Professional Documents
Culture Documents
Università Degli Studi Di Roma "La Sapienza"
Università Degli Studi Di Roma "La Sapienza"
Facoltà di Ingegneria
Dipartimento di Ingegneria Strutturale e Geotecnica
Dottorato di ricerca – XXI Ciclo
Structural integrity:
robustness assessment and
progressive collapse susceptibility
A.A. 2007/2008
Alle scale della facoltà,
che ho fatto sempre di corsa,
tranne una volta.
Acknowledgements
I would like to express my deepest gratitude to Prof. Franco Bontempi, for his indispensable
scientific contribution as well as for his valuable human advices and his constant and
wholehearted support.
A special thank is addressed to Professor Uwe Starossek, for the contribution given to this
research and for the warm hospitality at his department.
Professor Marcello Ciampoli and Professor Paolo Franchin, who revised this work and gave
me precious suggestions, are also gratefully acknowledged. Thanks are also due to Prof.
Rega, who coordinates Ph.D. activities.
I wish to thank Dr. Maren Wolff and Eng. Marco Haberland of Technical University of
Hamburg-Harburg for their collaboration and Engineers Simone Bolognesi and Viviana
Prisco for their valuable contribution.
Finally, I thank my colleagues Luca Sgambi, Fabio Giuliano, Sauro Manenti, who set me an
example to follow. Expecially, I wish to sincerely thank my colleagues and mates Francesco
Petrini, Chiara Crosti, Stefania Arangio and Konstantinos Gkoumas: for the help, the
sharing, the friendship.
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
1. THESIS OUTLINE
2. SYNTHESIS
3. ESSENTIAL LITERATURE
THESIS OUTLINE
INTRODUCTION 17
1 SAFETY AND STRUCTURAL INTEGRITY 19
2 RISK HANDLING 22
CONCLUSIONS 183
9 PROGRESSIVE COLLAPSE SUSCEPTIBILITY 185
9.1 LONG SUSPENSION BRIDGE ................................................................................................................................185
9.2 R.C. FRAMED BUILDING .....................................................................................................................................189
10 PROPOSED ROBUSTNESS ASSESSMENT METHOD 191
10.1 GENERAL FRAMEWORK......................................................................................................................................191
10.2 ROBUSTNESS CURVES ........................................................................................................................................192
APPENDICES 195
11 ROBALG CODE 197
11.1 GETTING STARTED .............................................................................................................................................197
11.2 MAIN FEATURES ................................................................................................................................................197
11.3 CODE .................................................................................................................................................................198
11.4 EXAMPLE FILE AND SCREENSHOT...........................................................................................................220
SYNTHESIS
ABSTRACT
The problem of structural safety is crucial issue in the field of civil engineering [1].
In order to ensure it, old codes and regulations only asked for the prevention of
structural failure, goal that can be easily achieved by means of local resistance
verifications (e.g. at a sectional level). Nowadays a general consensus seems yet to be
achieved among engineer community on the need of ensuring also a localized
resulting damage after the occurrence of a structural failure [2,3]. This assessment is
substantiated with two different considerations: not only is it impossible to prevent
every single failure in a structure (let think to human error in design or execution
phases, as well as extreme events that could not be directly considered in a common
design), but also even a small initial failure can result in a disproportionate structural
damage [4], as shown by several cases of building collapses in the past. Such a
behaviour is commonly understood as a lack of structural robustness [5], a primary
requirement for building and other civil works. In order to ensure a robust behaviour
and avoid the propagation of the damage throughout the structure [6], local
verifications are no more sufficient and specific investigations have to be carried out
on the whole structure.
A robust design should, for these reasons, take into account the different responses to
damage that result from different organizations of the composing elements in the
whole system [7] and cannot be limited to design the structure for additional
accidental load cases [8]. For this purpose a numerical evaluation of the robustness
degree of a structure [9], defined as the ratio between the decrement of ultimate
resistance and increment of the damage level, is suggested in this work: the presented
approach resorts a nonlinear analysis of the structure and is based on static or quasi-
static removal of the composing elements of the structure. The proposed method
takes the ultimate resistance as a parameter for the structural performance and the
number of failed elements as a parameter for the damage level: this means that
several configurations correspond to each damage level, since different combinations
of the removed elements are possible. Among these combinations, the best and the
worst configurations have been identified for each damage level (i.e. the
configurations that show the highest and the lowest response of the structure
respectively). By diagramming the trend of these two values corresponding to the
increment of damage levels, two curves are obtained, that provide a lower and upper
bound for the structure robustness.
Problems arise from the fact, that the exhaustive number of combinations increases
exponentially with the number of elements: in order to perform a feasible limited
number of analyses for one structure, optimization algorithms [10] have been used in
this work with the aim of moving the problem complexity from an exponential to a
polynomial form. The considered optimization techniques are an heuristic and a
probabilistic method. The first one is based on the assumption that the worst and the
best configuration for each damage level can be directly originated by an element
removal on the best and worst configurations of the previous damage level,
respectively; the second one avails a simulated annealing algorithms that starts from a
set move based on the most stressed or strained elements. In both case, the problem
becomes computationally feasible and the mentioned curves are obtained in a
polynomial time also for complex structural systems. These curves characterize the
response of the structure in term of robustness and provide useful information on the
investigated system; particularly, the method allows to clearly identify the critical
elements, enabling a significant improvement through a focused change in the static
scheme or a direct reinforcing of those element only.
In this respect, some applications on relatively simples structures are shown, aimed at
identifying the structural response in term of robustness as a direct consequences of
the different organization chosen for the system. Consequences of the static
indeterminacy level of a structure are presented and discussed, distinguishing
If these kind of static analysis can provide information on the robustness degree of
the structure, in order to trace the effective development of a progressive collapse in a
structure, more specific investigations have to be carried out [11, 12]. To asses the
progressive collapse susceptibility of a structure is quite an onerous task, since it
requires the consideration of many different aspects, starting from the knowledge of
real characteristics of the structure up to the analysis of the response, which generally
involves the consideration of dynamic actions, full material and geometric
nonlinearities and triggering of local mechanisms [13, 14]. The collapse mechanisms
can be very different, but particularly it is important to distinguish between primarily
horizontal load transfer structure and primarily vertical load transfer structures. In the
last section of this thesis, this kind of investigations are applied therefore to the case
of a framed reinforced concrete public building [15] and to a steel suspension bridge
with a long central main span [16]. Following a bottom-up approach, the critical
event modelling has been neglected and an initial failure has been a-priori assumed
on the structure. In the first construction the failure of a main column is assumed as
the starting event and the damage propagation is followed by means of an
incremental iterative nonlinear analysis. The structural system robustness is evaluated
and the contribution provided from the three-dimensional structure to the resistance
of the frame involved in the collapse is outlined. In the second case study an abrupt
rupture of some hanger ropes of the bridge suspension system is assumed as initial
damage and the response of the system to different damage levels and locations is
investigated. This kind of investigations avail some nonlinear dynamic analyses, that
directly account for the dynamic amplification caused from abrupt initial failures and
for the plastic reserves of the materials that the system can exploit during the collapse
evolution. Some considerations on the collapse modality are given and the difference
between sensibility to triggering and to the propagation of the collapse are
particularly outlined. From these investigations some indications on possible
measures that could arrest the collapse can be eventually deducted.
KEYWORDS
Robustness, damage, robustness curves, robustness indicators, combinatorial
optimization, heuristic optimization, simulated annealing optimization, static
indeterminacy, redundancy, element connection, structural continuity,
compartmentalization, progressive collapse, nonlinear analysis, iterative incremental
static analysis, dynamic investigation, collapse triggering and propagation.
Fig. 1: Two non robust behaviours: the Airfrance Concorde crashing and the Minnesota Bridge collapse.
ESSENTIAL LITERATURE
[1] Journal of the international association for bridge and structural engineering (IABSE) Special
edition, May 2006
[2] Unified Facilities Criteria (UFC): “Design of buildings to resist progressive collapse”, UFC 4-
023-03, 25 January 2005
[3] B. R. Ellingwood, R. Smilowitz, D. O. Dusenberry, D. Duthinh, H.S. Lew, N. J. Carino: “Best
practices for reducing the potential for progressive collapse in buildings”, NISTIR 7396,
National institute of Standard and Technology (NIST), February 2007
[4] Multi-hazard Mitigation Council (MMC) of the National institute of building science:
“Prevention of progressive collapse: Report on the July 2002 National Workshop and
recommendation for future efforts”, Washington D.C., 2003
[5] F. Bontempi: “Robustezza Strutturale”, invited lecture at Convegno Nazionale Crolli e
Affidabilità delle Strutture Civili, Università degli Studi di Messina, CRASC’06, Messina,
20-22 Aprile 2006 (in Italian)
[6] U. Starossek: “Progressiver Kollaps von Bauwerken, Beton Kalender, Sonderdruck Ernst &
Sons , 2008 (in German)
[7] J. Agarwal, D. Blockley, N. Woodman: “Vulnerability of structural systems”, Structural
Safety 25, 2003
[8] D.M Frangopol., M. Klisinski, 1992: ”Design for safety, serviceability and damage
tolerability”, Designing Concrete Structures for Serviceability and Safety, SP 133-12:225-
254, 1992
[9] U. Starossek, M. Haberland: “Approaches to measures of structural robustness”, 2008
Structures Congress – Crossing Borders, Vancouver, Canada, 24-26 April 2008
[10] Ausiello G., Crescenzi P., Gambosi G., Kann V., Marchetti Spaccamela A., Rotasi M.:
“Complexity and approximation”, Springer Verlag, 2003
[11] M. Wolff, U. Starossek: “Robustness assessment of a cable-stayed bridge”, 4th international
conference on bridge maintenance, safety and management, Seoul, Korea, 13-17 July 2008
[12] D. Yan, C. C. Chang: “ Vulnerability of long-span cable-stayed bridges under terrorist event”,
Proceedings of the International Conference on Bridge Engineering - Challenges in the 21st
Century, Hong Kong, 2006
[13] F. Bontempi, P.G. Malerba: “Il controllo della formulazione tangente per la soluzione di
problemi strutturali non lineari”, Studi e Ricerche, Vol.17,1996 (in Italian)
[14] G. Powell: “Progressive collapse: Case studies using nonlinear analysis”, presented at 2004
SEAOC annual convention, Monterey, August 2004
[15] ASCE: „The Oklahoma City Bombing: Improving building performance through multi-hazard
mitigation“, Report FEMA 277; FEMA Mitigation Directorate, Reston Virginia, 1996
[16] Bontempi, F., “Basis of design and expected performances for the Messina Strait Bridge”,
Proceedings of the International Conference on Bridge Engineering – Challenges in the 21st
Century, Hong Kong, 1-3 November 2006
INTRODUCTION
INTRODUCTION 17
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
A primary requirement for civil products is certainly the safety of the constructions.
Nevertheless, if a general consensus can be found among structural engineers on the
importance of achieving safety, not the same can be said on the ways of pursuing it. The
divergences start from the definition of the term “safety” itself, that is either referred
primarily to the safety of people as in [SIA260] or to the structural integrity of the building
as in [ISO 2394]. The first code states that:
“A structure can be declared safe if during a critical event, like impact, fire, downfall,
safety of people is granted”.
INTRODUCTION 19
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
possible failure of the system or of a part of it [T.U. 2005]. For instance the International
Standard above mentioned [ISO 2394] states that:
“Structures and structural elements should be designed, built and maintained in such a
way to be proper for their use during the structure’s design life in an economical way.
Particularly they should, with proper levels of reliability:
a. satisfy exercise ultimate state requirements
b. satisfy load ultimate state requirements
c. satisfy structural integrity requirements”.
Regulation:
SAFETY
Safety of Structural
people integrity
(SLS)
SECTIONS OR
SAFETY
ELEMENTS
Ultimate limit states
(ULS)
RESISTANCE R>S
verification on
If methods to satisfy the first two requirements are wide studied and documented,
regulations for the practical achievement of an acceptable level of structural integrity after a
failure are not as much codified. Moreover, if the respect of the first and the second limit
state implies that structural failures, meant as the overcoming of predetermined levels of
resistance and functionality respectively, have to be completely avoided, the respect of the
structural integrity limit state refers to the structural behaviour after the occurrence of a
INTRODUCTION 20
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
critical event, i.e. the overcoming of previously mentioned limit states has not to be averted
but assumed as possible and the consequence on the structure have to be evaluated.
In order to achieve a satisfying level of structural integrity after a failure, the structure
should be provided with intrinsic defences that could avoid the propagation of the damage.
This issue call in directly in question the structural design choices and the organization of
the elements in the structural system.
In this respect a further difference in the structural integrity achievement can be evidenced
in the assumption of possible structural damages or possible critical events that may cause
structural damages. In this last case, specific structural investigations can lead to a collapse
resistant design [Starossek & Wolff, 2005], i.e. a structure that will show a localized
damage but just under the occurrence of a particular assumed circumstance. A proper
number of investigations on the structure that exhibits different level of structural damage
could instead lead to more significant modification of the structural system, so that it could
be able to localize the damage prescinding from the event that could have caused it. To this
ability of the system organization of a structures refers more specifically the requirement of
robustness, a primary property for the structural integrity achievement of constructions and
more generally for a safe design of civil works. This theme will be better discussed in the
first part of this work, where some definition of robustness are provided and a state of the
art of current regulations and research is reviewed.
INTRODUCTION 21
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
2 RISK HANDLING
The problem of structural safety can be synthetically returned to that one of risk handling. If
the term safety has in fact just a qualitative denotation, it is often used as opposite of the
term risk, that can be instead quantitatively defined.
Intrinsic
ROBUSTNESS is presumed Damage
defenses
According to [Schneider, 1999] the risk is a measure for the magnitude of a hazard and it is
constituted by the probability P(F) of a damaging event F and the so-called average
expectation E(D|F) of the damage, should this event F take place. The simplest function
relating the two constituents of risk is the product of these quantities:
Referring more specifically to the risk of civil works, the first term P(F) represent the
probability of occurrence of an event that could lead to structural failures. Structural failures
can be determined by many different events:
- unfavourable combination of loads, material properties, geometrical parameters, all
resulting from standard actions and happenings;
- effects of critical but foreseeable events, such as collisions, climatic hazard etc;
INTRODUCTION 22
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
- unpredictable events.
The second term represents the consequences of structural failures and may be express in
terms of casualties or in monetary units, that indicate the cost due to service interruption,
damage repairing, lost goods and estates.
In order to reduce the risk (and to improve safety likewise) is possible to act on the first as
well as on the second component. As mentioned in the previous section, the approach is
substantially different, since in the first case the critical event has to be prevented while in
the second case has to be presumed. The safety measures discussed above can be better
identified and organised in this conceptual framework (see fig. 4).
If the issue of structural safety let itself to be easily framed from a conceptual point of view
in that one of risk handling, the practical assessment of risk for structural construction and
civil works is a quite complex task though. Both the probability evaluation and the damage
assessment present some difficulties, involving very high consequences, whose assessment
require onerous and complex investigations, that are though associated to very unlikely
occurrences.
Referring particularly to the probability evaluation, in a quite recent paper
[Ellingwood&Dusenberry 2005] described the probability that can be associated to a
structural collapse, as a chain of partial probabilities, according to the following expression:
The parameter Hi represents a distinct hazard, among them that can affect a structure, like,
for example in case of a building, occupancy loads and other demands, misuse, extreme
environmental effects, fires or abnormal loads.
The probability of one distinct hazard Hi to occur is therefore P[Hi]. The occurrence of this
potentially dangerous hazard doesn’t implies though that it will cause necessarily a local
damage in the structure. This event will be synthetically described by the conditional
probability P[D|Hi], i.e. the probability of a local damage in the structure, given that Hi
occurs. This probability considers actually two different aspects, the first one related to the
event and the second one related to the structural system. As a matter of fact, the hazard
could occur also without affecting the structure as well as it could involve some elements,
without anyone of them failing. Let think for instance to a bomb that explode near a
building. The explosion could also happen in a position where it has a minor impact on the
elements (for instance, too far from the building, or between two columns); on the contrary,
the impact area could involve significant elements like beams and columns, but they would
resist sturdily to the impact without showing any serious damage (i.e., in other words, the
INTRODUCTION 23
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
entity of impact would be too modest with respect to the resistance of the elements). The
first aspect is described by the exposure of the structure, while the second one refers directly
to the vulnerability of the structure. The structural vulnerability is intended here as the grade
of direct damage, that a dangerous event affecting the construction causes in its structural
system.
This ability differs, as better explained in the first part of this work, from the robustness of
the structure, that is instead the ability of the system to undergo a local damage without
suffering a major collapse and that is described from the last term of the equation P[F|DHi].
This term indicates the conditional probability that a collapse happens, given that hazard
and local damage have both occurred.
The term P[F] represents the overall probability of building failure and is the product of the
previous 3 terms exposed above.
[Haberland 2006] underlines that the collapse probability, referred as the collapse resistance
of the structure, i.e. the insensitivity of the structure to accidental circumstances, can be
achieved by acting on the first and second term only, being not possible to really control the
As a matter of fact, what could actually be controlled, would be the dangerousness of the
event, that means the probability that the occurred events would entangle the structure.
To better explain this, let refer to the three main components of risk that, according to a
more comprehensive risk analysis, can be identified in the hazardousness, vulnerability and
INTRODUCTION 24
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
exposure terms. As said before, the last two aspects are both contemplated in the second
factor of the mentioned equation. If it seems unfeasible to limit the hazard, i.e. probability
of occurrence of a potentially damaging event, it is generally possible tough to mitigate the
dangerousness of the event, by acting e.g. on the exposure of the building. In this manner,
when the event occurs, it won’t affect the structure too much. This could be done by
changing the location of a building (for instance, moving it away form the street, in order to
put more distance between a possible explosion or impact and the structural elements), as
well as by protecting the construction or its elements by means of protective barriers or
other casing systems. These strategies are not a direct concern of engineers though, since all
refers to non structural measures. Furthermore, the event control by means of the exposure
reduction of a constructions appears to be feasible for some type of hazards only: e.g.
terrorist attacks could be avoided or discouraged by means of sophisticated surveillance
system or security guards, but natural disasters appear to be less manageable in this sense.
Aspects related to the structural system are instead the vulnerability and robustness of the
structures. As mentioned before and better explained in paragraph 1.2 of the first part of this
work, both vulnerability and robustness are property of the structural system, but, while the
first property is a direct consequence of the mass and resistance of the structure, the second
property depends on the system organization, i.e. on the disposition of structural elements in
the system and their relative stiffness and resistance.
Another important difference between the two properties consists in the independence of the
second one from the event. Strategies aimed to mitigate the vulnerability of a construction
are structural measures, but are event-based. Strategies to improve the structural robustness
instead, referring to it as the last probability factor of previous equation, should start directly
from the assumption of a local damage and can therefore disregard the originating event. In
the paper of [Starossek&Haberland 2008] is underlined how strategies addressed to limit
this last factor, i.e. to increase the structural robustness, are seldom taken into account:
“Influencing abnormal events is generally not within the control of a structural
engineer whereas the probabilities P(D|H) and P(F|DH) are [respectively, the
conditional probability of a damage after the hazard and the conditional probability of
a failure after the occurrence of an initial damage]. The common standards usually
describe the safety of a structure as a function of the safety of all elements against local
failure, and therefore only take into account the probability P(D|H). The reaction of a
structure to the possible occurrence of a local failure, that is expressed by the
probability P(F|DH), is seldom investigated regularly.”
It’s important to underline that the mitigation of all these factors is always aimed to the
mitigation of the collapse probability P(F), since the main goal remains the achievement of
INTRODUCTION 25
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
the collapse resistance of the construction. The collapse probability P(F) must be limited to
a generally accepted value, i.e. a socially acceptable value identified through a combination
of professional practice and appropriate building regulation. In the above mentioned paper
of Ellingwood and Dusenberry, the value of 10-7/year is suggested as target value rate for
risk of collapse, even if it is recognised that a big problem of a completely probabilistic
approach is also represented from the difficulties of finding a general consensus on this
admissible risk.
HP-LC
qualitative risk
Deterministic
analysis
pragmatic
analysis of risk
scenarios
Probabilistic
Complexity
probabilistic
(quantitative)
risk analysis
LP-HC
Besides this problem, also other aspects arise from a probabilistic approach, as underlined
by [Starossek. 2006a].
[In the] current design methods […] low probability events and unforeseeable incidents
(accidental circumstances) are not taken into account. Within the scope of a
probabilistic design concept, such a simplification is necessary because the supporting
statistical data, derived from experience and observation, is unavailable. In the case of
a non-robust structure, however, this simplification becomes inadmissible. Let's
consider, for the sake of argument, a structure with primarily serial load transfer (say,
a high-rise building) where the local failure probabilities are statistically independent.
In that case, the probability of collapse will be in the order of the sum of the failure
probabilities of all the elements (say, the building's individual stories) of the structure.
If the number of elements is sufficiently large (simply, if the area of attack is large
enough), even very low probabilities of initial local failure can add up to a probability
INTRODUCTION 26
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
of global failure which is high enough to be taken seriously. (For structures with
primarily parallel load transfer, the probability of collapse is in the order of the product
of the failure probabilities of the elements and thus very low.) [Another problem of]
current design methods is that the underlying probabilistic concept requires
specification of an admissible probability of failure. Considering the extreme losses that
often result from progressive collapse, it seems difficult to reach a true societal
consensus on the admissible probability of such an outcome—a problem which risks of
the type “low probability / high consequence” are typically up against.
INTRODUCTION 27
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
The term robustness is intended in this work as defined by [SEI 2007, Starossek 2008], i.e.
as the structural insensitivity to local failure. It is important to underline that according to
this definition, the structural robustness is a property that does not depends on the particular
triggering cause, but is inherent to the structural system. The different robustness response
of two different structural system is shown in the following figure. System A shows a better
performance than system B when integer, but it is less robust, since when the structures are
damaged the structure B undergoes a lower decrement in its performance with respect to the
structure A.
P integer P P (performance)
damaged integer
damaged
ΔP ΔP ΔP
structure B
s s structure A
d
STRUCTURE A STRUCTURE B: (damage level)
more performant, less robust more performant, less resistant ROBUSTNESS CURVES
Fig. 8: Qualitative robustness slopes of non robust (A) and robust (B) structures.
when due to high continuity between elements, does permit a stress redistribution on other
elements and allows the load to find an alternative path to reach the ground, but is also
responsible of the rupture propagation in case the resulting alternate load path has not
sufficient resistance to bear the high stresses transmitted by this continuity [Starossek 2006].
3.1.2.1 Risk-based-approach
A framework for assessing robustness based on minimizing the risk associated with
potential failure is proposed by several authors. In [Faber, 2007a] the indirect consequences
of a structural failure are recognised as the major component of costs and therefore
included in the equation for the robustness measure. This indirect consequences are said to
be scenario-dependent and computed with the avail of an event tree chart (fig.9).
A robustness index is defined by the author as the ratio between direct risk and total risk
(direct + indirect), where direct and indirect risk of each branch of the tree are computed as:
Æ Indirect
Risk
Æ Direct
Risk
Direct Risk
Irob =
Direct + Indirect Risk
This index assumes value in [0,1], where greater values correspond to greater robustness
and measures relative risk only. This measure of robustness is dependent upon the
probability of damage occurrence and also dependent upon consequences.
Hence, according to this measure, robustness could be reduced also by reducing the
exposure of the structure (i.e. reducing the probability of occurrence of a potentially
damaging event) or its vulnerability (i.e. the probability of occurrence of local damage)
generally increasing its reliability. In this sense robustness is not just a characteristics of the
static system, which strongly distinguish this definition from other ones (e.g the topology-
or damage-based definition summarised in the following).
Another distinguish feature of the robustness as defined by the author consists in the
subordination to the risk evaluation. The risk analysis allows for a clear identification of all
aspects that contribute to a structural collapse, by enumerating a series of possible hazard
and decomposing its factor in the event tree. On the other hand, as already mentioned in the
introduction of this work and noted by [Starossek 2006a], both the determination of hazard
probability (due to the scarceness of statistical data for critical events not originated by
natural causes) and the assumption of costs associated to certain consequences (especially in
case of injuries and fatalities) could be challenging. The evaluation of the consequence
particularly, requires quite complex investigations for all the scenarios, whose number is
generally very large: in this respect, the probabilistic approach doesn’t allow for a reduction
of the required investigation in comparison to a deterministic one. What seems to be more
problematic yet, is to find a theoretical justification for the computation of risk, due to the
low probabilities and high consequences associated to structural collapses.
However, an intrinsic lack of analytical consistence can be often found in many engineering
issues and, as a matter of fact, several researchers have preferred a probabilistic approach
for assessing structural robustness: [Frangopol&Curley 1987] considered probabilistic
indices to measure structural redundancy, that are based on the relationship between damage
probability and system failure probability; more recently, [Lind, 1995] proposed another
generic measure of system damage tolerance, based on the increase in failure probability
resulting from the occurrence of damage; eventually, [Ellingwood 2005] and [Ciampoli
2008] underlined the importance of a probabilistic risk assessment in order to assess
robustness in a general manner.
given by other authors and also used in this work). A structure can be defined vulnerable
then, if any small damage produces disproportionate consequence. Starting from these
definitions, a complete vulnerability theory is developed, aimed to identify the most
vulnerable scenarios by focusing only on the form of the structure (e.g. it hierarchical
element organization) and leaving intentionally aside instead the possible damaging events.
SC DD F QF − QI
Vu ln erability index → VI = , where : RDD = and SC =
RDD DD max QI
N Eq. 4
M ∑ qi
Well − formedness → Q = ∑ Qi , where : Qi = i =1 and qi = K
N ii
i =1
number of operations required for the decomposition of a complex structure is not provided,
though the method can be automatized and an algorithm is described for this purpose. By
means of an organised search through the hierarchical model found for the structures,
different failure scenarios can be identified, that result in disproportionate consequence and
for each one a vulnerability index can be computed.
Since a very large number of failure scenarios is identifiable, the authors suggest to consider
only four particular scenarios plus a number of specific failure scenarios, that are deemed as
interesting for the particular design objectives or conditions of use of the structures. The
four type of failures indicated are very interesting, since they refer to some peculiar aspects
of robustness, that are better discussed in the following and may differ significantly between
distinct structural systems. The mentioned scenarios are listed in the following:
a. Total failure scenario: the total collapse of the structure is obtained with least effort. This
failure scenario is characterized by the highest structural consequence index (SC=1). If
different failure scenarios with index equal to 1 exist, then the failure scenario with the
lowest damage demand is chosen as the total failure scenario.
b. Maximum failure scenario: the vulnerability index is maximum.
c. Minimum failure scenario: the last added element in the hierarchical model is damaged
(i.e. the element that is less connected to the rest of the structure)
d. Minimum damage demand scenario: corresponding to the occurrence of the easiest
deterioration event.
The topologic approach presented by the authors seems to be a quite a comprehensive one
especially because it stresses the role played by the element organization and hierarchical
order of the structural system. The importance of element numbers and the element
connections is carefully recognised and the damage is identified at an intermediate
hierarchical level (cluster). Furthermore, the method has been also developed by the authors
in order to include dynamic effects of failure scenarios: in this case a different measure of
the attribute of well-formedness is provided, that considers not only stiffness but also inertia
and damping properties.
Even if the relevance of this method, especially for the methodological aspects, is not called
in question, from the point of view of a more practical application, it presents some
problematic issues. Firstly, the procedure of decomposition is quite onerous and this onus,
even when diverted from the user by developing an automatic procedure, is not completely
justified by the reduced number of scenarios, that will be in fact considered: even if the
approach starts from a bottom-up technique (see following sections), it avails then a top-
down choice of the scenarios to be considered. Further difficulties arise from the fact that
number and type of failure scenarios strongly depend on the chosen deteriorating event: a
suggested deteriorating event is the removal of one element (referred as “cut at both ends of
an element”) that is “a practical form of damage to use in search”. If different
deteriorating events are assumed, the vulnerability index looses expressiveness and doesn’t
allow to compare directly the robustness of different structures.
Vulnerability indices in terms of stiffness degradation are also proposed by the authors and
tested to assess the vulnerability of a cable-stayed bridge:
n tr (K i )
SD = ∑ Eq. 5
'
i =1 tr ( K i )
where:
- Ki and Ki’ are the i-th element’s stiffness matrix under the healthy and the damaged
conditions respectively;
- tr(•) represents the trace (the sum of the eigenvalues) of the matrix in the brackets;
- n is the number of damaged element.
The formulation proposed by the authors can be numbered among the so-called damage
based approaches, since it is based on the comparison between the nominal and damaged
configurations. The proposed indicator is based on the structural stiffness and seems to be
quite expressive, but requires onerous and specific investigations on the structure in order to
be calculated.
A measure of structural robustness, which also arises from the comparison of the structural
performance of the system in the original state and in a perturbed state, was proposed by
[Biondini&Restelli 2008]. The original state is intended as the configuration, in which the
structure is fully intact, while the perturbed state refers to a condition, in which a prescribed
damage scenario is applied.
The authors define robustness as “the ability of a system to suffer an amount of damage not
disproportionate with respect to the causes of the damage itself” and propose to directly
measure it by means of robustness indices, that are dimensionless functions of some
performance indicators, used as state variables, and vary in the range [0, 1]. The proposed
indicators can be divided into two groups, the first one connected with the structural
property of the static system alone (conditioning, determinant, trace, period) and the second
one considering also the load condition applied upon the system (displacement, stored
energy, vector of backward and forward pseudo-loads). The effectiveness of the indices is
investigated by the authors by means of several application and the indices ρT, ρs, ρΦ, and
ρ1 (see Eq. 5) are indicated as capable of providing a very effective measure of structural
robustness.
c0 d1 t1 T1
ρc = ; ρd = ; ρt = ; ρT =
c1 d0 t0 T0
Eq. 6
s ΔΦ 0 ΔΦ 0 ΔΦ1
ρs = 0 ; ρΦ = ; ρ0 = 1 − ; ρ1 = 1 −
s1 ΔΦ1 ΔΦ1 ΔΦ 0
where:
- the indices 0 and 1 refers respectively to the intact and damaged structure;
- c, d and t are respectively the conditioning number, the determinant, and the trace of
the stiffness matrix K associated with the considered structural configuration;
- T is the first vibration period associated with the mass matrix M and λi(K) denotes
the ith eigenvalue of the matrix K;
- s is the displacement vector of the structure
- ΔΦ0 and ΔΦ1 are the energy difference between intact and damaged system after
application of the backward pseudo-loads and forward pseudo load respectively.
Backward pseudo-loads represents the additional nodal forces that must be applied to
the damaged system to achieve the nodal displacements of the intact system, while
forward pseudo-loads represents the additional nodal forces that must be applied to
the intact system to achieve the nodal displacements of the damaged one.
Among these indicators, the one considering the ratio of stored energies ρφ is probably the
most expressive. The components of this ratio ΔΦ0 and ΔΦ1 are represented by the two
shaded area of fig. 10: the displacement of an intact structure under a determined force is
always smaller (point P0) than the displacement of a damaged structure (P1) as shown in the
bottom part of the figure. In order to obtain the same greater displacement from the integer
structure, a further load increment ḟ0 should be applied to the integer structure (that moves
to point Ṗ 0). The point Ṗ0 and P1 define a triangle (together with the axis origin) that
represents the work of the so-called “forward pseudo-loads”. Conversely, in order to bring
the response of a damaged configuration (Ṗ1) back to the displacement that it would have
suffered as undamaged (Ṗ0) under the same load ḟ0, a force decrement ḟ1 has to be applied to
the damaged structure. The shaded area of the triangle defined by the axis origin and the
points P0 and Ṗ1 represent the work of the so-called “backward pseudo loads”. The ratio
between this area and the previous defined one, represent a comparison between a scalar
entity that refers to the integer structure and a scalar entities that refers to the damaged one.
[…] the stiffness or damage-based robustness measures […] are either easy to
calculate or expressive. Energy-based approaches may be suitable to fulfil both
requirements equally. A possible formulation is based on the comparison between the
energy released by the initial failure and the energy required for a collapse
progression:
E r, j
R e = 1 − max Eq. 7
j E s,k
where:
- Es,k = energy required for the failure of the next structural element k.
This formulation does not include the design objective acceptable damage progression;
instead it verifies the possibility of a total collapse. [The equation above] has a simple
form and also appears to be expressive with regards to the possibility of a total
collapse. However, the determination of parameter Er,j presents a difficulty as it can be
both under and over estimated. The energy released by the initial failure of the
structural element is composed of several parts. For structures that have a tendency for
a pancake-like or domino-like collapse, a large or even dominant portion of the energy
will be kinetic energy transformed from the potential energy of the separated structural
elements. In such cases it is easy to carry out a numerical estimate. For structures with
a tendency for other types of collapse, the value can only be determined through a
comprehensive structural analysis. Furthermore, Er,j should only take into account that
energy part which leads to the damage of the next structural element k. This
requirement is most easily met with structures that have a tendency towards a pancake-
like or domino-like collapse. The separation and impact events which control the
mechanism here occur discretely and in a concentrated manner so that the re-use of the
released energy can also occur in a concentrated manner without large losses, and thus
be easily calculable. Overall, it can be stated that such structures are most likely
suitable for an application of energy-based robustness measures rather than anything
else.
[…] The expression also presumes that the energy of the triggering event is completely
absorbed by the damage of the initially failing component, or that at least the excess
energy does not contribute to the damage of the next element.
action. Referring to the case of the Oklahoma Building collapse for instance (that is further
described in the last part of this work) the explosion of a bomb caused a direct damage on
the construction, limited to the removal by brisance of one main column of the building, a
possible damage of the two adjoining columns and the failure of some part of the floor.
Neither people were directly injured by the explosion or by the column scattering deb, nor
the structural initial failures affected a significant area of the structures. Still, the indirect
consequences of the initial damages resulted in a progressive collapse that involved a major
part of the buildings and caused several human losses.
It seems important therefore to distinguish between direct and indirect effects on a structure,
caused by the considered critical event:
- one thing is the ability of a structure to undergo without significant failures or damages
an explosion, an impact or other critical circumstances, that directly affected some of its
elements;
- another one is the ability of the structure to maintain localized an initial damage or failure
caused by those circumstances, i.e. to take direct damages without developing major
indirect failures.
Considering that we usually call “invulnerable” something that cannot be in any way injured
or wounded (i.e. directly affected from an event), it seems appropriate to name the first
attribute with the term vulnerability [Faber 2006] and [Augusti et al. 2007] and the second
one with the one of robustness. On the other hand the term robustness is also used in
different fields like statistic, electronics, computer science, biology [Krakauer 2005], as the
insensitivity against small deviations in the nominal configuration, i.e. it refers to the
indirect consequence in a system, when an initial failure or change in configuration is
assumed.
Both the attributes of vulnerability and robustness, as here defined, are structural properties,
that depends on how the structure is designed and are therefore inherent to the structural
system. If the structure is invulnerable, it is also clearly resistant to the collapse, while the
contrary is generally not true. The invulnerability of a structure can be increased by a safe
design of key elements or a general over-dimensioning of elements. Generally, it would not
be considered as a desirable property for constructions, due to the cumbersome dimension
of elements that it would imply and to the economic effort required for the materials, that
could not be justified by the structural risk, being generally the probability of occurrence of
critical events quite small. In these cases, it seems more reasonable to achieve a good level
of robustness.
Different definitions can be found in literature for vulnerability and robustness of structures,
which can lead to different design strategies and goals, as discussed in the following section.
If a general consensus on the terms vulnerability and robustness given here cannot be found,
different terms could be as well used, provided that the two concepts remain distinguished.
The term collapse resistance could also be named differently, if a general consensus on this
term cannot be reached, but it seems important to clearly distinguish among structural
properties that depends on external events and properties that are characteristic of the
structure alone: robustness, as above defined, result to be a property inherent with the
structural system and independent of the possible causes of initial local failure. This
definition is in contrast to a broader definition of robustness [Faber, EN] which does include
possible causes of initial failure and should be therefore better addressed to the term
collapse resistance, that is a property that is influenced by numerous conditions including
both structural features and possible causes of initial failure.
It has to be underlined, that, accepting these two definitions of robustness and collapse
resistance, the aim of a safe design would be a collapse resistant design and not
compulsively a robust one. As reported by [Starossek, 2006a]:
It would intolerably limit the range of design possibilities, however, if only those
structural systems were permitted that are clearly robust. Nor is such a limitation
necessary because a structure whose system tends to promote collapse progression can
be made sufficiently collapse resistant by other measures such as an especially safe
design of key elements. Furthermore, collapse resistance may not be required for every
structure.
adjoining elements, propagating the ruptures with a “domino” effect and thus resulting in
the collapse of a major part of the system when not of the whole structure.
This phenomenon was experienced by several bridges and buildings, among which the case
of the A. P. Murrah Building, more in detail described in the following, was also numbered.
The cause that is at the basis of this phenomenon is called to be an insufficient redundancy
level of the structure. More precisely, a progression of rupture can be developed also in a
structure with a high level of continuity (and therefore potentially redundant), when the
alternate load path provided by redundancy is not provided with enough strength and
resistance to withstand the forces transmitted by continuity [Starossek 2006a].
collapse involved the full height of the tower, and falling debris also caused the
horizontal progressive collapse of an entire parking garage under construction adjacent to
the tower. [Leyendecker 1997]
Marine Barracks, Lebanon 1983: it was one of the first large scale terrorist attacks on
Department of Defences (DoD facilities causing significant collapse and causalities. The
estimated bomb equivalent weight consisted of 12,000 lbs TNT and caused the complete
destruction of the construction. Although no direct changes were made to U.S. national
building codes or DoD design criteria, the event began a history of attacks that eventually
influenced design to prevent progressive collapse.
Ambiance plaza, Connecticut, 1987: L'Ambiance plaza was planned to be a sixteen-story
building with thirteen apartment levels topping three parking levels. At the time of
collapse, the building was a little more than halfway completed. Suddenly, a slab fell on
to the slab below it, which was unable to support this added weight and in turn fell. The
entire structure collapsed, first the west tower and then the east tower, in 5 seconds, only
2.5 seconds longer than it would have taken an object to free fall from that height. The
exact cause of the collapse has never been established: the building had a number of
deficiencies and it is still not clear which one could have triggered the collapse, even if
many different theories have be so far developed by several authors [Korman 1987;
Cuoco 1992, McGuire 1992].
Murrah Federal buildings, Oklahoma 1995: on April 19, 1995, in a time span of about
three seconds, an 1800 kg car bomb blew up the Alfred P. Murrah Federal Building in
Oklahoma City. The bombing resulted in the death of 168 people and hundreds of
injuries. Even though this bomb was placed 3 to 5 meters away from the building, one
side of the building collapsed in consequence of the direct damage on one column (that
was removed by brisance) and a possible partial damage on the two adjoining columns
and floor slabs. This collapse and not the blast, caused eighty percent of the deaths in the
bombing [Prendergast 1995].
Hartford arena, 1998: on January 18, 1978 the Hartford Arena experienced the largest
snowstorm of its five-year life. At 4:15 A.M. with a loud crack the centre of the arena's
roof plummeted the 83-feet to the floor of the arena throwing the corners into the air. Just
hours earlier the arena had been packed for a hockey game. Luckily it was empty by the
time of the collapse, and no one was hurt. The society hired to investigate on the collapse
ascertained that the roof began failing as soon as it was completed due to design
deficiencies. Three major design errors, coupled with the underestimation of the dead
load by 20%, allowed the weight of the accumulated snow to collapse the roof [ENR
1979]. The collapse if this building is further treated in the last part of this work, where
an application is presented on a framed R.C. structure, whose model is based on that one
of the A. P. Murrah building [Giuliani&Prisco 2008].
I-35W Mississippi River bridge, 2007: this bridge, officially known simply as Bridge
9340, was an eight-lane steel truss arch bridge that carried Interstate 35W across the
Mississippi River in Minneapolis, Minnesota, United States. The bridge was completed
in 1967 and its maintenance was performed by the Minnesota Department of
Transportation. The bridge was Minnesota's fifth–busiest, carrying 140,000 vehicles
daily. The bridge catastrophically failed during the evening rush hour on August 1, 2007,
collapsing to the river and riverbanks beneath. Nowadays, the NTSB (National
Transportation Safety Board) has not determined the likely cause of the collapse, but did
identify a design error that may have contributed to the failure and particularly in the
design specified steel gusset plates that were undersized and inadequate to support the
intended load of the bridge, a load which had increased over time. This assertion was
made based on an interim report which calculated the Demand to Capacity ratio for the
gusset plates. The NTSB recommended that similar bridge designs be reviewed for this
problem.
Silver Bridge 1967: the Silver Bridge was an eyebar chain suspension bridge built in
1928, in order to connect Point Pleasant, West Virginia and Gallipolis, Ohio over the
Ohio River. On December 15, 1967, the Silver Bridge collapsed while it was choked with
rush hour traffic, resulting in the deaths of 46 people. Investigation of the wreckage
pointed to the cause of the collapse being the failure of a single eye-bar in a suspension
chain, due to a small defect only 0.1 inch (2.5 mm) deep. It was also noted that the bridge
was carrying much heavier loads than it was originally designed for and was poorly
maintained. The eyebars in the Silver Bridge were not redundant, as links were composed
of only two bars each, of high strength steel (more than twice as strong as common mild
steel), rather than a thick stack of thinner bars of modest material strength "combed"
together as is usual for redundancy. With only two bars, the failure of one could impose
excessive loading on the second, causing total failure—unlikely if more bars are used.
While a low-redundancy chain can be engineered to the design requirements, the safety is
completely dependent upon correct, high quality manufacturing and assembly.
Among the previous cases, in fig. 13 also the Meidum Pyramid is represented, even if that
doesn’t refer to a structural collapse. This pyramid though proved a significantly worse
behaviour that the other pyramids, that since millennia last almost unaltered, since it lost the
exterior cover. This was ascribed to a minor increment in the angulation of the walls that
coupled with a worse foundation ground (sand instead of rock) and design [Levy&Salvadori
1992]. Since a slight difference in the design add up to such a worse performance of the
structure, it seems meaningful to show it among other example of non robust behaviour.
In addition to the previously mentioned cases of collapses, also other cases are worth to be
briefly reported, where the structures maintained instead a sufficient integrity level after the
occurrence of impacts or explosions:
B17 collision, 1943: the airplane shown in the fig. 14 was a B-17F severely damaged
during the second world war in consequence of a collision with a Bf-109 in the Tunisian
skies. Even though the cargo tail was largely fractured, the plane managed to fly 90
minutes after the impact and landed safely.
Khobar towers, 1996: the Khobar Towers bombing was a terrorist attack on part of a
housing complex in the city of Khobar, Saudi Arabia, located near the national oil
company (Saudi Aramco) headquarters of Dhahran. In 1996, it was being used to house
foreign military personnel. On June 25, 1996 a fuel truck adjacent to Building #131
exploded in the housing complex. In all, 19 U.S. servicemen and one Saudi were killed
and 372 of many nationalities were wounded. The power of the blast was magnified
several ways. The truck itself shaped the charge by directing the blast toward the
building. Moreover, the relatively wide clearance between the truck and the ground gave
Progressive collapses are classified by the above mentioned author in 6 different types,
briefly recalled in the following:
Domino-type collapse: this collapse, experimented from a domino trail as the one
represented in the fig. 15, is characterised by the initial overturning of one element, that
rigidly falls over another element and, by means of a transformation of potential energy
into kinetic energy, causes an impact capable of triggering the following element
overturning. This is due to the horizontal loading of the impacting element, that result
from its tilting and motion. The collapse propagates in the overturning direction, that is
orthogonal to the direction in which the failing element was initially loaded. Primary
serial load transfer system are therefore inclined to this kind of collapse. The peculiarity
of this collapse consists as a matter of fact, in a weakness of the system towards forces
other than the principal forces. On the other hand, the risk of overturning implies that the
individual structures are sufficiently slender and unbraced. It has to be pointed out, that
the impact of one element onto the next element can also be indirect, i.e. mediated
through other elements, like for example in the case of the collapse of overhead
transmission lines towers. As a consequence, the propagating action does not need to be a
pushing force but can also be a pulling force.
Section-type collapse: this type of collapse is typically experimented by a beam under a
bending moment. When a part of the cross section is cut, the internal forces transmitted
by that part are redistributed into the remaining cross section. The corresponding increase
in stress at some locations can cause the rupture of further cross-sectional parts, and the
failure progresses throughout the entire cross section. This kind of failure is usually
referred as fast fracture, but it is useful to include it in this classification in order to
possibly exploit similarities and analogies. According to linear-elastic fracture
mechanics, the stresses in the vicinity of a crack tip tend to infinity as the distance from
the crack tip approaches zero. In a discrete system, the distance between consecutively
affected elements is non-zero. Still, there is a force concentration in elements
neighbouring a failing element. This effect is partly of dynamic nature which should be
considered when adopting methods of fracture mechanics.
Instability-type: instability of structures is characterized by small perturbations
(imperfections, transverse loading) leading to large deformations or collapse. Structures
are designed such that instability will not normally occur. The failure of a bracing
element due to some small triggering event, however, can make a system unstable and
results in collapse. An instability-type collapse is characterised by an initial failure of
elements which stabilize load-carrying elements in compression, that leads to the
instability of the elements in compression, who suddenly fails due to small perturbations
and triggers the failure progression. The progression of failure can vary. If the element
firstly affected by destabilization is one of a few primary components, like for instance
the leg of a truss tower, complete collapse can ensue immediately without cascading
failure of similar, consecutively affected elements (like in the other types of collapse
discussed before). Although strong disproportion between cause and effect is apparent in
such an event, it might be addressed to the case of disproportioned collapse and not to a
progressive collapse.
Mixed-type collapse: the types of collapse recalled above are relatively easily discerned
and described. Some collapses that have occurred in the past do not neatly fit into these
categories, however. The partial collapse of the Murrah Federal Building (Oklahoma
City, 1995) seems to have involved features of both a pancake-type and a domino-type
scenario. Such mixed-type collapses are less amenable to generalization because the
relative importance of the contributing basic categories of collapse can, in principle, vary.
Nevertheless, further study might lead to the definition of other well defined types of
collapse.
Besides, in a commentary of this section [ASCE 7-98 C1.4], the attention is focused on the
occurrence of progressive collapse, whose a definition is given:
“Through accident or misuse, properly designed structures may be subjected to
conditions that could lead to general or local collapse. Especially for specially
designed protective systems, it is impractical for a structure to resist general collapses
caused by gross misuse of a large part of the system or severe abnormal load acting
directly on a large portion of it. However precaution could be taken in the design of
structures to limit the effects of local collapses, that is to prevent progressive collapses,
which is the spread of an initial from element to element resulting, eventually, in the
collapse of an entire structure or a disproportionately part of it.”
Progressive collapses are clearly distinguished from general collapses, in which the
immediate failure of a major part of the structure cannot be averted and is therefore referred
as beyond the scope of the discussed provision, and from limited local collapses also, that
represent instead the aim of a design focused on structural integrity. Some exempla are
provided and also several factors that can possibly contribute to the risk of damage
propagation are listed. Furthermore, some design alternatives are discussed and divided
between direct and indirect methods. Among the direct design measures, the methods of
alternate load path and specific local resistance, originally presented in the renewed
technical report of [Leyendecker&Ellingwood 1997] are for the first time reported in an
official standard code together with some practical guidelines for the provision of general
structural integrity:
a. Direct design: explicit consideration of resistance to progressive collapse during the
design process, through either:
a. alternate path method: a method that allows local failure to occur but seeks to
provide alternate load paths so that the damage is absorbed and a major collapse
is averted;
b. specific local resistance method: a method that seeks to provide sufficient strength
to resist failure from accidents or misuse.
If the ASCE provisions was one of the first code to address the problem of progressive
collapse, the presented guidelines are more qualitative than quantitative. The commentary
contains a number of different design concepts and details to meet general structural
integrity requirements, but specific methods to achieve structural robustness or mitigating
progressive collapse are still not provided. On the contrary, the General Services
Administration (GSA) and the Department of Defence (DoD) have developed guidelines
that specifically address the vulnerability of buildings to progressive collapse, as part of
their overall building security requirements.
and the potential for progressive collapse is examined independently of any specific threat.
Vertical load-bearing members are removed to help introduce redundancy and resiliency
into the structure; this approach does not necessarily replicate the effects of any specific
abnormal load. The criteria in the GSA Guidelines are not part of a blast design or analysis
and the results based on these criteria should not be substituted for those obtained from a
substantiated blast analysis.
Flowcharts are used to determine whether a building is exempt from detailed consideration
for progressive collapse. A number of important attributes (including occupancy, type of
framing system, number of stories, and standoff distance) are used in the exemption process.
If it is determined that further consideration for progressive collapse is required, then either
a linear or nonlinear static/dynamic analysis procedure must be used. A linear analysis
approach is limited to low- to mid-rise buildings that are 10 stories or less in height with
relatively simple structural layouts. A more sophisticated nonlinear analysis usually is used
for taller buildings and/or buildings that have a typical structural configurations and can be
carried out only by qualified structural engineers, who are experienced in interpreting the
results of such an analysis.
Referring to linear analysis, the building must withstand as a minimum, the loss of one
primary vertical load-bearing member without causing progressive collapse. Both exterior
and interior scenarios must be examined for typical structural configurations. In the exterior
scenario, a framed building is to be analyzed separately for each of the following cases:
a. instantaneous loss of a column in the first story located near the middle of the short
side of the building;
b. instantaneous loss of a column in the first story located near the middle of the long
side of the building;
c. instantaneous loss of a column in the first story located at the corner of the building.
When performing a static analysis for the scenarios outlined above, the vertical load applied
to the structure accounts by means of a simplified factor of 2 for dynamic effects that
amplify the response when a column or wall is instantaneously removed from a building.
Furthermore, since the probability is small that full live load is present during a possible
progressive collapse event, 25 percent of the live load is used in the load combination:
DL = dead load
Load → L = 2 ⋅ (DL + 0.25LL) , where: Eq. 8
LL = live load
After the analysis has been performed, a demand-capacity ratio DCR is computed for each
of the structural members in the building. Failure of a structural member depends on the
Once the DCRs have been computed after the required number of analysis runs, it is
possible to determine the extent of collapse, if any. The maximum allowable extent of
collapse resulting from the instantaneous removal of an exterior column or wall must be
confined to the smaller of the following two areas:
a. the structural bays directly associated with the instantaneously removed column or
wall;
b. 1800 square-feet at the floor level directly above the instantaneously removed
column or wall.
Similar limits are given for allowable collapse areas based on the removal of interior
columns or walls. A high potential for progressive collapse is assumed to exist when
collapse areas that are determined from analysis are greater than the appropriate limiting
values prescribed in the GSA Guidelines. In new construction, several iterations in the
design of key structural members may be needed before the acceptance criteria are satisfied.
Although not required by the GSA Guidelines, procedures to obtain preliminary member
sizes are given in Appendix B. Existing facilities undergoing modernization are to be
upgraded or retrofitted to meet new construction requirements.
The level of progressive collapse design for a structure is correlated to the Level of
Protection (LOP) assigned to the building and the following three methods for mitigation of
progressive collapse are in detail examined:
1. Tie force method, that provide for a mechanical tying of the building. Minimum tie
forces, which vary with construction type and location in the structure, are typically
resisted by the structural members and connections that are designed for gravity and
lateral loads. The purpose of the horizontal and vertical ties is to enhance continuity and
ductility, and to develop alternate load paths in the structure. Internal and peripheral
horizontal ties must be provided, along with ties to external columns and walls. Specific
required tie strengths are given for reinforced concrete, structural steel, masonry, wood,
and cold-formed steel construction. In all cases, the paths of ties must be straight and
continuous and no changes in direction are permitted.
In a reinforced concrete structure, the flexural reinforcement in slabs, beams, and
girders can be used to satisfy the horizontal tie force requirements. Similarly,
longitudinal steel in concrete columns can be used to satisfy vertical tie force
requirements. Steel beams and girders are used as horizontal peripheral ties and column
ties in a steel structure, while either steel beams or the reinforcement in the concrete
slab are used as internal ties. Steel columns are instead used as vertical ties. The
connections between the steel members also must be designed to transfer the required
tie forces.
2. Alternate path method, which is applicable to buildings assigned to medium and high
LOPs, is similar to the method in the GSA Guidelines in that vertical load-bearing
elements are removed at various locations in the building. However, unlike the GSA
Guidelines where the members are removed at only the first floor level, in the DoD
alternate path method the vertical elements are removed at each floor level, one at a
time. For example, if there are three exterior columns that must be investigated and
there are five stories in the building, 15 alternate path analyses must be performed. The
details of the analysis are given for the different construction materials; including over
strength factors for material strength, which are analogous to the GSA strength-increase
factors. In a linear or nonlinear static analysis, the following amplified factored load
have to be applied referring to those bays immediately adjacent to the removed element
and at all floors above the removed element (first equation) and to the rest of the
structure (second equation):
tributary to the removed element. The damage limits for interior columns and walls is
two times that of exterior ones. Any collapse must not extend into the bays immediately
adjacent to the removed element.
4. Additional ductility requirements, which are applicable to all types of construction in
structures assigned to Medium and High Levels of Protection and is meant to ensure
that the failure mode for all external columns and walls in the ground floor is flexural
(ductile) rather than shear (brittle).
The report focused the attention on the difference between direct and indirect methods (this
distinction is further explained in the following section of this first part of the work) and
indicated different strategies to be used for the mitigation of progressive collapses. Among
the direct method, the strategies of alternate path method and the method of specific local
resistance, already mentioned by ASCE07 were reported, while among the indirect method,
the need of providing the structure with enough redundancy, continuity and ductility is
outlined.
In this respect though, it is worth noting that the same study reports the opinion of some
controlled demolition experts that noted how structures which possess inherent redundancy
to redistribute loads are easiest to bring down, fact that is not consistent with the indirect
methods proposed to increase the resistance against progressive collapses:
“Additional observation of the controlled demolition experts [...] indicated that grater
continuity may actually increase likelihood of progressive collapse. If the diaphragms
are too well tied together, a localize structural failure may actually pull down a greater
portion of the building. This observation contradict many engineering approaches to
structural integrity and needs to be studied further“.
Already in ’90, the Model Code underlined in paragraph 1.1.1 (general requirements) the
concept of proportionality between the causes and the effects [Model Code 1990]:
“Structures should with appropriate degrees of reliability, during their construction
and whole intended lifetime, perform adequately and more particularly:
a. withstand all actions and environmental influences, liable to occur
b. withstand accidental circumstance without damage disproportionate to the original
events (insensivity requirement)”
Reference to safety and structural integrity can also be found in Swiss [SIA260] and
German codes [DIN 1055-100, 2001], whereas a separated section should be devoted to the
Eurocodes.
3.3.2.1 Eurocodes
A significant part of Eurocode 1 [EN 1991-1-7: 2006] is devoted to accidental actions:
according to EN 1991-1-1-7, structural robustness is the ability of a structure to withstand
actions due to fires, explosions, impacts or consequences of human errors, without suffering
a damages disproportionate to the triggering causes. In these terms, robustness is intended
as property dependent on the particular triggering event considered and not as an intrinsic
property of the structural system [Baldassino 2006]. According to [Starossek&Wolff 2005]
instead, the dependence upon the causes let the property be more correctly defined as
collapse resistance.
From a general point of view, the new release of Eurocode 1 represented an inversion in the
European tendency to focus on the structural property in order to achieve a structural
insensitivity to damage, that counterpoised the main role played in the American regulations
and guidelines by loads and actions. This fact is clearly evidenced by comparing the
flowchart presented in the Eurocode and reported in fig. 16 with the flowchart of a draft
version, where the sufficient minimum robustness requirement was included into the second
strategy group (limiting the extent of localised failure) instead of appearing in the first one
(accidental action based) like now.
As better explained in the following sections, this difference in the meaning of robustness is
not just an academic distinction, since the definition of the term strongly and directly affect
the strategies and methods to be used for the robustness achievement.
ACCIDENTAL
DESIGN SITUATION
Apart form this considerations, the mentioned Eurocode 1 appears to be one of the most
detailed regulation for what concern the topic of robustness and structural integrity. In the
following, the main sections covering the topic are reported:
Accidental design situations - strategies for identified accidental actions
a. The accidental actions that should be taken into account depend upon:
a. the measures taken for preventing or reducing the severity of an accidental
action;
b. the probability of occurrence of the identified accidental action;
c. the consequences of failure due to the identified accidental action;
d. public perception;
e. the level of acceptable risk.
NOTE: in practice, the occurrence and consequences of accidental actions can be
associated with a certain level of risk level. If this level cannot be accepted,
additional measures are necessary. A zero risk level, however, is unlikely to be
reached and in most case is necessary to accept a certain level of risk. Such risk
level will be determined by various factor such the potential number of casualties
involved, economic consequences and the cost of safety measures etc.
b. A localized failure due to accidental actions may be acceptable, provided that it will
not endanger the stability of the whole structure and the overall load-bearing
capacity of the structure is maintained and allows necessary emergency measures to
be taken.
NOTE: for building structures such emergency measures may involve the safe
evacuation of persons from the premises and its surroundings.
For bridge structures such emergency measures may involve the closure of the road
or rail service within a specific limited period.
c. Measures should be taken to mitigate the risk of accidental actions and these
measures should include, as appropriate, one or more of the following strategies:
b. protecting the structure against the effects of an accidental action by reducing the
effects of the action on the structure (e.g. by protective bollards or safety
barriers);
c. ensuring that the structure has sufficient robustness by adopting one or more of
the following approaches:
Anyway, taking in account single action’s specificity, structural design must be oriented to
the whole resistant system and not only to the sizing and verification of single components.
“The robustness requirement must be tested during the design stage by applying
separately the following causes:
a. nominal loads, being these loads arbitrary but significant for the considered
scenario, in order to test the global behaviour: a fraction of loads acting in vertical
direction has to be somehow applied along an horizontal direction; this fraction,
unless otherwise stated by the designer, is assumed equal to 1% for construction
lower than 100 m, to 0.1% for construction higher than 200 m and to an interpolated
percentage for middle heights
Among the general purposes of the structural analysis, the assessment of the extreme values
of structural response and sensibility, is pointed out.
With respect to the assessment of the sensibility of the structural response particularly, in
the section devoted to non accidental actions (i.e. environmental and natural actions), the
need for particular analysis is highlighted: these analysis should in fact account not only for
the nominal, perfect configuration, but also for imperfections or ruptures in symmetry or
localized damages.
For this purpose buildings must be designed so that the principal structural system were
able to withstand local damages without entirely collapsing; building must undergo
proportionate deteriorations of resistance performance referring to the triggering
cause.
The updating of the mentioned regulation in 2008 cancelled unfortunately the majority of
the mentioned reference to robustness. In the new version of the Italian code [NTC 2008]
only the explicit reference to the verification of robustness among the other limit states is
left, together with the prescription of verifying the global behaviour of the structure by
considering conventional nominal actions:
“During the design phase, the robustness of the construction has to be checked by using
conventional nominal loads in addition to the usual explicit actions (excluding seismic
and wind actions).In order to verify the global behaviour of the construction, the loads
have to be applied along two orthogonal horizontal directions in the amount of 1% of
the system vertical loads”.
Some references to the methods for achieving robustness are done for different typology of
constructions. Some more detailed references are made for timber:
“The structural robustness requirements, mentioned in sections §§ 2.1 e 3.1.1 can be
also achieved by adopting proper design choice and adequate design measures, which,
in case of timber elements, should at least concern:
a. the protection of the structure and of its component elements against moistness;
b. the use of intrinsic ductile ties or binding systems with a ductile behaviour;
c. the exploitation of composite elements with a ductile global behaviour;
d. the reduction of timber zones which are stressed in tension orthogonally to the fiber
direction. The reduction apply especially if shear stresses are also present (like in
case of carvings) and, generally, when high level of moistness are to be considered
in the element during its service life.”
Prevention Presumption
Structural Specific
no yes avoided performed
measures analyses
The first choice for the achievement of structural integrity concerns the engagement of a
strategy of prevention or presumption of the structural failures. As discussed in the previous
sections, the first strategies leads to the reduction of the structural vulnerability or the
improvement of structural security (requirement which is better defined in the following),
depending respectively, on the adoption of structural measures or not The undertake of
second strategy instead contemplates only structural measures but require also an early
choice in the approach to be followed, i.e. the use of direct or indirect design methods,
which are describes and discussed in the following g section.
a. Direct measures
The structural behaviour under damaged condition is investigates. To this aim, some
assumptions have to be made concerning the modality of the analysis to be performed and
particularly, it has to be decided whether modelling or not the damage cause. If the action or
event producing the local failure is modelled, a deductive procedure is followed and a top-
down investigation is performed on the structure. If the study is made independent from the
failure-triggering cause, an inductive procedure and a bottom-up investigation is performed
on the structure, which identify the damage at low hierarchical level (e.g. composing
elements or sections, joints and connections, functional parts) and study directly the safety
and the performances of the damaged configuration. Particularly:
b. in a bottom-up investigation, the particular triggering cause is set aside and the
investigation proceeds from the lowest hierarchical structural level to the higher ones,
e.g. the consequences due to the failure of an element on the rest of the structure are
identified.
Referring to the top-down method, the assumption of possible accidental actions or critical
scenarios that can endanger the structure seems feasible just for some particular structural
typologies: e.g. bridges are prone to collisions on the pylon by ships or floating objects; tall
buildings had been more than was impacted by plans, embassy or military buildings are
exposed to explosions that can affect the main columns or to fires malevolently set in a
restricted area of the building. In these cases, the event modelling seems feasible, using
proper load scenarios to simulate the accidental action (impact, explosion etc.) and directly
verifying the structural response by means of a top-down investigation that follows the
propagation of ruptures from the highest hierarchical levels to the lowest ones.
In other cases instead, a clear identification of the damage causes seems less feasible: let
think for instance to human errors during the design phase or the execution of the work.
This difficulty has often induced to completely neglect the study of structural integrity.
The clear identification of the design objective inherent to the direct design (both in cases of
a direct or indirect approach) makes it preferable for a performance-based design.
Nevertheless the onus of the investigations required (which are quite complex ones, having
to account for large displacements and deformations, dynamic actions, nonlinear behaviour
of materials and possible development of local mechanism) has lead hitherto to prefer
instead the indirect approach further described.
b. Indirect measures
If an indirect design approach is followed, the structure has to be provided with an high
level of resistance, static indeterminacy and ductility, in order to allow for a good load
redistribution on the elements adjacent to the failed ones (structural redundancy).
Indirect measures, prescinding from specific analyses, are prone to be easily included in a
prescriptive set of code rules. Being independent from the critical event and from the type of
triggered local damage, they are also called “multi-objective measures”, since they can
generally be helpful also in case of other menaces or design problems, like, for instance
earthquakes, wind and other dynamic actions.
The major limit of an indirect design consists in the fact that the structural response to the
accidental action remains unknown. Without a specific structural investigations the promise
to maintain a localized damage under different possible conditions and accidents seems to
be in fact unfeasible. Furthermore, the high level of structural connection which is provided
in the structure (with this term meaning the continuity between the elements and the grade
of internal constraints) can prove to be not only ineffective but in some cases even harmful.
This aspect, already observed during some controlled demolitions of buildings [MMC 2005]
is better discussed in the following section and taken up in the last part of the work.
The role played by structural continuity seems particularly controversial: as a matter of fact,
favouring the stress redistribution from the failed elements to the adjacent ones, it allows the
load to develop alternative paths to reach the ground. On the other side though, an high level
of structural continuity can be also responsible of the propagation of ruptures, in case the
elements adjoining the failed ones are not resistant enough to bear the high stressed
transmitted to them. It is not to be expected all the elements to be provided with the required
resistance to resist any stress transmission, since it depends on the entity of the initial
damage and cannot be ensured therefore in all circumstances. Under exceptionally high
stresses then the failure would propagate through the most connected elements, causing a
great portion of the structure to collapse. This particular aspect was already highlighted in
the report of the [MMC 2005] and presented in the previous section of this part of the work.
With regards to this consideration, an alternative strategy to redundancy can be identified,
that is aimed to create some structural compartments, where the stress concentration and the
propagation of rupture have to be contained. This concept is somewhat opposite to that of
stress redistribution: by compartmentalizing a structure, a concentration of the stress
increments is sought, in order to obtain also a localization of ruptures. The sharp loss of a
restricted area prevent the adjacent element to fail too. The strategies is exemplified in the
figure below, by means of an analogy with the fusible protection inserted in a electric
circuit, where the transit of a too intense current is inhibited by the rupture of the fusible that
interrupts the circuit.
The isolation of structural parts can be obtained by a strengthening or a reduction of
continuity at the limit of the compartmentalized sections. In fig. 20 two example of
localized collapse are shown, in which the compartmentalization of a plane and of a
building was obtained, by means of a strengthening and a reduction of continuity
respectively.
fusible
b c alternate protection
2 path
1 4
3
5
a g d +
2 6 -
9
7 R
8
f e
includes the dependability means, i.e. the strategies and methods that can be followed in
order to achieve and maintain a dependable system.
Integrity: the attribute is referred to the absence of structural failure. This attribute
concerns therefore the structural state, in the sense that the maximum grade of structural
integrity is related to the nominal configuration of the structure, i.e. the undamaged one.
Most recent code and regulations [TU 2005] explicitly require to consider the structure in
the design phases not only in its nominal integer configuration, but also in a damaged
one, where a lack of structural elements can for any reason reduce the integrity level of
the system. It seems in fact not feasible to provide the structure of the ability to undergo
extreme accidental actions without suffering any damage. The strategy of structural
invulnerability would spare the onus of considering different level of structural integrity,
but it proves to be too much uneconomic considering that it refers to critical but very rare
events.
Reliability: the attribute is defined as the grade of confidence on the structure with
respect to environmental or human actions. It can be referred to a single element as well
as to a part or to the whole structural system. The attributes refers to the normal use of the
construction and can be therefore intended in a probabilistic way.
Security: the term is commonly intended as the vigilance and surveillance system. Here
it is instead more generally referred to the grade of confidence on the structure with
respect to malevolent (intentional) attacks. The security level of a construction can be in
this view enhanced also by means of non structural measures, like choosing a secure
location for the construction or providing it with vigilance or a proper escape system.
Damage tolerance: the term has to be referred to the ability of the structure to absorb,
continuously in time, local damage of small entity, such as those due to material
degradation or corrosion.
Collapse resistance: the attribute indicates the capacity of the structure to undergo
specific events of significant entity, with the whole system remaining stable
[Starossek&Wolff 2005]. It refers to the behaviour of the structure against a particular
critical event (such as an impact or an explosion). The structure can be designed for
single accidental circumstance: specific countermeasures can be adopted and calibrated
on the assumption of that particular fault event (top-down method); thus, this attribute
depends not directly on the structural system organization, but it is strongly influenced by
the actions involved in the considered accident.
Robustness: the attribute refers to the ability of the system not to propagate an initial
damage caused by a critical event. In a robust structure the damage remains bounded in a
limited portion of the structure and the entity of the collapse is proportional to the initial
failed elements directly affected by the accidental actions.
The conceptual organization of the attributes related to the structural safety is illustrated by
means of the flowchart at the left side of fig. 21: it is shown how a robust structure, being
this requirement referred to an intrinsic ability of the structural system to localize the
damage, is also damage tolerant and resistant to collapse, while the opposite cannot be
guaranteed.
Faults: with this term all physical threats are intended, that may or may not cause a
structural failure. They can be divided in two groups: the first one includes the external
fault, by this term meaning extreme environmental action or natural disaster as well as
accidental or intentional explosion, impact and attacks; the second groups includes
instead the intrinsic defects of the structure, like for example a crack in a concrete beam,
that doesn’t normally represent a structural problem, until non conventional loads
scenarios occur.
Errors: this term is here referred to the logical threats, i.e. the human errors in the
design, construction and life-cycle. In the case of error, the system is in an incorrect state
(potentially dangerous), that may or may not cause failure, e.g. when coupled with a fault
event such as earthquake or intense wind.
Failures: failures are intended as the manifestation of an error or a fault. In the case of a
failure, the service that the system offers deviates from compliance with the system
specifications. Furthermore, it has to be considered that a failure in a non robust structure,
even when localized, can lead to the failure of adjacent elements, overloaded by the force
redistribution caused by that failure or impacted during the collapse of the fist failed
elements. This can lead to a progressive collapse that involves a major part of the
structure.
occurrence of a fault or error. In this case the consequence of such an occurrence are
directly sought by means of top-down methods, that start from modelling the fault or the
error and avail specific investigation to highlight the response of the structure. Aim of
this mean is the achievement of a collapse-resistant structural system. It has to be
emphasizing that the modelling and the analyses required by such studies are quite
onerous; such studies seem therefore feasible only for particularly important or strategic
construction or complex structural system.
Failure presumption: this mean is aimed to provide the structure of a sufficient
robustness to suffer limited damage in consequence of structural failure. The failures are
then assumed possible to occur, independently by the causes that have determined them;
direct investigations on the structure have to be performed, that start from a damage that
involves the lowest hierarchical levels and studies the effects on the higher ones,
eventually determining the response of the overall structure (bottom-up methods).
Contrary to a top-down investigation, which can provide information on the structure
insensitivity just limited to a particular accidental situation, this approach could instead
supply information on the robustness degree of the structure, provided that a sufficient
number of failures are considered. Being this number usually very large, also in this case
the onus of such analyses is considerable, but it can be justified for some particular
construction, like the long suspension bridge presented in the following, whose unusual
characteristics could make the structural system susceptible of progressive collapse.
after a failure. Within a robust structure the damage is bounded and failures and has a
limited propagation, i.e. the entity of final collapse is proportional to the amplitude of the
initial damage determined directly by the accidental action. Analytically that means that the
ratio between the resistance decrement ΔR and the increment in the entity of the damage Δd
must be upper bounded:
ΔR
Robust structure ⇔ < L Eq. 10
Δd
To provide a quantitative evaluation of robustness, the damage has also to be quantified: the
extent of damage can be measured by computing the number of failed elements. Such a
number is referred as the “damage degree” of the structural configuration.
A method for an analytic quantitative definition of the robustness employing the damage
degree is provided, by performing a nonlinear static analysis on the integer structure first
and on the structure exhibiting different degrees of damage successively (i.e. removing
elements from the integer structure). For each damage degree all possible combinations of
removed elements on the structure should be considered and a maximum and minimum load
factor can be extrapolated, providing as result two curves that are a measure of the structure
robustness. This method can be qualitatively summarized in the following procedure, also
represented in fig. 22:
E!
Cd = Eq. 11
d!×(E − d )!
d : damage level
i : structural configuration
λu: ultimate load multiplier
N : # of combinations
D : maximum damage level
d := 0 0 1 2 3 4 5 d
λu,id
SI λud MAX Max {λu,id} NO SI
d := d +1 d<D i := i +1
Min {λu,i
i<N
λud MIN d}
NO
Curve of maxima
Curve of minima END
Fig. 22: Robustness evaluation chart and resulting structural response slopes.
By performing the analysis, particular attention has to be paid to local mechanisms possibly
developed, as well as to the choice of the load to be incremented. A distributed weight over
each elements should be to preferred, since it doesn’t require to replace point forces
whenever a loaded element is removed.
Another important parameter is the maximum damage level D, generally lower or at the
most equal to the number of element E. The number of the analysis to be performed
corresponds to all combinations of all the damage levels considered, that is:
D D E!
C = ∑ Cd = ∑ Eq. 12
d =0 d =0 d!×(E − d )!
When D is assumed equal to E (being conventionally the resistance of the null structure
equal to zero), the number C of all the possible combinations to be analyzed results
particularly simple to be expressed. This can be easily understood by ordering the elements
in a string with E positions, where the value 0 or 1 in each position respectively indicates
the absence or the presence of the element. The number of all possible strings is known by
the laws of the combinatory calculus to be:
E
D := E ⇒ C = ∑ C d = 2 E Eq. 13
d =0
This number increases exponentially with the complexity of the structure (number of
composing elements) and even for usual structure the number of all the possible
combinations is too large to allow an exhaustive analysis of all the configurations. For this
purpose, optimising algorithms have been developed in order to identify the most significant
combinations, as explained in detail in the second part of the work.
4.3.3.1 Resistance
The first assumption of this method concerns the parameter chosen for representing the
performance and the degradation of the quality of a structure, when the damage becomes
more significant. Several parameter could be used, as the stiffness of the structure or the
structural ductility or other properties related to displacement parameters. However, these
indicators are often too less sensitive to damage or not enough representative of the
structural behaviour [Biondini&Restelli, 2008]. The ultimate resistance is instead an
expressive measure of the structural capacity, that posses enough expressivity and
sensibility to damage. Furthermore, considering that the method avail do not avail dynamic
investigations, it seems more sensible to address the structural performances to the capacity
response, being this parameter directly obtainable as a result of a static analysis.
The reason for this scaling consists in the need of prescinding from the absolute nominal
capacity of the structure in the robustness evaluation, so that comparing two structures, the
biggest one would not always be the most robust too.
Conversely, a scaling of the abscissa has not been a-priori performed and it is discussed for
the definition of some robustness indicator proposed in the following. This choice can be
argued since structure with more elements could be favoured from the point of view of a
robustness assessment. Different solution for this problem have been considered and tested,
that are reported in the following.
This choice derives directly from the assumption of bounding the considered damage degree
at the level of elements in the structural hierarchy, assumption that has been widely
motivated in the previous section. On this premise, the number of total elements in the
structure counts also as absolute value for the structural robustness: in a structures with a
very large number of elements the failure of just one of them represent as a matter of fact a
minor event, that consistently leads to a more robust behaviour with respect to a structure in
which the same element has failed over an inconspicuous number of total elements. If a
scaling on the number of total elements would be performed, the response of the two
structure when one element is removed would be no more directly comparable.
a. Scaling by the total number of elements
This method of regularization is based on the scaling of damage level. In this case the
damage level will be divided by the total number of elements. In this way a structure with 9
elements that can suffer the loss of 3 of them remaining stable, will be not disadvantaged in
respect to a structure with 21 elements that can suffer the loss of 7 of them. That means, by
this scaling the relative damage entities will be considered and the number of element will
no more count as an absolute value. Still each element will count the same not depending on
his volume or area section. This aspect is better treated in the following section.
b. Scaling with respect to the element section
Let consider a multy-storey frame structure with 3 spans. According to the proposed
robustness quantification, the critical elements can be identified at glance as the 4 columns
of the first floor. It’s self evident that a failure of one these columns would have greater
consequence that the loss of one beam span for example and also that in the worst case, the
failure of just 4 elements is sufficient to drop the resistance to zero.
If the sections of these columns are incremented, the new structure can prove to be more or
less robust: changing the stiffness ratio between the different element, the modality of
collapse can also change and consequently also the form of the robustness curves and
indexes. So, for some force direction, the reinforcement of some elements could favour a
bigger force redistribution on other elements and lead to a plasticization of more sections
before a mechanism is triggered.
If a scaling of the damage level is performed on the basis of sectional dimension though, the
robustness of the second structures would be also favoured from the fact that a damage
increment of one would not imply the failure of one column, but just of a part of it,
contrarily to the case of original structure. Being the resistance referred to the nominal
value, the new structure would prove probably to be more robust of the original one even in
case of an unfavourable stiffness redistribution that triggers, as example, a local mechanism
and strongly reduces the structure capacity for subsequent damage levels.
In this way, the topological significance of the structural organization in the robustness
assessment would be loss in favour of the material quantity. The toughness provided by
incrementing the column sections should instead only influence the vulnerability of the
structure and its collapse resistance, due to an higher specific local resistance, but should not
directly affects its robustness.
Unfortunately, a unique indicator, capable of expressing all aspects of robustness could not
be identified and a number of indicators is proposed instead, each one referring to a
particular aspect of robustness. These indicators are meant to be jointly considered in a
robustness quantitative evaluation. Furthermore, being the robustness response bounded
between two curves, two different value of these indicators have to be considered,
depending on the curve which they are referred to, as better explained in the following
section.
In the fig.23 (top-left) the secant with the highest inclination is the first one, but hopefully
not many common structures would show the highest decrement of resistance for the first
damage level. According to the index measured though, the later an abrupt decrement in the
resistance occurs, the less it affects the structural robustness, being the inclination of the
secant mitigated by the absolute increment in the damage level (i.e. referred to the origin).
λ a λ b
0 1 2 3 4 5 6 d 0 1 2 3 4 5 6 d
λ λ
c d
0 1 2 3 4 5 6 d 0 1 2 3 4 5 6 d
d. Scattering from linear trend, calculated as the area subtended between the curves and the
straight line that connect the point of the integer structure with that one of the null one.
D ⎡ ⎛ d ⎞⎤
I d = ∑ ⎢ r (d ) − ⎜1 − ⎟ ⎥ ← - E ≤ Id ≤ E
d =0 ⎣ ⎝ E ⎠⎦ 2 2
e. Upper and lower bound for maximal damage that makes the structure unstable
I e = D low + k ⋅ D up , where : 0 < k < 1 and 1 ≤ D low ≤ D up ≤ E → 0 < I e < (1 + k ) ⋅ E
This property is consistent with the definition of robustness, that is focused on the
proportionality between the loss of quality of a structure and the damage entity that it should
suffer.
This parameter is the most simple and lend itself to be easily used in the engineering
practice, since it is perform-based: having a maximum damage of interest being fixed, the
worst response of the structure within that damage is measured and compared with the
expected performance or with an acceptable value determined as design objective: for
example, considering a two-storey framed building with four span in both directions, it
could be requested that the maximum decrement of resistance corresponding to a failure of
two elements at the most, doesn’t exceed the 40%. On the contrary it seems more difficult to
use this parameter for a quick reference in regulation, since the maximum damage level to
be considered can differ for different structures, as well as the maximum decrement to be
accepted within this damage level.
⎧1 − r (d ) ⎫
Verification : max ⎨ ⎬ ≤ L, ∀ 1 ≤ d ≤ D , L = acceptable value Eq. 15
⎩ d ⎭
Prescinding from the most common case of practice, this parameter can be considered also
from a more theoretical point of view, in case a more extensive investigation of the structure
is of interest or two structure have to be compared in term of robustness: analytically, this
indicator results to be bounded between the value of 1 (less robust structure: it occurs when
a mechanism is developed for the first damage level and the ultimate resistance drops
immediately to zero) and the value of 1/E (most robust structure). The minimum possible
value depends therefore on the number of elements of the structure.
⎧1 − r (d ) ⎫ 1 ≤ Ia ≤ 1
I a = max ⎨ ⎬ , ∀ 1≤ d ≤ D → Eq. 16
⎩ d ⎭ E
This peculiarity could be easily eliminated by referring the damage level to the maximum
number of element of the structure: in this case the index would be bound between the value
of E (least robust structure possible) and the value of 1 (most robust structure). As further
improvement also the difference in step level could be levelled, by interpolating the
obtained points with two curves and moving in this way from a discrete formulation to a
continuous one.
Nevertheless, the consideration of the role of elements numbers doesn’t appear to be
completely unreasonable, as it is a direct consequence of the choice of damage considered,
that has been bounded at an element level, prescinding from their size or the importance in
the structure. This assumption has been justified in the previous section; here it’s just worth
outlining that avoiding the indicator rescaling, the choice to use more elements or replicate
some of it for robustness purpose, could in some way be accounted (let think for example to
the doubling of bridge cables or hanger or to the redundancy of some bracing system). In
this case though, also the replacement of one big element with two smaller one would count
as element increment with a controversial effect on structural robustness.
b. Highest secant variation
The second proposed indicator consider the highest variation of the secant between two
subsequent points, excluding that one corresponding to the null structure (d=E), as better
explained in the following. Also here some continuous curves can be extrapolated from the
given points and in this case, the indicator assumes the meaning of the point of maximum
slopes.
This indicator is focused on the presence of critical element in the structure, since, as the
previous one, it also indicates an abrupt drop of the resistance. In this case importance is
given to the relative, not absolute, decrement of resistance, i.e. the decrement of resistance
referred not to the undamaged conditions, but to the previous damage level.
⎧ [r (d − 1) − r (d)] − [r (d ) − r (d + 1)] ⎫
I b = max ⎨ ⎬ , ∀ d ≤ D < E → 0 < I b < 0.5 Eq. 17
⎩ 2 ⎭
Again, the less robust structure is the one that becomes unstable by means of the removal of
just one element: for this structure the robust indicator assumes the minimum value of 0.5.
The most robust structure instead is the one that shows a linear decrement in the resistance,
giving raise to the maximum value of 0 for the indicator. Consequently, a structure that has
a concave robustness curve could not be said to be more robust of a structure that has a
linear decrement. This is consistent with the definition of robustness, that do not require
nothing more than a proportionate loss of quality. Contrarily, the concave robustness curve
structure would be less robust, since such a trend differ from a convex one just from the
point, in which a more sharp decrement is shown. In the first case the sharp drop off occurs
quite early (i.e. for the first damage level) and then the resistance level remains almost
constant or decrease very little; in the second case, a good resistance level is maintained at
first, but is abruptly dropped later.
This characteristic represents the main difference in comparison with the previous indicator,
that instead distinguishes the earliness or the lateness of an abrupt decrement in the
robustness trend. As can be seen in the figure below, according to the first indicator Ia
structure B would be considered more robust of structure A, where the sharp drop off occurs
earlier, whereas according to the second indicator Ib, both the structure would be considered
equally robust, being the entity of the decrement the same.
λ λ
Ia(A) > Ia (B)
0 1 2 3 4 5 6 0 1 2 3 4 5 6
d d
λ λ
Ib(A) = Ib (B)
0 1 2 3 4 5 6 0 1 2 3 4 5 6
d d
STRUCTURE A STRUCTURE B
Fig. 24: Comparison between the first and second indicator proposed
As a consequence, the indicator Ib is not directly affected by the number of elements in the
structure and doesn’t need therefore any calibration of the damage levels like the one
mentioned for the indicator Ia (even if the element dimensions will still influence the
indicator, since a removal of a bigger element count the same of a smaller one).
Furthermore this indicator could be easily contemplated in the regulation, by establishing
for each main structural type a maximum acceptable value for this indicator. Again, in case
of a performance-based verification, an exhaustive investigation that considers all the
damage levels can be avoided, by fixing a maximum damage level of interest, and
performed only in order to compare different solutions or in case a deeper investigation of
the structural behaviour is of interest.
It has to be pointed out that in case D = E the last point of the curves (corresponding to the
null structure) has to be excluded from the computation of this index. Without doing this, a
structure that maintains a very good resistance level up to the end and then abruptly drops to
zero when all the element are removed) would not be resistant at all. By disregarding the
last unavoidable drop off, only the previous trend is considered and the structure of the
previous example would be considered as very robust.
c. Subtended area
The third proposed indicator is based on the measure of the area of the histograms or the
area subtended by the curves that interpolate the histograms. The bigger the area subtended
by the minima curve, the more the structure is robust. The difference between the area
subtended by the curve of maxima and the one subtended from the curve of minima is
treated as measure of the exactness of the evaluation., since it represents the scattering
between the worst behaviour (considered for the assessment) and the best one.
The scattering area can be explicitly considered in the index by means of a reducing factor k
that assume values in (0,1). In this way, just one parameter is extrapolated, that accounts for
both the curve of minima and maxima, making the comparison between two structures
easier and immediately interpretable.
D
{ [
I c = ∑ r low (d ) + k ⋅ r up (d ) − r low (d )
d =0
]} , 0 ≤ k ≤1 ← 1 + k ≤ I c ≤ (1 + k) ⋅ E Eq. 18
In this case, both the number of elements and the lateness of abrupt decrement affect the
robustness of the structure, since they increase the area subtended to the curve. When
comparing two structures with the same element number, a structure with a concave trend
can be said to be more robust than a structure with a convex trend. Conversely, by
comparing two structure with the same trend, e.g. a linear decrement of resistance, the most
robust one is the one with more elements.
D ⎡ ⎛ d ⎞⎤
I d = ∑ ⎢ r ( d ) − ⎜1 − ⎟ ⎥ ← - E ≤ Id ≤ E Eq. 19
d =0 ⎣ ⎝ E ⎠⎦ 2 2
The indicator is focused on the scattering of the obtained trend from the linear one, i.e. on
the positivity of the second derivative of the curve. Nevertheless, being the indicator based
on the measure of an area, a structure with a longer curve (i.e. a structure that is capable of
undergoing more damage levels) would be favoured in the robustness evaluation than a
structure with a shorter curve.
Also in this case the abscissa could be referred to the total number of element in case the
robustness assessment is to be unbounded from the absolute number of elements.
Furthermore, a unique representation for the curve of minima and maxima could be
obtained, by means of a reducing factor k that determines the portion of the area related to
the curve of maxima has to be added to the area of the curve of minima, in case of a positive
contribution (while it seems more sensible to disregard a negative contribution of the area
related to the curve of maxima). In this case a conventional value of k between 0 and 1
should be a-priori fixed, and maintained equal for all the structures.
e. Upper and lower maximum damage
The last proposed indicator refers to the maximum achievable damage related to the curve
of maxima and to the curve of minima. Even if less handy for engineering practice, this
indicator highlights another important robustness aspect, that is the ability of the system to
last without the presence of several elements: the indicator represents the number of
elements that has to be removed for the triggering of a mechanism. In order to get the
information needed, the analysis should contemplate all possible damage levels, i.e. the
maximum damage level D should be set equal to the number of elements E. Also in this
case a parameter k has been considered in order to obtain a unique representative indicator
for both the curve of maxima and minima.
It’s important to note that this indicator is not always equal to the grade of static
indeterminacy: let us consider for example 4 elements highly connected (all plane 3 d.o.f.)
in a square shape and restrained with two hinge (simply supported system). The system has
3 grades of internal static indeterminacy, but the upper and lower maximum damages are
respectively 2 and 4. In the best case the system can be dismounted piece by piece, starting
from one vertical column, then removing the upper horizontal beam, then the other vertical
column and finally the lower simply supported beam. In the worst case instead, the
sequential removal of both horizontal beams leads to an unstable system.
Further example are reported in the second part of this work, where the relation between the
static indeterminacy and the robustness of a structure is widely discussed and investigated.
Here it’s just worth noting that the discrepancy between the value assumed by the indicator
and the grade of static indeterminacy level depends on the position and acting direction of
external restrains and on the type of connection between elements (hinge or clamped
connection). Furthermore, in case some local mechanism could be accepted and overcome
by an opportune analysis method, the upper maximum damage can be significantly
increased.
element resistance will remain the same, not affecting the computed failure progressions.
When not, the effective structural data have to be recovered from direct tests on the used
material samples, in order to avoid an erroneous estimation of the ultimate resistance and
failure progression.
The estimation of the loads is also not easy usually, since not the design actions are of
interest, but the values of the loads that were acting on the structure at the moment of the
failure. These value are obviously not recoverable form design specifications and need to be
evaluated by considering the plausible loads action on the structure during its service.
It needs to be remarked how difficulties arise from the fact that slight variations in the
estimation of relative loads or material characteristics can lead to different collapse
mechanisms and add up therefore to strong variation in the evaluated robustness response of
the construction.
Here just few words are worth to be spent about the opportunity of considering both these
nonlinearities simultaneously.
If it cannot be avoided to account for material nonlinearities, since yielding and breaking of
the structural elements are the purpose of the investigations, geometrical nonlinearities can
in principle be disregarded or at least limited to second order effects, accepting a slight
deterioration of the solution accuracy. This can be understandable, considering that the
displacements suffered from a structure before collapse are indeed quite large and could
hardly be considered “small” from the point of view of the usual cinematic simplifications.
This approximation would not represent a big deal though, when compared with the more
significant approximations due to the previous exposed estimations and assumptions and
would be mostly justified for the sake of the reduction of computational and modelling
onus.
The use of geometrical nonlinearities should be therefore limited to those cases, where their
disregard would entail an erroneous evaluation of the collapse mechanisms. For example, in
case of a collapse out of buckling, the second order effects cannot evidently be neglected.
The same can also be said if the moment-axial force interaction in a beam is of interested.
In some cases though, the effects of the displacement or bending on the axial stress are
sought. In this case considering second order effects only is not sufficient to catch this
aspect and a large displacement approach should be used. This is the case of some common
collapse mechanism, like the three-hinge mechanism development in a beam of a frame
subjected to a vertical action. In this case, as better explained in the following, the
consideration of the tension-stiffening effect plays a central role in the identification of the
collapse and, what is also very important, in the recognition of the beam rupture.
mechanism and then iterating the procedure used for the investigation of the initial damage
in the first place: the collapsing sections can be substituted with equivalent forces, which
can be applied on the rest of the structure in a further analysis.
• Viscosity
Iterative – incremental
nonlinear static analysis
• Positional loads procedures
Boundary NL • Unilateral restraints Nonlinear FEM solver
• Joint friction
In this section the attention is focalized on the first two types of nonlinearities and on the
analyses types that can account for such nonlinearities. In the application presented in the
third part of this work, the iterative-incremental NL static analysis procedure turned out to
be an useful investigation tool and therefore a brief resume of this kind of analysis is of
interest.
By means of second order effects collapse out of buckling can be evidenced, even if when
the effective post-critical behaviour is needed, large displacement have to be considered
(cinematic is no longer linearized). When a collapse out of buckling is to be modelled,
particular attention has to be paid also to some peculiar type of structural instability such as
the instability of arches. Phenomena like snap-through buckling can be though be also
considered when large displacement are taken into account.
• Equilibrium
• Congruence
• Constitutive law θ θ θ
PE = k/L PE = k · L PE = ½ k · L
Extra flexion due to N P/PE
BIFURCATION
II ORD
POINT
δ N 1
I
M = M - N⋅δ PE
θ
1 1
1 PCR < PE
N N
θ θ θ
In a real frame like the one investigated for the Oklahoma building collapse, the moment
redistribution is generally coupled with an increment in the axial force in the girder, so that
the section at mid-span can enter the plastic range when moment are still negative. Though,
for the sake of simplicity, let’s consider the girder of the example as fully horizontally
restrained. In this case the decrement of negative moment will continue until plastic limit is
reached for positive moment. Then a plastic hinge will form and a three-hinge mechanism
will develop.
At this point the static analysis cannot be continued anymore: the central point of the girder
will start to dynamically fall in vertical direction and huge tensile forces will soon develop
in the two beam now forming the girder. This tension-stiffening effect could at first be
considered to have a positive effect on the structure, since the transition from flexural to
tensile resistance provides the structure with further capacity, though it endows the
structural element of a prolonged structural connection, that can determine the overloading
of other structural elements too.
This mechanism is in fact in most cases a fatal mechanism that could just theoretically end
without consequence and usually leads instead in the best case to a local rupture only and in
the worst cases to the damage of adjoining elements and the progression of failures.
From a theoretical point of view, the beam collapsing sections could fall and get over the
static equilibrium point without the occurrence of any axial breakage of the girder fibres or
exhaustion of the plastic rotation capacity of the plastic hinges. In this case, when a distance
equal to two time the distance between the starting point and the equilibrium point has been
covered, no ruptures can occur anymore and the central joint would oscillate around the
static equilibrium point, where it would finally stop once the dynamic energy has been
dissipated from friction and viscosity.
Practically, one of two rupture (axial in the beams or rotational in the plastic hinges) always
occurs quite early, so at least a local damage in the collapsing area is to be expected. In case
not though, or if a complete detachment of the collapsing section has not occurred, the high
axial forces developed in the girder collapsing sections will be transferred by the residual
structural continuity to the adjoining elements. In a real frame, this would probably
determine the rising of high shear stress in the columns adjacent to the collapsing beam and
start therefore a new column failure that will end with the collapse of the major part of the
structure.
The considered mechanism is discussed also by [Powell, 2004], that notes how the
consideration of catenary action is fundamental when studying building structural collapse.
In the mentioned paper, a focus is also made on the different restraining action provided by
the lateral joints. In fig. 28 the image of the mentioned paper is reported, where the load-
deflection relationship for the considered mechanism (catenary action) is shown for three
different modelling of the problem:
Fig. 28: Load-deflection relationship for 3 models of catenary action [Powell, 2004]
1. Case 1 refers to the simplified model of rigid horizontal restraints at the outer columns
of the frame. The restraints in this example are roughly simplified, since no external
restraints would be present in an actual structure; though this model account in some
way for the anchorage that in a 3-dimensional structure would have to be provided by
the floor slab. In this case the beams have simple plastic hinges, with no P-M
interaction. In this case the beams are in tension at all levels and the resulting curve
shows how the anchorage greatly increases the structure strength and the amount of
energy absorbed, which could mean a substantial reduction in the maximum deflection.
2. In case 2 no horizontal restraint are modelled, i.e. the girders are restrained only from
bending of the columns. The beams have simple plastic hinges as in Case 1 and plastic
hinges can also form in the columns. Here, the beams in the lowest floor are in tension,
and the beams in the upper floor and the roof are in compression. The anchorage
resulting only from the bending stiffness and strength of the columns, provide a much
smaller, but still significant, strength increase with respect to case 1.
3. In case 3 rigid horizontal restraint at the outer columns of the frame are considered, as
in case 1, but the beams have plastic hinges that account for P-M interaction. Due to this
interaction, the axial tension forces reduce the moment strength of the beam, and hence
reduce the basic strength of the structure. Also, the beam hinges extend plastically,
which reduces the catenary effect. The comparison shows that it is easy to overestimate
the catenary effect.
This mechanism is also presented in other studies [Palma et al. 2003; Rahimian&Moazami
2008]. In these papers the effects of catenary action are also extended to the functioning of
two-dimensional panels and floor slabs.
b. Instability in plastic range
Speaking of coupling effects between geometrical and mechanical nonlinearities, a recall
has to be made to the fact, that when vertical bearing columns or compressed elements enter
the plastic range, the critical load then cause these elements to buckle becomes significantly
smaller than the Euler’s load, the elements are usually verified for.
+ σU ET
B σK
σ0 σ0 A
+ +
E
σS
ε θ
A B C
M σ
All 4 type are
possible,
1D lumped hinges
usually work
χ well.
ε
• Full spread
On. preprocessing
• The section law F-δ has to be computed • Just the material law σ-ε must be known
Fast calculation
plasticity
LUMPED
small number of elements and by the absence of other computational challenging aspects
such as load-dropping or geometrical nonlinearities, that maintained a low computational
onus. In the last applications presented instead, progressive collapse investigations are
performed with the avail of lumped plasticity, considered by means of plastic hinges
assigned at discrete points in the structure and whose sectional behaviour was computed
from the user from section properties and material law. This choice is the less onerous for
the program but, as discussed below, requires particular attention from the user in the pre-
processing phase.
a. Lumped plasticity
In case the material law is defined at a sectional level, the section law F-d has to be
computed from section properties and material law and be assigned to the plastic hinges,
that have to be distributed at discrete points along the elements. Plastic deformation is
computed by means of an assigned hinge length, as better explained in the following
section.
In case the material laws are assigned at a fiber level, a less onus is required from the user,
but the section as to be explicitly given to the program, i.e. general section cannot be used.
The plastic behaviour is still considered lumped at some location, i.e. just a defined part of
the element only behaves nonlinear and a fiber hinge length has to be defined. Stress-strain
state is checked at discrete points of the section (whose location and tributary area are
assigned).
b. Spread plasticity
Also in case the material law is defined at a sectional level, the section law F-d has to be
computed from section properties and material law, but in this case the element will behave
nonlinear all his extent long and the plastic deformations are computed by an integration
over the element length. The onus required from the user is smaller than in the previous
cases but the computational onus increases. In case the material laws are assigned at a fiber
level, the onus required from the user is minor, since just the material laws must be given as
input to the program. Also in this case all the elements behave as nonlinear and the stress-
strain state is also checked at discrete points of every element section (whose location and
tributary area are assigned).
length and a fitting placement of the plastic hinge location throughout the elements along
the element length.
HIGH
DISCRETIZATION
Intrinsic loss of objectivity
(due to model choice)
POSSIBLE CONFLICTS due to
Defect of regularization
(due to model calibration)
Fig. 31: Plastic hinge calibration factors: choice of hinge type, length and location.
Problems arise from defining a correct value for this equivalent length, that would lead to a
realistic plastic rotation, as better explained in the following section.
LUMPED PLASTICITY:
PLASTICITY: Plastic hinge ÅÆ M-χ diagram bilinearized LUMPED PLASTICITY:
PLASTICITY: • Ductility is strongly underestimated
1) HINGE TYPE 1) HINGE TYPE
M M SOLUTION
My ∞ • Resistance is duly computed My
1 Plastic benefit is modeled
• Ductility is strongly underestimated
Me Me
Plasticity is lumped in a single point: A more realistic behavior is considered, in order to let the
plastic deformation to develop over the plastic length Lp
χ as result, the plastic deformation can’t develop. χ according to the section structural factor β = My /Me.
χe χ y χe χy χu
Approximation is better for section with a low 1
structural factor β = My/Me (non compact sections) λu P λu P
My My
χy
χe
θ u = θy ! χy
θu = θ y + θp
χu θp = ½(χe + χy) × Lp
χ∞ Lp
PLASTICITY:
LUMPED PLASTICITY: • Ductility is strongly underestimated LUMPED PLASTICITY:
PLASTICITY: • Ductility is strongly underestimated
1) HINGE TYPE 1) HINGE TYPE
M SOLUTION M SOLUTION
Mu
My My
a b 2 Material hardening is considered 3 Hinge “fictive”
fictive” length
Me Me
Not only the amount of hardening influences the Over Lh! An ultimate plastic deformation is computed from an assumed
displacement, but also the form that is assumed for the ultimate elongation εu pertinent to the material. This
hardening: in the second case (b) the plastic rotation is deformation is considered to develop on an EQUIVALENT
χ χ hinge length Lh, leading to a realistic value of plastic rotation.
considerably greater than in the first case (a).
χe χ y χh χu χe χ y χu
3
2b 2a λu P
λu P λu P
My
Mu My Mu My
χy
χy χy
χh χu θ u = θy + θp
θ u = θy + θ p θu = θ y + θp
χu χu θp = (χu –χy) ×Lh
θp = ½(χh + χu) × Lp θp = ½(χy + χu) × Lp Lh
Lp Lp
Fig. 32: Hinge type choices: plastic benefit, material hardening and hinge length.
In order to assume a correct length for the plastic hinge, many expressions can be found in
literature or in current regulations. Most of them, shown in fig. 33, account in some way
also for the load distribution, by means of the parameter z, that represents the distance of the
section to the contraflexure point.
PLASTIC SEGMENT Lp
Structural factor β (i.e. on SECTION PROPERTY)
Sawyer (1964) Lh = ¼ h + 0,075 · z k3 = 0,6 (fc’ =35,2 MPa); 0,9 (if fc’ =11,7 MPa)
fc = concrete compression strength
Mattock (1965) Lh = ½ h + 0,05 · z
fc’ = 0,85 fc
Corley (1966) Lh = ½ h + 0,2 h½ · (z/h) H = section height
h = section effective height
z = section distance to contraflexure point
Lh = 0,08 · z + /60 asl · db · fy
1
z
θ y = χ y + 0,0025 + α sl db = bar diameter (middle value)
Ord. 3274 (2003) 3 (d − d ') f c fy = yielding strength
θ u = θ y + (χ u − χ y )⋅ Lh ⋅ ⎜1 −
⎛ 0.5Lh ⎞ χy = curvature at yielding
⎟ χu = ultimate curvature
⎝ z ⎠
purpose of identifying the effective location of the plastic zone and possibly also in order to
find the effective plastic segment extent, one could be tempted to use a dense distribution of
hinges along elements. This can lead though to possible conflicts due mainly either to an
intrinsic loss of objectivity resulting from the refined mesh assumed, or to defect of
regularization in the calibration of the model. The aspects are better discussed in the
following section, where the main objectivity and regularization problems are presented,
together with some proposed solutions.
when, after entering the plastic range and exhausting the sectional ductility, elements break
releasing the residual resistance stored with plastic deformation. When plasticity is
modelled by means of plastic lumped hinges, the negative branch has to be considered for
the hinge behaviour at a sectional level. Different models can be used, in order to favour the
convergence of the analysis or the realism of the simulation.
Some possible models for plastic hinge load-dropping are discussed in the investigation on
the R.C. framed building presented in the third part of the work, while algorithms that can
be used in order to catch a negative slope in the response curve of the whole structure are
discussed in the following sections. Here it’s just worth outlining that, as said above, the
modelling of material softening or load dropping give rise to several numerical problems
and should be avoided whenever possible.
For example, in case just the ultimate resistance of the structure is of interest, like in the
robustness curve study presented in the first part of the work and tested on some simple
structures in the third one, the ductility of sections can also be considered infinite, since it
won’t affect the structural resistance (provided that no particular situations occur, where the
fragile behaviour of some section prevent the stress redistribution over other elements and
refrain the exhaustion of the resistance stock of the whole structural system).
SOLUTION OBJECTIVITY:
OBJECTIVITY: how the adopted mesh affects the result A
λ
(Bontempi, 1986) λu
A 1 element only: LE = L δ
− Δλ
λF
δy 2 δy δ
−2εy −εy
εy 2 εy
(stable)
− Δλ − Δλ
− σy
δ1y δ1 δ2y - δ1y δ2−δ1
B λ λ
δ1 δ2 λu λu
Homogeneous
(unstable)
− Δλ − Δλ
λF
In case the aim of the investigation is tracing the evolution of the system and following the
failure progression and the chain rupture, the modelling of dropping-load plastic hinges
cannot be avoided and attention has to be paid to the calibration of analysis parameters in
order to favour the analysis convergence.
SOLUTION REGULARIZATION:
REGULARIZATION: how the adopted mesh affects the result
λp
No reduction of event
a
lumping tolerance value
Errors due to
L
Me My Me
b Idealized elastic -perfectly
plastic M-χ diagram
χe χe
Elementary structure: BASIC SOLUTION: one hinge assigned in the middle of the span
• Elastic-perfectly plastic behavior δ Hinge rupture
M Hinge activation
(no hardening) My λu
• Symmetry (Location of max
Me
moment is known)
• Static determined system
χ δ
(just one hinge makes the
χe χ y χu δe δu
structure unstable)
Me
REFINED DISCRETIZATION:
χ
one hinge is assigned in the middle of each element
χe χy χu
In the presence of distributed load, the moment distribution along an element
No reduction of event
a slightly varies around the positive maximum. Being hinges densely
lumping tolerance value distributed along the element, only a pretty low value of the event lumping
tolerance (E.L.T.) allows to recognize which hinge should actually be
activate, whereas a standard value of E.L.T. would let many hinges form
around the point of maximum moment: the apparent plastic zone extent
ductility..
would be completely fictive and lead to erroneous evaluations of ductility
Discretizzazione 1,2,3
Actual Lp
Fig. 35: Solution regularization: cause of error and solution a (E.L.T. value)
SOLUTION REGULARIZATION:
REGULARIZATION: how the adopted mesh affects the result M M
y
Me
REFINED DISCRETIZATION:
χ
one hinge assigned to each element, hinge hardening behavior is considered
χe χy χu
Despite the high meshing (being the E.L.T. correctly calibrated) the plastic
b Idealized elastic -perfectly zone result to be localized (only one plastic hinge activates), due to the
M-χ diagram
plastic M-χ idealized diagram assumed. In order to obtain the effective extent of plastic
zone, an hardening behavior has to be considered for the plastic hinge.
SAP2000 analysis
With hardening segment Without hardening segment
Actual Lp Actual Lp
SOLUTION REGULARIZATION:
REGULARIZATION: how the adopted mesh affects the result My
Me
REFINED DISCRETIZATION:
χ
one hinge is assigned in the middle of each element
χe χy χu
Hinge length has to be modified depending on the hinges that will form:
c No regularization of the when only one hinge activates, would be meaningless to assign an hinge
hinge length length regularized on the assumed mesh (ductility would result strongly
underestimated). On the contrary, if hardening is considered and more hinges
activate, the length assigned to each hinge has to be regularized, otherwise the
ductility will result strongly overestimated (Lh = LE / #HE).
SAP2000 analysis
Effective hinge length =
Curve
Curve di pushover
pushover alal variare
di capacità variare del
del n°
n° di
di elementi
elementi
= # active hinges × single hinge length 180
180
160
160 22
Trave semplicemente appoggiata - sezione HEB240 44
140
140
lunghezza trave [mm] 4000 88
120
120
lunghezza cerniera [mm] 360 (Bresler, Lin, Scalzi; 1968). 16
16
100
100 11 22 44 66 1122
Discretizzazione
c arico
32
o
32
caric
ccer
ernn cceerrnn cceerrnn ccer
ernn cceer
r nn
80
80 64
64
numero di elementi 2 4 8 16 32 64 128
128
128
60 11 22 44 66
lunghezza elemento [mm] 2000 1000 500 250 125 62,5 31,3 60 12
12
teorico
teorico
40
numero cerniere formate 2 2 2 2 4 6 12 40
20
zona plasticizzata [mm] 2000 1000 500 250 375 312,5 343,8 20
00
lunghezza tot. cerniere [mm] 720 720 720 720 1440 2160 4320 00 10
10 20
20 30
30 40
40 50
50 60
60 70
70 80
80 90
90 100
100 110
110 120
120 130
130 140
140 150
150 160
160 170
170 180
180 190
190 200
200 210
210 220
220
lung.tot. / lung.teorica 2 2 2 2 4 6 12 freccia
freccia
specific ground motions compatible with the seismic hazard spectrum for the site must be
simulated, task that is not easy to be accomplished.
the second one is the computation demand of a nonlinear time-history analysis, that,
notwithstanding the significant increase in computing power witnessed in recent years,
remains very high, especially when distributed inelasticity is considered for complex
structure with many elements (let consider a long bridge or a tall building or even a
multy-story 3D building like the one investigated in the third part of this work);
thirdly, errors in the definition or assemblage of a finite elements model are difficult to
detect from dynamic analysis results, whilst they tend to be relatively evident from the
output pushover runs.
This difficulties are less compelling in a static iterative incremental analysis, that is also
useful for different purposes. Static analysis provides very important structural response
information such as [Pinho 2007a]:
a. identification of critical regions, where large inelastic deformations may occur;
b. individuation of strength irregularities in plan and elevation that might cause important
changes in the inelastic dynamic response characteristics;
c. evaluation of the force demand in potentially brittle elements;
d. prediction of the sequence of yielding or failure of structural members.
In addition, the explicit insight that pushover-derived base shear vs. top displacement
capacity curves provide into the stiffness, strength and ductility of a given structure,
constitutes the type of qualitative data that is always most informative and useful within a
design application, even when time-history analysis is then employed for the definitive
verifications.
This aspect is particularly important in view of a robustness investigation, since when just
the point of structural collapse is of interest, like for the construction of the robustness curve
presented, a static incremental analysis allows the immediate identification of the structural
response by means of an expressive and easy to compute parameter, that is the ultimate
resistance. The use of a dynamic analysis for the robustness curve assessment would be not
much difficult to implement in the algorithm written for the sequential removal of structural
element, but would entail some questions in the post-processing phase about the parameter
to use in order to identify the structural capacity.
intensity levels [Pinho 2007b]. In fig. 37 the response of a structure is shown, as can be
obtained by means of a much more complex and time-consuming incremental dynamic
analysis procedure, where a structure is subjected to a series of nonlinear time-history
analyses of increasing intensity with the objective of attaining an accurate indication of the
“true” dynamic response of the structure
Traditional pushover analysis is performed subjecting the structure to monotonically
increasing lateral forces with invariant distribution until a target displacement is reached.
Both the force distribution and target displacement are hence based on the assumption that
the response is controlled by a fundamental mode, that remains unchanged throughout.
However, such invariant force distributions cannot account for the redistribution of inertia
forces caused by structural yielding and the associated changes in the vibration properties,
including the increase of higher-mode participation. In order to overcome such drawbacks,
different solution have been developed.
An alternative type of nonlinear static analysis involves running multiple pushover analyses
separately (Multi-modal push-over). Each running corresponds to a given modal distribution
and the structural response is then estimated by combining the action effects derived from
each of the modal responses (i.e. each displacement force pair derived from such procedures
does not actually correspond to an equilibrated structural stress state).
A further refinement of such multiple-pushover procedures, with the aim to account for the
alteration of local resistance and modal characteristics of the structure induced by the
frame can in this way remain undetected and leads therefore to a complete erroneous
evaluation of the collapse failure, but also of the structural capacity. That is the reason
why, as aforementioned, a displacement-based pushover can be sensible just in case
some adaptive pushover techniques are used.
3. In the third case a response controlled analysis is instead used. The incremental analysis
is always force-based, i.e. the load vector is incremented like in case 1, but the load
increment is computed starting from the displacement control of a given joint of the
system. That means, that only one displacement value (and not all the displacement
vector like in case 2) is imposed; then the corresponding load multiplier is computed
and applied to the force vector. The computed load multiplier can result to be also a
negative value, thereby allowing to evidence the softening branch of the response curve
without imposing the collapse mode.
NL STATIC ANALYSIS F λ i F2
Force control
λ i F1
Δλ i
Load (vector) is incremented:
softening branch is not caught. δ
F λ i u2
Displacement control
λ i u1
Displacement (vector) is incremented:
collapse-mode is imposed, not found!
δ
δTARGET
Δλi
Response control F λ(Ui) F2
δi
Δλi
A control node displacement (value) λ(Ui) F1
U is incremented (up to a target):
the load multiplier λ(Ui) is computed
and applied to the force vector. δ
δTARGET
neglected, since it seems less pertinent to the aim of robustness assessment and progressive
collapse studies.
Fig. 39: Newton-Raphson (left) and modified Newton-Raphson (right) methods [Pinho 2007 a]
Different convergence check schemes, which may make use of three distinct criteria
(displacement/rotation, force/moment, energy based), can be employed to check the
convergence of the solution at the end of each iteration. The displacement/rotation criterion
provides a direct local control over the precision obtained in the solution of the problem,
usually ensuring overall accuracy. The force/moment criteria are suggested if a
displacement convergence check is not sufficient for the internal forces of the elements to
be adequately balanced. The maximum accuracy and solution control is obtained by
combining the displacement and force convergence check criteria, while the maximum
analysis stability is obtained if convergence is achieved for one of the two criteria checked,
at a price of lower analytical precision. Tolerances in convergence criteria should be
carefully defined.
The solution algorithm may feature a hybrid incremental algorithm, obtained from a
combination of the Newton-Raphson and the modified Newton-Raphson procedures,
whereby the stiffness matrix is updated only in the first few iterations of a load step, thus
obtaining an acceptable compromise between velocity in achieving convergence and
required computational effort [Cook et al. 1989].
Fig. 40: Arc-length method with generic constraint surface (a) and orthogonal iterative changes (b)
[Wempner 1971, Risk 1972]
Two approaches, fixed arc-length and varying arc-length are generally used. In the former
the arc-length is kept fixed for current increment, whereas in the latter case, new arc-length
is evaluated at the beginning of each load step to ensure the achievement of the solution.
Simplification of the constraint equation leads to a quadratic equation, whose roots are used
for determining the load-factor. Proper selection of root is one of the key issues of the
method, whose details will be discussed in subsequent sections. Generally, for the first
increment, the trial value of the load-factor is assumed as 1/5 or 1/10 of total load. For
further increments the load-factor is computed according to the rate of convergence of the
solution process. In case of divergence from the solution path, the arc-length is reduced and
computations are done again. The computation time of the solution process is also of major
concern in finite element analysis, so a maximum number of iterations are preset; and if
solution does not converge in the specified number of iterations then the load step is
reduced and the process is started over again.
In the first part of this work, a method has been presented for the measure of the robustness
degree of a structure, that is based on the investigation of the response of all possible
damaged configuration of the structural system. As previously mentioned, problems arise
from the fact, that the number of investigations to be performed in order to obtain the exact
robustness curves increases exponentially with the number of elements: this is not just a
speculative issue, since it involves the computational feasibility of the problem.
As better discussed in the following, some kind of problems requires an effort in term of
time or memory, that makes impossible to find a solution for them, even if more and more
fast and powerful computer were at hand. In order to better explain this concept and define
the computational complexity for the problem of robustness curves, some basilar concept of
are presented in the following.
This example explains that even wide jumps ahead in computational power don't transform
unfeasible problems in feasible ones. In this respect, some brief concepts from theoretical
computer science are briefly recalled in the following sections, in order to define the
complexity related to the problem of robustness evaluation, as introduced in the first part of
this work.
where:
- γ is the tape symbols alphabet;
where:
- δN is the (partial) transition function defined as:
δ: (Q-F) × (γ ∪{Ь}) Æ P (Q × (γ ∪{Ь})× {r, l, i})
It's worth noting that the transition function of the NDTM can drive the automaton on a set
of states with various configurations of the tape and head. This means that a NDTM can
operate various computations in parallel at the same time.
The focus of this paragraph will be the decision problems. A decision problem is a
particular case of the more general optimization problems. Optimization problems will be
discussed later in the thesis, while for the sake of brevity, decision problems will be referred
simply referred as problem in the following of this section.
A decision problem can be informally defined in this way: the input of the problems is a
structure and an integer (e.g. a graph and an integer k); the goal of the problem is to answer
the question: does exist a solution for the problem, whose measure is less (greater) than k?
Now it's possible to define some complexity classes for what concerns the computational
time. It's necessary to point out that given a polynomial:
m
[
p( x ) = ∑ a k ⋅ x k
k =0
] Eq. 23
only the monic term xm will be taken into account. For a deeper understanding of why this
suffices, refer to [Ausiello et al. 2003b]
A problem p belongs to the class DTIME[f(n)] if and only if there exists a DTM (or, and it's
the same, a deterministic algorithm) that, for any possible input instance of dimension n,
solves the problem in time f(n). The complexity class P is defined as:
P = U∞ [ k
k =1 DTIME( n ) ] Eq. 24
This means that P is the class of all the problems solvable by deterministic polynomial time
algorithms for any given input instance.
A problem p belongs to the class NTIME[f(n)] if and only if there exists a NDTM (or, and
it's the same, a non deterministic algorithm) that, for any possible input instance of
dimension n, solves the problem in time f(n).
The complexity class NP is defined as:
NP = U∞ [ k
k =1 NTIME( n ) ] Eq. 25
This means that NP is the class of all the problems solvable by non deterministic
polynomial time algorithms for any given input instance.
One important question is if it’s possible to solve a problem p in NP using a deterministic
algorithm, and how much time is needed in order to find the solution. The answer is that it's
always possible to solve p in exponential time using a deterministic algorithm, since it's not
known any deterministic polynomial time simulation of a NDTM using a DTM. This
doesn’t mean that it's necessary to use exponential time to actually solve a problem
belonging to NP, but that no deterministic algorithms are known capable of solving such a
kind of problems in polynomial time.
Before going further in the complexity matters, an introductory description of the problem
of solving problems in NP can be useful. Suppose that, given a certain number of cities and
roads connecting each pair of them, one road per couple of cities, along with the road's
length, one is interested in computing if there exists a tour that touches all the towns one and
only one time and that has a total length smaller than a certain distance.
This problem, better known as traveling salesman problem (TSP), looks like a very simple
and natural one [Campagna, 2007]. But thinking carefully about an algorithm solving it,
would probably bring to an enumeration of a certain number of possible paths before
finding one that satisfies the constraint. Suppose that no one of such a cycle exists: the
algorithm would output all of the possible cycles before answering “no”. The number of
existing cycles, for an arbitrary set of towns, can be exponential in the cardinality of the set.
This means that even for small sets, the number of computational steps necessary to answer
the posed question would be unfeasible for any computer existing in the world.
A subclass of NP problems are the so called NP-complete problems. These problem are
among the most complex ones because of their inner mathematical structure. NP-complete
problems are very important, both for the theoretical challenge they pose and for the
practical importance they have.
In order to define this kind of problems, it's necessary to introduce the concept of Karp-
reduction between problems. A problem p is Karp-reducible to a problem q if and only if
there exists a Karp reduction, that is an algorithm A that transforms every instance x of p in
an instance y of q such that x belongs to positive instances of p (instances with a solution) if
and only if y belongs to the positive instances of q. If p is Karp reducible to q, than the
notation p ≤ q.
For problems in NP, the interesting Karp-reductions are those carried out in polynomial
time, so in the following p ≤ q is to be intended as a polynomial time Karp reduction.
A problem p is NP-hard if and only if for any given q in NP, q ≤ p.
A problem p is NP-complete, if and only if p belongs to NP and it’s NP-hard.
A non-exhaustive list of known NP-complete problems is reported in the following.
- Satisfiability problem (SAT) - Subset sum problem
Just as a notice, it's worth to point out that the question whether P = NP or not is an open
one. There's a consistent prize for anyone who will provide a solution to this question, but
even if it would be sufficient to find a polynomial time algorithm for just one specific NP-
complete problem in order to prove that P = NP, because of the property of NP-
completeness, it seems unlikely that someone will ever be able to provide such an answer.
For these kinds of problems new complexity classes and new reductions have to be defined,
even though they're mainly an extension of the previous ones.
Before doing that it's necessary to define a problem in a formal way: a problem p is
described by the following 4-tuple :
where:
- I is the set of input instances for p;
- SOL is a function that associates to any input instance x in I the set of feasible solutions
of x;
- goal specifies whether P is a minimalization or a maximalization problem.
A problem P belongs to the class NPO if and only if the following holds:
1. the set of instances I is recognizable in polynomial time;
2. a polynomial q exists, such that, given x in I, for any y in SOL(x), |y| <= q(|x|) and
furthermore, for any given y such that |y| <= q(|x|), it is decidable in polynomial time
whether y is in SOL(x);
3. the measure function m is computable in polynomial time.
As said before, an optimization is, somewhat, an extension of a decision problem, and for
any given optimization problem p holds that if it belongs to NPO, the corresponding
decision problem belongs to NP.
A class for “simple” problems exists in the optimization framework, as well as for
“complex” problems: a problem p belongs to PO if and only if an algorithm A exists, that
computes in polynomial time an optimal solution y in SOL(X) together with its value m(x),
for any given x in I.
At this point it's necessary to introduce the concept of Turing reduction. A precise definition
is out of the scope of the present work, so let’s consider it for sake of simplicity an
extension of a Karp reduction. In the following the syimbol ≤ means Turing reducible.
Given the Turing reductions, it's possible to introduce the concept of NP-hardness (in an
optimization scenario): an optimization problem p is called NP-hard, if and only if for every
decision problem q in NP, q ≤ p. Since this definition is not an operative one, it's worth to
point out that, roughly speaking, an optimization problem P is called NP-hard, if it's
associated decision problem is NP-complete.
2 4
2 3 4 3 4 4
1 1 1
3
4
3 4 4 4 Computational tree of a
non-deterministic Turing
3 3
machine:
3
4
all possible configurations
d=3 2 2 4 2 4 S123 S134 S134 S134
are computed in
1 1 1 polynomial time
3 4 (# computational steps s =
# system elements E).
d=4 2 4
S1234
1
Fig. 41: Computational complexity for identifying all damaged configuration of a structure.
Given the above consideration on the problem complexity, an exact solution cannot be
provided, since it would be unfeasible to perform all the required investigations. A
a random variable defined over the set of possible choices of the algorithm [Ausiello et al
2003b].
It’s possible to identify two fundamental classes of probabilistic algorithms: the first one
comprises the so called Monte Carlo algorithms, that represents the most used method and
has a very fast running time, but cannot guarantee the exactness of the solution (i.e. it
produces an answer that is correct with a bounded probability); the second one is composed
by algorithms called Las Vegas, that may not return an answer at all (i.e. the algorithm
doesn’t terminate), but in case a solution is provided, it is guaranteed to be correct. The
latter group presents also the inconvenient the time required for the solution tit’s not ensured
to be sustainable (i.e. the runtime for each input is a random variable whose expectation is
bounded).
In the framework of algorithms the so called search algorithms are the ones that operate a
scan of the state space, looking for a particular one that minimizes or maximizes a certain
measure. This algorithms can mainly be of two major classes: informed search algorithms,
both deterministic and randomized, relying mainly on heuristics, and local search
algorithms, that instead of searching through the search space and exploring paths from an
initial state like the former do, evaluate and modify a state that is completely known (the
same distinction actually apply also for deterministic algorithms).
Referring to the former class, example algorithms are given by the greedy best search first
A* and variants.
With respect to the second group, the main aspect of local search algorithms is that they are
instead not interested in computing a path to the goal. The only interesting thing is the state
configuration that has a certain measure. The focus is then on the present state and the
technique used is a modification of the state itself. This kind of algorithms have two main
advantages:
1. use of little memory, often constant (the state needs to be tracked);
2. they can usually find reasonable solutions for large (or even infinite, continuous) state
spaces. These algorithms are suitable for solving pure optimization problems, since the
goal is to find the best state according to an objective function.
An interesting example for local search is the so called “state space landscape”, in which the
x-axis contains the state configuration and the y-axis represent the measure. Observing the
graph reported in fig.42, it’s clear how moving through spaces makes the measure change.
Among the local search algorithms, it’s worth to number the hill climbing, which uses kind
of a greedy approach, local beam search, which proceeds keeping track of k parallel states in
spite of using just one, genetic algorithms, which relies on a set of evolving rules in order to
compute successive states and simulated annealing, which will be covered in the following.
While genetic algorithms as well as simulated annealing, are randomized algorithms, both
hill climbing and local beam can be used in deterministic or stochastic versions.
objective function
shoulder
plateaux
local minimum
state space
As better explain in the following, in this work the simulated annealing technique has been
deepened, in order to solve in an efficient way the robustness curve problem, while a
different approach that avails the use of genetic algorithms can be found in [Bolognesi,
2000].
Iterative search methods are generally useful in all cases in which some combinatorial
structures have to be optimized, by means of a cost function. Usually in these cases, the
search among all possible structures is untreatable and deep first search approaches are too
expensive.
Since there is no know algorithms for finding the optima solution efficiently, the iterative
improvement methods is to be used: the basic intuition consist in considering the
configurations to be laid on a surface and aiming to find the highest point. All these
methods start at a random configuration and repeatedly consider various move, accepting or
rejecting some of them, until getting stuck and restarting. A move-set should be defined that
describes what moves have to be considered from any configuration.
proceeds from state s(j) by choosing a trial state r and comparing the energy at r, f(r), with
the energy at the current state, f[s(j)]. If the energy of the new configuration is lower than
that of the previous one. This means that if f(r) < f[s(j)]), the new configuration is
immediately accepted; if the new configuration has a larger energy, the new state is
accepted with some nonzero probability that depends on the amount of the increase in
energy, f(r) − f[s(j)]. In simulated annealing, the probability of moving to a state with larger
energy also depends on a “temperature” parameter T. When the temperature is high, the
probability of acceptance of any given point is high. When the temperature is low, however,
the probability of accepting any given point is low.
Although the technique is heuristically related to the cooling of a metal, as in the original
application, it can be used in other kinds of optimization problems. The objective function
of the general optimization problem is used in place of the energy function of the original
application. Simulated annealing is particularly useful in optimization problems that involve
configurations of a discrete set, such as a set of particles whose configuration can
continuously change, or a set of cities in which the interest is an ordering for shortest
distance of traversal (TSP). Various application of SA can be found in literature and the
method has become widely used.
To use simulated annealing, a “cooling schedule” must be chosen; i.e. the initial temperature
has to be fixed and the type of temperature decrement with the time has to be decided too.
If T(0) is the initial temperature, a simple schedule is T(k + 1) = qT (k), for some fixed q,
such that 0 < q < 1. A good cooling schedule depends on the problem being solved, and
often a few passes through the algorithm are used to choose a schedule. Another choice that
must be made is how to update the state of the system; that is, at any point, how to choose
the next trial point.
The probability of accepting a higher energy state must also be chosen. Although this
probability could be chosen in various ways, it is usually taken as:
⎡f ( r ) − f (s ( j) )] / T ⎤
P = e ⎢⎣ ⎥⎦
Eq. 27
certain number of state changes have been made at the given temperature even if the limit
on the number of trial states to consider has not been reached. Finally, there must be some
kind of overall stopping criterion.
The main steps of a simulated annealing algorithms have been reported by [Gentle 2003] as
follows:
1. Set k = 1 and initialize state s.
2. Compute T(k) based on a cooling schedule.
3. Set i = 0 and j = 0.
4. Generate state r, and compute δf = f(r) − f(s).
5. Based on δf, decide whether to move from state s to state r.
If δf ≤ 0, accept and set i = i +1; otherwise, accept with a probability P[δf, T(k)].
6. If i is equal to the limit for the number of successes at a given temperature imax, go to
step 1.
7. Set j = j + 1. If j is less than the limit for the number of iterations at the current
temperature, jmax, go to step 3.
8. If i = 0, deliver s as the optimum;
otherwise, if k < kmax, set k = k + 1 and go to step 1;
otherwise, issue message that “algorithm did not converge in kmax iterations”.
The travelling salesperson problem (TSP) can serve as a prototype of the problems in which
the simulated annealing method has had good success. In this problem, a state is an ordered
list of points (“cities”), and the objective function is the total distance between all points in
the order given plus the return distance from the last point to the first point. One simple
rearrangement of the list is the reversal of a sub-list; that is, for example,
(1, 2, 3, 4, 5, 6, 7, 8, 9) Æ (1, 6, 5, 4, 3, 2, 7, 8, 9).
Another simple rearrangement is the movement of a sub-list to some other chosen point in
the list; for example,
(1, 2, 3, 4, 5, 6, 7, 8, 9) Æ (1, 7, 8, 2, 3, 4, 5, 6, 9).
prevents the search to be stuck in a local minimum and makes therefore SA particularly
suitable for discrete combinatorial problems, where high variation of values corresponding
to subsequent states are usual (see fig.44a).
As the cooling procedure proceeds and relatively low temperature are reached, worse
solution are less frequently accepted and the algorithm tends to behave as a steepest descent
algorithm, where the search of the optimum is localize in a subset of neighbouring states.
This good capability of bypassing points of local minimum has induced to chose the SA as
optimization algorithm to reduce the number of damaged configuration to be investigated in
the proposed method for the robustness evaluation of a system. In the following figure, the
flowchart of a SA algorithm for a generic optimization problem is shown (on the right) and
its translation for the specific structural problem of interest is reported next (on the left).
N
r ≤ Px N
r ≤ Px
N N
Y halting condition Y
t ≤ tmax t ≤ tmax halting condition
STOP STOP
Fig. 43: General SA algorithm and application to the considered structural problem
In order to apply the SA to the problem of robustness curves, three main aspects has to be
considered, that are summarized in the following.
strong
reduction
0 11 2 2 3 34 54 d damage level d
3
1. Chose an arbitrary numeration of elements
λui d=2 2 4
2. For each damage level d, start with a number
that has d sequential ciphers (startinf from 1) 1
given d Æ C1 = 1 2 3 … d (starting configuration)
The considered procedure for generating perturbations represents the most simple
identification of move set and is based on a organization of the damaged configurations,
which depends on the arbitrary enumeration of the elements composing the system.
However, as better explained in the following section, the choice of move set is a central
point in simulated annealing algorithms: the individuation of a convenient move set is the
real ingenuity, more that the decision of using one algorithm instead of another. Other
critical points can be the evaluation of function design and the choice of the annealing
schedule and other parameter of the algorithm, which are briefly discussed in the following.
c. Calibration of SA parameter
With respect to the calibration of SA parameters, three main aspects can be identified: the
choice of a proper cooling schedule; the definition of a criterion for the increment of
temperature; and the identification of an halting parameter.
As can be easily understood, the choice of the cooling schedule influences the ability of the
algorithms to find a good solution and it’s therefore quite important. It is usual to
distinguish three possible cooling schedule, reported below according to an descent order of
steepness of the curve. For the usual application anyway, the first expression is usually
used, which has been also considered for the application to the proposed structural problem.
T ( t ) = α ⋅ T ( t − 1) , where : 1 > α ≥ 0,8
α
T( t ) = T(0) ⋅ ⎛⎜1 − t ⎞
⎟ , where : 1≤ α ≤ 4 Eq. 28
⎝ t max ⎠
of the problem moves from an exponential to a polynomial form, i.e. the robustness
investigation can actually be performed. To better explain this method, let consider as
example a structure of 8 element, like the one exemplified in fig. 47.
An exhaustive investigation should at first perform a push-over analysis on the integer
structure (d = 0); then all the 8 configuration where an element is missing have to be
analyzed for damage level d =1; afterwards, all the 28 configurations without 2 elements for
d = 2, and so on up to the maximum damage D. In case this maximum damages has been
fixed equal to the number of the elements, the last damage level corresponds to the unique
configuration of the null-structure (i.e. a structure without element, that has zero resistance)
and the total number of analyses is 2E. that for the structure considered correspond to 256
combinations.
d>1
= 2 ×[E-(d-1)]
for D = E
D
Ctot = Σd Cd = E2 + 1
1 0
polynomial
Exhaustive combinations
E!
Cd =
d! × (E-d)!
ERROR! exponential
D
(non cons.) Ctot = Σd Cd = 2E
0 1 2 --- 8 0
d for D = E
investigated reduces while the damage increases (6 configuration for d = 3, 5 for d = 4 and
so on) and the total number of combination becomes:
D
C min/ max = 1 + E + ∑ [E − (d − 1)] Eq. 29
d =2
which is equal to 37 for the structure with 8 elements considered. It important to point out
that, contrary to the exhaustive investigations, the procedure has to be performed separately
for the curve of maxima and for the curve of minima, being the optimization different
depending on the optimum type. The number of total combination for obtaining both curves
is therefore given by the following expression:
D
C tot = 1 + E + 2 ∑ [E − (d − 1)] ⎯⎯⎯→ C tot = E 2 + 1
D=E
Eq. 30
d=2
For the considered structure taken as example, the total number of reduced combinations to
be investigated results to be equal to 65, which is a significant reduction when compared
with the number of 256 combinations calculated above whit respect to the exhaustive
investigation (ca. 75%). This reduction increases as percent value with the number of
composing structural elements: if we consider a structure with 16 elements, the number of
exhaustive combinations assuming D = E is equal to 65636, whereas the number of reduced
combinations is 257 corresponding to a reduction of more than 99%. In the following table,
the number of exhaustive and reduced combination is reported for a number of structures.
Observing the table, it is interesting to notice that for the first structure, the number of
heuristic-based reduced configurations in greater than the number of exhaustive
combinations. This apply to structures with a number of elements E ≤ 4 and is due to the
fact that, as mentioned before, the considered optimization requires to be performed
separately for the search of the curves of maxima and minima. This fact makes the
optimization inefficient for structure that have a very low number of elements.
Another peculiar aspect can be observed by comparing the number of exhaustive and
reduced combinations related to a maximum damage level taken equal to the half of the
element number: in case of exhaustive investigations, the total number of combinations for
D= E / 2 is of the order of half of the combinations for D = E (it is the exact half when E is
odd and D = (E-1) / 2, which correspond to the half of the maximum possible damage
levels, recalling that the first damage level is equal to zero); when the considered
optimization instead is used, the number of combinations for D = E / 2 is significantly
greater than the number of combinations for D =E. This fact can be better understood by
observing the graphics of fig. 48: the trend of the heuristically reduced combinations with
the damage don’t respect the symmetry about the middle damage value, contrarily to what
happen to the trend of the exhaustive combinations.
This fact assumes importance from the consideration that complex structures like high
building or long bridges have usually a significant number of elements (say about half
thousand): in these cases, the computation of the robustness curves for all possible damage
levels (D = E) even if computationally feasible when the heuristic optimization is used,
requires a very high number of investigations (about two hundred thousand). On the other
hand, it seems reasonable to investigate the robustness of such complex structures just up to
a number of damage levels (say D < E/2), being the behaviour of a heavily damaged or
almost destroyed structure not of interest in the practice. For this reason, a stronger
reduction of combinations in the first half of the maximum damage level D, instead of in the
second half, would be more profitable, even though not compelling.
removed (numbered sequentially from 1 to E). Among the E value of the ultimate
resistance, both the maximum and minimum values are saved for the maxima and minima
curve respectively and the element identification numbers corresponding to these
configurations are registered.
The damage level can be now incremented again. From this point on, the procedure repeats
all the same for the following damage levels. First the analysis for the curve of maxima are
performed, then the integer structure is restored and the analysis for the curve of minima is
then started. The two algorithm macros are very similar and only differ from the fact that in
one case the maximum value of resistance is sought among those corresponding to the same
damage level, while in the other case the value sought is the minimum one.
for e =1 to E for e =1 to E
Remove element e
Remove element e
if e <> edMIN Pushover analysis λu,ed
Pushover analysis λu,e1
Restore element e λudMIN,
Restore element e edMIN
λu1MAX, e1MAX
CALL “CURVE MAX”
λu1MIN, e1MIN
MAX” Curve of minima
In the figure above, the macro related to the minima curve is shown: for each damage level,
the element registered in correspondence to the previous minimum found is removed. In this
way the minimum configuration of the previous damage level is restored. After that, the
other elements are removed one by time and each time a pushover analysis is performed,
then each element is restored. Among the analyses performed, the minimum value of the
resistance is found and the corresponding element removal is registered, so that it can be
removed to restore the minimum configuration found for the analysis of the following
damage level.
Some practical applications developed with the avail of this implemented algorithm are
shown and discussed in the third part of the presentation, while the code of the program is
reported and commented in the first Appendix.
The analysis parameter can be calibrated by the user, by means of a proper macro that
require input from the user. Then the analysis on the integer structure can be repeatedly
launched and the graphical full result can be reviewed. These results include the stress and
strain distributions in the elements, that will be showed step-by-step following the trend of
the incremental analysis, as well as the so-called pushover curve (force-displacement curve).
By reviewing this result the user can see if the analysis was correct and go on with the
analyses of damaged structures or change the analyses parameter. In this case a new analysis
on the integer structure will be performed until the analyses parameter are correctly
calibrated.
Let consider for example the structure represented in the following figure: the curve of
maxima show a final increment in the resistance, when element 2 is removed after element 3
and 1, for the third damage level. This is due to an erroneous evaluation of the structural
resistance of the configuration chosen as maximum for the second damage level, where only
the two the vertical elements 2 and 4 are present. In this case the structure is a mechanism
and therefore this configuration should not appear in the maxima curve, having a null
ultimate resistance. The program though calculate a non-null resistance for this
configuration, since it, being present only vertical force and no imperfections, it consider the
structure stable in the direction of interest.
Hence, the modelling of imperfections at least in the vertical elements is generally required
to avoid incorrect results, even if geometrical effects are not considered. In the first figure
below, the problem has been overcome by considering a slight inclination on the vertical
elements. The inclination considered is opposite for the two columns, in order to maintain
the structural symmetry. This measure will generally work fine for most structure, even if
some cases can occur, in which the imperfection of orthogonality cannot work: this is the
case for example of a structure like the one represented in the second figure below, where
all the node of the vertical elements are restrained. Here again, the structure without
imperfection will show an uncorrected trend of the maxima curve, since the ultimate
resistance corresponding to the third damage level and to a configuration where only the
vertical columns are left is greater than the resistance corresponding to the second damage
level. This is due again to the fact that the structure is not recognised as unstable: in order to
let the program consider this fact, two small horizontal force have been applied to the nodes
on the top of the columns, since the element inclination would not affect the restrained
nodes.
It is worth to be noted, that would the structure be fully restrained in all the nodes (i.e. with
hinges instead of movable supports), no kind of imperfection modelling would resolve the
problem of a monotonic response. In this case though, the damaged system composed by
the columns only would actually have a great resistance. This example falls therefore in the
case of local mechanism problems, since the non monotonic response can be addressed to
an underestimation in the resistance of the previous configuration, more than to an
overestimation of the configuration with the two columns only.
The integer structure for example will be considered unstable when all plastic hinges have
occurred in the horizontal and diagonal members. In this situation, the vertical elements are
though still in the elastic range and would react to further increment of the vertical acting
forces.
λg
2 4
without 1
slight inclination of
imperfection vertical elements
PU [kN]
PU [kN]
1
1
3 2
0,75
1 3
0,5
0,5
4
0,25
4 4 2
1 4 1
1
0
0 1 2 3 4 0 1 2 3 4
Damage Level Damage Level
Fig. 48: Overcoming of problems due to disregarding the imperfection modelling (a)
λg
4 2
5
without slight horiz. forces
imperfection 1 on upper supports
I3V STRUCTURE ROBUSTNESS MAX MIN I3V STRUCTURE ROBUSTNESS MAX MIN
PU [kN]
PU [kN]
5
1
5 1 2
1 1
3
1 3
2
0,5
0,5
4 2
4 4 4
3 3
5 5
0
0 1 2 3 4 5 0 1 2 3 4 5
Damage Level Damage Level
Fig. 49: Overcoming of problems due to disregarding the imperfection modelling (b)
and especially the fourth indicator consider positively a concave decrement with respect to a
linear one. The last indicator instead, measuring the endurance of the structures to damages,
is therefore not influenced by this characteristic of the response.
In order to test the functioning and expressivity of the proposed indicators, their values are
reported in the following, referring to the rough approximation of both histograms with a
linear decrement.
a. Decay: Ia = 1 Å 1 Ia 8 (for both histograms)
b. Abrupt decrement: Ib = 0 Å 0 Ib ½ (for both histograms)
c. Subtended area: Ic = 4 Å 1 Ic 8 (for each k)
d. Scattering: Id = 0 Å -4 Id +4 (for each k)
e. Survival: Ie = 8 Å 1 Ie 8 (for both histograms)
Damage
# of elements: MIN CFG
E = 8MAX CFG ROBUSTNESS INDICATOR:
d El. ID Pumin El. ID Pumax
static Ia maximum secant starting from the 1st pt. (account for the
0 indetermin.
0 level:
1 i = 210 1
1 1 0,831745 3 0,896185 lateness of an abrupt decrement)
lower max. damage: D =8
2 5 0,648331 min7 0,832096 ⎧ r (d ) ⎫
upper I a = max ⎨ ⎬ ∀0≤d≤E →1 ≤ I b ≤ E (isostatic)
3 max.4damage: 0,523591 Dmax2 = 8 0,684849
⎩d E ⎭
4 max. damage:
fixed 8 0,376882 D =88 0,580928
5 2 0,269581 6 0,429733
Ib maximum slope pt. tangent (measure the abrupt decrement
exhaust.
6 combination:
6 0,16434 Cex=4 2560,320632 due to the removal of critical element)
⎧ (r −r ) − (ri −ri +1) ⎫
Ib = max⎨ i −1 i
7 7 0,042504 C 1= 65 0,261956
reduced combination: red ⎬ ∀i < D → (linear) 0 ≤ I b ≤ 0.5 (isostatic)
8 3 0 5 0 ⎩ 2 ⎭
STAR STRUCTURE ROBUSTNESS MAX MIN STATIC INDETERMINANCY:
high restrain grade
PU [ad] 1
⎧ ∂R ⎫
R ′ = max ⎨ ⎬ = 0 .25 ← cos t
⎩ ∂d d =1,..., D ⎭
Ia ≈ 1
8 1 2
Ib ≈ 0 !
0,5
7 3
λg
6 5 4
0
0 1 2 3 4 5 6 7 8
Damage Level
grade of structural continuity cannot be very high. The structure therefore is made 4 time
static indeterminate by means of a redundant number of elements. The total number of
elements in the structure is 21, which corresponds to a number of 685960 combinations for
the damaged configurations. The implemented heuristic optimizations reduced the number
of structural configurations to be investigated to just 286.
The robustness curve obtained for this structure are shown in the figure, together with the
value of the robustness indicators evaluated for the maximum and minimum histogram.
Among all the indicators, particularly interesting is that one which refers to the maximum
damage level endured in the best and worse case: while the curve of minima shows a quite
steep decrement of resistance and the structure is unstable after the first two damage levels,
the curve of maxima manifests a lower decrement of resistance and tolerates the removal of
a very high number of elements before becoming unstable. On the right of the figure, the
last stable configuration for maximum and minimum histograms are shown. In this respect,
it’s interesting to notice that due to the analysis ability to undergo local mechanism, the
maximum damage level exceeds the static indeterminacy grade in the measure of 4.
Damage MIN CFG MAX CFG 11 9 10 12
# of elements: E = 21
d El. ID Pumin El. ID Pumax
0 indetermin.
static 0 1
level: i = 04 1
1 9 0,471351 1 0,888941
lower 13 15 17 16 14
2 max. 3damage:0 Dmin2 = 2 0,888941
upper max. damage: Dmax11= 9 0,628625
3 10 0 7 5 3 1 2 4 6 8
4 17 0 12 0,628625
18 20 21 19
fixed
5 max. 16damage:0 D =59 0,628625
6 1 0 6 0,628625 λ
exhaust.
7 combination:
5 0 Cex=13 695860
0,628625
8
reduced 11
combination: 0 Cred14= 2860,628625
9 6 0 17 0
13 15 17 16 14
4,76 > Ia > 1,11 7 5 3 1 2 4 6 8
1 2
0.75
18 20 21 19
11 12 5 6 13 14
0.5
MAX MIN
9
0.25
11 9 10 12
0,26 > Ib > 0,13
3 17 13 15 17 16 14
7 5 3 1 2 4 6 8
0
18 20 21 19
0 1 2 3 4 5 6 7 8 9
Damage Level
The values assumed from the robustness indicators are reported in the following. It’s worth
noting, that they are calculated separately for the maximum and minimum histogram. That’s
apply also for the last three indicators, which could possibly be calculated as single value by
defining a parameter k, that calibrates the influence of the maximum histogram.
# of elements: E = 13 8 6 7 9
4
8,12 > Ia > 2,03 2 1 3 5
1
0,75
12 10 11 13
λ
0,31 > Icr > 0,09
2
0,5
MAX MIN
6
0,25
3 8 6 7 9
10 6 7 8 9 4 5 10 4 2 1 3 5
0
12 10 11 13
0 1 2 3 4 5 6 7 8 9 10
Damage Level
In the figure the obtained resulting maximum and minimum histograms are shown, together
with the representation of the last stable maximum and minimum configurations referring to
the curve of maxima and to the curve of minima. Also in this case, the removal of just two
elements makes the structure unstable, while the maximum sustainable damage for the
histogram of maxima shows a slight higher value than in the previous structure.
i=4 i = 12
High element number High element connection
11 9 10 12 8 6 7 9
13 15 17 16 14
14 4 2 1 3 5
7 5
5 3
3 11 22 44 66 88
18 20 21 19 12 10 11 13
λ λ
λ
PU [ad] TRUSS STRUCTURE ROBUSTNESS MAX MIN PU [ad] VIERENDEEL STRUCTURE ROBUSTNESS MAX MIN
1
2
0,5
0,5
11 12 5 6 13 14
9 6
0,26 > Ib > 0,13 0,31 > Ib > 0,09
0,25
3 17 10 6 7 8 9 4 5 10
0
0
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 10
Damage Level Damage Level
Therefore, contrarily to what could be maybe expected, the Vierendel structures doesn’t
prove to be more robust than the truss one, even if its static indeterminacy grade is three
times that of the truss structure. This could be partially explained considering that if an high
level of continuity can provide for a redistribution of stresses on the other elements, it can
also transmit the failure to other elements, when the alternate load paths are not enough
resistant to withstand the high stress transmission. In this case a compartmentalized
structure could prove to be more robust, localizing the collapse between two low connected
joints.
Redundancy Compartmentalization
TRIGGERING OF SUDDEN/EARLY
FEASIBLE ALTERNATE COLLAPSE
CHAIN RUPTURE COLLAPSE
LOAD PATH STANDSTILL
PROGRESSIVE COLLAPSE ROBUSTNESS
The comparison between these two structure is also useful for testing the robustness
indicators: while the first and the third one seems to correctly indicate a clear prevalence of
the robustness of the truss scheme, the second one responds too slowly to response variation
and the fourth one penalizes too much the structure with element redundancy. This is due to
the fact that this indicator is computed as difference between the area subtended to the curve
and the area subtended to a straight line connecting the resistance of the integer structure
with that one of the null structure, i.e. the with the best possible response of a structure with
a number E of elements. This characteristic determines a strong advantages of structures
with less elements and leads to scarcely expressive representation of the robustness, as the
direct superposition of the curves obtained for the two structure shown.
PU [kN]
1
elements
0,75
2 4
λg
0,5
1
STATIC INDETERMINANCY
0,25
I3V STRUCTURE ROBUSTNESS MAX MIN
PU [kN]
1
0
0 1 2 3 4
Damage Level
3
restrains
0,5
4 2
5 λg
1
I3E STRUCTURE ROBUSTNESS MAX MIN
PU [kN]
1
0
0 1 2 3 4 5
Damage Level
0,75
6 5
12 11
connection
λg
7 4
0,5
16
13 15 10
0,25
8 14 3
1 2
0
0 1 2 3 4 5
Damage Level
9. HIGH CONNECTED: the first one its composed of four beams disposed in a square
and highly connected (transmission of force and moments). The resulting robustness
curves shows though an abrupt decrement of resistance for the first damage level, due to
the fact, that the failure of whichever element would open the square that as integer
show an high ultimate resistance just in function of its closed resisting form.
10. STRONG RESTRAINED: the second structure is composed by 5 bars (transmission of
axial force only, through internal hinge connections), disposed as a braced square. The
structure is internally static determined and the static indeterminacy level is all due to
the external restraints, that are supports applied at each node. As expected, the structure
shows a very good behaviour in term of robustness, since the two curves doesn’t deviate
very much from a linear trend and present moreover a low scattering, i.e. the robustness
of the structure is bounded between two similar behaviours.
11. MANY ELEMENTS: the third structure is also a truss structure composed of bar and
hinges. This structure, contrarily to the previous one, is externally static determined, i.e.
the rigid body motion is avoided by means of a simply support of two nodes at the basis
of the truss. The internal static indeterminacy is all to be ascribed to the redundant
number of elements. The robust behaviour of the structure is described, as common for
this typology of truss structure, by a almost horizontal trend of the maxima curve and a
very steep trend of the minima curve, i.e. with an high scattering between the two
extremal behaviours.
λg
19 20 21 SHEAR-TYPE FRAME ROBUSTNESS MAX MIN 19 20 21
9 10
P U [ad]
11 12
1
9 10 1 12 16 17 18
λg 1
PU
5 6 7 8
13 14 15
16 17 18 1 2 3 4
STRUCTURAL BEHAVIOUR
0,5
5 6 7 8
λg
19 20 21
13 14 15 9 10 11 12
16 17 18
5 6 7 8
1 2 3 4 13 14 15
0
1
17 1 2 3 4 5 6 7 8 9 10 2 3 4
Damage Level
9 10
1
11 12
9 10 1 12 16 17 18
λg 1
5 6 7 8
0,75
13 14 15
16 17 18
1 2 3 4
0,5
5 6 7 8
λg
19 20 21
13 14 15 9 10 11 12
0,25
16 17 18
5 6 7 8
1 2 3 4 13 14 15
0
1 2 3 4
0 1 2 3 4 5 6 7 8 9 10
Damage Level
Fig. 56: Robustness comparison between a shear-type and flexural type design
In the figure above two building frame are shown. The geometry is the same for both the
system, but the structural system is different since in one case the frame has a shear-type
behaviour obtained by increasing the stiffness of the beams, and in the other case the frame
has a flexural behaviour obtained by increasing the sections of the columns.
As can be seen in the figure, the response of the shear-type frame should be preferred in
term of robustness to that one of the flexural-type frame, especially in consideration of the
minimum response, that shows an higher value of the resistance for the fist damage levels.
before after
Fig. 57: The A.P. Murrah Federal building befor and after the collapse [FEMA 277]
Shortly after the collapse, the Federal Emergency Management Agency (FEMA) deployed a
team to investigate the damages caused by the Murrah Building bombing and to identify
engineering strategies for reducing such damage to new and existing buildings. In the
investigation report [FEMA 277 1996] the damages caused by the blast are reviewed and a
possible failure mechanism for the building is identified. It is deduced from visual
inspection and analysis of the damages that the column named G20 in the original drawings
(see Fig. 60) was in all likelihood abruptly removed by brisance in consequence of the bomb
explosion, while a possible damage caused by the detonation to the two adjoining columns
remains ambiguous [Coreley et al. 1998].
It is anyway concluded in the report that: “progressive collapse extended the damage
somewhat beyond that caused directly from the blast” and that “the type of damage that
occurred and the resulting collapse of nearly half the building is consistent with what would
be expected for an ordinary moment frame building of the type and detailing available in the
mid-1970’s”.
ROOF GIRDER
TYPICAL
GIRDERS
TRANSFER GIRDER
• Plan dimension: 60 m x 21 m
COLUMN BLASTED AWAY BY MAIN SECONDARY
• Main frames: 3 longitudinal frames of 10 THE EXPLOSION (G20) COLUMNS COLUMNS
spans, 6 m each span.
• Elevation: 10 floors for a total height of about 35 m
DISCONTINUITY IN THE
• Peculiarities: 5 columns of the main frame stop at 3° floor on a transfer girder, BOTTOM REINFORCEMENT
that has a bottom reinforcement discontinuity at supports OF THE TRANSFER GIRDER
Fig. 58: Original drawings of the A.P. Murrah building design [FEMA 277]
Stair well:
3D model 12” concrete stiffening core
(30,48 cm) modeled with
SHELL elements (4×4 mesh).
Floors:
6” concrete slab (16,4 cm)
modeled with SHELL
elements subdivided with a
4×4 mesh.
Girder:
3 groups for main girders plus
one of secondary girder,
modeled with FRAME
elements subdivided with a 4
elements mesh.
Columns:
Ancillary side buildings walls: Basis restraints:
12 groups, modeled with
24” concrete walls (60,96 cm), All d.o.f. are restrained at FRAME elements subdivided
SHELL elements (4×4 mesh). column and wall basis. with a 2 element mesh.
- Dynamic aspects are generally to be taken into account, directly or by means of a quasi-
static approach.
- Nonlinear aspects have to be considered, due to the large deformation occurring during a
collapse: in particular, referring to the mechanical behaviour of the material, not only a
elastic-plastic law has to be considered, since exceeding the material in all likelihood the
elastic behaviour, but also the possible rupture of sections and the consequent section
unloading has to be included.
- The possible development of local mechanism has to be checked and the consequences
have to be evaluated.
With respect to the lack of knowledge mentioned in the first point, some assumptions had to
be taken both for the geometrical and mechanical aspects of the model. Referring to
geometrical aspects, the disposition and dimension of elements where directly inferred from
that ones of the Murrah building reported in [FEMA 277 2006] and just some assumption
had to be made about some missing information, such as the definition of a plausible
thickness for the walls of the staircase and of the two adjoining ancillary buildings. The
amount and disposition of reinforcement instead differ from those used in the Murrah
Building, since in the presented model the calculation of the reinforcement was performed
by means of the automatic design of a current commercial code, according to Eurocodes 2
[EN 1992]. Particularly, this choice eliminated the peculiarity of the bottom reinforcement
discontinuity in the transfer girder, so that the investigated building can be assimilated to a
building like it would be currently designed rather than to the real one. Referring to the
mechanical characteristics of the material, the design value of the resistance were used for
the building design, while the characteristic value were considered instead in the collapse
investigation. Starting from these information and assumptions, a full three-dimensional
model of the building was made (see fig. 61) and loaded with design permanent weight and
service overloads computed according to the previously mentioned Eurocodes 2.
In this model the static reaction of the column G20 at the interception with the transfer
girder was identified and used in an further model to substitute the column with this
equivalent force. As already mentioned above in fact, in this investigation the modelling of
the explosion is discarded and a direct structural failure is assumed, by completely removing
the column G20.
The investigation is carried out by incrementing a vertical pushing force opposite to the
column reaction. The dynamic aspect of the column removal can be therefore just be limited
to a quasi-static approach, in the sense that in order to guarantee the exhaustion of the
dynamic peak this pushing force is expected to reach two times the value of the equivalent
force that simulates the column reaction.
Lack of knowledge Assumptions and choices ON THE BASIS OF Nonlinear static analyses are carried out:
F.E.M.A. REPORT
• Full 3D geometry (the Ancillary building and
original drawings reported staircase core modeled as Concrete: HOW ON
by F.E.M.A. were not walls of plausible thickness • E= 32858 Mpa
complete) • Pushing load is a vertical force • 8 two dimensional substructures
• fck=4000 psi= 27,6 MPa
(directed to the ground) applied to (segments of main girders and
• Actual loading at the Design loads are assumed Long. steel: the removed column joint secondary columns next to G20
moment of the explosion in order to perform the line)
• E= 28500 psi=196500 MPa • Vertical displacement is monitored
reinforcement design
• fck=60000 psi= 413,7 MPa at the considered joint • Two-dimensional frame model
Automatic design of current
Transv. steel: • Flexure plastic hinges (lumped • Three-dimensional building
• Effective disposition of all commercial code performed
according to EC2 standards • E= 28500 psi=196500 MPa plasticity) are assigned at beam model
reinforcements
(continuous reinforcement) • fck=34780 psi= 239,8 MPa ends
Characteristic resistance Vertical 3thÆ9th Roof • Different type of M-χ diagrams are
• Real resistance value of the assumed for the hinge behaviors
values are considered for loads [kN/m2] [kN/m2]
materials
nonlinear analyses Slab 3,8 3,8
Overload 2,0 1,0 Extrapolated
Only one column (G20) is
Live 3,0 0,5
• Elements directly hit by assumed to be destroyed by
Sum 8,8 5,3 Drop Residual
the explosion the explosion (statically
Horizontal 5% of to zero branch
removing from the system) λ
loads vertical loads
Fig. 60: Assumptions and choices for the building modelling process.
Referring to the nonlinear aspects to be considered in the analyses, the nonlinear behaviour
of the material was accounted by means of flexural plastic hinges located at the end of each
girder span, whose hinge length was computed according to the Italian seismic regulation
[Ord. 3274, 2003]. According to the regulation, the moment curvature diagram of the
sections are schematized by three segment (I stage up to the concrete cracking, II stage up to
the bar yielding and III stage up to the section rupture), but three different form were
considered for the hinge unloading branch: in the first one the hinge does not drop load
when the ultimate curvature is reached and a perfectly plastic behaviour is extrapolated
starting from the hinge ultimate moment; this hinge form doesn’t entail a big computational
effort, but also doesn’t permit to evidence the resistance drops in the response curve of the
structure consequent to the rupture of some sections; therefore a second hinge form was
considered, in which the plastic moment completely drops to zero as the ultimate curvature
is reached; a third hinge model was eventually considered, in which a residual sloped
branch is considered after the reaching of the ultimate moment. The use of this type of hinge
allowed for a more realistic model of the hinge unloading since the residual branch slope is
obtained drawing a line that starts from the highest point of the diagram (ultimate moment
of the section) and cross the point that refers to the ultimate moment and curvature of the
mere reinforcement of the section.
Several nonlinear static analyses where performed (see fig. 63), first on 8 substructures
extrapolated from the two-dimensional model of the first frame, then on the whole two-
dimensional model of the first frame and then on the complete three-dimensional model,
whose most significant outcomes are summarized in the following section.
8.2.2.1 Results
The analyses on the substructure of the first frame, that refer to segments of main girders
and secondary columns next to G20 line, permitted to evidence that the typical girders and
the roof girder contribute equally to the ultimate resistance of the structure, whereas the
transfer girder alone take more than the 45% of the whole resistance.
Resistance [kN]
E F G H 3000 B 2090,8 56,0
2500 C 2342,5 62,7
2000 D 2649,8 70,9
1500 E 78,7
2940,9
1000
F 3237,2 86,7
500
G 3533,6 94,6
0
-170 -120 -70 Displacement -20 [mm] H 3735,6 100
A B C D
E F G H
A significant difference on the ultimate resistance is also shown by the comparison of the
response curves of the two-dimensional and three-dimensional model. The ultimate
resistance of the tree-dimensional model results to be significantly greater then that one of
the model that account for the first frame alone (6200 vs. 4000 kN). Hence an
underestimation of the 35% would be committed neglecting the full consideration of the full
three-dimensional behaviour of the buildings. Also when the complete three-dimensional
model for the building is duly considered, the ultimate resistance cannot even reach the
static value of the column reaction (8800 kN). That means that when the building is
statically deprived of the column, it is not even able to sustain its weight (self weight and
service overloading); even more if a dynamic amplification factor would be considered (that
even if lower than the value 2 because of the energy dissipation provided by the many
plasticized sections, would be however greater than 1). A more refined estimation of the
dynamic amplification factor or an explicit consideration of the dynamic aspects would
prove to be not very significant and has been therefore disregarded.
More interesting can be instead a review of the progression of the subsequent plasticization
and of the structural parts involved in the collapse: all the outcomes resulting from the
previously described investigations show the development of a mechanisms that involves
the central span of all the girders between the columns adjoining the removed one (named
G16 and G24 in the original drawings).
This mechanism proves to be very dangerous, since it determines the rising of high axial
stresses in these girders, before the involved plastic hinges would exhaust their ductility.
These huge axial forces are transmitted by the structural continuity to the adjoining
elements and lead to an early collapse of the adjoining columns overloaded in shear and
moments.
Resistance [kN]
8800kN
8000
6000
interested by the
explosion, the
structure is not
2000
-180 -130 -80 -30 Load carrying capacity of the system without the column
Displacement [mm]
2D FRAME ANALYSIS: plastic hinges development M I plastic hinge form 3D FRAME ANALYSIS: plastic hinge development M I plastic hinge form
c c
c c
6000
4000
5000
3000 4000
2000 3000
2000
1000
1000
0 0
-300 -250 -200 -150 -100 -50 0 -180 -130 -80 -30
Displacement [mm] Displacement [mm]
Fig. 62: 3D response of the 2d and 3d building model and result comparison
8.2.2.2 Findings
The review of the results provided by the performed investigations permits to identify some
characteristics that influence the progressive collapse susceptibility of the building that are
intrinsic to the structural system and its organization.
It can be concluded, that the development of a mechanisms like the one shown by the
analysis, that involves the central span of all girders between lines 16 and 24, as shown in
the review of investigation outcomes, could not be bore by the structure. The attempt of
designing these columns to resist this mechanism seems in fact to be unfeasible. On the
other side, an effort aimed to compartmentalize the structure would also be ineffective: a
modification of the column-girder connections could be for example considered, that
contemplates special controls joint able of providing for an early detachment of the
collapsing sections, but in consequence of the chosen static scheme (particularly referring to
the main column positioning) the collapsed area could be limited at the lowest to the frame
portion between the main column lines (G16 and G24), i.e. not even to a third of the frame,
which makes the measure not worth of the effort.
The development of this mechanism should therefore be avoided. Limiting the consideration
to structural measure only, this purpose can be achieved at least in two different ways:
would an alteration of the static scheme be possible, a reduction of the span length (e.g.
increasing the main column number or extending the secondary column to the ground)
would with all likelihood strongly limit the damaged area. Otherwise strengthening the
transfer girder could also work, provided that the girder section is designed to resist the
abrupt removal of the column without exceeding from the elastic range at least in one of the
section interested by the mechanism. As a matter of fact, as long as the static scheme
remains unmodified, the development of three-hinge mechanism would just be delayed to
an higher value of the dynamic amplification factor. The progressive collapse susceptibility
could be therefore mitigated just in case the section hardening would be sufficient to rise
this value in proximity to 2. It’s interesting to notice that a section hardening that involves
just the amount of the reinforcement (like suggested in the previously mentioned report
[FEMA 277, 2006] by means of a special moment frame detailing and design) would be not
adequate to this purpose and a more effective oversizing of the section dimension and of the
reinforcement would be instead required.
peninsula with the Sicily island. This project results particularly interesting from the point
of view of a robust investigation, since the structural system consists in a 3300 m long main
central span over a total length of 3666 m. The bridge suspension system is provided by two
pairs of prestressed steel cables, each measuring 1,24 m for the diameter section. The cables
have a total length of 5330 m and rely on two 383 m high towers and on the anchorage
blocks. The deck is formed by three continuous box-girders (the outer ones assigned to the
railway and the inner one devoted to the railway) supported by 30 m apart transversal
girders set 30 m away one from the other and suspended to the cables through a system of
121 pairs of rope hangers.
TOWER A MID-SPAN
345 m
B
900 m
C 450 m
As regards the load scenarios, the hanger failures are always considered to occur on the
unloaded structure, i.e. neither the train nor the car loads are assumed on the bridge, which
results therefore deformed only under the self weight.
Different starting schemes were considered for the investigations, each referring to the
above mentioned damaged configurations of the bridge. In each structural model the
hangers that are intended to be removed during the analysis were modelled by means of
equivalent forces: so it’s possible to perform the element killing needed during the analysis
just by activating a time-history function on those forces, as in the following section better
explained. A preliminary static analysis has therefore been performed on the integer
structure, in order to compute the reactions exerted by the hangers on the rest of the
structure. Then, for each considered contingency scenarios, the interesting hangers have
been substituted in the static scheme with its previously computed reactions: a static
analysis on the model that consider the load case assigned to these reaction should provide
the same results of the integer bridge analysis, ensuring that the nominal configuration is
correctly restored by the equivalent forces; a static analysis where this load case is neglected
would instead be able to show the final equilibrated configuration for the damaged bridge
only if no subsequent plasticizations occurred due to dynamic amplification after the
equilibrium point.
hardening and to drop load when ultimate sectional ductility is reached. It’s worth
noting that in the performed analyses the deck is considered to behave elastically and
only some qualitative considerations about a possible yielding of the continuous box-
girders of the deck are provided.
3. Dynamic analysis: The abrupt failure of one or more hangers is a dynamic event, that
can hardly be considered through a quasi-static approach by means of an amplification
factor: the estimation of this coefficient would indeed not be easy for such a complex
structural system, since it depends on many factors, like the ratio between the time taken
by the failure event and the period of the vibration mode solicited from that failure (i.e.
how much the failure event is felt as “dynamic” or as “quasi-static” from the structure)
and the energy absorbed from possibly occurring plasticization of elements, thing that
cannot be known in advance. Moreover, if it is still true that in a first evaluation (aimed
to exclude the possibility of damage propagation), only the first peak of the dynamic
wave is relevant (if the structure survive it without any subsequent damage, it will resist
all the more the following dumped waves), not the same can be said when other
elements collapse as result of the triggering events. In this case, as better shown in the
following, a full dynamic analysis is required to correctly investigate the propagation of
collapse.
The hanger failure has been therefore considered through a sharp ramp function (the time
needed for the failure is taken equal to 1/100 second) assigned to the equivalent forces that
simulated the presence of the considered hangers in the structure. The dynamic analysis
performed consists in a direct-integration time-history analysis where no damping is
assigned and all the previous exposed nonlinearities are considered. The time integration
method used for the time-history analysis was the Hilbert-Hughes-Taylor (HHT) method,
that use a single parameter alpha, that can assume value in [-1/3, 0] and that was set to zero
in the performed analyses, letting the method to be equivalent to the so called trapezoidal
rule method (i.e. method of the average acceleration).
In order to let the hanger failure occurs in a self-weight equilibrated configuration, the
dynamic analysis starts at the end of a previous static case that provides for the initial
conditions of the subsequent time-history case.
In order to cause the rupture of the two hangers that trigger the chain ruptures to the left
and to the right of the damaged zone, 9 hangers have to abruptly fail together. Also in
this case the collapse propagation triggers in direction of the bridge centre first (after 2
second since the removal) and the hanger to the left of the damaged zone (LH)
surmount the rupture of 12 hangers to the right and resist up to 4 seconds after the
rupture of the first right hanger before breaking and triggering the collapse also toward
the tower (following figure).
It can be also observed that, in the direction of the bridge centre, the hangers yielding
and rupture speed up with time: in the first second after the first rupture just 2 hangers
break, while in the next second the chain rupture involve the following 4 hangers
(following figure). The velocity of collapse propagation, even if not very high at first
becomes rapidly hasty.
0 0 Æ Ruptures propagation
2 2
1 1
0 0
2 2
RH (Right Hangers)
1 1
0 0
First rupture occurs in the RH
2 2
Rupture propagation speeds up:
1 1
0 0
• 2 hangers break in 1 sec. then
2 2
• 4 break in the following second
1 1
0 0
2 2 LH (Left Hangers)
First LH break 3 seconds after the
1 1
0 0
first RH rupture, when 12 RHs
2 2
have already broken
Rupture propagation does not
1 1
9 8 7 6 5 4 3 2 1 0
0 0
0 1 2 3 4 5 6 7 8 9 speed up:
Time [s] Å Chain ruptures propagate Æ Time [s] • 2 hangers break each second
Fig. 65: Rupture propagation to the right and to the left of the removed hangers, for asymmetric damage in
zone B.
4. Initial failures in zone D: The centre of the bridge proves to be the most sensible zone
to a damage involving the hanger system: in this zone the asymmetrical abrupt failure of
just 7 hangers and the symmetrical failure of 5 hangers are sufficient for the triggering
of collapse. Furthermore the failure propagation results very fast: referring to the
propagation toward the centre, 7 hangers break in just three second (fig. 70), while in
the previous exposed case of asymmetrical failure in zone B (fig. 69) one second more
was needed to let the same number of hangers break in the same direction.
Even if the first rupture trigger very soon and propagates also very rapidly, though it
does not accelerate, contrarily to what happened in the previous exposed case: by
carefully comparing the graphics above (fig. 70) it can be observed that two hangers
break in each half second after the first rupture. Nevertheless the collapse doesn’t come
to a stop, not even in correspondence of the higher resistant hangers (section type 2).
These hangers are very far from the centre though, so when the ruptures get so far, the
unsupported deck involves the major of the central span, making very difficult (and also
1 1
0 0
Æ Ruptures propagation
2
2
1
1
0
RH - LH propagation
0
1
1
simultaneously in the RH & LH
0
0
2 2
Rupture trigger very soon (in
the first one and half second)
and propagate with an high
1 1
0 0
velocity (7 hangers break in 3
2 2
seconds) compared with AF-AF-zB.
zB.
1 1
1 1
Progressive collapse doesn’
doesn’t
come to an halt though, not
even when encountering
0 0
3 2 1 0 0 1 2 3
Time [s] Å Chain ruptures propagate Æ Time [s] hanger greater sections
Fig. 67: Rupture propagation to the right and to the left of the removed hangers, for asymmetric damage in
zone D.
effects on the structure and very 7 hangers must be symmetrically removed on both
sides in order to trigger the propagation of the ruptures on the adjoining hangers.
collapse propagation
2
growing element ductility
Symmetrical
Asymmetrical
2. Preferential direction for the collapse propagation: to the higher damage sensibility
of the bridge central zone counterpoises a lower acceleration of the collapse progression
triggered by central ruptures, with respect to that one triggered by lateral ruptures.
This effect is due to the particular configuration of the structural system that requires a
growing hanger length from the centre to the sides of the bridge: when a chain rupture
triggers, the ultimate elongation required to the hangers adjoining the failed ones
increases as the collapse propagates (because the unsupported deck length also
increases).
If the initial damage occurs at mid-span, it involves the shortest hangers and the
collapse propagation is partially slowed down from the growing element ductility of
sideward hangers. On the contrary, a more intense initial damage is required sideways
to trigger chain ruptures, but then the hanger breakdowns speeds up when moving
toward the centre, where the hanger length decreases.
3. Qualitative measure that could possibly lead the collapse to an halt: in the case of a
central rupture a closer increment in the section of the hangers (that remain instead the
same for about 5/6 of the span length) could possibly provide for a collapse standstill. In
the case of a chain rupture triggered in a lateral zone the preferential direction showed
by the progressive collapse would probably make less effective such a measure.
4. Sensibility to modality of damage (asymmetrical or symmetrical failure): another
consideration about the possible collapse standstill concerns the higher susceptibility of
the bridge to an unsymmetrical hanger failure than to a symmetrical one: in the last case
the symmetrical hinge formations determines a symmetrical moment increment on the
deck box-girders, thus possibly allowing for an early deck segment detachment that
would arrest the collapse.
CONCLUSIONS
CONCLUSIONS 183
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
wm − w0
DAF =
we − wo Eq. 31
CONCLUSIONS 185
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
CONCLUSIONS 186
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
hangers and the collapse propagation is partially slowed down from the growing element
ductility of sideward hangers. On the contrary, a more intense initial damage is required
sideways to trigger chain ruptures, but then the hanger breakdowns speeds up when moving
toward the centre, where the hanger length decreases.
A preferential direction for the collapse propagation is then shown, that could address some
countermeasure aimed to mitigate the progressive collapse susceptibility of the structure.
Particularly, in the case of a central rupture a little increment in the section of the hangers
(that remain instead the same for about 5/6 of the span length) could possibly provide for a
collapse standstill, while in the case of a chain rupture triggered in a lateral zone instead, the
preferential direction showed by the progressive collapse would probably make less
effective such a measure and stronger modification of the structural system would be
probably required.
CONCLUSIONS 187
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
interacting plastic hinges. With these premises it is however possible to support five
possible effects of this mechanism, that are:
a. The hanger breaks in consequence of an axial overloading transmitted by the cinematic
chain of the deck segments. In this case the ruptures would propagate through the hangers
by a zipper type collapse [Starossek 2007], just like in the cases previously examined.
Possible countermeasure could be aimed to increase the resistance or the section of
predetermined hangers, placed at a mutual distance that should be not greater than the
considered collapsing deck segment length. A collapse standstill would be anyway not a-
priori guaranteed by such countermeasure, since the effects b, c, d, e could occur
afterwards.
b. The deck breaks at hanger level in consequence of shear overloading: the axial stress of
the collapsing deck segments is not entirely taken from the hangers but is also transmitted
by shear force to the adjoining deck portion. The brittle rupture consequent to a shear
failure of the section could provide for a detachment of the collapsing deck portion, but
in consideration of the relative shear resistance between the deck and the hangers, it
seems unlikely that such a failure could actually occur in this system.
c. The deck breaks out for axial overloading: if the hangers and the deck section adjoining
the collapsing segment would resist up to the axial rupture of the latter, a detachment of
the collapsing section could be assumed that would arrest the collapse. Such high axial
stress in the deck collapsing section however, normally require quite low inclination
angles to occur, so that the horizontal component of the axial force, that is undertaken
from the adjoining deck segment, would also not be distant from the yielding value. Such
exiguous safety margin doesn’t permit to exclude an involvement of these elements in the
collapse.
d. The moment redistribution following the flexural failure of the three sections of the
considered deck portion determines the flexural overloading of other adjoining sections
and the failures propagate through the deck instead through the suspension system.
e. The hinge exhaust their ductility and breaks: when the ultimate rotation is reached, the
moment stored in the plastic hinge is dropped and the deck section is no longer able to
sustain and transmit any moment; however, the axial and shear loads could be still
transmitted, at least partially, to the adjoining sections. The rupture of flexural plastic
hinge cannot therefore indicate a simultaneous arrest of all stress transmission How and
how much axial and shear stresses are transmitted to the other elements is an aspect that
strongly depends on the joint detailing and on the joint physical realization and cannot
therefore be easily accounted by means of an unrefined model. Furthermore, the rupture
CONCLUSIONS 188
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
of a plastic hinge is a dynamic event, whose effects cannot be accounted just considering
the static flexural unloading of the involved sections. A possible countermeasure
addressed in this direction could attempt to obtain an early separation of this deck
segment. This aim could be accomplished by interrupting the continuity of the deck at
each 6th span, by means of special control joints, capable to abruptly disconnect the deck
segment when a predetermined joint rotation value is reached.
CONCLUSIONS 189
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
adequate to this purpose and a more effective over-sizing of the section dimension and of
the reinforcement would be instead required.
CONCLUSIONS 190
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
CONCLUSIONS 191
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
methods have been presented and discussed and a direct bottom-up method, that starts has
been indicated as effective for the achievement of structural robustness particularly.
CONCLUSIONS 192
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
at a sectional level. Similarly, the verification of a section pleads the analysis of its fibres. A
too localized initial damage would seem therefore to be inconsistent with the other
verification and with the common engineering practice. Furthermore it would require a
significantly bigger calculus onus that would not be justified from a clear improvement in
the expressivity of the robustness curves. The investigation of the structural robustness is a
quite arduous task, that need to be kept as simplest as possible also in consideration of the
result interpretability and of the need to promote the acceptance of the method. Eventually,
a clear identification of the critical elements could not be provided anymore, neither an
abrupt resistance decrement would similarly represent a critical section, since the damage
level would no more be bounded with physical entities.
With respect to the first problem, some possible modification of the algorithm were
nevertheless presented and discussed, in order to take into account in the damage level the
entity of the element sections, by means of a simple rescaling of the robustness curve axis as
well as a more incisive modification in the procedure of element removal.
Another problem arises from the fact that the exhaustive number of combinations of failed
elements to be considered for an exhaustive investigations increases exponentially with the
number of elements. In order to perform a feasible limited number of analyses for one
structure, optimization algorithms have therefore been used in this work with the aim of
moving the problem complexity from an exponential to a polynomial form. The
optimization techniques considered are a heuristic and a probabilistic method. The first one,
that has been implemented and applied in the investigations of some simple structures, is
based on the assumption that the worst and the best configuration for each damage level can
be directly originated by an element removal on the best and worst configurations of the
previous damage level, respectively; the second one avails a simulated annealing algorithms
that starts from a set move based on the most stressed or strained elements. In both case, the
problem becomes feasible and the mentioned curves can be obtained in a polynomial time
also for complex structural system.
The obtained curves characterize the response of the structure in term of robustness and
provide useful information on the investigated system; particularly, the method allows to
clearly identify the critical elements, enabling a significant improvement through a focused
change in the static scheme or a direct reinforcing of only those elements.
CONCLUSIONS 193
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
CONCLUSIONS 194
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
APPENDICES
APPENDICES 195
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
11 ROBALG CODE
In this appendix, the code written for the calculus of the robustness curve is reported. The
code is written to work with ANSYS, that use a proprietary language called Ansys
parametric design language (APDL), quite similar to the well-known FORTRAN. The
algorithm avail the use of static nonlinear analysis of ANSYS and is conceptually divided in
ten main parts, that are briefly commented beneath. The program was meant to work fully
automatically and allows a simple and flexible usage by the users. In order to a proper
usage, just some requirement on the starting model should be accomplished, that are
recalled in the following sections.
ULIB,ROBALG7,TXT
USE,MAIN
APPENDICES 197
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
By reviewing this result the user can see if the analysis was correct and go on with the
analyses of damaged structures or change the analyses parameter. In this case a new analysis
on the integer structure will be performed until the analyses parameter are correctly
calibrated.
11.3 CODE
=================================================================================
CODICE APDL PER IL CALCOLO DI CURVE DI ROBUSTEZZA
AUTORE: LUISA GIULIANI DATA REVISIONE: Settembre 2008
UNIVERSITA’ DI ROMA LA SAPIENZA – DIPARTIMENTO DISG
================================================================================
MAIN!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
!__________________MODELLO STRUTTURA E PARAMETRI ANALISI________________________
*USE,MODEL ! Modello geometrico della struttura
*USE,INPAR ! Inizializzazione parametri e vettori
*USE,MESH ! con vincoli, meshatura e carichi
*USE,CARICHI
:REPLAY
*USE,SOLVPAR ! Definizione parametri per l'analisi
!______________________RISOLVO STRUTTURA INTEGRA________________________________
*USE,SOLVINT ! Analisi della struttura integra
*USE,VEDIRIS ! Visualizzazione dei risultati (curva
! di pushover, andamenti e carico ult.)
!------ Scelgo se modificare e ripetere l'analisi della struttura integra ------
*ASK,GOON,Change analysis parameter and resolve integer structure again?,'n'
*IF,GOON,EQ,'y',THEN
JUMPLINE='REPLAY'
*ELSE
JUMPLINE='AVANTI'
*ENDIF
*GO,:%JUMPLINE%
:AVANTI
!-------- Scelgo se procedere con l'analisi delle strutture danneggiate ---------
TM=TSTOP-TSTART ! Tempo di analisi della str. integra
*USE,CALCTIME,TM ! Stimo il tempo tot per le analisi succ
*ASK,GOON,Proceed with analyses of damaged structures? (y/n),'y'
*IF,GOON,EQ,'n',THEN
JUMPLINE='TERMIN'
*ELSE
JUMPLINE='CONTIN'
*ENDIF
APPENDICES 198
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
*GO,:%JUMPLINE%
:CONTIN
!____________________RISOLVO LE STRUTTURE DANNEGGIATE__________________________
*USE,SAVENGO ! Salvo i risultati e procedo
*ASK,PRIMA,Which curve you want to be calculate first? (MAX/MIN),'MAX'
*IF,PRIMA,EQ,'MAX',THEN
*USE,CURVMAX
*USE,RESETINT ! Reimposto la config. integra
*USE,CURVMIN ! Risolvo il problema dei minimi
*ELSEIF,PRIMA,EQ,'MIN',THEN
*USE,CURVMIN ! Risolvo il problema dei minimi
*USE,RESETINT ! Reimposto la config. integra
*USE,CURVMAX ! Risolvo il problema dei massimi
*ENDIF
*USE,RESULT ! Registro e plotto i risultati
:TERMIN
/EOF!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
MODEL!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
!___________________INIZIALIZZAZIONE PARAMETRI MODELLO_________________________
NV=0
NT=0
E=0
ET=0
DMAX=0
!____________________________MODELLO DELLA STRUTTURA____________________________
*ASK,RISPOSTA,'Use wizard to build the model? (y/n)','n'
*IF,RISPOSTA,EQ,'n',THEN
*USE,INFOFILE
*ASK,INPUTPATH,'Name of input file','INPUT'
*ASK,INPUTEXT,'Extension of input file','TXT'
/INPUT,%INPUTPATH%,%INPUTEXT%,,,1
*ELSE
*USE,MODASK,E,NV,DMAX
*ENDIF
EPLOT
/PSF,DEFA, ,1,0,1
/PBF,DEFA, ,1
/PIC,DEFA, ,1
/PSYMB,DOT,1
/PBC,ALL, ,1
/REPLOT
/EOF!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
APPENDICES 199
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
INPAR!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
!_____________________DICHIARAZIONE PARAMETRI E VETTORI_________________________
!--------------------- Parametri scalari del modello ---------------------------
*ASK,TITLE,Analysis title,'ROBALG' ! Chiedo il titolo
/TITLE,%TITLE% ! Chiedo il # di nodi totali e il # di
*IF,NT,EQ,0,THEN
*ASK,NT,# of total nodes,9
*ENDIF ! Chiedo il # di nodi vincolati IN DIR.
*IF,NV,EQ,0,THEN ! INTERESSE (ovvero la direzione della
*ASK,NV,# of restrained nodes,8 ! forza agente). Nel modello tali nodi
*ENDIF ! devvono essere stati numerati in modo
*IF,ET,EQ,0,THEN ! sequenziale a partire da 1 fino ad NV.
*ASK,ET,# of total elements,8 ! Chiedo il # degli el. totali del mod.
*ENDIF
*IF,E,EQ,0,THEN ! Chiedo il # degli el. da rimuovere che
*ASK,E,# of removable elements,%ET%!
*ENDIF ! Tali el. devono essere stati numerati
*IF,DMAX,EQ,0,THEN ! nel modello in modo seq. da 1 ad E
*ASK,DMAX,Maximum damage level,%E%
*ENDIF ! Chiedo il danno massimo
*USE,CALCCOMB ! calcolo il # di combinazioni da analiz
!----------- Array dei carichi ultimi per il liv. di danno corrente ------------
*DIM,PUD,ARRAY,E,3,1 ! Dichiaro un vettore PUD che contiene
! i carichi ultimi delle configurazioni
! analizzate per il livello di danno D
*DO,I,1,E,1 ! Inizializzo il vettore PU, inserendo
PUD(I,1) = I ! nella prima colonna il numero
PUD(I,2) = -1 ! identificativo di ogni elemento e
PUD(I,3) = -1 ! nelle altre due il valore iniziale -1
*ENDDO ! NB: L'ID di el. serve per l'EKILL iniz
!------------------- Array PU dei carichi ultimi critici -----------------------
*DIM,PU,ARRAY,(DMAX+1),3,1 ! Dichiaro una matrice PU che conterrà
*DO,I,1,(DMAX+1),1 ! le inf. della config critica per ogni
*DO,J,1,3,1 ! liv. di danno da 0 a DMAX, prima per
PU(I,J)=-1 ! la curva dei MIN e poi per quella dei
*ENDDO ! MAX. La matrice è inizializzata con
*ENDDO ! il valore -1 in tutte le posizioni.
!------------------ Array ROBCUR delle curve di robustezza ---------------------
*DIM,ROBCUR,ARRAY,(DMAX+1),7,1
*DO,I,0,DMAX
ROBCUR((I+1),1)=%I%
APPENDICES 200
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
*DO,J,2,7,1
ROBCUR((I+1),J)=-1
*ENDDO
*ENDDO
/EOF!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
MESH!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
/PREP7!============================== ! Entro nel preprocessore
*ASK,MESHING,'Do you want to refine element mesh?','n'
*IF,MESHING,EQ,'y',THEN
*GET,ET,ELEM,0,COUNT ! Prendo il n. tot. di el. def.
*GET,NT,NODE,0,COUNT ! Prendo il n. tot. di nodi def.
*ASK,MESH,Number of element subdivision? (odd),3
*DIM,ELMESH,ARRAY,%MESH%,%ET%,1 ! Creo un vettore per salvare
*DO,EE,1,ET,1 ! le inf. sugli el. meshati
*GET,NMAX,NODE,0,NUM,MAXD ! Max n. assegnato ai nodi def.
*GET,EMAX,ELEM,0,NUM,MAXD ! Max n. assegnato agli el. def.
*GET,NI,ELEM,%EE%,NODE,1 ! N. assegn. al nodo iniz. di EE
*GET,NJ,ELEM,%EE%,NODE,2 ! N. assegn. al nodo fin. di EE
EDELE,EE ! Elimino l'elemento da discr.
! Creo MESH-1 nodi interni tra
! l'iniz. e il fin. dell'el.
! eliminato. Li numero consec.
FILL,%NI%,%NJ%,%MESH-1%,%NMAX+1%,1! da NMAX+1 a NMAX+MESH-1
EN,%EMAX+1%,%NI%,%NMAX+1% ! Primo sottoelemento
ELMESH(1,%EE%,1)= %EMAX+1%
EN,%EMAX+2%,%NJ%,%NMAX+MESH-1% ! Ultimo sottoelemento
ELMESH(%MESH%,%EE%,1)= %EMAX+2%
*DO,I,1,%(MESH-3)/2%,1 ! Sottoelementi interni
J=(I*2)+1
EN,%EMAX+J%,%NMAX+I%,%NMAX+I+1%
ELMESH(I+1,%EE%,1)= %EMAX+J%
EN,%EMAX+J+1%,%NMAX+MESH-I%,%NMAX+MESH-I-1%
ELMESH(%MESH-I%,%EE%,1)= %EMAX+J+1%
*ENDDO
! Creo il sottoel. centrale che
! rinomino EE (sarà rimosso)
EN,%EE%,%NMAX+((MESH-1)/2)%,%NMAX+((MESH+1)/2)%
ELMESH(%(MESH+1)/2%,%EE%,1)= %EE%
*ENDDO
*GET,ET,ELEM,0,COUNT ! Prendo il n. tot. di el. def.
*GET,NT,NODE,0,COUNT ! Prendo il n. tot. di nodi def.
APPENDICES 201
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
*ENDIF
/PNUM,NODE,1
/PNUM,ELEM,1
/REPLOT
FINISH!============================== ! Fine dati preprocessore
/EOF!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
CARICHI!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
/SOLU!=============================== ! Entro nel solutore
!-------Definisco i carichi----------
*MSG,UI
Definire i carichi con cui portare a collasso la struttura. %/&
N.B.: non utilizzare peso proprio o carichi distribuiti per elementi asta!
/WAIT,2
*ASK,LOADTYPE,'Apply: 1.Self weight, or 2.Distributed load, or 3.Pinned load?',1
*IF,LOADTYPE,EQ,1,THEN ! Definisco l'accelerazione
*ASK,DIRACC,Acceleration direction? (X/Y),'Y'
*ASK,MULTACC,Specify the load multiplier for self weight,10E003
*IF,DIRACC,EQ,'X',THEN
ACEL,(-9.81E-003*MULTACC)
*ELSEIF,DIRACC,EQ,'Y',THEN
ACEL,,(9.81E-003*MULTACC) ! il valore + è in dir. -y)
*ENDIF ! Se è definità una densità
*ELSEIF,LOADTYPE,EQ,2,THEN ! applica all'el. il p.p.
*ASK,DIRDIS,Force direction (X/Y),'Y'
*ASK,VALDIS,Force value,10E003
!multipro,'start',2
! *cset,1,3,DIRDIS,'Force direction (X/Y)','Y'
! *cset,4,6,VALDIS,'Force value',-10E005
!multipro,'end'
*IF,DIRDIS,EQ,'X',THEN ! Applico una press. sulla faccia
SFBEAM,ALL,3,PRESS,%VALDIS%! orizzontale dell'elemento trave
*ELSEIF,DIRDIS,EQ,'Y',THEN ! Applico un carico distribuito
SFBEAM,ALL,1,PRESS,%-VALDIS%! sugli elementi travi in direz.
*ENDIF ! perp. alla faccia 1 (-y)
*ELSEIF,LOADTYPE,EQ,3,THEN
*ASK,LN,Number of loaded joints,1
*DO,I,1,LN,1
*ASK,LNN,Loaded node #,%NV+1%
*ASK,DIRPIN,Force direction at given node (X/Y),'Y'
*ASK,VALPIN,Force value,-10E005
APPENDICES 202
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
!F,%LNN%,F%DIRPIN%,%VALPIN%
*IF,DIRPIN,EQ,'X',THEN
F,%LNN%,FX,%VALPIN%
*ELSEIF,DIRPIN,EQ,'Y',THEN
F,%LNN%,FY,%VALPIN%
*ENDIF
*ENDDO
*ENDIF
SAVE ! Salvo il modello
FINISH!============================== ! Fine dati solutore
/EOF!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
SOLVPAR!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
/SOLU!===============================
ANTYPE,0 ! Analisi statica
OUTRES,ALL,ALL ! Salvo risultati di ogni passo
*ASK,NC,Number of control node,%NV+1%
*ASK,FDIR,Monitored direction at given node (X/Y),'Y'
*ASK,PS,'First step load ratio',5000
*ASK,AL,Use arc-length method (y/n),'n'
*IF,AL,EQ,'y',THEN
NSUBST,%PS%,0,0 ! Fisso il primo passo (1/5 FY)
ARCLEN,1,0,0 ! Metodo Arc Length ON
*ASK,SMAX,Max value of monitored displacement,0.005
ARCTRM,U,%SMAX%,%NC%,U%FDIR% ! spostamento massimo ammesso
AUTOTS,-1.0 ! Automatic substepping OFF
NCNV,2 ! (richiesto per l'arc-length)
*ELSE ! Termina l'analisi alla mancata conv.
ARCLEN,OFF ! ma non arresta l'esecuzione del progr.
*ASK,PSMAX,Max step number,10000
*ASK,PSMIN,Min step number,500
NSUBST,%PS%,%PSMAX%,%PSMIN%
AUTOTS,ON ! Automatic substepping on
LNSRCH,ON ! Line search on
TIME,1
NCNV,1,10E010 ! Termina analisi e programma alla
*ENDIF
RESCONTROL,,NONE
FINISH!============================== ! Fine dati solutore
/EOF!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§ù
APPENDICES 203
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
SOLVINT!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
/SOLU!===============================
D=0 ! Indice struttura integra
*GET,TSTART,ACTIVE,0,TIME,WALL,CPU ! Prendo il tempo di CPU
SOLVE ! Risolvo la struttura integra
*GET,TSTOP,ACTIVE,0,TIME,WALL,CPU ! Calcolo il tempo di analisi
!SAVE
FINISH!==============================
/POST1!============================== ! Entro nel postproc. generale
PU(1,1)=0 ! La config della struttura integra ha
SET,LAST ! tutti gli elementi presenti, quindi
*GET,CV,ACTIVE,0,SOLU,CNVG ! indico come el. assente l'el. 0
PU((D+1),3)=CV ! Salvo l'inf. sulla conv. in PU
*GET,TSN,ACTIVE,,SET,NSET ! Chiedo il numero totale di passi TSN
*IF,CV,EQ,0,THEN ! e fisso SN pari al numero di passi
SN=TSN-1 ! andati a convergenza. Se SN=0, non ci
*ELSE ! sono risultati validi, (es. struttura
SN=TSN ! labile)quindi il carico ultimo va
*ENDIF ! fissato a zero. Altrimenti quando SN
*IF,SN,GT,0,THEN ! SN>0 prendo il carico ultimo con FSUM
*IF,CV,EQ,0,THEN ! dall'ultimo passo a convergenza, dopo
SET,PREVIOUS ! aver selezionato i soli nodi vincolati
*ENDIF
NSEL,S,NODE,,1,NV,1 ! Seleziono i nodi vincolati
FSUM,0,ALL ! Calcolo il carico ultimo
*GET,PU((D+1),2),FSUM,0,ITEM,F%FDIR% ! Ne salvo il modulo in PU
NSEL,ALL ! Riseleziono tutti i nodi
*ELSE
PU((D+1),2)=0
*ENDIF
*IF,PU((D+1),2),LT,0,THEN
PU((D+1),2)=-1*PU((D+1),2)
*ENDIF
NSEL,ALL ! Riseleziono tutti i nodi
RESET
FINISH!============================== ! Fine dati postproc. generale
/EOF!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
VEDIRIS!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
!La parte di inizializzazione dei parameri TSN,SN,CV ecc. viene ripetuta solo
!per poter chiamare manualmente la routine quando SOLVINT non è stata chiamata
!in precedenza (ad es. nell'analisi manuale di strutture danneggiate).
APPENDICES 204
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
APPENDICES 205
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
*ENDDO
*ENDIF
!---------Plastic strain---------
*ASK,RISPOSTA,'Review plastic strain? (y/n)','y'
*IF,RISPOSTA,EQ,'y',THEN
/CONT,1,9,AUTO
/ESHAPE,%ESHAPE%
*DO,I,1,SN,%INC%
SET,,,,,,,I
PLNSOL,EPPL,EQV, 1,1.0 ! Plotta le def. plast.
/WAIT,0.05 ! (Von Mises)
*ENDDO
*ENDIF
!-----------Axial stress---------
*ASK,RISPOSTA,'Review axial stress? (y/n)','y'
*IF,RISPOSTA,EQ,'y',THEN
*ASK,SIGMA,'Elastic stress?',240000
SALLOW,%SIGMA% ! Define maximum stress
/CONT,1,8,%-SIGMA%, ,%SIGMA%
/ESHAPE,%ESHAPE%
*DO,I,1,SN,%INC%
SET,,,,,,,I
PLNSOL, S,EQV, 1,1.0 ! Plotta le tensioni
!PLNSOL,S,1, 1,1.0 ! di Von Mises ad ogni
/WAIT,0.05 ! passo (o l'andamento
*ENDDO ! della tens. princ.)
*ENDIF
*ASK,RISPOSTA,'Review axial hinge progression? (y/n)','y'
*IF,RISPOSTA,EQ,'y',THEN
*DIM,PHP,ARRAY,ET,SN+1,1 ! Define a matrix named
*DO,J,1,ET,1 ! PHP with ET (el. n.)
PHP(J,1)=J ! rows SN columns
*ENDDO ! (n. of substeps)
!-----------Axial hinge---------
*DO,I,1,SN,%INC%
SET,,,,,,,I ! Set the interest step
ETABLE,SAXL,LS,1 ! Store el. ax. stress
! in a table named SAXL
SFCALC,SF,SAXL,' ',3 ! Calculate and store
! in table named SF the
! work-rate of each el.
*VGET,PHP(1,I+1),ELEM,all,ETAB,SF, ,2
APPENDICES 206
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
APPENDICES 207
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
CALCTIME!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
!Manca la trasformazione corretta in secondi dei tick della cpu!
Tred=TM*Cred
!Forse un clock tick corrisponde a 400 s. Quindi per avere il tempo in min
!devo moltiplicare per 400 e poi dividere per 60.
Tred1=Tred*400/60
!Metodo alternativo
/RUNST
RTIMST
*GET,TM2,RUNST,0,RTIMST,TOTAL
TM2=TM2*70 !Ritaratura
Tred2=TM2*Cred/60 !Tempo in minuti
*MSG,UI,'reduced',Tred,'red converted',Tred1,'red
forecast',Tred2,'exhaustive',(Cex/Cred)
Eximated time for analyses: %/&
- %1X %C %3X = %1X %G %1X clock ticks or%/&
- %1X %C %3X = %1X %G %1X min %/&
- %1X %C %1X = %1X %G %1X min; %/&
- %1X %C %5X = %1X %G %1X time red time.
/WAIT,2
/EOF!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
SAVENGO!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
!------------------------- Salvo i risultati in ROBCUR -------------------------
*DO,I,1,3 ! Salvo i risultati della struttura
ROBCUR((D+1),(I+1))=PU((D+1),I) ! integra sia nella I che nella II
ROBCUR((D+1),(I+4))=PU((D+1),I) ! parte dell'array finale (la cfg
*ENDDO ! è unica e vale come MIN e MAX)
!---------- Resetto l'analisi e procedo con le analisi seguenti ----------------
/POST1!============================== ! Entro nel postproc. generale
RESET ! Resetto l'analisi: SERVE??
FINISH!============================== ! Fine dati postproc. Generale
!---------- Salvo per le analisi seguenti solo i risultati base-----------------
/SOLU!===============================
APPENDICES 208
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
CURVMAX!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
INDEX='MAX'
*IF,PRIMA,EQ,'MAX',THEN
START=1
*ELSEIF,PRIMA,EQ,'MIN',THEN
START=2
*ENDIF
*USE,SOLDAMSTR
/EOF!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
RESETINT!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
!_________________REIMPOSTO LA CONFIGURAZIONE INTEGRA___________________________
!--------------------- Riattivo tutti gli elementi -----------------------------
/SOLU!=======================
EALIVE,ALL
!*DO,K,1,E
! *IF,MESHING,EQ,'y',THEN
! *DO,I,1,MESH,1
! EALIVE,%ELMESH(I,K,1)%
! *ENDDO
! *ELSE
! EALIVE,K
! *ENDIF
!*ENDDO
FINISH!======================
!--------------------------Reset dell' array PU---------------------------------
*IF,PRIMA,EQ,'MAX',THEN
*DO,J,1,3 ! Serve per sapere quale el. è
PU(2,J)=ROBCUR(2,(1+J)) ! stato individuato come critico
*DO,I,3,DMAX+1,1 ! al LD precedente e rimuoverlo
PU(I,J)=-1 ! all'inizio del nuovo ciclo di
*ENDDO ! analisi per il nuovo LD. Va
*ENDDO
*ELSEIF,PRIMA,EQ,'MIN',THEN
*DO,J,1,3 ! Serve per sapere quale el. è
PU(2,J)=ROBCUR(2,(4+J)) ! stato individuato come critico
*DO,I,3,DMAX+1,1 ! al LD precedente e rimuoverlo
PU(I,J)=-1 ! all'inizio del nuovo ciclo di
APPENDICES 209
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
CURVMIN!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
!________________________RISOLVO PROBLEMA DEI MINIMI____________________________
INDEX='MIN'
*IF,PRIMA,EQ,'MIN',THEN
START = 1
*ELSEIF,PRIMA,EQ,'MAX',THEN
START = 2
*ENDIF
*USE,SOLDAMSTR
/EOF!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
RESULT!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
!______________________REGISTRO E PLOTTO I RISULTATI____________________________
SAVE
*VPLOT,ROBCUR(1,1),ROBCUR(1,3),6 ! Plotto i PU MAX MIN vs i LD
*USE,WRITEOUT
/EOF!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
CALCCOMB!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
!*IF,DMAX,NE,E,THEN
Cex = 0 ! inizializ. la var. da calc.
EF=1 ! Fattoriale di n. di el.(EF=E!)
*DO,F,1,E
APPO = F
EF=EF*APPO
*ENDDO
*DO,LD,0,DMAX
LDF=1 ! Fattoriale del livello di
*DO,F,1,LD ! danno corrente (LDF=D!)
APPO=F
LDF=LDF*APPO
*ENDDO
EMLDF=1 ! Fattoriale di (E-D)!
*DO,F,1,(E-LD)
APPO=F
EMLDF=EMLDF*APPO
*ENDDO
CD = EF/(LDF*EMLDF)
Cex = Cex+CD
APPENDICES 210
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
*ENDDO
Cred = ((2*DMAX-1)*E)-(DMAX*(DMAX-1))+1 ! numero di combinazioni ridotte
!ELSE ! Non c'è bisogno dell'else dato
! Cex=2**DMAX ! che le formule sopra sono più
! Cred=E^2+1 ! generali e comprendono anche
!*ENDIF ! il caso DMAX=E per cui però
! le espres. sono più semplici
*MSG,UI,'exaustive',Cex,'reduced',Cred
Number of combination: %/&
- %1X %C %1X = %1X %I %/&
- %1X %C %3X = %2X %I
/WAIT,2
/EOF!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
INFOFILE!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
*MSG,UI
Caricare un file dati con la geometria della struttura e le proprietà%/&
della sezione e del maeriale ELASTO-PLASTICO, ricordando di: %/&
1) numerare gli elementi da rimuovere in sequenza da 1 ad E %/&
2) numerare i nodi vincolati in direz. della forza in seq. da 1 a NV %/&
3) riportare con il nome indicato i parametri seguenti: %/&
%2X - ETYPE per l'elemento scelto e Area, Jz, H per le car. sezionali%/&
%2X - Ex, SigmaY, Etx per le proprietà del materiale %/&
%2X - NT, ET per il # di nodi e di elementi totali usati nel modello %/&
%4X - DMAX per il livello di danno massimo da considerare.
/WAIT,2
/EOF!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
ERRORMSG!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
*MSG,UI
The pushover curve cannot be computed, due to the %/&
excessive number of node reactions to be summed.
/WAIT,2
/EOF!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
SOLDAMSTR!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
*DO,D,%START%,DMAX,1
/SOLU!=======================
*IF,PU(D,1),GT,0,THEN
*IF,MESHING,EQ,'y',THEN
*DO,I,1,MESH,1
DCW = %PU(D,1)%
APPENDICES 211
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
DEW = ELMESH(I,%DCW%,1)
EKILL,%DEW%
*ENDDO
*ELSE
EKILL,%PU(D,1)% ! Sopprimo l'el. individ
*ENDIF ! come critico nel liv.
*ENDIF ! di danno precedente
FINISH!====================== ! (NO PER D=1)
!_______RICERCA CONFIG CRITICA MINIMA PER IL LIVELLO DI DANNO D_________
!--------------------- Aggiorno i vettori ------------------------------
*DO,I,1,E,1
PUD(I,2)=-1 ! Resetto il vettore PUD
PUD(I,3)=-1
*ENDDO
!---------------- Sopprimo un elemento alla volta ----------------------
*DO,K,1,E,1
MATCH=0 ! Resetto la variabile
*DO,J,1,DMAX,1 ! In questo modo evito
*IF,K,EQ,PU(J,1),THEN ! di sopprimere elementi
MATCH = -1 ! che sono già soppressi
*ENDIF ! in config. di partenza
*ENDDO
*IF,MATCH,EQ,0,THEN
/SOLU!====================== ! Sopprimo l'elemento K
*IF,MESHING,EQ,'y',THEN
*DO,I,1,MESH,1
DEW = ELMESH(I,K,1)
EKILL,%DEW%
*ENDDO
*ELSE
EKILL,K ! (un el. per volta)
*ENDIF
/NERR,0 ! Evito il prompt alla
SOLVE ! mancata convergenza
FINISH!======================
/POST1!======================
/NERR,DEFA ! Riattivo i msg di err.
SET,LAST ! Chiedo ris. ult. passo
*GET,TSN,ACTIVE,,SET,NSET ! Get the total step number
*GET,CV,ACTIVE,0,SOLU,CNVG ! and store it in TSN variable
PUD(K,3)=CV ! Salvo l'inf. della CV in PUD
*IF,CV,EQ,0,THEN ! Se SN = 0 vuol dire che la struttura è
APPENDICES 212
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
APPENDICES 213
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
FINDMAX!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
*IF,D,EQ,1,AND,PRIMA,EQ,'MAX',THEN ! Per D=1 trovo prima
*USE,FINDMIN ! il MIN e lo salvo in
*ENDIF ! ROBCUR. Poi il MAX:
PU(D+1,2)=-1 !non 0 sennò non scrive l'ultimo ris. ! Inizializzo il carico
*DO,N,1,E,1 ! fino ad E non a DMAX!!! ! ultimo con un valore
*IF,PUD(N,2),GT,PU(D+1,2),THEN ! minimo (zero) e lo
*DO,I,1,3,1 ! sostituisco con ogni
PU(D+1,I)=PUD(N,I) ! valore più alto che
*ENDDO ! trovo in PUD, cioè
*ENDIF ! alla fine ci salvo il
*ENDDO ! il PU max di PUD
*DO,R,1,3 ! Copio la riga di PU
ROBCUR(D+1,R+4)=PU(D+1,R) ! nella II parte della
*ENDDO ! matrice ROBCUR (MAX)
/EOF!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
FINDMIN!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
*IF,D,EQ,1,AND,PRIMA,EQ,'MIN',THEN ! Per D=1 trovo prima
*USE,FINDMAX ! il MAX e lo salvo in
*ENDIF ! ROBCUR. Poi il MIN:
PU(D+1,2)=10**10 ! Inizializzo il carico
*DO,N,1,E,1 ! fino ad E non a DMAX!!! ! ultimo con un valore
*IF,PUD(N,2),LT,PU(D+1,2),AND,PUD(N,2),GT,-1,THEN! non superabile e lo
*DO,I,1,3,1 ! sostituisco con ogni
PU(D+1,I)=PUD(N,I) ! valore più basso che
*ENDDO ! trovo in PUD, cioè
*ENDIF ! alla fine ci salvo il
*ENDDO ! il PU min di PUD
*DO,R,1,3 ! Copio la riga di PU
ROBCUR(D+1,R+1)=PU(D+1,R) ! nella I parte della
*ENDDO ! matrice ROBCUR (MIN).
/EOF!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
MODASK!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
*ASK,NT,# of total nodes,33
*ASK,ET,# of total elements,32
!*ASK,NV,# of restrained nodes,8 !Passati dal file principale
!*ASK,E,# of damageble element,8
/PREP7!============================== ! Entro nel preprocessore
!_______Definisco il materiale_______
*MSG,UI
Inserire le proprietà del materiale ELASTO-PLASTICO.
APPENDICES 214
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
/WAIT,2
*ASK,Ex,'Young modulus',2.1E008
*ASK,PR,'Poisson ratio',0.3
*ASK,DENS,'Density',7849
*ASK,SigmaY,'Yield stress',2.4E005
*ASK,ExT,'Tangent Modulus',2.1E005
MP,Ex,1,%Ex% ! Modulo elastico acciaio
MP,PRxy,1,%PR% ! Modulo di Poisson
MP,DENS,1,%DENS% ! Densità dell'acciaio in kg
TB,BISO,1,1,2, ! Materiale bilin. inelastico
TBTEMP,0 ! Temp. per i dati seguenti
TBDATA,,%SigmaY%,%ExT%, ! Snervamento e incrudimento
!________Definisco l'elemento _______
*MSG,UI
Indicare il tipo di elemento (TRAVE/ASTA) e le %/&
caratteristiche della sezione da utilizzare.
/WAIT,2
*ASK,EType,'Joint DOF for elements? (2/3)','3'
*ASK,Area,'Section area',0.18
*IF,%EType%,EQ,3,THEN
*ASK,Jz,'Rotational inertia',5.4E-003
*ASK,H,'Section height',0.6
ET,1,BEAM23 ! Elemento trave inelastico 2D
! (3 g.d.l. per ogni nodo)
R,1,%Area%,%Jz%,%H% ! Definisco il set di costanti
! (R,1) per l'el.: Area, Jz, H
! per la sezione rettangolare
*ELSE
ET,1,LINK1 ! Elemento asta inelastico 2D
R,1,%Area% ! Area in mq (diameter = 0,1 m)
*ENDIF
!_______Definisco la geometria_______
!--------Coord. nodi vincolati-------
*MSG,UI,%NV%
Inserire le coordinate dei nodi vincolati. %/&
Tali nodi vengono automaticamente numerati %/&
da 1 ad NV = %I
/WAIT,2
*DO,I,1,NV ! Chiedo le coordinate dei nodi
*ASK,X,Coord X of node #%I%,%I% ! vincolati (numer. da 1 ad NV)
*ASK,Y,Coord Y of node #%I%,%I% ! Per la stella inserire:
APPENDICES 215
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
N,I,%X%,%Y%
*ENDDO
!---------Coord. altri nodi-----------
*MSG,UI,(NT-NV)
Inserire numerazione e coordinate degli altri %I nodi
/WAIT,2
*DO,I,NV+1,NT,1
*ASK,NN,Node #,%NV+1+I% ! Chiedo numerazione e coord.
*ASK,X,Coord X of node #%NN%,%I%! degli altri nodi dichiarati
*ASK,Y,Coord Y of node #%NN%,%I%
N,NN,%X%,%Y%
*ENDDO
!---Def. elem. removibili----
/PNUM,NODE,1 ! Visualizzo numeri nodi
/REPLOT
*MSG,UI,E
Inserire gli elementi removibili, definendone i nodi estremi. %/&
Gli elementi removibili vengono automaticamente numerati in %/&
modo sequenziale da 1 ad E=%I.
/WAIT,2
NUMST,ELEM,1 ! Parti a numerare gli el. da 1
*DO,I,1,E,1 ! Chiedo posiz. el. princ.
*ASK,M,1st node for element %I%,%NV+1%
*ASK,N,2nd node for element %I%,%I%
APPENDICES 216
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
!______Condizioni al contorno________
!-------Definisco i vincoli----------
*MSG,UI,NV
Definire il grado di vincolo dei %I nodi vicolati %/&
(1 carrello, 2 cerniera, 3 incastro)
/WAIT,2
*DO,I,1,NV,1 ! Chiedo il grado di vincolo dei
*ASK,GV,Restrain at node #%I%,3 ! nodi vincolati
*IF,GV,EQ,3,THEN ! Vincolo i nodi con un incastro
D,I,ALL,0 ! se GV = 3, con una cerniera se
*ELSEIF,GV,EQ,2,THEN ! GV=2 e con un carrello se GV=1
D,I,UX,0
D,I,UY,0
*ELSE
*ASK,VDIR,Restrained direction #%I%? (X/Y),'Y' ! nodi vincolati
D,I,U%VDIR%,0
*ENDIF
*ENDDO
!__________Salvo il modello__________
SAVE
FINISH!============================== ! Fine dati preprocessore
/EOF!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
!§§ §§
!§§----------------- UNITA' DI MISURA ---[KN - m] --------------------§§
!§§ Attenzione: dato che la massa è sempre in kg, la densità del mat. §§
!§§ va espressa in kg/mc e l'accelerazione deve tener conto di questo §§
!§§ §§
!§§-------------------- PROCEDURA GENERALE -------------------------- §§
!§§ Salvare questo file nella cartella di lavoro di Ansys insieme ad §§
!§§ un file di input per il modello (altrimenti è possibile crearlo §§
!§§ tramite wizard durante l'esecuzione del programma). Poi aprire §§
!§§ Ansys e digitare in sequenza nella barra di comando: §§
!§§ *ULIB,ROBALG,TXT §§
!§§ *USE,MAIN §§
!§§ §§
!§§-------------------- PROCEDURA GENERALE -------------------------- §§
!§§ Listato per l'analisi statica non lineare di una struttura 2D al §§
!§§ variare del suo stato di danneggiamento. La struttura di esempio 觧
!§§ composta da 8 travi disposte a stella e incastrate agli estremi e §§
!§§ è soggetta ad una forza verticale crescente nel nodo centrale. Si §§
APPENDICES 217
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
APPENDICES 218
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
!§§ non vengono ripetute, quindi la prima riga di PU (D=0) non viene §§
!§§ sovrascritta, mentre la seconda (D=1) viene sovrascritta senza §§
!§§ fare analisi ma basandosi sul risultato del max per D=1 che è §§
!§§ stato opportunamente salvato in precedenza. In dettaglio, l'array §§
!§§ PU ha un numero di righe pari ai livelli di danno DMAX+1 (ovvero §§
!§§ da 0 a DAMX) ed un numero di colonne pari a 3: nella prima colonna§§
!§§ è indicato il numero dell'elemento critico (ovvero da rimuovere §§
!§§ rispetto alla configurazione critica precedente per avere una §§
!§§ nuova congigurazione critica per il nuovo livello di danno; nella §§
!§§ seconda colonna viene salvato il carico ultimo relativo a quella §§
!§§ configurazione; nella terza colonna viene salvata l'informazione §§
!§§ relativa alla convergenza dell'analisi (0 = no conv, 1= conv.). §§
!§§ §§
!§§ Aggiornati INDEX='MIN' INDEX='MAX' §§
!§§ a D = 2 ______________ _____________ §§
!§§ EL PU CV EL PU CV §§
!§§ Liv. danno ______________ _____________ §§
!§§ 0 0 263 1 0 263 1 §§
!§§ 1 1 192 1 4 198 1 §§
!§§ 2 -1 -1 -1 -1 -1 -1 §§
!§§ D ... ... ... ... ... ... §§
!§§ DMAX -1 -1 -1 -1 -1 -1 §§
!§§ §§
!§§------------------------- ARRAY PUD ------------------------------ §§
!§§ E' un array di dimensioni E*3 inizializzato ad ogni nuovo ciclo §§
!§§ in cui aumenta il livello di danno con valori pari a -1 nella §§
!§§ seconda e terza colonna, mentre nella prima rimangono sempre §§
!§§ elencati i numeri identificativi da 1 ad E degli elementi che ad §§
!§§ ogni nuovo ciclo di danno vengono rimossi uno ad uno nelle analisi§§
!§§ NB: all'aumentare del livello di danno le analisi da effettuare §§
!§§ diminuiscono, quindi il vettore, che viene inizializzato ad ogni §§
!§§ nuovo ciclo di danno, viene riempito su un numero sempre inferiore§§
!§§ di righe. In particolare, gli elementi rimossi permanentemente, §§
!§§ ovvero gli elementi che fanno parte della configurazione critica §§
!§§ individuata per il livello di danno precedente, non verranno §§
!§§ considerati per la rimozione nel livello di danno corrente, perciò§§
!§§ le righe del vettore PUD corrispondenti a questi elementi non §§
!§§ verranno sovrascritte e conterranno quindi sempre i valori -1 di §§
!§§ inizializzazione nella seconda e terza posizione. Tramite questo §§
!§§ vettore quindi è possibile individuare le configurazione critica §§
!§§ per il livello di danno precedente, rimuovendo idealmente dalla §§
!§§ struttura integra gli elementi indicati dal numero presente nella §§
APPENDICES 219
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
!§§ prima colonna di PUD che sono seguiti da -1 nelle colonne seguenti§§
!§§ §§
!§§ _____________ ______________ ______________ §§
!§§ D=0 EL PU CV D=1 EL PU CV D=2 EL PU CV §§
!§§ _____________ ______________ ______________ §§
!§§ 1 192 1 1 -1 -1 1 -1 -1 §§
!§§ 2 194 0 2 110 0 2 50 1 §§
!§§ 3 197 1 3 100 1 3 -1 -1 §§
!§§ ... ... ... ... ... ... ... ... ... §§
!§§ E 197 0 E 197 0 E 40 0 §§
!§§----------------------- ARRAY ELMESH ----------------------------- §§
!§§ e' un array che contiene le informazioni sulla meshatura degli el.§§
!§§ operata in automatico dal programma. In particolare, ogni el. da §§
!§§ rimuovere (numerato da 1 ad E) viene suddiviso in un numero MESH §§
!§§ (dispari) di sottoelementi. Il sottoelemento centrale viene per §§
!§§ chiarezza rinominato come l'elemento principale, mentre gli altri §§
!§§ sottoelementi seguono una numerazione progressiva. Tutti i nuovi §§
!§§ sottoelementi vengono registrati nella matrice ELMESH lungo la §§
!§§ colonna di indice corrispondente all'elemento principale suddiviso§§
!§§ in modo che quando nel corso delle analisi è prevista la rimozione§§
!§§ dell'el. principale E, vengono killati insieme tutti i sottoelem. §§
!§§ in cui è stato suddiviso, registrati nella colonna corrispondente.§§
!§§ | da rimuovere | fissi | §§
!§§ ET(iniz) = 6 _______________________ §§
!§§ E = 4 1 2 3 4 5 6 §§
!§§ MESHING = 'y' _______________________ §§
!§§ MESH = 5 7 11 15 19 23 27 §§
!§§ | 9 13 17 21 25 29 §§
!§§ V 1 2 3 4 5 6 §§
!§§ ET(new) = ET(iniz)*MESH = 30 10 14 18 22 26 30 §§
!§§ ELMESH = MESH * ET = 5 * 6 8 12 16 20 24 28 §§
!§§ §§
!§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
APPENDICES 220
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
APPENDICES 221
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
N,5,0,-1
N,6,-1,-1
N,7,-1,0
N,8,-1,1
!-----Coord. altri nodi------
N,9,0,0 ! Nodo centrale
!---Def. elem. principali----
NUMST,ELEM,1 ! Parti a numerare da 1
*DO,N,1,NT-1,1
E,9,N ! Numerazione da 1 ad E
*ENDDO
!________Definisco i vincoli_________
*DO,N,1,NV,1
D,N,ALL,0 ! Incastro i nodi da vincolare
*ENDDO
FINISH!============================== ! Fine dati preprocessore
/EOF !§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
APPENDICES 222
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
The analysis parameter are also reported, that have been asked by the program wizard
immediately before the analysis running: PS is the width of the first step force, while PSMAX
and PSMIN are the maximum and minimum step number to be used in the automatic time-
stepping performed from Ansys solution algorithm. When these two parameters are reported
to be zero, then the arc-length method has been used as solving algorithm and the parameter
AL will be equal to ‘y’. In this case the automatic time stepping cannot be used, but a
further parameter has to be specified, i.e. the maximum fixed displacement, reported as
SMAX. in this case the NC parameter indicates the identification number of the control node,
which this maximum displacement SMAX is assigned to. If no arc-length method was used
instead, the SMAX parameter will be empty and the value of zero will be reported, as in the
output file showed here, and the NC parameter indicate only the identification number of
the node used to trace the pushover-curve provided as control output after the analysis of the
integer structure, but has no particular role in the analysis.
Eventually, in the last part of the file the analysis outcome are listed: the first column
represent the different damage levels considered, form zero (integer structure) to DMAX. The
first group of three column refers to the analysis performed for the curve of minima, while
the second group refers to the analysis for the curve of maxima. For each group the
configuration of element, the ultimate resistance and the convergence of the analyses are
listed. In the case used as example the maximum damage level has been set equal to the
number of elements, i.e. the 8 beams forming the star. Therefore not only the values of
ultimate resistance corresponding to damage level d = 0 will be the same for the maxima
and minima columns (since it refers in both case to the integer structure), but also the value
corresponding to damage level d=8, since both refers to the null structure.
Due to the heuristic used, explained in detail in the second part of this work, the minima or
maxima structural configuration corresponding to each damage level d can be obtained by
removing from the integer structure all the elements listed in the correspondent column (the
second or the fifth column respectively) from the first damage level to the considered one d.
The number listed under the columns CV indicate the convergence of each analysis. The
value 0 indicates that the last analysis step had non convergence, while the value 1 indicates
a converged final step. This parameter can be used as control test when the arc-length
method has been used, since in this case each analysis should reach the convergence. When
instead no displacement control is performed, the analysis stops when the apex of the
response curve is reached, therefore the CV parameter will be always 0. This lack in
convergence however will not affect the ultimate resistance, since it will be computed from
the last converged step anyway.
APPENDICES 223
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
______________________FILE TITLE_______________________
---------FILE TITLE-------|-----------JOBNAME----------|
STELLA PP ROBALG
____________________MODEL PARAMETERS___________________
---NV--|---E---|-DMAX--|-INPUT-|--MESH-|---NT--|---ET--|
8 8 8 STELLA 5 41 40
___________________ANALISYS PARAMETERS__________________
---PS--|-PSMAX-|-PSMIN-|---NC--|--FDIR-|---AL--|--SMAX-|
500 1000 100 9 Y n 0.00000
________________________ROBCUR____________________________
-----|---------CFG MIN---------|--------CFG MAX----------|
--D--|-EL--|-----UL------|-CV--|-EL--|-----UL------|-CV--|
0 0 316465.814 0 0 316465.814 0
1 1 269259.491 0 3 287964.361 0
2 5 217067.864 0 7 260930.870 0
3 2 176972.425 0 2 217832.099 0
4 6 134520.175 0 8 179332.082 0
5 4 95029.264 0 4 135622.378 0
6 8 53424.792 0 6 96046.396 0
7 3 13276.250 0 1 48023.198 0
8 7 0.000 1 5 0.000 1
11.4.3 SCREENSHOT
In the following figure some screenshots from the program running are reported. The first
image refer to the pre-processing phase, where the model is created or imported from a text
file and then the element meshing is performed in the structure. Then the three following
images shows the stress and strain development related to the incremental static analysis
performed on the integer structure, while the fifth image is the corresponding pushover
curve obtained. In the last image the robustness histogram obtained at the end are shown.
APPENDICES 224
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
APPENDICES 225
STRUCTURAL INTEGRITY: ROBUSTNESS ASSESSMENT & PC SUSCEPTIBILITY
1. INDEX OF CONTENTS
2. LIST OF FIGURES
3. LIST OF TABLE AND EQUATION
4. LITERATURE
INDEX OF CONTENTS
SYNTHESIS OF THE WORK 7
INTRODUCTION 17
1 SAFETY AND STRUCTURAL INTEGRITY 19
2 RISK HANDLING 22
CONCLUSIONS 183
9 PROGRESSIVE COLLAPSE SUSCEPTIBILITY 185
9.1 LONG SUSPENSION BRIDGE ................................................................................................................................185
9.1.1 Dynamic amplification...........................................................................................................................185
9.1.2 Triggering and propagation ..................................................................................................................186
9.1.3 Collapse standstill and further developments........................................................................................187
9.2 R.C. FRAMED BUILDING .....................................................................................................................................189
10 PROPOSED ROBUSTNESS ASSESSMENT METHOD 191
10.1 GENERAL FRAMEWORK......................................................................................................................................191
10.2 ROBUSTNESS CURVES ........................................................................................................................................192
10.2.1 Main aspects and limitations .................................................................................................................192
10.2.2 Procedure summary and further efforts.................................................................................................193
APPENDICES 195
11 ROBALG CODE 197
11.1 GETTING STARTED .............................................................................................................................................197
11.2 MAIN FEATURES ................................................................................................................................................197
11.3 CODE .................................................................................................................................................................198
11.4 EXAMPLE FILE AND SCREENSHOT...........................................................................................................220
11.4.1 INPUT FILE ..........................................................................................................................................221
11.4.2 OUTPUT FILE ......................................................................................................................................222
11.4.3 SCREENSHOT.......................................................................................................................................224
LIST OF FIGURES
Fig. 1: Two non robust behaviours: the Airfrance Concorde crashing and the Minnesota Bridge collapse._________ 14
Fig. 2: Safety and structural integrity _______________________________________________________________ 20
Fig. 3: Main requirements for the three limit states defined. _____________________________________________ 20
Fig. 4: Handling with risk: prevent or presume? ______________________________________________________ 22
Fig. 5: Decomposition of collapse probability factors __________________________________________________ 24
Fig. 6: Deterministic, probabilistic and pragmatic approach_____________________________________________ 26
Fig. 7: Qualitative definitions of structural robustness__________________________________________________ 31
Fig. 8: Qualitative robustness slopes of non robust (A) and robust (B) structures. ____________________________ 32
Fig. 9: Event tree for risk-based assessment of robustness [Faber, 2007b] __________________________________ 33
Fig. 10: Energy based robustness indicator [Biondini&Restelli, 2008] _____________________________________ 39
Fig. 11: Exposure, vulnerability and robustness according to [Faber, 2006b] _______________________________ 42
Fig. 12: A domino collapse _______________________________________________________________________ 43
Fig. 13: Emblematic cases of structural collapses _____________________________________________________ 47
Fig. 14: Collapse resistant structures _______________________________________________________________ 48
Fig. 15: Disproportionate collapse typologies ________________________________________________________ 50
Fig. 16: Accidental design situation [EN 1992-1-1-7] __________________________________________________ 60
Fig. 17: Strategies and methods for structural integrity _________________________________________________ 67
Fig. 18: Direct and indirect methods _______________________________________________________________ 68
Fig. 19: Redundant road network and fusible-protected electric circuit. ____________________________________ 71
Fig. 20: Structural integrity achieved by redundancy (above) or compartmentalization (below). _________________ 72
Fig. 21: Dependability main components and their logical interrelation. ___________________________________ 74
Fig. 22: Robustness evaluation chart and resulting structural response slopes. ______________________________ 80
Fig. 23: Indicators proposed for the quantitative evaluation of robustness.__________________________________ 85
Fig. 24: Comparison between the first and second indicator proposed _____________________________________ 88
Fig. 25: Problems to face in a structural collapse investigation___________________________________________ 95
Fig. 26: Nonlinearity sources _____________________________________________________________________ 99
Fig. 27: Geometrical nonlinearities _______________________________________________________________ 100
Fig. 28: Load-deflection relationship for 3 models of catenary action [Powell, 2004] ________________________ 102
Fig. 29: Instability in plastic range________________________________________________________________ 103
Fig. 30: Mechanical nonlinearities ________________________________________________________________ 104
Fig. 31: Plastic hinge calibration factors: choice of hinge type, length and location. _________________________ 106
Fig. 32: Hinge type choices: plastic benefit, material hardening and hinge length.___________________________ 107
Fig. 33: Plastic hinge length estimation ____________________________________________________________ 108
Fig. 34: Solution objectivity _____________________________________________________________________ 110
Fig. 35: Solution regularization: cause of error and solution a (E.L.T. value) ______________________________ 112
Fig. 36: Solution regularization: solution b (hardening) and c (hinge length)_______________________________ 113
Fig. 37: Pushover analysis aim [Pinho 2007 b] ______________________________________________________ 116
Fig. 38: Nonlinear static analysis typologies ________________________________________________________ 118
Fig. 39: Newton-Raphson (left) and modified Newton-Raphson (right) methods [Pinho 2007 a] ________________ 119
Fig. 40: Arc-length method with generic constraint surface (a) and orthogonal iterative changes (b) ____________ 120
Fig. 41: Computational complexity for identifying all damaged configuration of a structure.___________________ 128
Fig. 42: element of difficulty in search of optimum by probabilistic algorithms______________________________ 131
Fig. 43: General SA algorithm and application to the considered structural problem_________________________ 136
Fig. 44: Main aspects of SA application to the considered structural problem ______________________________ 137
Fig. 47: Heuristic optimization for the robustness curve problem.________________________________________ 142
Fig. 48: Number of exhaustive and heuristically reduced combinations ___________________________________ 144
Fig. 49: Developed algorithm for reduced analyses. __________________________________________________ 146
Fig. 50: Overcoming of problems due to disregarding the imperfection modelling (a) ________________________ 151
Fig. 51: Overcoming of problems due to disregarding the imperfection modelling (b) ________________________ 151
Fig. 52: Star structure__________________________________________________________________________ 156
Fig. 53: Truss structure_________________________________________________________________________ 157
Fig. 54: Vierendel structure _____________________________________________________________________ 158
Fig. 55: Comparison of static indeterminacy level ____________________________________________________ 159
Fig. 56: High and low connection_________________________________________________________________ 160
Fig. 57: Different type of static indeterminacy level ___________________________________________________ 161
LITERATURE
Agarwal et al. 2005 Agarwaal J., Blockley D., Woodman N., 2005: “Vulnerability of
structural systems”, Structural Safety 25, 2003
ANSYS 2007 “Advanced analysis techniques guide, R. 11.0”, Ansys Inc. & Ansys
Europe, Jan 2007
ASCE 7-02 “Minimum design loads for buildings and other structures”,
American Society of Civil Engineers, Reston, VA, 2002.
ASCE 7-98 C1.4 “Minimum design loads for buildings and other structures”,
COMMENTARY C1.4: GENERAL STRUCTURAL INTEGRITY
Astaneh-Asl 2003 Astaneh-Asl A.:“Progressive collapse prevention in new and
existing buildings”, Proc. of the 9th Arab Structural Engineering
Conference, Abu Dhabi, UAE, Nov. 29 – Dec. 01, 2003
Augusti et al. 1984 Augusti G.,Baratta A., Casciati F., 1984: “Probabilistic methods in
structural engineering”, Chapman and Hall ed., 1984.
Augusti et al. 2005 Augusti G., Schuëller G.I., Ciampoli M.: “Safety and Reliability of
Engineering Systems and Structures”, Proc. of the 9th International
Conference on Structural Safety and Reliability (ICOSSAR '05),
Rome, 2005.
Augusti et al. 2007 Augusti G., Borri C., Niemann H.: “Is Aeolian risk as significant as
other environmental risk?”, 2007.
Ausiello&al. 2003a Ausiello G., D'Amore F., Gambosi G.: “Linguaggi modelli
complessità”. Franco Angeli Ed., 2003
Ausiello&al. 2003b Ausiello G., Crescenzi P., Gambosi G., Kann V., Marchetti
Spaccamela A., Rotasi M.: “Complexity and approximation”,
Springer Verlag, 2003
Avizienis et al. 2004 Avizienis A., Laprie J-C., Randell B.: “Dependability and its
Threats: A Taxonomy”. Proc. IFIP 18th World Computer Congress,
Building the Information Society, pp.91-120. ed. R. Jacquart,
Kluwer Academic Publishers, Toulouse, France, 22-27 August
2004.
Baldassino 2006 Baldassino N.: “Some notes concerning robustness” 3rd national
congress on collapse and reliability of civil structures (CRASC’06),
“Università degli Studi di Messina”, Messina, Italy, 20-22 April
2006.
Bangert 2005 Bangert P.: “Downhill simplex methods for optimizing simulated
annealing”, Proceedings of ALGORITMY 2005, pp. 341,347, 2005.
Bathe 1996 Bathe K.J.: “Finite Element Procedures”, Prentice-Hall, 1996.
Biodini et al 1999 Biondini F., Bontempi F., Malerba P.G.: “Analisi di affidabilità di
strutture di c.a./c.a.p. mediante simulazione in campo non lineare”,
Studi e Ricerche, Scuola di specializzazione in Costruzioni in C.A.,
Fratelli Pesenti, 1999
Giuliani&Prisco 2008 Giuliani L., Prisco V., “Nonlinear analysis for progressive collapse
investigation on reinforced concrete framed structures”, 2008
Structures Congress – Crossing Borders, Vancouver, Canada, 24-26
April 2008.
Giuliani&Wolff 2006 Giuliani L., Wolff M.: “Strategie per il conseguimento della
robustezza strutturale: connessione e compartimentazione”, 3rd
national congress on collapse and reliability of civil structures
(CRASC’06), “Università degli Studi di Messina”, Messina, Italy,
20-22 April 2006.
Grierson et al. Grierson D.E., Xu L.. Liu Y.: “Progressive failure analysis of
buildings subjected to abnormal loading”, 2005.
GSA 2003 GSA Guidelines - Progressive Collapse Analysis and Design
Guidelines for New Federal Office Buildings and Major Renovation
Projects, June 2003.
Gulvanessian 2006 Gulvanessian H., Vrouwenvelder T.: “ Robustness and the
eurocodes”, Structural Engineering international, SEI Vol. 16, No.
2. May 2006.
Haberland 2007 Haberland M.: “Progressiver Kollaps und Robustheit”, Master
thesis at Hamburg University of Technology, Structural Analysis
and Steel Structures Institute, Hamburg 2007.
Hamby 1990 Hamby E.C.: “ Bridge deck behaviour”, E&FN SPON , Chapman &
Hall Printer, second edition, Sept. 1990.
Holt&Hartmann 2008 Holt R., Hartmann J.: "Adequacy of the U10 & L11 Gusset Plate
Designs for the Minnesota Bridge No. 9340 (I-35W over the
Mississippi River)", Interim Report, National Transportation Safety
Board, January 2008.
ISO 2394: International Standard ISO/FDIS 2394: “General. principles on
reliability for structures”, 1998.
Kirkpatric et al. 1983 Kirkpatrick S., Gelatt C. D., Vecchi M. P., 1983 “Optimization by
Simulated Annealing”, Science, Vol 220, Number 4598, pages 671-
680, 1983.
Korman 1987 Korman R.: "Flawed Connection Detail Triggered Fatal
L'Ambiance Plaza Collapse." ENR, October 29, 1987.
Krakauer 2005 Krakauer, D.C.: “Robustness in Biological Systems: a provisional
taxonomy”, Complex Systems Science in Biomedicine. Kluwer,
2005.
Lamont et al. 2006 Lamont S., Lane B., Jowsey A., Torerro J., Usmani A., Flint G.,
2006 “ Innovative structural engineering for tall buildings in fire”,
Structural Engineering international, SEI Vol. 16, No. 2. May 2006.
Laprie 1992 Laprie J.-C.: “Dependability: Basic Concepts and Terminology”,
Springer-Verlag, 1992.
Levy&Salvadori 1992 Levy M., Salvadori M.: “Why buildings fall down – how structures
fail”, W. W. Norton & Company, 1994.
Smith 2006 Smith J.W.: “Structural robustness analysis and the fast fracture
analogy”, Structural Engineering international, SEI Vol. 16, No. 2.
May 2006.
Sørensen&Christensen 2006 Sørensen J.D., Christensen H.H.: “Danish requirements for
robustness of structures : background and implementation”,
Structural Engineering international, SEI Vol. 16, No. 2. May 2006.
Sozen et al. 1998 Sozen M.A., Thornton C. H., Corley W.G.: “The Oklaoma city
bombing: structure and mechanisms of the Murrah building“,
ASCE 120 Journal of performance of constructed facilities, Aug
1998.
Starossek 1997 Starossek U.: “Zum progressiven Kollaps mehrfeldriger
Brückentragwerke“, Bautechnik, pp. 443-453, July 1997.
Starossek&Wolff 2005 Starossek, U., Wolff, M. 2005: “Design of collapse-resistant
structures”, presented at 2005 Workshop “Robustness of Structures”
organised by the JCSS & IABSE WC 1, Garston, Watford, UK
November, 28-29, 2005.
Starossek 2006 a Starossek U.: “Progressive collapse of bridges - aspects if analysis
and design”, invited lecture, International symposium on sea-
crossing long-span bridges, Mokpo, Korea, Feb. 15-17, 2006.
Starossek 2006 b Starossek U.,: Progressive collapse of structures: Nomenclature and
procedures. Structural Engineering International 16(2): 113-117,
2006.
Starossek 2007 Starossek U.: “Typology of progressive collapse”, Engineering
Structures 29(9): 2302-2307, 2007.
Starossek 2008 Starossek U.: “Progressiver Kollaps von Bauwerken” (Progressive
collapse of structures). In Betonkalender 2008: Part VIII. Berlin:
Wiley-VCH, Ernst & Sohn, (in German).
T.U. 2005 Testo Unitario - Norme tecniche per le costruzioni, D.M. 14/09/05,
Italy 2005 (in Italian).
Turing 1937 Turing A.M. : “On computable numbers, with an application to the
Entscheidungsproblem”, Lond. Math. Soc. 42, pp. 230-265, 1937)
UFC 4-023-03 “Design Of Buildings To Resist Progressive Collapse, Unified
Facilities Criteria” (UFC), 25 January 2005.
Val&Val 2006 Val D.V., Val E.G.: “Robustness of frame structures”, Structural
Engineering international, SEI Vol. 16, No. 2. May 2006.
Vlassis et al. 2006 Vlassis A.G., Izzuddin B.A., Elghazouli A.Y., Nethercot D.A.:
“Design oriented approach for progressive collapse assessment of
steel framed buildings”. Structural Engineering international, SEI
Vol. 16, No. 2. May 2006.
Wempner 1971 Wempner, G.A.: “Discrete approximation related to nonlinear
theories of solids”, International Journal of Solids and Structures,
7:1581-1599, 1971.
Wada et al. 2006 Wada A., Ohi K., Suzuki H., Konho M., Sakumoto Y.: “A study on
the collapse control design method for high-rise steel buildings”,
Structural Engineering international, SEI Vol. 16, No. 2. May 2006.
Wisniewsky et al. 2006 Wisniewsky D., Casas J.R., Ghosn M., 2006: “Load capacity
evaluation of existing railway bridges on robustness quantification”,
Structural Engineering international, SEI Vol. 16, No. 2. May 2006.
Wolff&Starossek 2008 Wolff M., Starossek U.: “Robustness assessment of a cable-stayed
bridge”, 4th international conference on bridge maintenance, safety
and management, Seoul, Korea, 13-17 July 2008.
Yan&Chang 2006 Yan D., Chang C.: “Vulnerability of long-span cable-stayed bridges
under terrorist event”, Proceedings of the International Conference
on Bridge Engineering - Challenges in the 21st Century, Hong
Kong, 2006.