You are on page 1of 9

Engineering Structures 30 (2008) 1408–1416

www.elsevier.com/locate/engstruct

Probability based modelling and assessment of an existing post-tensioned


concrete slab bridge
A. O’ Connor a,∗ , Ib Enevoldsen b
a Department of Civil, Structural and Environmental Engineering, Trinity College Dublin, Ireland
b Ramboll Consulting Engineers, Virum, Denmark

Received 18 December 2006; received in revised form 5 July 2007; accepted 13 July 2007
Available online 10 September 2007

Abstract

This paper presents the application of probabilistic techniques to the load capacity assessment of a post-tensioned concrete slab bridge which
was built in 1959 outside Copenhagen. The bridge failed to demonstrate the required capacity following a deterministic assessment and as such
was deemed to require repair or replacement. Following the Danish Roads Directorates stated policy of performing probabilistic assessment,
where deemed beneficial, prior to repairing or replacing, the structure was probabilistically assessed according to the newly developed Danish
guideline for Reliability Based Classification of the Load Carrying Capacity of Existing Bridges. The paper presents the techniques employed
in modelling the critical limit state, as well as the statistical techniques employed in modelling loads, resistance and associated uncertainties.
Ultimately, the structure was shown to have sufficient load carrying capacity and as such the scenario of expensive repairs or replacement was
avoided, with significant financial benefits to the bridge owner.
c 2007 Elsevier Ltd. All rights reserved.

Keywords: Bridge; Probabilistic assessment; Traffic modelling; Load effect; Stochastic

1. Introduction deterministic assessment to identify the critical limit state


for the structure. This deterministic assessment is performed
A common problem among bridge owners/managers is the according to the generalised rules and using generalised
need to reduce spending whilst attempting to operate and partial factors provided by assessment codes. Next a bridge
maintain an increasingly ageing bridge stock. In response to specific probabilistic assessment free from the aforementioned
this challenge the past decade has seen increased interest generalisations results in the determination of a formal
by European bridge owners and managers in the use of probability of failure, p f , for the structure at the critical limit
probabilistic methods for the assessment/management of
state. Comparison of the calculated p f with that specified
their bridges. Employed once a deterministic assessment of
by legal requirements and demonstration that it exceeds these
a structure has rendered a repair/rehabilitate/replace now
requirements is deemed sufficient to validate the safety of the
scenario, the methods have been demonstrated to provide
structure. It is important to stress that at no stage is the safety
significant cost savings where the required safety of the
of the structure compromised rather it is optimised through
structure can be demonstrated by probabilistic methods.
The basic principle is to derive a bridge specific code and the provision of an bridge specific ‘individualised’ safety
consequently a bridge specific safety rating. This involves rating free from the imposed generalisations of deterministic
statistical modelling of load and resistance parameters obtained assessment codes which must necessarily be conservative in
through on-site measurements and from as-built drawings. order to be widely applicable.
Updating of these models may be performed where additional The policy of the DRD has for the past number of years been
information is available. The first stage of the process involves to perform probability based assessment, where it is deemed
to be beneficial, on any structure which failed to demonstrate
∗ Corresponding author. Tel.: +353 18961822; fax: +353 16773072. sufficient load carrying capacity by deterministic means. The
E-mail address: alan.oconnor@tcd.ie (A. O’ Connor). policy has proven extremely beneficial financially through

c 2007 Elsevier Ltd. All rights reserved.


0141-0296/$ - see front matter
doi:10.1016/j.engstruct.2007.07.023
A. O’ Connor, I. Enevoldsen / Engineering Structures 30 (2008) 1408–1416 1409

Table 1
Bridge classification under 4 passage types for case considered

Passage type Class Passage type definition


Normal passage 80 No restrictions on vehicle positions on structure, full dynamic factor applied to vehicles.
Conditional passage I 150 Vehicles positioned in traffic lanes on structure, full dynamic factor applied to vehicles.
Conditional passage II 175 Vehicles positioned in traffic lanes on structure, dynamic factor applied to one vehicle only.
Conditional passage III 200 Only heavy vehicle positioned on structure, no dynamic factor applied.

the avoidance of unnecessary repair/replacement of structures


which had been classed deficient following deterministic
assessment but which are demonstrated to have sufficient
capacity in a probabilistic assessment.
The following paper presents through example the use
of the probabilistic techniques in the assessment of a post-
tensioned concrete slab bridge. The structure failed to achieve
the required load rating for the motorway network in a
deterministic assessment and as such was proposed for
probabilistic assessment at the controlling limit state. The
assessment is performed according to DRD guideline for
reliability based classification of the load carrying capacity of
existing bridges.

2. Bridge classification in Denmark

In Denmark in recent years considerable effort has been Fig. 1. Standard vehicle silhouettes.
put into the development of a rigorous methodology for
the assessment and classification of existing bridges under limit states that are believed to be conservatively modelled
the control of the Danish Roads Directorate (DRD). The according to the deterministic code. If, however, the probability
assessments are mostly dealing with the passage of heavy based assessment is not sufficient for obtaining the required
vehicles, where a heavy vehicle is defined as a vehicle with a classification, the strengthening project can be probability
weight higher than 48 t which needs special permission from based, which under normal circumstances proves less costly
the police. The classification system [1] is based on a set than a strengthening project based on traditional assessment
of standard vehicles representing vehicles with a total weight methods [2].
ranging from 20 t to 200 t. Fig. 1 illustrates some of the standard In the probability based analysis approach the uncertainties
vehicles defined. In a regular deterministic assessment of the of the specific individual bridge and the local traffic situation
bridge class, a traffic load combination is applied comprising can be taken into account consistently. These uncertainties
a standard vehicle of 50 t with a second special permit heavy arise from physical uncertainties in the use and identification
vehicle (weight 50–200 t). The resulting bridge class is equal to of materials, from the load and truck models, from statistical
the weight in tons of the heaviest standard vehicle in the load sources, and finally from the simplification in the structural
combination. Further bridge classes are analysed for passages evaluation models. Furthermore, bridge specific traffic loads
of heavy vehicles with imposed restrictions such as speed limits and traffic frequency can be taken into account. It is obvious
and exclusion of other traffic, Table 1. that the chance of a heavy truck meeting another heavy truck is
In line with the outlined classification system, a special lower for a bridge with a relatively low heavy truck frequency
‘blue’ road network has been defined. Inclusion on the than for a bridge with a high heavy truck frequency. This
network requires initially that bridges have been assessed modelling of limit states and stochastic variables can be used
by applying general deterministic methods with elastic directly to determine the formal annual probability of failure,
or plastic analysis according to the Danish guideline for p f , or the directly related reliability index, β, applying standard
classification of bridges [1]. Where the obtained bridge class techniques such as First Order Reliability Methods (FORM) [3–
is insufficient (< class 100), traditionally the alternatives of 5]. Furthermore, the modelling is the core in all the other
expensive rehabilitation or replacement have been considered, probability based safety management applications such as
following the traditional decision process illustrated in safety updating due to new information, inspection and test
Fig. 2 (left). As previously discussed, in recent years the planning.
DRD have consistently considered an alternative decision The probability based approach is often combined with
process, Fig. 2 (right), i.e. probability based approaches, advanced response models to fully optimise capacity. The
before an expensive strengthening or rehabilitation project results of the probability based assessment have often been
is implemented. Probability based assessment will in many found to considerably raise the bridge class achieved from
cases be fruitful, especially if the bridge has problems with deterministic assessment [2,6–9]. In many cases, these analyses
1410 A. O’ Connor, I. Enevoldsen / Engineering Structures 30 (2008) 1408–1416

Fig. 2. Traditional decision process for upgrading of existing bridges (left) and revised decision process including consideration of the probabilistic based approach
(right).

have resulted in a satisfactory bridge class (i.e. >100) thereby


minimising or avoiding strengthening projects. It is important to
stress that at no stage is the safety of the structure compromised;
rather the bridge specific safety is calculated. Therefore, these
methods have proven to be very beneficial for bridge managers
with large cost savings as results.

3. Deterministic assessment of example bridge

The structure, illustrated in Fig. 3, consists of three spans


of 6.39, 17.72 and 6.39 m, giving a total length of 33.88 m
(including side span overhang of 1.69 m). The overpass bridge,
skewed at 62.67◦ carries 4 lanes of traffic, (2 × carriageways
of 2 lanes each plus an accelerating/merging lane) on the
E47 at Sakskøbing, Fig. 4. The width of the structure is 28
m. Transversely the bridge is supported on 10 columns at
spacings of 3.24 m (centre to centre). The main structure Fig. 3. Nystedvej bridge elevation.
consists of a 0.5 m deep longitudinally post-tensioned slab. Post
tensioning is provided in groups of 8 tendons, with each tendon assessment, and (3) standard SLS checks on deflection, steel
composed of 12 No. 7 mm diameter wires. The tendons have stresses, crack widths etc. Detailed discussion of the results of
a parabolic profile longitudinally. Mild steel reinforcement is deterministic assessment is not provided here.
provided longitudinally and transversely. Shear reinforcement In normal passage the structure was found to have
is not provided in the slab. insufficient hogging moment capacity at the outermost
A deterministic load carrying capacity analysis of the column–slab intersection location (indicated on Fig. 4) to
structure was performed according to the guidelines set out by achieve a Class 100 rating. At this location the classification
the Danish Roads Directorate and national standards [1,10,11]. of the structure was limited to Class 80, i.e. a Class 80
A finite element (FE) model of the structure was developed with and a Class 50 vehicle side by side on the structure, see
the width of modelled deck corresponding to that supported by Fig. 1 for vehicle silhouettes. The plastic capacity of the
6 columns. This width of the structure was reduced to optimise section was calculated using typical stress–strain relationships
solver time for the FE model. It was based upon the required for the concrete, reinforcement and prestressing steel. The
width for positioning of the vehicles and adequate distribution relationships employed for the reinforcing and prestressing
of the sectional forces. The slab stiffness was modelled as steels are illustrated in Fig. 5. The characteristic concrete
orthotropic, with E y = 1/5E x, having been calculated as an compressive strength was 27.5 MPa, with the maximum (i.e.
appropriate ratio. strain corresponding to characteristic compressive strength) and
The deterministic assessment investigated the capacity of the ultimate compressive strains limited to ε y = 0.20% and εult =
structure in terms of (1) plastic bending capacity assessment of 0.35% respectively. The characteristic steel strength, f ytk , was
the slab at the Ultimate Limit State (ULS), where assessment of 510 MPa as was the assumed value for the characteristic rupture
the foundation forces and bearing capacity was also performed, tensile stress, f utk . The characteristic tensile yield stress of the
(2) shear capacity of the slab at ULS, including punching shear second yield point of the prestressing steel, f ptk , was 1422
A. O’ Connor, I. Enevoldsen / Engineering Structures 30 (2008) 1408–1416 1411

Fig. 4. Nystedvej bridge plan.

(a) Mild steel. (b) Prestressing steel.

Fig. 5. Modelled reinforcement stress–strain relationships.

MPa, with the first yield point located at k ∗ f ptk = 70% of The revised decision process outlined in Fig. 2 (right)
f ptk , and the rupture tensile stress, f utk , was 1618 MPa. The suggests that where a structure fails a traditional assessment
plastic strain from the second yield point of the steel, ε y0t = that a probability based assessment should only be carried out
0.1%. The safety factors for concrete, mild and prestressing where it is likely to result in a beneficial assessment. In the case
steel, rupture tensile stress for mild and prestressed steel and E of the structure considered, the capacity is well in excess of the
modulus for mild and prestressed steel were γc = 1.815, γs = minimum required in all of the conditional passage cases and
1.43, γu = 1.43 and γ E = 1.43 respectively [11]. Whereas it is only in the Normal Passage case that it fails to achieve
full development of plastic sectional forces is permitted by the necessary rating. Considering this fact, in addition to the
the Danish codes, plastic re-distribution of sectional forces is restrictive partial safety factors employed in the deterministic
limited to 10%–15% for prestressed concrete where sufficient assessment, and that the transition from a Class 80 to Class
rotational capacity could be demonstrated. 100 vehicle implies only a 22% increase in the weight of the
The classifications calculated for conditional passage types controlling axles of the vehicle (i.e. rear group of axles) it
I, II and III were 150, 175 and 200 respectively, well in excess seems that this structure is a good candidate for probabilistic
of the required Class 100, Table 1. The difference in these assessment.
results is explained by the fact that in Normal Passage the
vehicles may theoretically be placed right at the edge of the
4. Bridge probability based classification
slab (in the area indicated as hard shoulder in Fig. 4) while in
conditional passage the vehicles must remain within the marked
driving lanes. The structure considered is significantly wider The procedure for the probability based safety assessment
than the marked lanes (Fig. 4) with no kerbs preventing traffic of an existing bridge according to the DRD guideline [12] is
from travelling close to the edge. discussed here:
1412 A. O’ Connor, I. Enevoldsen / Engineering Structures 30 (2008) 1408–1416

4.1. Formal requirements for safety assessment employed to describe the loading on, and resistance of the
structure. First, consideration is given to the required safety
The first stage in the process is the selection of the level of the structure.
classification case for which the structure is to be considered
(i.e. normal passage or conditional passage based upon the 5.1. Requirements for safety level
result of the deterministic assessment), and by implication the
formal requirements for a specific safety assessment. In the case The requirement in the ultimate limit state for structural
of the structure considered the required classification is Class safety is specified with reference to failure types and failure
100 in Normal Passage: consequences, i.e. safety class with requirements for the formal
yearly probability of failure p f . The definition of the safety
4.2. Determine critical limit state violation from deterministic
index, β, in this regard is taken from the DRD guideline [12].
analysis
The safety index, β, is formally defined in terms of the
The critical limit state identified by deterministic assessment probability of failure [3–5]:
is exceedance of the ULS slab plastic hogging moment capacity
β = −Φ −1 p f

at the location of the outermost column support in normal (3)
passage. Mathematically exceedance of the flexure limit state for which Φ −1 (·) is the inverse function of the standardised
may be described as: normal distribution. The guideline specifies values, according
g≤0 where g = Mcap − Mapplied (1) to the type of failure, for a 1-year reference period. Failure Type
I deals with ductile failures with the requirements that an extra
with: Mcap the moment capacity of the section and Mapplied the carrying capacity beyond the defined resistance is available, i.e.
induced moment. in the form of strain hardening, for which βt ≥ 4.26. Failure
In the probabilistic assessment the ULS plastic bending Type II includes ductile failure without any extra carrying
capacity of the post-tensioned slab, Mcap , is determined from capacity, for which βt ≥ 4.75. Failure Type III includes states
randomly generated values of the modelled resistance variables.
of brittle failure and instability failures, for which βt ≥ 5.20.
In the example considered the governing random variables of
The failure classes may be directly related to the target lifetime
capacity were identified as the characteristic concrete strength,
(50 year) β values specified for ‘great’ consequence of failure
f cu , and the prestressing steel characteristic strength, f ptk , i.e.
listed in the ISO standard on General Principles on Reliability
Mcap f cu , f ptk . Dependent upon the section analysed other

for Structures [13] and in Eurocode EN 1990 — Basis for
variables such as variations in tendon locations at the critical
Structural Design [14].
cross section may also be considered.
For the post-tensioned slab bridge deck considered, the
In the deterministic assessment, values of the plastic
critical limit state, i.e. flexure, represents a ductile failure
capacity are determined for discrete values of the resistance
mode. The structure is not required to have remaining capacity
properties of the cross section. To obtain values of Mcap in
following exceedance of the plastic ULS capacity and as such
the probabilistic analysis, for randomly generated values of the
Failure Type II — Ductile failure without remaining capacity,
modelled stochastic variables, a response surface is required.
which requires βt ≥ 4.75 is specified.
A response surface is a closed form and differentiable limit
state surface constructed using a polynomial or other suitable The probability of failure p f (and by implication β) is
function fitted to the results obtained from a limited number of computed using the First Order Reliability Method (FORM).
discrete numerical analyses [3]. Computation is performed using the proprietary software
The applied moment: package PROBAN [15].

Mapplied = MDL + MSDL + MLL + Mpar (2) 5.2. Traffic load modelling
where MDL represents the moment due to dead load, MSDL
For a two-lane short span (or short influence length) bridge,
represents the moment due to superimposed dead load MLL
represents the moment due to live load, and Mpar represents in the majority of cases the critical loading events in the extreme
the parasitic prestress moment induced at the critical location. occur due to the meeting events between ordinary trucks or
Discussion of statistical assessment of MDL , MSDL and Mpar is meeting events involving heavy transports with ordinary trucks.
contained in subsequent sections. As such the extreme distribution Fmax of the considered load
The magnitude of MLL is a function of the randomly effect q can be obtained from [16–18]:
generated vehicle weights and positions on the structure. To Fmax (q) = exp(−(ν1 − ν12 )T (1 − F1 (q)))
facilitate evaluation of MLL for these random combinations of
· exp(−(ν2 − ν12 )T (1 − F2 (q)))
weight and location in the probabilistic assessment an influence
surface (i.e. the 2d equivalent of an influence line) is employed. · exp(−ν12 T (1 − F12 (q))) (4)

5. Probabilistic classification and modelling of loads where ν1 and ν2 are the intensities (i.e. number vehicles/day)
of the considered traffic in lane 1 and 2, respectively. ν12 is the
The following section outlines the modelling of the input intensity of meeting events, i.e. of the considered traffic in both
stochastic variables and deterministic parameters, which were lanes at the same time, defined as:
A. O’ Connor, I. Enevoldsen / Engineering Structures 30 (2008) 1408–1416 1413
 
L 1 + l1 L 2 + l2 in free flowing conditions (i.e. normal passage) deterministic
ν12 = ν21 = ν1 ν2 + (5)
V1 V2 modelling of the speed of the special transport and ordinary
transport as 60 km/h and 80 km/h respectively was
where L is the influence length, l is the vehicle length and V the
conservative, i.e. producing a higher probability of meeting
speed, with index 1 or 2 referring to lanes 1 or 2 respectively.
events.
The influence length L is formally defined as the length of the
The transverse location of the vehicles is modelled as a
structure which when loaded contributes to the magnitude of the
normally distributed random variable with an expected value
sectional force at a specified location (where relieving zones are
equivalent to the location of the centre of the lane and a
excluded from this length).
standard deviation of 0.24 m in accordance with the specified
In Eq. (4) T is the considered reference period for the
requirements for normal passage [12]. The longitudinal location
extreme distribution (i.e. one traffic year). The distribution
is selected, conservatively, as the most severe longitudinal
for the load effect in lane 1, F1 (q), lane 2, F2 (q), and
locations, l1 and l2 , in the two lanes. These longitudinal
the distribution of load effects due to simultaneous traffic
locations are taken from the deterministic analysis. The
load in both lanes, F12 (q), must be determined. These three
conservatism of this assumption is demonstrated by the results
distributions do, in general, include modelling of (1) the
of a sensitivity study performed as part of the probabilistic
number, configuration and weight of trucks, (2) the longitudinal
analysis (δβ/δ parameter) where moving the vehicles a small
and transverse appearance in bridge lanes (3) the dynamic
distance away from the critical longitudinal location is seen to
amplification of the static truck load, (4) the mechanical models
considerably increase the computed value of β.
for the relation between traffic load and traffic load effects and
The dynamic factor is modelled as, K s = 1 + ε, where
(5) the relative importance of load in the actual lanes.
ε is the dynamic supplement, which for vehicles in normal
Having determined the overall load on the structures, within
passage is modelled as two independent normally distributed
the influence length, the Inverse FORM method is employed
stochastic variables with N (41.5/W ; 41.5/W ), where W is
to determine the individual vehicle weights in each lane [12].
the total vehicle weight in kN [12]. Modelling the dynamic
The overall sectional forces are then determined using the
amplification factor in this manner implicitly assumes (a) an
aforementioned influence surface.
inverse proportionality between the dynamic amplification and
5.3. Stochastic modelling of the traffic load parameters vehicle weight and (b) a reduction in the coefficient of variation
(CoV) with increasing weight, which compare well with the
In the traffic load model, values are entered for the literature [22–24].
following: the traffic load of the vehicles, flow rates and Model uncertainty in the traffic load considers the
proportions, vehicles transverse location, dynamic impact uncertainty associated with the definition of load itself and
factors and model uncertainty for the vehicle weight. with uncertainty regarding the transformation of the load to
The traffic loading for the heavy transport vehicles is a load effect. The expected value for model uncertainty in
modelled with an axle configuration equivalent to the class 100 the traffic load is defined in terms of a judgement factor, I f ,
vehicle illustrated in Fig. 1. Studies have demonstrated that which is multiplied by the load. I f is modelled as normally
total vehicle weight is well modelled as a normally distributed distributed with a mean value equal to 1.0 and a coefficient
stochastic variable [16,19–21]. The expected value for this class of variation, VI f , dependent upon the uncertainty associated
of vehicle has been determined from background studies of with the load model. VI f is taken as 10%, 15% or 20% for a
Danish traffic data [19] as 1072 kN. The ordinary transport level of uncertainty, considering the loading, assumed as small,
vehicle is composed of 6 axles, has a total weight of 520.9 kN medium and large respectively [12,25,26]. The uncertainty of
and a prescribed load distribution to the axles [19]. The total the load model can as a starting point be taken as small for
vehicle weight is modelled as a normally distributed stochastic conditional passage and for the normal passage situations in
variable. The standard deviation of 49.1 kN is chosen for all which the relative influence of the ordinary traffic on safety is
classes of vehicle so that the relative uncertainty declines with minor; as medium for normal passage in usual situations and as
increasing class [19]. The value was specified as owners are large for passage situations solely involving ordinary traffic and
required to specify vehicle total weights within +/− 8 t with for unusual normal passage situations (e.g. curved ramp bridges
90% confidence. As a result the given standard deviations are and bridges sensitive to eccentric load). In the case considered,
expected to be conservative in some cases. normal passage, the uncertainty is taken as VI f = 0.15.
The annual frequency of the heavy transport vehicles is
specified dependent upon the road class and vehicle weight. 5.4. Stochastic modelling of the permanent and quasi-
The daily frequency of ordinary vehicles is determined from permanent load effects
site specific traffic flow records. The cumulative duration of the
day over which trucks are assumed to use the road network in The load effect due to permanent loads is modelled as
modelled deterministically as 15 h, a value which is typical of normally distributed, with a mean value taken as to 1.0
European motorways [21]. × Permanent Effect and a CoV of 7.1% (i.e. a CoV of
The speeds of the vehicles have an influence on the 5% is associated with the permanent action itself while the
likelihood of a meeting event taking place as evidenced by Eq. CoV of the model uncertainty is taken as 5%, the overall
(5). Analysis of site specific traffic records demonstrated that CoV for the permanent effect is therefore computed as =
1414 A. O’ Connor, I. Enevoldsen / Engineering Structures 30 (2008) 1408–1416

Table 2 structure) that only small variations can arise, (ii) where
Model uncertainty factors [25] attention has been paid to eccentricities, secondary moments,
Accuracy of the calculation model etc., (iii) where the model has been verified for the bridge
in question, or (iv) where an improved model has resulted
Good Normal Poor
V I1 0.04 0.06 0.09 in a reduction of the uncertainty of an important stochastic
ρ1 −0.3 0.0 0.3 variable; (b) Normal calculation accuracy is used in situations
where computation models are used that are generally accepted
Material property deviations
as being in conformity with normal practice and (c) a Poor
Small Medium Large computation model is one that has been excessively simplified
V I2 0.04 0.06 0.09 and does not meet the requirements for a model of normal
ρ2 −0.3 0.0 0.3
accuracy [25].
Material identity The uncertainty associated with determining material
Good Normal Poor parameters is dependent upon the amount of information
V I3 0.04 0.06 0.09 available on the materials of the structure and on the availability
ρ3 −0.3 0.0 0.3 of test results etc.
For uncertainty factors associated with material identity
√ it should be noted that: (a) Good material identity can be
5%2 + 5%2 ). For the variable loads the mean value is taken as
1.0 × Variable Effect with a coefficient of variation including assumed if the identity of the materials has been verified on the
model uncertainty of 11.2% (i.e. surfacing CoV 10%, model bridge or if the identity of materials used subsequently can be
uncertainty CoV 5%). The parasitic moment induced by the documented, (e.g. “as built” drawings); (b) Normally material
prestress is modelled as normally distributed with mean value identity is assumed when the materials are assigned on the basis
equal to 1.0 × Parasitic Effect and a CoV of 7.1% (i.e. CoV 5% of the project material and there is no reason to doubt that the
associated with the parasitic action itself and model uncertainty bridge in question was not built in accordance with the project
CoV of 5%) [25,26]. material and (c) Poor material identity arises when estimated
values are used or where the project material is dubious or
6. Modelling of capacity incomplete [25].

The two strength parameters which are modelled as 6.1.1. Concrete compressive strength f cu
stochastic in the analysis are (1) the concrete compressive The concrete compressive strength is modelled as a
strength, f cu and (2) the strength of the prestressing steel, f ptk . log-normally distributed variable [25,26]. It is specified in
It is important to consider how the effects of model uncertainty the original 1959 construction drawings with a 28-day
are included with respect to the resistance parameters. characteristic cube strength σT = 350 kg/cm2 . In the
1949 code [27] to which the structure was designed the
6.1. Model uncertainty for capacity characteristic concrete strength is said to correspond to the
mean compressive strength [12]. In performing an assessment
The model uncertainty takes account of: (1) the accuracy of an existing structure, it is vital that original documentation,
of the calculation model, I1 , (2) possible deviations from the including design codes at the time of design are consulted as
strength of material properties in the structure involved as relationships such as documented here, between characteristic
compared with that derived from control specimens, I2 , and and mean parameters, whose definition may have changed
(3) the degree of control on site (i.e. materials identity), I3 . with subsequent codes, can prove extremely important. The
The model uncertainty is taken into account by introducing cylinder strength σC = 0.8 ∗ σT = 280 kg/cm2 [12]. The
judgement factors Im related to the material properties. The CoV, V fcu (=VM ), is fixed at 0.18 in the background of the
judgement factor Im , which is assumed to be log-normally requirements for concrete testing given in [28]. For intact
distributed with a mean value equal to 1.0 and CoV, VIm , is concrete structures, a deterministic increase in the compressive
introduced by multiplying the basic material variables by Im . strength of the concrete with time can be assumed in the
VIm , is calculated as [12]: absence of contra indications. The strength can be taken as
q 25% higher than the original 28-day strength if the bridge in
VIm = VI21 + VI22 + VI23 + 2 ρ1 VI1 + ρ2 VI2 + ρ3 VI3 · VM

question is over 5 years old [28]. Thus f cu (and by implication
(6) E[ f cu ]) at 28 days may be increased by 25%, whereby 1.25
E[ f cu ] = 29.4 MPa.
where the variation and correlation coefficients, VIi and ρi Based upon the discussion of the model uncertainties
respectively, are as specified in Table 2, and VM is the CoV presented above, the accuracy of the calculation model is
of the basic material variable. taken as Normal, the uncertainty of the material properties
In considering Table 2, for uncertainty factors associated in construction is classified as Medium while the material
with the accuracy of the calculation model it should be noted identity is classified as Normal. Thus the coefficient of variation
that: (a) Good computation models can for example be used including model uncertainty is taken as 19.90%. The resulting
(i) where the model is so simple (corresponding to a simple standard deviation is 5.85 MPa.
A. O’ Connor, I. Enevoldsen / Engineering Structures 30 (2008) 1408–1416 1415

6.1.2. Strength of the prestressing steel f ptk The Guideline for Reliability Based Classification of
The strength of the prestressing steel is modelled as a log- the Load Carrying Capacity of Existing Bridges may be
normally distributed variable [26,27]. Test data regarding the downloaded (in English) from the website of the Danish Roads
material properties of the prestressing steel for the bridge is Directorate at http://www.vd.dk.
not available. Information contained on the drawings indicates
7 mm wires were used. The guaranteed strength of such References
wires is estimated at the 0.1% fractile. From the structural
drawings the characteristic tensile yield stress of the second [1] Vejdirektoratet. Beregningsregler for eksisterende broers bæreevne (In
Danish: Guideline for assessment of the load carrying capacity of bridges)
yield point of the prestressing steel, f ptk , was 1422 MPa, and
1996.
the rupture tensile stress, f utk , was 1618 MPa. The CoV on the [2] Jensen FM, Knudsen A, Enevoldsen I, Stoltzner E. In: Ryall MJ,
strands at rupture is specified as 4% (=VM ) [29]. Modelling Parke GAR, Harding JE, editors. Probabilistic-based bridge management
the variable using a log-normal distribution this corresponds implemented at Skovdiget west bridge. Bridge management, vol. 4.
to an expected value, E[ f ptk ] = 1490 MPa and a standard Thomas Telford; 2002. p. 223–30.
[3] Melchers R. Structural reliability analysis and prediction. 2nd ed. Wiley;
deviation of 59.6 MPa. In determining the model uncertainty
1999.
the accuracy of the calculation model was taken as Normal, [4] Madsen HO, Krenk S, Lind NC. Methods of structural safety. Prentice-
the uncertainty for the material resistance in the construction Hall; 1986.
were assumed as Medium and the material identity as Normal. [5] Ditlevsen O, Madsen HO. Structural reliability methods. John Wiley;
The CoV including model uncertainty was thus 11.14%. This is 1996.
equivalent to a deviation of 165.92 MPa. [6] Enevoldsen I. Experience with probabilistic-based assessment of bridges.
Structural Engineering International (SEI) 2001;11(4).
[7] Enevoldsen I, Jensen FM. Safety-based bridge management. In:
7. Classification from safety index analysis Proceedings 16th IABSE congress. 2000.
[8] Enevoldsen I, Jensen FM, Bjerrum J. Implementation of safety-based
Safety assessment of the carrying capacity of the bridge was maintenance management for the sidewalk at Storstroem bridge. In:
performed probabilistically, at the governing ULS limit state, in Proceedings, IABMAS 2002. first int. conf. on bridge maintenance, safety
and management. 2002.
combination with plastic material response models as outlined
[9] O’ Connor AJ, Enevoldsen I, Bjerrum J. Probabilistic-based assessment of
above. The analysis returned a safety index β = 5.06 which is the Klovtofte bridges. In: Proceedings, IABMAS 2004. Second int. conf.
greater than the required value of (βt =) 4.75. It was therefore on bridge maintenance, safety and management. 2004.
concluded that the bridge can receive the classification CLASS [10] Vejdirektoratet. Beregningsregler for eksisterende broers bæreevne.
100 for Normal passage. Thereby the rating of Class 80 for Revision. (In Danish: Guideline for assessment of the load carrying
Normal Passage resulting from the deterministic assessment is capacity of bridges - Revision) 2002.
[11] DS411:1999. Norm for betonkonstruktioner, Dansk Standard, 1999. (In
revised. Danish: Design code for concrete structures) 1999.
The main reasons behind this revised classification are [12] Vejdirektoratet. Beregningsregler for pålidelighedsbaseret klassificering
considered to be that: (a) the probabilistic live loading model for eksistende broer. (In Danish: Guideline for reliability based
is more realistic for the structure under consideration, as it classification of the load carrying capacity of existing bridges) 2004.
not only takes account of the weights of the vehicles but also Download http://www.vd.dk.
[13] ISO 2394. General principles on reliability for structures. 2nd ed. 1998.
their frequency of occurrence on the individual structure and
[14] EN 1990:2002. Eurocode – Basis for structural design. CEN; 2002.
the probability of their position on the structure and (b) the [15] DNV Software. Veritasvejen 1, PO Box 300, N-1332 Hovik, Norway;
safety based formulation of capacity assessment reduces the 2006 (http://www.dnv.com).
conservatism inherent in the deterministic approach for the [16] O’ Connor A, O’ Brien E. Mathematical traffic load modelling and factors
limit state considered. influencing the accuracy of predicted extremes. Canadian Journal of Civil
Engineering 2005;32:270–8.
8. Conclusion [17] Ditlevsen O. Traffic loads on large bridges modeled as white noise fields.
Journal of Engineering Mechanics 1994;120(4):681–94.
[18] Ditlevsen O, Madsen HO. Stochastic vehicle queue load model for large
Following a deterministic assessment it was concluded that bridges. Journal of Engineering Mechanics 1994;120(9):1829–47.
strengthening or rehabilitation of the structure presented was [19] Vejdirektoratet. Karakteristisk Trafilkast i Danmark, Totalvaegt og
required if the bridge was to be included on the Danish Akseltryk af Koretojer i Danmark. (In Danish: Characteristic traffic loads
Motorway Network for heavy transports. A probabilistic in Denmark, total weight and axle weights of vehicles in Denmark) 2001.
[20] Jacob BA. Methods for the prediction of extreme vehicular loads and
analysis of the structure at the critical limit state identified
load effects on bridges. Report of Subgroup 8, Eurocode 1.3, LCPC Paris.
by deterministic assessment demonstrated that it did indeed 1991.
have the capacity to be included on the Network. The result [21] O’ Connor AJ. Probabilistic traffic load modelling for highway bridges,
represented a significant saving for the bridge owner both in Ph.D. Thesis. Department of Civil Engineering, Trinity College Dublin;
terms of the direct replacement cost and of the indirect costs, 2001.
[22] Kirkegaard PH, Nielsen SRK, Enevoldsen I. Heavy vehicles on minor
which would have been incurred in replacing the structure. It is
highway bridges – dynamic modelling of vehicles and bridges. Structural
important to stress once again that at no stage has the safety reliability theory paper no. 171. Aalborg University, ISSN 1395-7953-
of the structure been compromised, rather a bridge specific R9721. 1997.
safety assessment, free from the generalisations of deterministic [23] Hwang E-S, Nowak AS. Simulation of dynamic loads on bridges. Journal
codes, has resulted in considerable savings for the owner. of Structural Engineering 1991;119(6):853–67.
1416 A. O’ Connor, I. Enevoldsen / Engineering Structures 30 (2008) 1408–1416

[24] Nassif HH, Nowak AS. Dynamic loads for girder bridges under normal Recommendations for loading- and safety regulations for structural
traffic. Archives of Civil Engineering 1996;XIII(4). design) 1987.
[25] Nordisk Komité for Bygningsbestemmelser. Retningslinier for last- [27] DS411:1949. Norm for betonkonstruktioner, Dansk Standard, 1949. (In
og sikkerhedsbestemmelser for bærende konstruktioner. NKB-rapport Danish: Design Code for Concrete structures) 1949.
nr. 35. The Nordic Committee on Building Regulations. (In Danish: [28] Poulsen PE. Vurdering af betons styrke ved provning af udborende
Recommendations for loading- and safety regulations for structural kerner. Rapport no. 75:64. Ren og Anvendt Mekanik, Danmarks
design) 1978. Ingeniorakademi, Bygningsafdelingen. (In Danish: Evaluation of concrete
[26] Nordisk Komité for Bygningsbestemmelser. Retningslinier for last- og strength from core tests) 1975.
sikkerhedsbestemmelser for bærende konstruktioner. NKB-rapport nr. [29] Jensen GC, Olsen K. Teknisk Stabi, 14. udgave, Teknisk Forlag. (In
55. The Nordic Committee on Building Regulations. (In Danish: Danish: Technical guideline) 1976.

You might also like