Parte C delle lezioni del
Corso di Dottorato sull'OTTIMIZZAZIONE STRUTTURALE
Prof. Ing. Franco Bontempi
Aprile - Maggio 2015,
Facolta' di Ingegneria Civile e Industriale
Universita' degli Studi di Roma La Sapienza
14. 14
STRUCTURAL
MODELING
CODE
Global Frame Models Local Models
Frame
Work
Substruct-
ured Models
STRUCTURAL
MODELING
CODE
Global Frame Models Local Models
Frame
Work
Substruct-
ured Models
structural configurations
specificity of the modeling
commercial
codes
16. Design of Experiments (DOE)
• In general usage, design of experiments (DOE) or
experimental design is the design of any information-
gathering exercises where variation is present, whether under
the full control of the experimenter or not. However, in
statistics, these terms are usually used for controlled
experiments.
• Formal planned experimentation is often used in evaluating
physical objects, chemical formulations, structures,
components, and materials. Other types of study, and their
design, are discussed in the articles on computer
experiments, opinion polls and statistical surveys (which are
types of observational study), natural experiments and quasi-
experiments (for example, quasi-experimental design).
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
16
21. The nature of optimum (1)
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
21
22. The nature of optimum (2)
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
22
A sub-optimal solution
to a problem is one
that is less than perfect.
Slack situation: loose and not pulled tight.
26. Bounded Rationality
Bounded rationality is the idea that in decision-making, rationality
of individuals is limited by the information they have, the
cognitive limitations of their minds, and the finite amount of time
they have to make a decision. It was proposed by Herbert A.
Simon as an alternative basis for the mathematical modeling of
decision making, as used in economics, political science and
related disciplines; it complements rationality as optimization,
which views decision-making as a fully rational process of finding
an optimal choice given the information available. Another way to
look at bounded rationality is that, because decision-makers lack
the ability and resources to arrive at the optimal solution, they
instead apply their rationality only after having greatly simplified
the choices available. Thus the decision-maker is a satisfier, one
seeking a satisfactory solution rather than the optimal one.
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
26
27. Model Extensions
• Ariel Rubinstein proposed to model bounded rationality by
explicitly specifying decision-making procedures..
• Gerd Gigerenzer opines that decision theorists have not really
adhered to Simon's original ideas and proposes and shows
that simple heuristics often lead to better decisions than
theoretically optimal procedures.
• Huw Dixon later argues that it may not be necessary to
analyze in detail the process of reasoning underlying bounded
rationality. If we believe that agents will choose an action that
gets them "close" to the optimum, then we can use the notion
of epsilon-optimization, that means you choose your actions
so that the payoff is within epsilon of the optimum. The notion
of strict rationality is then a special case (ε=0).
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
27
28.
29. εὑρίσκω
• Heuristic (/hjʉˈrɪstɨk/; Greek:
"Εὑρίσκω", "find" or "discover") refers
to experience-based techniques for
problem solving, learning, and
discovery that give a solution which is
not guaranteed to be optimal. Where
the exhaustive search is impractical,
heuristic methods are used to speed
up the process of finding a satisfactory
solution via mental shortcuts to ease
the cognitive load of making a
decision. Examples of this method
include using a rule of thumb, an
educated guess, an intuitive judgment,
stereotyping, or common sense.
• In more precise terms, heuristics are
strategies using readily accessible,
though loosely applicable, information
to control problem solving in human
beings and machines.
• L'euristica (dalla lingua greca εὑρίσκω,
letteralmente "scopro" o "trovo") è una
parte dell'epistemologia e del metodo
scientifico.
• Si definisce procedimento euristico, un
metodo di approccio alla soluzione dei
problemi che non segue un chiaro
percorso, ma che si affida all'intuito e
allo stato temporaneo delle
circostanze, al fine di generare nuova
conoscenza. È opposto al
procedimento algoritmico. In
particolare, l'euristica di una teoria
dovrebbe indicare le strade e le
possibilità da approfondire nel
tentativo di rendere una teoria
progressiva.
31. Simulated Annealing (Metropolis)
• Simulated annealing (SA) is a generic probabilistic heuristic for the
global optimization problem of locating a good approximation to the
global optimum of a given function in a large search space.
• The name and inspiration come from annealing in metallurgy, a
technique involving heating and controlled cooling of a material to
increase the size of its crystals and reduce their defects.
• This notion of slow cooling is implemented in the Simulated
Annealing algorithm as a slow decrease in the probability of
accepting worse solutions as it explores the solution space.
Accepting worse solutions is a fundamental property of heuristics
because it allows for a more extensive search for the optimum.
• The method is an adaptation of the Metropolis-Hastings algorithm, a
Monte Carlo method to generate sample states of a thermodynamic
system, invented by M.N. Rosenbluth and published in a paper by
N. Metropolis et al. in 1953.
40. Nelder-Mead Method (Amoeba)
• The Nelder–Mead method or downhill simplex
method or amoeba method is a commonly used
nonlinear optimization technique, which is a
well-defined numerical method for problems for
which derivatives may not be known.
• The Nelder–Mead technique is a heuristic
search method that was proposed by John
Nelder & Roger Mead (1965) for minimizing an
objective function in a many-dimensional space.
48. Genetic Algorithm (GA)
• The original motivation for the GA approach was a biological
analogy. In the selective breeding of plants or animals, for example,
offspring are sought that have certain desirable characteristics,
characteristics that are determined at the genetic level by the way
the parents’ chromosomes combine. In the case of GAs, a
population of strings is used, i.e. chromosomes.
• The recombination of strings is carried out using analogies of
genetic crossover and mutation, and the search is guided by the
results of evaluating the objective function f for each string in the
population.
• Based on this evaluation, strings that have higher fitness (i.e.,
represent better solutions) can be identified, and these are given
more opportunity to breed.
52. Coding
• One of the distinctive features of the GA approach is to
allow the separation of the representation of the problem
from the actual variables in which it was originally
formulated. In line with biological usage of the terms, it
has become customary to distinguish the ‘genotype’—
the encoded representation of the variables, from the
‘phenotype’—the set of variables themselves.
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
52
53. Genotype space = {0,1}L
(mappa)
Phenotype space
(territorio)
Encoding
(representation)
Decoding
(inverse representation)
01101001
01001001
10010010
10010001
Translation
63. Tensile crack phenomena in HCS
(splitting, bursting, spalling).
• splitting cracks: caused by stresses resulting from
the development of prestressing in the anchorage
zone, that may generate traction stresses in the
concrete.
• bursting cracks: a local effect, generated by the
strand slippage into the slab end while the former
widens slightly on being cut.
• spalling cracks: occurring above the axis of the
strands in the HCS end zone, caused also by the
development of prestressing in the concrete at the
slab ends where only the lower part holding the
strands begins to be prestressed.Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
63
66. Cross-section of the reference HCS
and numerical model
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
66
67. Tensile deformations in the vertical
directions for the spalling effect
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
67
68. The binary coding of the geometry
characteristics of the section
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
68
69. • The fitness function F includes terms to represent
the weight of the slab.
• First, functions gi(x), represents the geometry
constraints, implicitly satisfied during the definition of
the variable space.
• Functions hi(x) represent the structural safety
constraints. In this study, two checks are carried out:
1. the first one on the bending stress, carried out after the
initial structural analyses on the meso-scale model.
2. the second one, on the spalling stress, carried out on the
micro-scale model.
• If both checks are positive, the individual is fitting the
constrain conditions, otherwise, it is discarded and a
different element is introduced in the population.
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
69
129. Uses of genetic algorithm
• To perform the stochastic exploration of
the load space;
• To handle the uncertainties related to the
definition of the loads;
• To investigate the global behavior of the
structure by means of the definition of the
envelope diagram of the performances;
• To define the worst load combination;
• To scrutinize the exact value of a specific
performance.
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
129
134. Performance in relation to the
return periods of the actions.
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
134
135. S N
Geometry of the long-span
suspension bridge considered.
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
135
136. The design variables and
the performance levels
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
136
137. A genetic algorithm approach for
performance assessment
• The performance of a long-span suspension bridge is
investigated by means of a GA approach.
• Focus is given to three aspects of the structural behavior of
the bridge:
1) maximum vertical displacement;
2) maximum longitudinal and transversal slope;
3) maximum tension in main cable and in the tower legs.
• The load scenarios that lead to the most severe performance
metrics are explored in the space of the load variables by an
optimization process based on GA’s.
• The implementation of a GA based optimization is essential
since the traditional optimization techniques are rather
ineffective, due to the high number of dimensions of the load
variables space and the presence of numerous local optimum
points.
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
137
138. Loading systems considered in the
genetic algorithm analysis.
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
138
139. Remarks on loading system
• Traffic and train loads are directed vertically but the
possibility to have a longitudinal component due to the
acceleration (A) or the deceleration (D) is also taken into
account.
• In addition, a torque is present if the traffic loads are not
positioned on the axis of the respective box girder
section.
• The wind action, assumed always present and flowing
transversally to the longitudinal axis of the bridge,
produces lift, drag and torque.
• In order to represent analytically the entire loading
system, 16 variables are needed.
• Since each of the girders is formed by 123 finite
elements, the position of the loads will be defined by
integer variables, ranging, in general, from 0 to 123.
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
139
140. Variables considered for the
definition of the loading system.
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
140
141. Description of loading system
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
141
142. Binary coding
• The position variables are implemented in binary
code with a dimension of eight bits (the minimum
dimension able to represent the position of the
loads on the bridge deck):
• In this row vector, x1 defines the position of the
train on the bridge deck, in binary code: for
example, if the train load starts from the fifteenth
element on the deck, the variable x1 is:
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
142
143. Population
• The GA starts by considering an initial population of N
row vector x created assigning random values to the
unknown variables; each row of the matrix X represents
the chromosome of one solution:
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
143
144. Target functions
• In order to evaluate the performance of the bridge, the
following six target functions are considered:
1. the vertical displacement (negative) for the bridge deck;
2. the horizontal displacement (positive) for the bridge
deck;
3. the longitudinal slope for the bridge deck;
4. the transversal slope for the bridge deck;
5. the axial tension for the main cables;
6. the stress state induced by the axial action and the two
bending moments for the bridge tower legs.
• Each performance is measured by the peak value over
all nodes of the bridge deck.
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
144
145. Evolution of population
• For each target function, the genetic algorithm
creates new populations of N row vector x in order
to find the worse configurations of the considered
loads.
• The genetic algorithm works by evaluating the target
function in correspondence with each assumed
vector x.
• If the population contains a N number of x vectors,
the best N/2 vectors are saved in a new population
while the other vectors are erased.
• To complete the new population, additional N/2
vectors are formed from the saved vectors using the
genetic operator of mutation and crossover.
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
145
146. Mutation
• The mutation on the generic vector i of the
population n changes a single bit of a
randomly selected chromosome; for example
provides the change from 1 to 0:
• As a result a new vector k is obtained for the
population n+1.
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
146
147. Crossover
• The crossover on the generic vectors i and j of
population n is provided in this example:
• where a group of cells of chromosomes i and j is
selected and the respective states changed.
• As a result there are two new vectors k and l for the
population n+1. Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
147
148. Remarks
• When N/2 new vectors are created, the genetic algorithm
restarts with the evaluation of the target function for each
vector xn+1.
• It should be observed that a genetic algorithm is a
stochastic evolutionary procedure because the operators
of mutation and crossover are not deterministic but there
is a probability of occurrence for each operator.
• Usually the probability of occurrence of the mutation
operator is low (0 – 5%) while the probability of
occurrence of the crossover operator is high (70 – 90%).
• What makes this procedure attractive is the fact that
usually there is a large interdependence between the
quality of results and of the choice of these parameters.
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
148
150. • The FE model consists of 1614 elements (beams, no
compression cable elements and gaps) and 1140 nodes.
• For each of the six previously chosen performance
metrics (target functions), GA analysis is performed with
an initial randomly chosen population of 100
chromosomes. For each chromosome the structural
analysis, accounting for geometrical and material
nonlinearities, is developed using ADINA, starting each
time from the reference configurations under permanent
loads and adding the traffic and wind loads.
• The custom software reads the output evaluation and
performs the genetic recombination of the chromosomes
to get a new generation of chromosomes: 100 cycles of
regeneration are considered for a total of 10000 different
load scenarios, each of them leading to a nonlinear
structural analysis.
• The probability of occurrence of the crossover operator
is of 80% while the probability of occurrence of the
mutation operator is of 2%.
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
150
152. Remarks
• It is clear that the convergence of the variables that
define the train position (A) is better than the one that
defines the position of the light traffic load (B).
• From a design point of view, it means that the influence
of the railway load in defining the vertical displacement is
much higher than the traffic load.
• In addition, it can be observed that the railway loads
converge towards two different edges (North and South).
This is due to the fact that the geometry of the bridge is
almost symmetrical.
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
152
163. Knowledge Growth Process
KNOWLEDGE
REQUIRED
BY AN EVOLUTIVE
DESIGN
NEW KNOWLEDGE
REQUIRED BY
AN INNOVATIVE
DESIGN
ACTUAL
KNOWLEDGE BASIS
163Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
179. Failure due to unexpected facts
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
179
180. Causes of system failure
100%
Time
%offailure
Unknown phenomena
Known phenomena
Research
level
Design code
level
past present future
A
BB B
C
Humanerrors
Ottimizzazione Strutturale
franco.bontempi@uniroma1.it
180
193. Dalian, June 2008 193
0
500
1000
1500
2000
2500
3000
3500
SPAN 1100 1298 1385 1410 1624 1991 3300
BISA
N-
VER
RAZZ
JIAN
GYN
HUM
BER
GRE
AT
AKA
SHI
MES
SINA