Documente Academic
Documente Profesional
Documente Cultură
by
Norbert S. Que
February, 2003
i
ABSTRACT
Actual experimental data are used to validate and determine the most suitable
algorithm for parameter identification. It was found that the smoothing-based method
is by far superior than other schemes. As the problem is nonconvex and the nonlinear
program can only guarantee a local or stationary point, global optimisation procedures
are introduced in order to verify the accuracy of the solutions obtained by the
algorithm.
Two evolutionary search methods capable of finding the global optimum are
implemented for parameter identification. The results generated by the evolutionary
search techniques confirm the reliability of the solutions identified by the best
nonlinear programming algorithm. All computations carried out in the thesis suggest
the suitability and robustness of the selected algorithm for parameter identification.
ii
ACKNOWLEDGEMENTS
I wish to thank the Department of Science and Technology for offering the
scholarship which allowed me take up doctoral studies at the University of New South
Wales, Australia. I also appreciate the support given to me by the University of the
Philippines. I would also like to acknowledge the assistance extended to me by Mr.
Francois Fernandes.
I wish to thank my friends in the Mascot care group for their prayers. I would
also like to extend my gratitude to all my friends at the Punchbowl Baptist Memorial
Church and the Community Bible Church (formerly known as St. Matthias@UNSW).
I would also like to take this opportunity to thank my brothers and sisters, most
especially my mother and my in-laws in Cebu, for their encouragement.
Above all, I would like to dedicate this thesis to Jesus Christ for sustaining my
family and me all these years. To Him be the glory.
iii
TABLE OF CONTENTS
ABSTRACT i
ACKNOWLEDGMENTS ii
NOTATION x
LIST OF TABLES xx
1 INTRODUCTION 1
1.1 Background 1
2.1 Introduction 11
2.6 Summary 45
iv
3 STATE PROBLEM 47
3.1 Introduction 47
5.5 Time Reduction Techniques for the Parameter Identification Problem 192
- DE/rand/1 236
- DE/best/1 236
- DE/rand-to-best/1 236
- DE/rand/2 237
- DE/best/2 237
REFERENCES 282
x
NOTATION
ft tensile strength
Lp least-power norm
N size of population
p point load
tc tensile strength
ta , tb breakpoint strength
w crack width
w vector of displacement discontinuities at the crack interface
σu modulus of rupture
υ Poisson’s ratio
ω objective function, error norm or identified error
ω̂ real error
Ω boundary domain
xii
LIST OF FIGURES
Figure 2.1 Relative sizes of the fracture process zone for (a) brittle, (b)
ductile, (c) quasibrittle materials. 14
Figure 2.2 A generic nonlinear softening curve. 16
Figure 2.3 Piecewise linear stress-strain curve for the crack band
model. 17
Figure 2.4 Definition of the two-parameter fracture model. 18
Figure 2.5 Size effect law as defined in Equation (2.5). 20
Figure 2.6 Dugdale’s plastic zone model. 21
Figure 2.7 Stress-deformation behaviour of a quasibrittle specimen in
tension. 22
Figure 2.8 Definition of the cohesive crack model. 24
Figure 2.9 A schematic illustration of a uniaxial tensile test. 28
Figure 2.10 A schematic illustration of a tensile splitting test. 31
Figure 2.11 Illustration of a three-point bend test. 32
Figure 2.12 Work of fracture as an area in the load-displacement (p-δ)
curve. 33
Figure 2.13 Common wedge splitting shapes (a) cubical – mould
fabricated, (b) and (c) cylindrical – cored from existing
structures (Brühwiler and Wittmann, 1990). 34
Figure 2.14 Schematic illustration of a wedge splitting test (Brühwiler
and Wittmann, 1990). 35
Figure 4.1 Diagram of the feasible solution space for Example 4.1. 118
Figure 4.2 Plot of the value of the objective function along the x-axis 119
Figure 4.3 Diagram of the feasible solution space for Example 4.2. 120
Figure 4.4 Smoothing of complementarity conditions. 126
Figure 4.5 Three-point bend model used to generate the pseudo data. 131
Figure 4.6 Perfect pseudo data points generated from a three-point
bend model. 132
Figure 4.7 Wedge splitting model used to generate the pseudo data. 133
Figure 4.8 Perfect pseudo data points generated from a wedge splitting
model. 133
Figure 4.9 Data 1: comparison of predicted with actual p – u
responses. 136
Figure 4.10 Data 2: comparison of predicted with actual p – u
responses. 138
Figure 4.11 Data 2: details of p – u responses at the peak load. 138
Figure 4.12 Perturbed data set. 143
Figure 4.13 Data set with outlier. 144
Figure 4.14 Perfect data: comparison of predicted with actual p – u
response. 145
Figure 4.15 Perturbed data: comparison of predicted with actual p – u
response. 146
Figure 4.16 Perturbed data: details of p – u responses at the peak load. 147
Figure 4.17 Identified p – u response for a data with outlier using L1 –
L2 norm. 148
Figure 4.18 Data with outlier: comparison of predicted with actual p – u
response. 149
Figure 4.19 Data with outlier: details of p – u responses at the peak
load. 149
Figure 5.1 Milan three-point bend test showing grid points where
horizontal displacements are measured. 156
Figure 5.2 Dimensions for the structural model of Milan three-point
bend test. 157
Figure 5.4 Details of the LMC/EPFL wedge splitting test. 158
xv
Figure 5.3 Plot of the 48 recorded data points and the chosen subset of
32 data points for the Milan three-point bend test. 157
Figure 5.5 Plot of the 128 recorded data points of the LMC/EPFL
wedge splitting test. 159
Figure 5.6 Plot of the 128 recorded data points and the chosen subset
of 32 data points for the LMC/EPFL wedge splitting test. 160
Figure 5.7 Dimensions for the structural model of Danish three-point
bend test. 161
Figure 5.8 Data points for the three-point bend test Danish normal
strength concrete. 162
Figure 5.9 Data points for the three-point bend test Danish high
strength concrete. 162
Figure 5.10 Milan data: comparison of predicted with actual p – u
responses. 165
Figure 5.11 Milan data: details of p – u responses at the peak load. 166
Figure 5.12 Milan data: two-branch softening laws determined by the
NLP-based algorithms. 166
Figure 5.13 LMC/EPFL data: comparison of predicted with actual p – u
responses. 169
Figure 5.14 LMC/EPFL data: details of p – u responses at the peak
load. 169
Figure 5.15 LMC/EPFL data: two-branch softening laws determined by
the NLP-based algorithms. 170
Figure 5.16 LMC/EPFL data: comparison of p – u curves predicted by
the smoothing and penalty algorithms. 171
Figure 5.17 LMC/EPFL data: two-branch softening laws determined by
smoothing and penalty algorithms. 171
Figure 5.18 Danish 1 data: comparison of predicted with actual p – u
responses. 173
Figure 5.19 Danish 1 data: two-branch softening laws determined by
the NLP-based algorithms. 174
Figure 5.20 Danish 1 data: details of p – u responses at the peak load. 174
Figure 5.21 Danish 2 data: comparison of predicted with actual p – u
responses. 176
Figure 5.22 Danish 2 data: details of p – u responses at the peak load. 177
xvi
Figure 6.1 Pseudo data: GA iteration history of the power law. 232
Figure 6.2 Pseudo data: fitness contour of the power law. 233
Figure 6.3 Pseudo data: fitness landscape of the power law. 233
Figure 6.4 Pseudo data: GA iteration history of the two-branch law. 234
Figure 6.5 DE procedure in generating a trial vector. 240
Figure 6.6 Pseudo data: iteration histories for DE/best/2/bin and
DE/best/2/exp strategies using a two-branch law. 243
Figure 6.7 Pseudo data: DE iteration history of the power law. 244
Figure 6.8 Milan data: GA and DE iteration histories of the power law. 246
Figure 6.9 Milan data: fitness contour of the power law. 247
Figure 6.10 Milan data: fitness landscape of the power law. 248
Figure 6.11 Milan data: GA and DE iteration histories of the two-
branch law. 249
Figure 6.12 Milan data: identified p-u responses of smoothing, GA and
DE methods for a three-branch law. 250
Figure 6.13 Milan data: identified three-branch law for smoothing, GA
and DE methods. 251
Figure 6.14 Milan data: GA and DE iteration histories of the three-
branch law. 252
Figure 6.15 LMC/EPFL data: GA and DE iteration histories of the
power law. 253
Figure 6.16 LMC/EPFL data: fitness contour of the power law. 254
Figure 6.17 LMC/EPFL data: fitness landscape of the power law. 254
Figure 6.18 LMC/EPFL data: fitness contour of the power-exponential
law. 255
xviii
Figure 6.38 Danish 2 data: fitness landscape of the power law. 272
Figure 6.39 Danish 2 data: GA and DE iteration histories of the two-
branch law. 273
Figure 6.40 Danish 2 data: identified p-u responses of smoothing, GA
and DE methods for a three-branch law. 274
Figure 6.41 Danish 2 data: identified three-branch law for smoothing,
GA and DE methods. 275
xx
LIST OF TABLES