Professional Documents
Culture Documents
ABSTRACT: The architectural preservation process is generally based on a sequence of anamnesis and anal-
ysis, diagnosis, therapy, control and prognosis. In the analysis phase, an objective way to assess the safety
of the structure is essential. The outline of the assessment process is described both in the ISCARSAH
Recommendations and ISO13822, referring to determining the actual safety or reliability of existing (histori-
cal) structures. This paper focuses on the use of reliability based assessment techniques, using a probabilistic
analysis accounting for the uncertainty within the analysis in an objective manner. As a result the user obtains a
value for the actual reliability or failure probability. Besides a general description of its possibilities within
the analysis of historical structures, an application is treated to illustrate the methodology as well as its
limitations.
1321
Change of use, Routine in-
Cause spection, Degradation,…
Phase Responsible
structural engi-
Visual inspection neer alone
Preliminary
evaluation
Study of actions to
be taken
Responsible
Anamnesis
Phase structural
Detailed investigation, engineer alone
Doubt yes
Site investigation,
about Detailed structural Research
safety analysis institute-specialists
no Responsible
engineer with a
team of experts
Diagnosis
no yes Phase
Safety Large
Inspect structure
sufficient conse-
Confer together
quences
and decide as a team
yes no
No action taken
Therapy
Intensify inspections
Phase Phase
Continuous assessment, Reliability based design
reliability analysis
1322
to various parameters. An overview is given elsewhere
Random Variables (Schueremans, 2001):
R, E, a, θ
– Number of people put to danger (np );
– the type of possible damage (life injury, economi-
FORM/ FEM: cal damage, social-cultural damage, environmental
SORM
damage);
Reliability Structural Nascom – the preset service life of the construction (tL );
analysis analysis
the level in which people are exposed to risk (public
Numerical
PERMAS buildings, bridges, off-shore constructions, ...) and;
integration
– the level in which people are warned beforehand
Diana and can be put to safety (gradual failure with
Monte Carlo
β, pf visible damage against sudden collapse without
&
Ansys warning) (W ).
Directional
Sampling
Nastran,...
βT, pfT
is the actual reliability sufficient? It is in this detailed The formula proposed in eq. 1 to determine nominal
analysis phase that a reliability based analysis delivers target failure probabilities is a mix of proposals pre-
added value. sented by different authors. It is very suitable as it
accounts for a social criterion (Sc ) that can be rein-
terpreted to encapsulate the importance of historical
buildings or the preservation value. This is indicated
3 RELIABILITY BASED ASSESSMENT
in Table 1 (in Italic).
3.1 Framework
3.3 Reliability algorithms and (commercial)
To be able to make a probabilistic design or assess-
software
ment according to a preset target failure probability,
following requirements need to be fulfilled, Figure 2: The generalized reliability problem can be outlined
starting from the basic reliability problem. The basic
– accurate reliability algorithms in a user friendly reliability problem considers only one load (E) and one
environment; resistance (R). Starting point is that both the load effect
– knowledge of random variables for the design vari- (E) as the resistance (R) are random variables. Each
ables (R: resistance, E: sollicitions, a: geometry, is described by a known probability density function:
θ: model uncertainty) and availability of materials fE (e) and fR (r). In general, both the load effect (E) and
models and related data; the resistance (R) are a function of time t. The safety
– appropriate structural models; limit state will be violated when at a certain moment
– preset target failure probabilities (pf ,T ) or corre- in time, R(t) − E(t) < 0. The chance or probability that
sponding target reliability values (βT ). this will happen equals the probability of failure pf .
As both R and E are a function of time, pf is also a
function of time. Because of mathematical complex-
3.2 Target failure probabilities pf ,T ity, the probability density functions are transformed
into time-invariant probability density functions. As a
For the decision process that follows up on a safety
consequence, the reliability analysis is performed for
assessment of an existing structure, it is important to
a preset reference period or so-called design service
define target safety levels that can serve as decision
life tL :
criteria. In relation to fulfilling such criteria, existing
(historical) structures will differ from new structures at
the design stage. Furthermore, determining the target
probability of failure is not a technical problem solely. Eq. (2) is generalized as follows:
Whether or not an historical building should meet
the target probability of failure value pfT = 5.10−4
according to Eurocode (EN1990, 2002), is subject of
discussion. Several authors suggest widening the dis- in which g(R, E) is called the limit state function
cussion and proposing a differentiation with respect (LSF); the probability of failure is identical with the
1323
Table 1. Factors influencing the nominal target probability Table 2. Development of integrated software for structural
of failure (used values in bold). reliability analysis.
1324
– Sensitivity coefficients (αi ) are obtained for all ran- Table 3. Reference probability density functions for the
dom variables included within the reliability analy- main design variables.
sis.These sensitivity coefficients directly reflect the
importance of a specific parameter within the anal- Mean Standard
Name of value deviation
ysis. And therefore, these parameters are to be con-
variable Symbol Dim. PDF µx σx
sidered with priority when gathering more accurate
information or designing a structural intervention Action Effects (E )
procedure;
– The impact of structural interventions can be Permanent G kN/m2 Normal Gk 0.10µX
assessed in a pre-design phase in a similar man- Imposed – Q kN/m2 Gumbel 0.2 Qk 1.10µX
ner, allowing different alternatives to be judged in 5 year
an objective manner. Imposed – Q kN/m2 Gumbel 0.6 Qk 0.35µX
50 year
The major drawbacks are: Wind – W kN/m2 Gumbel 0.5 Wk 0.40µX
1 year
– For complex structures, an automatic interface is
Wind – W kN/m2 Gumbel 0.7 Wk 0.25µX
required with an accurate unbiased structural model 50 year
of the structure. Often, this will be a finite element
model of the structure, in which use is made of Resistance (R)
typical non-linear constitutive relations to simulate Steel R kN/m2 Lognormal Rk + 2σR 0.08µR
the real structural behavior. This interface is not Timber Rm kN/m2 Lognormal Rm,0 0.25µR
Concrete fc kN/m2 Lognormal fc,0 0.06–
available for all (generic) finite elements codes or
0.30µR
numerical tools used for assessment at this moment;
– The outcome of the reliability analysis thus is Masonry fc kN/m2 Lognormal / /
related to the accuracy of the underlying structural
Geometry (a)
model. Model uncertainty is often added to cover Geometry a / Normal a0 Variable
the uncertainty within the accuracy of the numerical
model; Model uncertainty (θ )
– The outcome of the reliability analysis is directly On action θE / Normal 1.00 0.10
related to the accuracy of the probabilistic models On resistance θR / Normal 1.10 0.07
of the random variables involved. Often, the num-
ber of data available is (too) limited to establish
reliable probabilistic load or material models;
– The processing time is related to the complexity
no probabilistic material model is available at this
of the problem, the shape of the limit state func-
moment. Some preliminary proposals have been made
tion, the resulting reliability and the number of
by few authors (Schueremans and Van Gemert, 2001,
random variables included within the analysis. Pro-
2007; Proske et al. 2006).
cessing time increases in cases of discontinuous
For the load effect, an international consent is avail-
limit state functions with strong curvatures and
able on the probability density functions to be used.
large dimensional problems having a small fail-
A non-limitative reference list of probability density
ure probability. This however is often the case in
functions for the main design variables is given in
historical structures.
Table 3 (Diamantidis, 1999, 2001; JCSS, 2006).
3.5 Probabilistic models for random variables
3.6 Developments in adjacent fields of
The reliability based assessment requires probabilis-
research/application
tic models for the different random variables involved
within the analysis. Certainly the Resistance variables Probabilistic techniques are not common use in daily
(R) pose difficulties within the context of histori- design or assessment. The methodology is mainly used
cal structures. These are related to the probabilistic when the complexity of the problem is beyond the cal-
modeling of material properties, required within the ibration domain of current codes, when the safety of
constitutive modeling. the structure is beyond the traditional design or when
The joint committee of Structural Safety has set the consequences of failure are large. Remark that
up several workgroups who take action to establish these boundary conditions often apply for historical
probabilistic material models. For several construction buildings.
materials, models have been established, for exam- To put the methodology in a broader reference
ple: steel (2002), concrete (2002), soil and timber frame, several examples of various application
(2006). For masonry, and more specifically historical domains demonstrate its applicability in design
masonry with its irregular possibly multi-leaf layout, (T2881, 1999; Diamantidis, 2001): design of ships and
1325
Figure 3. Global overview of the critical part A – city side.
off-shore structures; design of tunnel segments (TNO, Figure 4. Reconstruction of structural layout of city wall.
Delft, Nl); safety assessment of the London Eye (UK,
TNO Bouw); risc analysis of the derailment guidance These are no longer present. On the city side as well
for high speed line (Netherlands, Simtech); Risc- as on the rural side, the wall was lined with a sloped
analysis Jamuna Bridge (Haskoning, Bangladesh), embankment (F, E) covering the foundation pillars and
service life prediction LNG Terminal (Dabhol, India, arches. This embankment is no longer present at the
KULeuven); infrastructure dams such as Dijkring or rural side, and only partly at the city side. For the con-
Oosterschelde-stormvloedkering (Bouwdienst Rijk- struction of the wall, a local type of lime-sandstone
swaterstaat, Nl). (Diegemse Zandsteen) was used. An iron containing
sandstone (Diestiaan sandstone) was mainly used for
decorative purposes, for example in the inner arcades.
4 APPLICATION
4.2 Safety assessment
Possible applications are multiple. Up till now, applica-
tions cover mainly structural components or structures A single arcade – the repetitive structural element - is
that can be simulated with relatively simple mod- taken as an individual control unit. To clarify the ben-
els (with respect to: structural models, number of efit of using a reliability based assessment, different
random variables covering the problem), or complex levels to assess the safety are used:
structures are reduced to models with acceptable com- – Level 0: To obtain a measure for the remaining
plexity (Schueremans et al., 2001; Schueremans & Van safety margin, the structure is checked using nomi-
Gemert, 2004). nal values for the applied loads and resistances. The
The example treated, clearly illustrates the added resistance-load ratio (r) is used as a measure for the
value of the application of this type of analysis both in remaining safety margin with respect to a certain
the assessment of the existing historical structure and ultimate limit state;
in the pre-design phase of possible interventions. – Level I: The analysis is performed according to the
partial safety factor method;
– Level III: The analysis is performed using prob-
4.1 Romanesque city wall of Leuven
abilistic evaluation algorithms based on sampling
The Romanesque city wall of Leuven (B) dates from techniques. An accurate value for the system
1150. The medieval wall is in a severe state of decay. reliability is obtained. The same limit state func-
The part studied in this paper is a piece of the edifice tions that are used for the Level I analysis, are
between the former Biest- and Minneporte (2 gates). accounted for.
This part of the Romanesque city wall near the river
The probability density function and the parame-
Dijle has a total length of about 150 m and comprises
ters for the different random variables are listed in
2 towers.
Table 4. These cover the different types of uncertain-
This example only covers the most critical part of
ties encountered during the survey: accuracy of the
the wall, Figure 3. A global view of the structural lay-
geometry, the uncertainty on the material properties
out of the wall is given in Figure 4. The wall consists
(subsoil and stone masonry), actual loads and future
of an inner or city side (B) and an outer or rural side
loads.
(A). Round foundation arches of 2 m high and 3.5 to
Following limit states are checked, Table 5, (defini-
4 m wide carry the outer wall. The inner wall is a con-
tion of symbols, see Figure 5):
tinuous arcade of 4 m wide round arches with their
tops 3 m above the outer arches. A walkway of 0.90 m – Rotational-equilibrium. The centre of gravity is
width (C) is present on top of the arcades. The outer determined (yg,tot ), accounting for the structural
wall was equipped with shooting holes centred in the geometry and the slant of the wall. As long as
arches of the inner part (D) and crowned by a parapet the value is positive, the centre of gravity remains
also bearing shooting holes. within the cross section, thus the rotation limit state
1326
Table 4. Romanesque city wall – Random variables.
Mean
Random variable PDF value Std dev.
Initial assessment – n = 23
Load:
ρm [kN/m3 ]:density of masonry N 19 1.9
Geometry (n = 17)
Geometry of wall N nom 0.05
Resistance of subsoil:
c [kN/m2 ]: cohesion LN 30.11 15.29
ϕ [◦ ]: friction coefficient N 25.56 4.59
γdr [kN/m3 ]: dry density N 16 1.6
Resistance:
fm : stone masonry strength LN 23.2 4.3
Figure 5. Stress distribution in the subsoil at foundation
Uncertainty: level – left: original situation; right: with foundation strength-
ε [m]: model uncertainty N 0.0 0.01 ening.
Strengthening with reinforced concrete foundation
slab – additional random variables: n = 23 + 5
– Compressive stresses are calculated in the masonry
Load: (σ m,max ) as well as in the soil at foundation level
P [kN]: permanent load N 40 4
(σ gr,max,el and σ gr,max,pl ). Maximum stresses are
q [kN/m2 ]: floor load LN 5 2
ρb [kN/m3 ]: proper weight N 25 1
found at the foundation tip of the pillars, due to the
slant of the wall. For both materials, a non-tension
Geometry (reinforced concrete slab): material model is used. For the soil, a linear-elastic
lslab [m]: length of slab N 4 0.05 (el) as well as an elastic-plastic (pl) material model
wslab [m]: width of slab N 4 0.05 is used.
Legend: N: Gaussian; LN: LogNormal: PDF: probability The results of the safety assessment of part A of the
density function. wall are listed in Table 5. From the Level 0 and Level
I analysis, it is clear that the structural stability is in
Table 5. Results of structural reliability – original situation. doubt. The remaining safety ratio (r = 1.04) is limited.
The limit state function of the stresses in the subsoil is
Stresses Stresses violated. The probabilistic method offers an objective
LSF// Rotation in the in the way to assess the remaining safety, accounting for the
assessment equilibrium masonry subsoil
level [m] [N/mm2 ] [N/mm2 ] present uncertainties: pf = 0.11. This value does not
meet the preset target value.
Level 0 σm,max = 0.74 < σgr,max,pl = 0.50 <
fc,m = 23.2 dg = 0.52
r = 30.5 r = 1.04
4.3 Strengthening – reliability based design
Level I yg,tot = 0.35 > 0 σm,max,d = 1.00 < σgr,max,el,d =
(OK) fc,m,d = 16.2 0.90 > The lack of safety originates from the limited load-
etot = 0.53 > OK dg,d = 0.19 bearing capacity of the soil in combination with
d/6 = 0.29
NOT OK NOT OK the large slant of the wall. This is partly caused by the
removal of the original sloped embankments at the
Level III System reliability β = 1.2 < βT = 4.2 rural and city side. At present, the top part of the foun-
Corresponding failure probability:
pf = 0.11 > pf ,T = 1.7 10−5
dation is above the original ground level. Thus, the
depth of the foundation decreased significantly. To
increase the safety to an acceptable level, two options
are available:
is met. The eccentricity (etot ) is checked. Whenever – Widen the foundation at the support. This will
the eccentricity exceeds the mid-third (d/6), part of reduce the soil stresses, see Figure 5;
the cross-section is in tension. As use is made of a – Increase the load-bearing capacity of the soil by
non-tension material model, the force equilibrium restoring the original sloped embankment at rural
is met using compressive stresses only. This leads side. The latter possibility however was not chosen
to an increased stress level; by the responsible authorities.
1327
5 CHALLENGES
1328
ACKNOWLEDGEMENTS Proske D., Lieberwirth P., Van Gelder P. 2006. Sicher-
heitsbeurteiling historische Bogenbrücken. 16. Dresdner
The authors express their thanks to the Flemish Fund Brückenbausymposium.
for Scientific Research (FWO) for the doctoral grant, Schueremans L, Figeys W., Heirman G., Brosens K., Van
Gemert D. 2006, Application of probabilistic concepts
offered to Els Verstrynge.
in evaluation and strengthening of constructions. WTA-
Tag in: WTA-Almanach 2006, Restauration and Building
Physics, 327–350.
REFERENCES Schueremans L. & Van Gemert D. 2005. Benefit of Splines
and Neural Networks in simulation based structural relia-
Diamantidis D. 2001. Probabilistic Assessment of Existing bility analysis. Structural Safety. 27(3), 246–261.
Structures., A publication of The Joint Committee on Schueremans L. 2001. Probabilistic evaluation of struc-
Structural Safety JCSS. Rilem Publications S.a.r.l. tural unreinforced masonry. Ph. D. Thesis„ KULeuven,
Diamantidisd D. 1999. Koordination und Entwicklung eines Belgium.
probabilistischen Sicherheidskonzepts für neue und beste- Schueremans L., Van Gemert D. & Smars P. 2001 Safety
hende Tragwerke, T2881, Fraunhofer IRB Verlag, ISBN assessment of masonry arches using probabilistic meth-
3-8167-5451-1. ods. International Journal for Restoration of Buildings
EN1990:2002.Basis for design of structures, Eurocode, and Monuments. Aedificatio Verlag – Fraunhofer IRB
European Standard. Verlag, Heft 5, 517–538.
ISCARSAH 2005. Recommendations for the analysis, con- Schueremans L., Van Gemert D. 2004. Assessing the safety
servation and structural restoration of architectural her- of existing structures: reliability based assessment frame-
itage. Draft Revised Version to Align with the ISO 13822 work, examples and application. JCEM Journal of Civil
International Standard (June 13, 2005). ICOMOS. Engineering and Management, 10(2), 131–141.
ISO13822:2003. Bases for design of structures – Assess- Schueremans L., Van Gemert D. 2006. Probability density
ment of existing structures. International Standards functions for masonry material parameters – a way to go
Organization. ?, SAHC 5, New Delhi, India, Vol 2., 921–928.
ISO2394:1998. General principles on Reliability for Struc- Verstrynge E., Ignoul S., Schueremans L., Van Gemert D.
tures. International Standards Organization. 2008. Modelling of damage accumulation in masonry sub-
JCSS 2006. Probabilistic Model Code. Joint Commit- jected to a long-term compressive load. SAHC 6. Full
tee on Structural Safety. Internet publication: URL: paper submitted for publication.
http://www.jcss.ethz.ch/.
Lourenco P.B. 1996. Computational strategies for masonry
structures. Ph.D. Thesis, TUDelft, 206 pages.
1329