the conventional reduced order method is illustrated in Figure
6.23, where it is shown that without employing the selective addition/replacement
procedure the coupling scheme starts to diverge after the third coupling iteration.
234
Chapter 6 Reduced Order Method
Table 6.5: Number of required coupling iterations for 500 time-steps ( t 0.01s )
with a tolerance of 1e-4 m
235
Chapter 6 Reduced Order Method
236
Chapter 6 Reduced Order Method
237
Chapter 6 Reduced Order Method
238
Chapter 6 Reduced Order Method
6.3.2.6 Example 3: Linear SoilStructure Interaction
239
Chapter 6 Reduced Order Method
On the other hand, with the same problem modelled such that the foundation beam is
only modelled in the partitioned structure sub-domain, the convergence
characteristics become poor, as expected and as shown in Figure 6.26. However, it is
clear that the reduced order method with the selective addition/replacement
outperforms adaptive relaxation by far. Moreover, without employing the selective
addition/replacement procedure, the reduced order coupling scheme starts to diverge
after the 10th coupling iteration (noting that there are 10 degrees of freedom at the
soil-structure interface).
According to Equations (6.60) and (6.67), the reduced order method converges when
a good approximation of the condensed tangent stiffness is achieved. Considering
Figure 6.26, it can be seen that this is achieved through eleven cycles of
displacement/force history data. By using the reduced order method, it is guaranteed
that reasonable approximations of the condensed interface stiffness matrices can be
constructed during coupling iterations, and indeed the more displacement/force
240
Chapter 6 Reduced Order Method
history data (i.e. the more coupling iterations) the better the approximation of the
tangent stiffness matrix.
It is however worth observing that the required number of coupling iterations for
convergence using the reduced order method is still relatively large in comparison
with what would be necessary in a typical monolithic treatment. Furthermore,
considering the convergence behaviour of the reduced order method, it is clear that
as a result of having a poor approximation of condensed interface stiffness matrices
at the first coupling iterations, the compatibility errors are very big. This could be a
problematic issue in highly nonlinear problems where the high compatibility error of
the starting coupling iterations could cause divergence in the nonlinear solution
procedure. Although in problems such as the one presented this could be avoided by
reallocating the interface stiffness between the two sub-domains, in problems where
such partitioning strategy cannot be employed (such as modelling of a retaining wall)
there is a need for a more robust coupling technique with improved convergence
241
Chapter 6 Reduced Order Method
Towards this end, a mixed reduced order method is proposed in the next section. The
approach is based on approximating initial linear condensed stiffness and flexibility
matrices, which are then constantly updated using the iterative displacement and
force modes obtained during successive coupling iterations.
A further enhancement of the reduced order method is proposed here, which utilises
an initial linear condensed stiffness matrix of the partitioned soil and structure sub-
domains at the beginning of soil-structure interaction coupling procedure. During
subsequent coupling iterations, these initial approximated matrices are continuously
updated based on the iterative history of the interface displacement/force increments,
thus achieving improved approximation in nonlinear analysis. The general
formulation for establishing the initial stiffness matrix is similar to the procedure
presented in Section 6.3.1. In the following, these initial stiffness matrices are
obtained via a reduced order method formulation, which gives identical initial
quantities.
Considering the structure partitioned sub-domain and assuming that there are N
degrees of freedom at the interface of the soil-structure interaction problem, the
following N+1 displacement vectors are prescribed at the interface of the structure
sub-domain:
U
T
i
T ui1 ,..., uij ,..., uiN , i 0,..., N (6.117)
i
where:
0 i j or i 0
uij (6.118)
u i j and i 1
242
Chapter 6 Reduced Order Method
F
T
i
T i f i1 ,..., fij ,..., f iN , i 0,..., N (6.119)
the initial condensed stiffness matrix of the structure sub-domain can be formulated
as:
1
KTC FT U T . U T UT
T T
(6.120)
0
where:
such that:
I n
I
n
J
n
KTC U Ti U Ti FTi FTi
n
I J
(6.124)
Taking:
U U Ti UTi
i I ,J I J
T (6.125)
243
Chapter 6 Reduced Order Method
F i I ,J
FTi FTi
I J
T (6.126)
(6.128)
(6.129)
with:
F J ,I
FTi KTC U Ti
J ,I J ,I
T (6.130)
0
the corrective stiffness matrix of Equation (6.123) can now be obtained in the normal
way from the reduced order method according to the following:
F I 1
U
I 1
T KTC i
T (6.131)
U
I 1
i
T UT δ (6.132)
F T
I 1
FT δ (6.133)
With:
T
δ 0 , , J , , I 1 (6.134)
244
Chapter 6 Reduced Order Method
and solving Equation (6.132) for δ through minimisation of the error norm, the
following can be obtained from Equation (6.133):
1
F I 1
UT U Ti
I 1
FT UT UT
T T
T (6.135)
Comparing Equation (6.135) with Equation (6.131), the updated condensed stiffness
matrix of the partitioned structure sub-domain given by Equation (6.123) takes the
following form:
1
KTC KTC FT UT UT UT
T T
(6.136)
I 0
F
T
i
B i fi1 ,..., f ij ,..., f iN , i 0,..., N (6.137)
where:
0 i j or i 0
f ij (6.138)
f i j and i 1
F
T
i
B i fi1 ,..., fij ,..., f iN , i 0,..., N (6.139)
the initial condensed flexibility matrix of the soil partitioned sub-domain can be
formulated as:
1 1
K BC U B FB . FB FB
T T
(6.140)
0
Now the condensed tangent flexibility matrix of the soil sub-domain at iteration I can
be approximated by using a correction to the initial flexibility matrix:
245
Chapter 6 Reduced Order Method
1 1 1
K BC K BC K BC (6.141)
I 0
such that:
K BC
I F F U U
i I
B n
i J
B n
i I
B n
i J
B n (6.142)
Taking:
U i I ,J
U Bi U Bi
I J
B (6.143)
F i I ,J
FBi FBi
I J
B (6.144)
K BC FBi U Bi F
1 I ,J I ,J 1 i I ,J
K BC B (6.145)
0
with:
U J ,I
U Bi K BC FBi
J ,I J ,I
B (6.148)
0
the updated condensed flexibility matrix of the partitioned soil sub-domain can be
obtained as:
246
Chapter 6 Reduced Order Method
1 1 1
K BC K BC U B FB FB FB
T T
(6.149)
I 0
Using the above presented mixed reduced order method coupling technique, the
same problem of Section 6.3.2.6 is considered with the foundation beam completely
modelled in the structure sub-domain, which is more onerous than the case where the
foundation beam is apportioned between the two sub-domains. Figure 6.27,
compares the convergence rate of the mixed reduced order method with the
conventional reduced order method, highlighting its potential superiority in iterative
coupling of FEM-FEM coupled problems.
247
Chapter 6 Reduced Order Method
STEP 2: The structural solver computes the response of the structure for : FTi
I
IF I ≥1 then :
i. Selective addition/replacement of displacement/force vectors
ii. Construct UT and FT
1
KTC KTC FT UT UT UT
T T
iii.
I 0
STEP 3: The corresponding interface forces at the soil domain can be calculated by
applying equilibrium: FTi FBi 0
I I
n n
STEP 4: Based on these forces and the external loading applied to the soil domain,
the soil solver computes the response of the soil domain for: U Bi
I
IF I ≥1 then :
i. Selective addition/replacement of displacement/force vectors
ii. Construct FB and U B
1 1 1
K BC K BC U B FB FB FB
T T
iii.
I 0
STEP 6: If convergence to compatibility has not been achieved, the following new
estimation of the displacements will be applied to the structure sub-domain and the
iteration will continue (I=I+1) until convergence:
U U
1
U U Ti I K BC KTC
i I 1 I 1 i I i I
T n n I I B n T n
248
Chapter 6 Reduced Order Method
The building is designed for office purposes and is assumed to be loaded equally on
each floor. The footing is partly modelled in soil sub-domain and partially in the
structure sub-domain and the interface degrees of freedom are assumed to be at
nodes that belong to both of the partially modelled footings with a total number of
degrees of freedom equal to 30. The soil-structure interaction analysis is carried out
assuming plane strain conditions, where the developed domain decomposition
approach is employed utilising ADAPTIC and ICFEP.
249
Chapter 6 Reduced Order Method
Table 6.7: Geometric and material properties of the partitioned soil-structure system
Table 6.8 presents the required number of coupling iterations for convergence to the
prescribed tolerance for the reduced order method, mixed reduced order method and
adaptive relaxation scheme. Figure 6.29 shows the number of coupling iterations
required in each increment for the different coupling schemes. The convergence rates
of the three approaches in the first, fifth and sixth load increments are also illustrated
in Figures 6.30 to 6.32, respectively. This demonstrates that the mixed reduced order
method achieves a faster convergence rate than other coupling schemes.
46 31 22
Table 6.8: Number of required coupling iterations for different coupling schemes
250
Chapter 6 Reduced Order Method
Figure 6.29: convergence behaviour over full range of response for different schemes
251
Chapter 6 Reduced Order Method
252
Chapter 6 Reduced Order Method
6.5 Conclusion
It is shown that by using the condensed tangent interface stiffness matrices of the
partitioned sub-domains in the update of boundary conditions, superior convergence
characteristics could be achieved. In this respect, convergence to compatibility
occurs for linear problems immediately at the first iteration. Although this immediate
convergence does not normally occur for nonlinear problems, employing this method
for the update of interface Dirichlet boundary conditions ensures a high convergence
rate. This brings the performance of the proposed coupling approach close to the
monolithic treatment.
Although the condensed tangent stiffness may be readily available with some
nonlinear field modelling tools, a more general approximation for the condensed
stiffness matrices is desirable and can indeed be achieved during the course of
coupling iterations, thus avoiding the need for explicit determination and extraction
of the stiffness matrices. In this respect, the condensed tangent stiffness matrix is
approximated via reduced order models, building on a previous approach by
Vierendeels et al. (2007) concerned with fluid–structure interaction problems.
Various significant modifications to this approach are proposed here, leading to a
versatile and efficient approach for coupled modelling of nonlinear soil-structure
interaction problems. In this regard, a major pitfall associated with the original
method causing divergence is overcome by a new selective addition/replacement
procedure of force and displacement mode vectors, where the applicability and
advantages of this modification are demonstrated by means of several examples.
Moreover, different to the method proposed by Vierendeels et al. (2007), the reduced
253
Chapter 6 Reduced Order Method
order method proposed here is solved only once to obtain the prescribed Dirichlet
data to structure sub-domain, while the prescribed Neumann data to the soil sub-
domain is obtained by enforcing equilibrium at all iterations.
254
Chapter 7
Case Studies
7.1 Introduction
The problems under consideration are treated by partitioned analysis, where the
coupling is carried out through coupling of discipline-oriented solvers, ADAPTIC
and ICFEP, for nonlinear structural and geotechnical analysis as outlined in
Chapter 4.
In this example, the in-plane nonlinear behaviour of the unbraced single-bay pitched-
roof steel frame of Figure 7.1, resting on a flexible soil is modelled. The effect of
soil-structure interaction at the foundation level is taken into account using the
partitioned treatment. The effect of the soil-structure interaction on the overall
response is demonstrated by comparing the results obtained by the partitioned
treatment to those of a non-interactive case (i.e. rigid soil base).
The considered steel frame (Figure 7.1) utilises the following material characteristics
and member cross-sections: E= 210 GPa (Young’s modulus), fy=350 MPa (yield
strength), HEB280 columns, and HEB240 rafters. The frame is modelled with
ADAPTIC using elasto-plastic cubic beam-column elements (Izzuddin & Elnashai,
1993), which enable the modelling of geometric and material nonlinearity, while the
material behaviour is assumed to be bilinear elasto-plastic with kinematic strain
hardening of 1%. The soil sub-domain is modelled with ICFEP using an associated
elasto-plastic Mohr-Coulomb constitutive model and discretised using 8-noded
isoparametric quadrilateral elements. Young’s Modulus E varies with depth z
256
Chapter 7 Case Studies
according to (E = 10,000 + 5,000z kN/m2), Poisson’s ratio μ=0.2, the bulk unit
weight γ = 20 kN/m3, the angle of shearing resistance, φ′=20° and cohesion c′=20
kPa.
The interface degrees of freedom are assumed to be at nodes that belong to the
footings with a height of 0.5 m and an effective out of plane depth of 1m. There are
20 degrees of interface freedoms in total. The loading system consists of a constant
horizontal force of H=22 kN and a varying vertical load of magnitude w 0.25 ,
which are modelled in the structure partitioned sub-domain and applied in 34 load
increments.
The results from the partitioned treatment are presented in Figures 7.2 and 7.3, which
depict the nonlinear variation of the vertical settlements under footings A and E of
the steel frame with respect to the vertical load factor. Although the generated level
of the vertical settlements is relatively small, different settlements of the left end and
right end of each foundation generate significant rigid foundation tilting (Figures 7.2,
7.3). This can provide different equilibrium paths and change the force distribution in
the structure compared to non-interactive case. As a result the response of the
interactive case is significantly different compared to the non-interactive case, for
which the rotation of the base is not taken into account. This is demonstrated in
Figures 7.4 and 7.5, showing the horizontal displacement of nodes B and D of the
steel frame with respect to the load factor, for both interactive and non-interactive
cases, respectively. The horizontal and vertical displacements of the roof top of the
frame (node C) is also depicted in Figures 7.6 and 7.7, showing the significant
difference in the response of the interactive case compared to the non-interactive
case. It is worth noting that due to the small indeterminacy degree of the steel frame
and high distance between the foundations, the horizontal and vertical reactions at
the foundations in both interactive and non-interactive case are almost similar.
However, due to the rotation of the foundations the generated moment at the base as
shown in Figures 7.8 and 7.9 varies significantly. The variation of the moment at
node C with respect to the load factor for both interactive and non-interactive case is
also depicted in Figure 7.10.
257
Chapter 7 Case Studies
258
Chapter 7 Case Studies
259
Chapter 7 Case Studies
260
Chapter 7 Case Studies
261
Chapter 7 Case Studies
In Figure 7.11, the deformed shape and the bending moment contours of the
partitioned structure sub-domain in the final load step are shown and compared for
both interactive and non interactive case.
Finally, Figure 7.12 shows contours of stress level at the end of analysis under the
applied load on the structure (load step = 34). The stress level is the ratio, at the same
mean effective stress, of the current deviatoric stress to the deviatoric stress at
failure. It therefore varies from 0 to 1, where 1 indicates full plasticity and failure. It
is evident from Figure 7.12 that the applied loading conditions have mobilised a
plastic zone underneath the footings.
262
Chapter 7 Case Studies
Figure 7.11: Deformed shape (scale=5.0) and bending moment (kN-m) in final load
step for a) non-interactive case b) interactive case
Figure 7.12: Contours of stress level in soil sub-domain (at final increment)
263
Chapter 7 Case Studies
The spread of plasticity under both footings (Figure 7.12) shows the significant rigid
tilting of both footings A and E. This is further evident from the deformed shape of
the structure which shows the footings experiencing rigid tilting. As a consequence,
there is no plastic hinge formation at the base of right column, as depicted in
Figure 7.11, compared to the non-interactive case.
264
Chapter 7 Case Studies
In this section, the response of a multi-storey five-bay steel frame (Figure 7.13)
resting on a flexible soil is modelled using the partitioned analysis. The benefits of
utilising a fully coupled soil-structure analysis via the partitioned approach are
demonstrated, and the results are compared to those from a field elimination
technique where the soil sub-domain is modelled with linear transitional and
rotational springs (Winkler foundation).
265
Chapter 7 Case Studies
freedom are assumed to be at nodes that belong to both the footings and soil
underneath. The total number of interface degrees of freedom is 60 for this case.
Table 7.1: Geometric and material properties of the partitioned soil-structure system
To further assess the merits of the various coupling algorithms presented in this
work, the above problem is analysed using different update techniques for a
tolerance of 10 4 m, where the total number of required coupling iterations (for 10
increments) is listed in Table 7.2. As expected, the constant relaxation scheme has
the worst convergence rate, while the proposed mixed reduced order coupling
scheme has the highest convergence rate.
However, the mixed reduced order scheme requires a greater number of calls to soil
and structural solvers in this case, since it requires the determination of the initial
condensed stiffness matrix prior to the coupling iterations, which involves more
interface-related computations than the reduced order method. Hence, one might
consider the trade-off for different types of problems under consideration, where the
266
Chapter 7 Case Studies
proposed mixed reduced order scheme attains even greater superiority for problems
involving more load steps and/or greater nonlinearity.
The frame structure with its applied loads transmits the loading to the soil, which in
turn deforms due to cumulative action of these loads. As a result, the soil below the
footings goes under vertical and differential settlements. Consequently, the
deformation of soil surface beneath the foundation could cause significant
redistribution of the loads in the frame structure. Moreover, in cases where the
differential settlements are considerable this could cause significant damage to the
infill walls in the frame structure.
On the other hand, it is shown here that by using a fully coupled partitioned
approach, both the structure and soil behaviour could be effectively captured to the
267
Chapter 7 Case Studies
The results of both approaches for the variation of the vertical settlement under
column C2 (where the maximum vertical settlement is observed) with respect to the
load factor are presented in Figure 7.14. Clearly, except for the first increment where
the response of both models is linear, the real settlement of the footing is expected to
be much higher (up to 500%) than predicted by the simple field elimination approach
when the nonlinearity of the soil is taken into account.
The vertical deformation profile of the soil surface, for different load-steps, obtained
from coupled analysis is given in Figure 7.15, where it is clear that the six troughs
correspond to the locations of the footings showing their vertical settlement.
Considering the generated level of vertical settlement and the rigid tilting of the
268
Chapter 7 Case Studies
A vector plot of displacements in the soil sub-domain in the vicinity of the structure
is also shown in Figure 7.16. The absolute magnitudes of these vectors are not
important, though their relative magnitude shows the mechanism of ground
deformation. Contours of stress level in the soil partitioned sub-domain for the final
load-step are also depicted in Figure 7.17, where it is evident that the applied loading
conditions have mobilised an extensive plastic zone underneath the building. This
zone is, however, smaller and shallower under the right hand side footing, which is
in agreement with the previous figures that show most of the deformation and load
concentration nearer the left side footings.
269
Chapter 7 Case Studies
In Figure 7.18, the variation of bending moment at the base of column C1 for
different load levels is shown and compared with that of the field elimination
technique. This shows the significant effect of taking into account the nonlinear
behaviour of soil sub-domain in soil-structure interaction analysis. It can be clearly
observed that the bending moment of the structural elements in the fully coupled
analysis is significantly higher than that of field elimination analysis. This fact is
further demonstrated in Figure 7.19, where the bending moment in the middle of
beam B1 obtained by partitioned approach is compared with that of field elimination
for different load-steps.
270
Chapter 7 Case Studies
Figure 7.18: Bending moment (kN-m) at the base of C1 for different load-steps
Figure 7.19: Bending moment at the middle of beam B1 for different load-steps
271
Chapter 7 Case Studies
The deformed shape and the bending moment contours of the partitioned structure
sub-domain at the final load step are shown in Figure 7.20b. The same quantities are
also obtained and presented in Figure 7.20a using the field elimination approach. It is
evident from the deformed shape of the structure in fully coupled interaction analysis
and from the vectors underneath each of the three footings in Figure 7.16 that these
experience rigid tilting and significant vertical settlements.
Figure 7.20: Deformed shape (scale=5.0) and bending moment (kN-m) in final load
step for a) linear Winkler foundation b) nonlinear partitioned analysis
272
Chapter 7 Case Studies
273
Chapter 7 Case Studies
Finally, the deformed shapes of beam B1 at the final load-step are presented in
Figure 7.23 for both partitioned and field elimination approaches, where
significantly larger deflections are confirmed with the partitioned approach.
Open cuts and excavations in a limited urban space are gradually increasing in
frequency because of the development and upgrade of infrastructures and the
construction of new buildings. At the same time, public concerns have risen over the
effects of excavation-induced ground movements on adjacent structures and utilities.
Excavation inevitably results in deformation of the adjacent ground and settlement of
adjacent buildings behind an excavation wall, causing problems such as loss of
invaluable historic property, third party impact, construction delay, and substantial
increase of project cost (Son et al., 2005; Aye et al., 2006; Boone et al., 1999; Seok
et al., 2001).
274
Chapter 7 Case Studies
The case study under consideration in this section represents a typical urban
situation, where ground excavation can often induce significant movements and
damage to the nearby structures. It is shown that by utilizing a fully coupled soil-
structure interaction model using the partitioned treatment, such nonlinear behaviour
of both structure and soil could be accurately captured. This shows the high potential
of using a fully coupled soil-structure model towards providing reliable assessment
and minimizing the associated damage in such problems. The example considers a
steel frame resting on a soil subjected to ground excavation, where nonlinear elasto-
plastic constitutive behaviour of the soil, as well as geometric and material
nonlinearity of the structure, are taken into account. Figure 7.24 depicts the problem,
where the left hand side boundary is assumed to be consistent with an axis of
symmetry. The plan view of the analysed building frame is also shown in Figure
7.25.
275
Chapter 7 Case Studies
276
Chapter 7 Case Studies
The soil-structure interaction analysis is carried out assuming plane strain conditions
in the soil using an effective out-of-plane width of 1m, where the developed domain
decomposition approach is employed utilising ADAPTIC and ICFEP. The
considered soil-structure system is partitioned physically into two sub-domains, soil
and structure, where each sub-domain is discretised separately according to its
characteristics as shown in Table 7.3.
Table 7.3: Geometric and material properties of the partitioned soil-structure system
The frame structure is modelled with ADAPTIC using cubic elasto-plastic beam-
column elements (Izzuddin & Elnashai, 1993) using 10 elements per member for
both columns and beams, and the material behaviour is assumed to be bilinear elasto-
plastic with kinematic strain hardening. The footings are discretised using 4 elements
per member.
The soil sub-domain and the un-braced excavation are modelled with ICFEP using
an elasto-plastic Mohr-Coulomb constitutive model, with parameters chosen to
represent the behaviour of London clay (Table 7.3). The nonlinear solution
procedure employed for analysing the soil sub-domain is based on a Modified
277
Chapter 7 Case Studies
The above problem is analysed for various scenarios with respect to the loading
applied to the structure (which is assumed to be loaded equally on each floor with a
total gravity load equal to λ×5 kN/m2), the excavation depth (He) and the distance of
the structure from the excavation wall (Le).
Table 7.4 lists various loading scenarios considered for analysing the above problem
with respect to the load factor (λ) applied in structure sub-domain, and the
excavation depth (He) in the soil sub-domain. Considering Table 7.4, the loads on
the structure are applied in the first six increments, and from increment 7 to 16 the
soil is excavated while the loading in the structure is assumed to be constant.
Figures 7.27, 7.28 and 7.29, show the vertical settlement at the centre of the left,
middle and right footings of the analysed frame (see Figure 7.24), respectively, for
the various loading scenarios, where it is assumed Le=2m. The vertical displacement
of the ground surface is also depicted in Figure 7.30.
278
Chapter 7 Case Studies
Figure 7.27: Vertical settlement of the left footing for different load cases
Figure 7.28: Vertical settlement of the middle footing for different load cases
279
Chapter 7 Case Studies
Figure 7.29: Vertical settlement of the right footing for different load cases
Figure 7.30: Cumulative vertical displacement of the ground surface (last increment)
280
Chapter 7 Case Studies
The effect of the structural loads on the horizontal displacement of the excavation
wall is also depicted in Figure 7.31. It is a consequence of these lateral ground
movements that the structure undergoes additional settlements due to its weight.
The effect of the excavation depth on the vertical deformation profile of the soil
surface and the horizontal displacement of the excavation wall is also depicted in
Figures 7.32 and 7.33, respectively, for Case 6 (Table 7.3).
281
Chapter 7 Case Studies
Figure 7.32: Vertical displacement of ground surface for different excavation depths
(Case 6)
Figure 7.33: Horizontal displacement of the excavation wall for different excavation
depths (Case 6)
282
Chapter 7 Case Studies
The effect of distance of the structure from the excavation (Le) on the vertical
settlements of the footings is also depicted in Figures 7.34, 7.35 and 7.36 for model
Case 6.
Figure 7.34: Vertical settlement of the left footing for different Le (Case 6)
283
Chapter 7 Case Studies
Figure 7.35: Vertical settlement of the right footing for different Le (Case 6)
Figure 7.36: Vertical settlement of the right footing for different Le (Case 6)
284
Chapter 7 Case Studies
All the above coupled analyses were carried out using reduced order method with a
tolerance of 10-4 m. However, to further assess the merits of different coupling
algorithms presented in this thesis, Case 6 is analysed using different update
techniques for the same tolerance, where the corresponding total number of required
coupling iterations is listed in Table 7.5. As expected, the constant relaxation
scheme has the worst convergence rate, while the proposed mixed reduced order
coupling scheme outperforms the other coupling schemes.
Figure 7.38 and 7.40 show contours of stress level before and after the full
excavation to 6m depth under the applied load on the structure. It is evident from
Figure 7.40 that the applied loading conditions have mobilised an extensive plastic
zone underneath the building. This zone is, however, smaller and shallower under
the right hand side footing, which is in agreement with the previous figures that
show most of the deformation and load concentration nearer the excavation.
285
Chapter 7 Case Studies
Figure 7.38: Contour plots of stress levels and plasticity induced in soil sub-domain
in increment 6 (Case 6)
286
Chapter 7 Case Studies
Figure 7.40: Contour plots of stress levels and plasticity induced in soil sub-domain
in increment 12 (Case 6)
287
Chapter 7 Case Studies
The deformed shape and bending moment variation in the structure sub-domain for
different increments are shown in Figure 7.41. It can be clearly observed that the
maximum bending moments of the structural elements after the excavation are
significantly higher than before the excavation. The comparison of the bending
moment values for four selected regions A, B, C and D, as shown in Figure 7.41d, is
presented for both the 6th and 12th load increments in Table 7.6. It is evident from the
deformed shape of the structure after excavation and also from the vectors
underneath each of the three footings (Figure 7.39), that after 6m of excavation the
footings experience rigid tilting and significant vertical settlements. However, the
footing nearest to the excavation has the smallest tilting, as its deformation is also
dominated by the horizontal movement towards the unsupported excavation.
Figure 7.41: Deformed shape (scale=5) and bending moment (kN-m) of structure for
(a) 1st, (b) 6rd , (c) 7th and (d) 12th increment
288
Chapter 7 Case Studies
Mz (kN-m) Mz (kN-m) %
Region
Increment 6 Increment 12 Increase
A 120 215 80%
B -120 -191 60%
C -21 -72 340%
D 21 153 720%
With the new Eurocodes coming into force, and their introduction of unified
procedures for limit state design of structural elements, plastic bending of steel sheet
piles will be allowed. According to Bourne-Webb et al. (2007), the limit state design
approach is well established for conventional building structures but has not
generally been applied to the design of earth-retaining structures — at least not in a
unified manner. One area of interest is the use of limit state principles in the design
of steel sheet pile retaining walls and, in particular, whether it is safe to allow the
formation of plastic hinges at the ultimate limit state and, if so, how to verify that the
behaviour of the wall zone undergoing plastic deformation is within acceptable
limits. In this section, the soil-structure interaction problem under consideration is a
2D simulation of a cantilever steel sheet pile retaining wall, as depicted in Figure
7.42, where the left hand side boundary of the problem is assumed to be consistent
with an axis of symmetry.
289
Chapter 7 Case Studies
Increment Number
1 2 3 4 5 6 7 8 9
He (m) He=8 He=8 He=8 He=8 He=8 He=8 He=8 He=8 He=8
W(kPa) W=20 W=40 W=60 W=80 W=100 W=120 W=140 W=160 W=180
290
Chapter 7 Case Studies
In this example, employing the relaxation scheme is very expensive since the
relaxation parameter tends to very small values, in order of 10-3, to ensure
convergence, and as a result both constant and adaptive schemes become extremely
computationally inefficient. Moreover, by employing the standard reduced order
method in this case, appropriate approximation of the condensed stiffness matrix,
and thus convergence, can only be achieved when the coupling iterations exceed the
number of interface degrees of freedom which in this case is 42.
The applicability and benefits of the developed coupling technique in nonlinear soil-
structure interaction analysis can be established by considering the results of this
case study. These results show that the excavation has mobilized a significant plastic
deformation near the retaining wall.
Importantly in the final increment, the soil deformations due to excavation have
caused the retaining wall to reach its plastic moment as illustrated in Figure 7.44,
where the bending moment distribution and the deflected shape of the retaining wall
are depicted for the different incremental stages of the analysis given in Figure 7.44.
291
Chapter 7 Case Studies
Figure 7.44: Deflection (scale: 5) and bending moment (kN-m/m) distribution of the
retaining wall for (a) 1st, (b) 4th, (c) 7th, (d) 8th and (e) 9th increment
292
Chapter 7 Case Studies
Increment number
1 2 3 4 5
He (m) He=8 He=8 He=8 He=8 He=8
W(kPa) W=20 W=40 W=60 W=80 W=100
293
Chapter 7 Case Studies
The obtained results for the proposed wall show that the excavation has mobilized a
significant plastic deformation near the retaining wall. The vector plot of
displacements in the soil sub-domain in the vicinity of the excavation is presented in
Figure 7.46, which shows a completely different mechanism in the partitioned soil
sub-domain compared to the previous case.
Importantly in the final increment, the soil deformations due to excavation cause the
retaining wall to reach its plastic moment as illustrated in Figure 7.47, where the
bending moment distribution and the deflected shape of the retaining wall are
depicted for different incremental stages of the analysis outlined in Table 7.8.
Finally, the change in the horizontal force in the lateral support is also depicted in
Figure 7.48.
294
Chapter 7 Case Studies
Figure 7.47: Deflection (scale: 5) and bending moment (kN-m/m) distribution of the
retaining wall for (a) 1st, (b) 2nd , (c) 3rd , (d) 4th and (e) 5th increment
295
Chapter 7 Case Studies
7.6 Conclusion
296
Chapter 8
Conclusion
8.1 Introduction
This work has been primarily motivated by the lack of sophisticated monolithic tools
for modelling nonlinear soil–structure interaction problems, while recognising the
existence of advanced tools for nonlinear analysis of structure and soil in isolation.
Although coupled modelling of soil-structure interaction problems may be achieved
using a monolithic treatment, the partitioned treatment has been advocated as
offering major benefits in the context of coupled modelling of nonlinear soil-
structure interaction.
Accordingly, the aim of this work has been to develop advanced numerical methods
for nonlinear coupling of static and dynamic soil-structure interaction problems,
where the partitioned approach is adopted as a framework for coupling field-specific
tools with minimal intrusion into codes.
In this respect, the developed approach is believed to offer great potential towards
providing an integrated interdisciplinary computational framework for coupled
modelling of soil-structure interaction problems.
8.2 Conclusions
This work has shown that the partitioned treatment is a feasible and realistic
approach for coupled modelling of nonlinear soil-structure interaction problems.
Unlike the monolithic approach, the partitioned approach offers major benefits,
including i) allowing field-specific discretisation and solution procedures that have
proven performance for each partitioned domain, and ii) facilitating the reuse of
298
Chapter 8 Conclusion
existing nonlinear analysis software with all the resource savings that this brings.
The success of the partitioned approach, however, has been shown to hinge on the
adopted coupling algorithm. Indeed, a principal contribution of this work has been
the investigation of the performance of existing coupling algorithms, and the
development of more powerful algorithms that address convergence issues,
particularly in the context of nonlinear soil-structure interaction problems. Towards
this end, this work has successfully developed novel iterative coupling approaches,
which are based on the reduced order method, yet offering major improvements of
the convergence characteristics and computational performance. Furthermore, the
applicability of the developed partitioned approaches to soil-structure interaction
problems, exhibiting significant nonlinearity in both structure and soil, has been
demonstrated by means of several case studies.
Hereafter, the main conclusions from various parts of this work are summarised.
It has been shown that coupling of partitioned sub-domains may be achieved using
the staggered approach, though this approach should be used with great care in
relation to both stability and accuracy. The main conclusions in this regard are:
299
Chapter 8 Conclusion
Iterative coupling methods have been shown to offer major enhancement over the
staggered approach in the context of partitioned analysis of soil-structure interaction
problems. Considering the applicability requirements for both dynamic and static
analysis, the algorithms that are considered to be more suitable for soil-structure
interaction coupling are:
Sequential/Parallel Dirichlet-Neumann.
Parallel Dirichlet-Dirichlet.
Particular emphasis has been placed in this work on Dirichlet-Neumann (D-N) type
of iterative schemes, specifically the sequential D-N algorithms.
Within the family of sequential D-N algorithms, the convergence performance over a
load/time step has been identified as the most important feature. In this respect, it has
been demonstrated that:
300
Chapter 8 Conclusion
where the structure of the data exchange between the various codes is also
elaborated.
Due to the lack of general convergence analysis for the relaxation update technique
in FEM-FEM coupling, various mathematical and computational characteristics of
the coupling method, including the governing convergence rate and choice of
constant relaxation parameter, have been established. The work undertaken has
shown that:
301
Chapter 8 Conclusion
An adaptive relaxation scheme has been developed for enhancing the performance of
iterative coupling algorithms, where the choice of the relaxation parameter is guided
by the iterative corrections of Dirichlet entities at the interface. It has been shown
that the adaptive relaxation scheme:
Avoids the trial and error procedure for the selection of an optimum, even
adequate, constant relaxation parameter.
Improves the convergence rate of constant relaxation in both linear and
nonlinear analysis significantly.
Finally, although using the adaptive scheme removes significant difficulties in the
conventional relaxation iterative coupling scheme, there is a pitfall associated with
both constant and adaptive relaxation schemes when the partitioned sub-domain
parameters dictate very small values of the relaxation parameter for convergence. In
such cases, relaxation methods break down with poor convergence rates and
significant computational inefficiency.
It has been proposed in this work that the performance of iterative coupling methods
may be effectively enhanced for nonlinear analysis through the use of the condensed
interface stiffness matrices of the structure and soil partitioned sub-domains,
providing an effective first-order guide to iterative displacements at the soil-structure
interface. This would bring the performance of the proposed coupling approach very
close to the monolithic treatment.
302
Chapter 8 Conclusion
Although the condensed tangent stiffness may be readily available with some
nonlinear field modelling tools, more general approximation for the condensed
stiffness matrices is desirable. Towards this end, the condensed tangent stiffness
matrix has been approximated in this work via reduced order models, building on a
previous approach by Vierendeels et al. (2007). Nevertheless, major modifications to
this approach have been proposed in this work, leading to a more versatile and
efficient approach for coupled modelling of nonlinear soil-structure interaction
problems. In this respect:
303
Chapter 8 Conclusion
method is proposed, which has a much higher convergence rate than the
conventional reduced order method.
The applicability of the presented coupling techniques has been demonstrated for
nonlinear soil-structure interaction, where the superior convergence rate of the
presented reduced order schemes, particularly the proposed mixed reduced order
method, compared to relaxation scheme is highlighted.
Although iterative coupling method proposed in this work, utilising the condensed
tangent stiffness matrices of the partitioned sub-domains, possesses a high
convergence rate, its performance may be further enhanced through future research
in the following areas:
Combined coupling/field iterations: Since the field models for structure and
soil are nonlinear, iterations are typically performed in such models to
determine the state of the corresponding sub-domains. Therefore, there are
304
Chapter 8 Conclusion
305
Chapter 8 Conclusion
306
References
Aye, Z. Z., Karki, D. & Schulz, C. 2006. Ground movement prediction and building
damage risk-assesment for the deep excavations and tunneling works in
bangkok subsoil. International Symposium on Underground Excavation and
Tunneleing. Bangkok, Thailand.
Bull, J. W. 1988. Finite element analysis of thin-walled structures, Taylor & Francis.
Collenz, A., Bona, F. D., Gugliotta, A. & Soma, A. 2004. Large deflections of
microbeams under electrostatic loads. Journal of Micromechanics and
Microengineering 14, 365-373.
Dutta, S. C. & Roy, R. 2002. A critical review on idealization and modeling for
interaction among soil–foundation–structure system. Computers &
structures, 80, 1579-1594.
308
References
Dutta, S. C., Bhattacharya, K. & Roy, R. 2004. Response of low-rise buildings under
seismic ground excitation incorporating soil–structure interaction. Soil
Dynamics and Earthquake Engineering, 24, 893-914.
Farhat, C. & Lesoinne, M. 2000. Two efficient staggered algorithms for the serial
and parallel solution of three-dimensional nonlinear transient aeroelastic
problems. Computer Methods in Applied Mechanics and Engineering, 182,
499-515.
Felippa, C. A. & Park, K. C. 2004. Synthesis tools for structural dynamics and
partitioned analysis of coupled systems. Multi-Physics and Multi-Scale
Computer Models in Nonlinear Analysis and Optimal Design of Engineering
Structures under Extreme Conditions. Ljubliana, Slovenia: Proceedings
NATO-ARW PST ARW980268.
309
References
Funaro, D., Quarteroni, A. & Zanolli, P. 1988. An iterative procedure with interface
relaxation for domain decomposition methods. SIAM Journal on Numerical
Analysis, 25, 1213-1236.
Gaba, A. R., Simpson, B., Powrie, W. & Beadman, D. R. 2002. Embedded retaining
walls: Guidance for economic design, research project: 629. London:
Construction Industry Information and Research Association.
Glowinski, R., Dinh, Q. V. & Periaux, J. 1983. Domain decomposition methods for
nonlinear problems in fluid dynamics. Computer Methods in Applied
Mechanics and Engineering, 40, 27-109.
Harr, M. E. 1966. Foundation of theoretical soil mechanics, New York, Mcgraw Hill
book co.
Harr, M. E., Davidson, J. L., Ho, H. O., Pombo, L. E., Ramaswamy, S. V. & Rosner,
J. C. 1969. Euler beams on a twoparameter elastic foundation model. Journal
of Geotechnical and Geoenvironmental Engineering, 95,933-948.
Hoffman, J. D. 2001. Numerical methods for engineers and scientists, CRC Press.
310
References
Inaba, T., Dohi, H., Okuta, K., Sato, T. & Akagi, H. 2000. Nonlinear response of
surface soil and NTT building due to soil-structure interaction during the
1995 Hyogo-ken Nanbu (Kobe) earthquake. Soil Dynamics and Earthquake
Engineering, 20, 289-300.
Jardine, R. J., Potts, D. M., Fourie, A. B. & Burland, J. B. 1986. Studies of the
influence of non-linear stress-strain characteristics in soil-structure
interaction. Geotechnique, 36, 377-396.
Jin, W.-L., Song, J., Gong, S.-F. & Lu, Y. 2005. Evaluation of damage to offshore
platform structures due to collision of large barge. Engineering Structures,
27, 1317-1326.
Kamiya, N. & Iwase, H. 1997. BEM and FEM combination parallel analysis using
conjugate gradient and condensation. Engineering Analysis with Boundary
Elements, 20, 319-326.
Kamiya, N., Iwase, H. & Kita, E. 1996. Parallel implementation of boundary element
method with domain decomposition. Engineering Analysis with Boundary
Elements, 18, 209-216.
Krabbenhoft, K., Damkilde, L. & Krabbenhoft, S. 2005. Ultimate limit state design
of sheet pile walls by finite elements and nonlinear programming. Computers
& structures, 83, 383-393.
Kurian, N. P. & Manjokumar, N. G. 2001. A new continuous winkler model for soil-
structure interaction. Journal of Structural Engineering, 27, 269-276.
311
References
Mu, M. 1999. Solving composite problems with interface relaxation. SIAM Journal
on Scientific Computing, 20, 1394-1416.
Noda, T., Fernado, G. S. K., Asaoka, A. 2000. Delayed failure in soft clay
foundations. Journal of the Japanese Geotechnical Society of Soils and
Foundations, 40, 85-97.
Nogami, T. & Lain, Y. C. 1987. Two parameter layer model for analysis of slab on
elastic foundation. Journal of Engineering Mechanics, ASCE, 1131, 1279-
1291.
312
References
Ravaska, O. 2002. A sheet pile wall design according to Eurocode 7 and Plaxis. In:
MESTAT, P. (ed.) Numerical methods in geotechnical engineering. Paris:
Presses de l’ENPC/LCPC.
Rizos, D. C. & Wang, Z. 2002. Coupled BEM–FEM solutions for direct time domain
soil–structure interaction analysis. Engineering Analysis with Boundary
Elements, 26, 877-888.
Rugonyi, S. & Bathe, K. J. 2001. On finite element analysis of fluid flows fully
coupled with structural interactions. Computer Modelling in Engineering &
Sciences, 2, 195-212.
Seok, J. W., Kim, O. Y., Chung, C. K. & Kim, M. M. 2001. Evaluation of ground
and building settlement near braced excavation sites by model testing.
Canadian Goetechnical Journal, 38, 1127-1133.
Smith, B. S. & Coull, A. 1991. Tall building structures: Analysis and design, Wiley-
Interscience.
313
References
Taylor, D. W. 1964. Fundamentals of soil mechanics, New York, John Wiley and
Sons.
Tian, L. & Li, Z.-X. 2008. Dynamic response analysis of a building structure
subjected to ground shock from a tunnel explosion. International Journal of
Impact Engineering, 35, 1164-1178.
Vallabhan, C. V. G. & Das, Y. C. 1991. Modified vlasov model for beams on elastic
founadtions. Journal of Geotechnical Engineering, 117, 956-966.
Vierendeels, J., Lanoye, L., Degroote, J. & Verdonck, P. 2007. Implicit coupling of
partitioned fluid–structure interaction problems with reduced order models.
Computers & Structures, 85, 970-976.
Viladkar, M. N., Godbole, P. N. & Noorzaei, J. 1994. Modelling of interface for soil-
structure interaction studies. Computers & Structures, 52, 765-779.
314
References
Vlahos, G., Cassidy, M. J. & Byrne, B. W. 2006. The behaviour of spudcan footings
on clay subjected to combined cyclic loading Applied Ocean Research, 28,
209-221.
Wang, C. M., Xiang, Y. & Wang, Q. 2001. Axisymmetric buckling of reddy circular
plates on pasternak foundation. Journal of Engineering Mechanics, ASCE,
127, 254-259.
Winkler, E. 1867. Die lehre von der elasticitaet und festigkeit [the systematic
teaching of elasticity and strength], Prag,Czechoslovakia, H. Domenicus, ed.
Zienkiewicz, O. C., Taylor R. L., Jhu J. Z. 2005. The finite element method: Its basis
and fundamentals, Butterworth-Heinemann.
Zienkiewicz, O. C. & Taylor, R. L. 1991. Finite element method solid and fluid
mechanics: Dynamics and nonlinearity, New York, MCGraw-Hill.
Zienkiewicz, O. C. & Taylor, R. L. 2005. The finite element method for solid and
structural mechanics, Butterworth-Heinemann.
315
Appendix A
316
Appendix B
317
Appendix C
In the following the critical algorithmic step (the update of iterative boundary
conditions at the interface) of various iterative coupling schemes (presented in
Chapter 4) are presented, where the condensed interface tangent stiffness matrices
of partitioned sub-domains are used for the update of boundary conditions in
successive iterations enforcing convergence to compatibility/equilibrium at the
interface.
In the above, vectors U XX and FXX correspond to the displacements and external
forces for the non-interface degrees of freedom, while U Xi and FXi correspond to
displacements and forces for the interface degrees of freedom, respectively.
Assume that the compatibility and equilibrium defaults at the interface of the
structure and soil sub-domains for iteration number I of load/time step n take the
form:
U n U Bi n U Ti n
I I I
(C.3)
F n FBi n FTi n
I I I
(C.4)
Assume that KTC and K BC are the condensed tangent stiffness matrices at the
interface of the structure and soil sub-domains.
STEP 5: If convergence to equilibrium has not been achieved, the new estimation of
the interface forces according to the Equation (C.5) is applied to the structure
domain, and iteration continues (I=I+1) from STEP 2 until convergence to
equilibrium is achieved.
F i I 1
T n
I
n
FTi K BC KTC
1
K
1 1
U
I
n
C
B
1
F n
I
(C.5)
319
Appendix C
STEP 3: If convergence to compatibility and equilibrium has not been achieved, the
new estimation of the interface displacements and forces according to Equations
(C.6) and (C.7) is applied to the structure and soil sub-domains respectively, and
iteration continues (I=I+1) from STEP 2 until convergence to compatibility and
equilibrium is achieved.
U
U Ti KTC K BC K
I 1 1
U n F n
i I C I I
T B (C.6)
n n
F
i I 1
B n
I
n
FBi K BC KTC
1 1 1
K
U
I
n
C
T
1
F n
I
(C.7)
STEP 3: If convergence to compatibility and equilibrium has not been achieved, the
new estimation of the interface displacements and forces according to Equations
(C.8) and (C.9) is applied to the soil and structure sub-domains respectively, and
iteration continues (I=I+1) from STEP 2 until convergence to compatibility and
equilibrium is achieved.
U
U Bi KTC K BC K
1
i I 1
U n F n
I C I I
B n T (C.8)
n
F
i I 1
T n
I
n
FTi K BC KTC
1 1 1
K
U
I
n
C
B
1
F n
I
(C.9)
320
Appendix C
U
U Ti KTC K BC K
I 1 1
U n F n
i I C I I
T B (C.10)
n n
U
U Bi KTC K BC K
1
i I 1
U n F n
I C I I
B n T (C.11)
n
F
i I 1
T n
I
n
FTi K BC KTC
1 1 1
K
U
I
n
C
B
1
F n
I
(C.12)
F
i I 1
B n
I
n
FBi K BC KTC
1 1 1
K
U
I
n
C
T
1
F n
I
(C.13)
321
Appendix D
Numerical Example
In the following, the presented example in Chapter 4 (Figure D.1) is coupled using
various coupling algorithms presented in Appendix C.
K K i Pi
K
K j Pj
To facilitate the verification of the various coupling schemes presented hereafter, the
coupled problem is initially modelled and solved monolithically. Performing global
structural analysis using a monolithic approach results in the formation and solution
of the following global system of equations:
40 10 0 0 0 2 0
10 40 10 20 0 20
3
0 10 25 0 15 4 0 (D.1)
0 20 0 35 15 5 0
0 0 15 15 40 6 60
67
231
2 268
231
3 541
4 (D.2)
231
5 463
6 231
241
77
In order to perform the partitioned analysis using the condensed interface stiffness
approach, the partitioned sub-domains must be modelled in isolation, and their
corresponding condensed interface stiffness matrices must be obtained.
written in the form of Equation (D.3). In the following PXi and Xi correspond
T T
, respectively.
323
Appendix D Numerical Example
0 15 4 T P4 T
i i
15
0
15 15 5i P5i (D.3)
T T
15 15 40 60
6
75 45
15 0 15 1 8
KTC 15 15 8 (D.4)
0 15 15 40 45 75
8 8
written as:
40 10 0 0 2 0
10 40 10 20 3 20
i i
0 4 B P4 B
(D.5)
0 10 10
0 20 0 20 5i P5i
B B
In the above PXi and Xi correspond to the forces and displacement of the
B B
22 16
1
10 0 0 10 40 10 0 0 3 3
K BC (D.6)
0 20 0 20 10 40 10 20 16 28
3 3
324
Appendix D Numerical Example
Prescribing an initial guess for the interface Dirichlet data at the interface of T ,
i I 0
0 and 5i
I 0
4 T 0 , for the first iteration (I=0):
T
3
i I 0 0 P i I 0
15 0 15 4 T 4 T 6 2
i I 0 0 P i I 0 P i I 0 45
5 T 4 T
0 5 T
15 15 (D.7)
2
15 15 40 6 60 P5i 45
I 0
T
2
Prescribing an initial guess for the interface Neumann data at the interface of B (
P
i I 0
0 and P5i
I 0
4 B 0 ), for the first iteration (I=0):
B
0 2
40 10 0 0 2 2 / 3
10 40 10 20 3 20 3 8 / 3
i I 0 i I 0
i I 0
0 4 B 4 B
(D.8)
0 10 10 P 0
4 B 8 / 3
20 i I 0 P i I 0 0 i I 0 8 / 3
5 B 5 B 5 B
0 20 0
Comparing P4i and P5i with P4i and P5i , and comparing 4i
I 0 I 0 I 0 I 0 I 0
and
B B T T B
i I 0
with 4i and 5i
I 0 I 0
5 B , it is clear that convergence to either equilibrium or
T T
325
Appendix D Numerical Example
i I 1
4 T 0
I 1
5i 0
T
1 (D.9)
75 45 22 16 22 16 8 45 541
8
8 3 3 3
3 3 2 231
.
45 75 16 28 16 28 8 45 463
8 8 3 3 3 3 3 2 231
P i I 1
4 B 0
I 1
P5i 0
B
1
75 45
1
22 16
1
8 75 45 45 130 (D.10)
1
3
8 8 3 3 8 8 2 11
.
45 75 16 28 8 45 75 45 1300
8 8 3 3 3 8 8 2 77
Prescribing the new estimate for the interface Dirichlet data at the interface of T at
i I 1 541 241
4 T 231 P4i 77
I 1
15 0 15 T 6
0 i i I 1 i I 1 130
15 15 5 P5 T P4 T
I 1 463
(D.11)
T 231 11
15 15 40
60 P5i 1300
6
I 1
T
77
Prescribing the new estimate for the interface Neumann data at the interface of B
326
Appendix D Numerical Example
67
0
2 2 231
40 10 0 0 20 3
268
10 40 10 20 3
i I 1 P i I 1 130 i I 1 231
0 4 B 4 B
11 4 B 541
(D.12)
0 10 10
20 i I 1 i I 1 1300 i I 1 231
5 B P5 B 77 5 B 463
0 20 0
231
It is clear that at the first coupling iteration (I=1) convergence to equilibrium and
compatibility is achieved. Moreover, the obtained results by the above coupling
procedure are identical to those obtained by the monolithic treatment.
Prescribing an initial guess for the interface Dirichlet data at the interface of T (
i I 0
0 and 5i
I 0
4 T 0 ), for the first iteration (I=0) gives:
T
3
i I 0 0 P i I 0
15 0 15 4 T 4 T 6 2
0 i I 0 0 P i I 0 P i I 0 45
15 15 5 T 5 T 4 T (4.13)
15 15 40 2
6 60 P5i 45
I 0
T
2
Prescribing an initial guess for the interface Dirichlet data at the interface of B (
i I 0
0 and 5i
I 0
4 B 0 ), for the first iteration (I=0):
B
327
Appendix D Numerical Example
2
2 0 2 15
40 10 0 0 20 8
10 40 10 20 3 3 15
i I 0 I 0
i I 0
0 4 B 0 P4 B P4 B 16
i (D.14)
0 10 10
20 i I 0 0 P i I 0 P i I 0 3
5 B 5 B 5 B 32
0 20 0
3
1
22 16 75 45 103 541
i I 1
4 T 0 3 3 8 8 6 231
I 1
(D.15)
5i 0 16 28 45
75 71 463
T
3 3 8 8 6 231
Prescribing the new estimate for the interface Dirichlet data at the interface of T at
i I 0 541 241
4 T 231 P4i 77
I 0
15 0 15 T 6
0 i I 0 463 i I 0 i I 0 130
15 15 5 P5 T P4 T (4.16)
T 231 11
15 15 40
60 P5i 1300
I 0
6
T
77
Prescribing the new estimate for the interface Dirichlet data, by applying
compatibility, at the interface of B at iteration I=1 and solving the partitioned sub-
domain B gives:
328
Appendix D Numerical Example
67
2
0 2 231
40 10 0 0 3
10 40 10 20 20 3 268
i I 0 541 i I 0 i I 0 231
0 10 10 0 4 B 231 4 B 4 B 130
P P
(4.17)
463 P i I 0 P i I 0 11
5 B 5 B 1300
0 20 0
5 B
20 i I 0
231
77
Prescribing an initial guess for the interface Neumann data at the interface of T (
P
i I 0
0 and P5i
I 0
4 T 0 ), for the first iteration (I=0) gives:
T
i I 0 P i I 0 0 i I 0
15 0 15 4 T 4 T 4 T 6
0 i i i I 0
5 T 6
I 0 I 0
15 15 5 P5 0 (D.18)
T T
15 15 40 6
6 60 6
Prescribing an initial guess for the interface Neumann data at the interface of B (
P
i I 0
0 and P5i
I 0
4 B 0 ), for the first iteration (I=0) gives:
B
329
Appendix D Numerical Example
2
2 0 2 3
40 10 0 0 8
10 40 10 20 3 20 3
i I 0 i I 0
i I 0
3
0 4 B P4 B 0 4 B 8
(D.19)
0 10 10
20 i I 0 P i I 0 0 i I 0 3
5 B 5 B 5 B 8
0 20 0
3
1 1
22 16 75
1
45 26 130
P
i I 1
0 3 13 11
4 T 3 8 8
I 1
P5i 0 16 28 45 75 26 1300
T
3 3 8 8 13 77
Prescribing the new estimate for the interface Neumann data at the interface of T
i I 1 130 541
i I 0 P4 T i I 1
15 0 15 4 T 11
4 T
231
0 I 0 1300 i I 1 463
15 15 5i P5i 5 T
I 1
(D.20)
T
T 77 231
15 15 40
6 60 6 241
77
Prescribing the new estimate for the interface Neumann data at the interface of B
330
Appendix D Numerical Example
67
0
2 2 231
40 10 0 0 20 3
268
10 40 10 20 3
i I 1 P i I 1 130 i I 1 231
0 4 B 4 B
11 4 B 541
(D.21)
0 10 10
20 i I 1 i I 1 1300 i I 1 231
5 B P5 B 77 5 B 463
0 20 0
231
331