You are on page 1of 13

Engineering Structures 22 (2000) 1677–1689

www.elsevier.com/locate/engstruct

Proof load testing for bridge assessment and upgrading


Michael H. Faber a, Dimitri V. Val b,*
, Mark G. Stewart b

a
COWI, Consulting Engineers and Planners AS, Lyngby, DK-2800, Denmark
b
Department of Civil, Surveying and Environmental Engineering, The University of Newcastle, Newcastle, NSW 2308, Australia

Received 27 April 1999; received in revised form 27 September 1999; accepted 10 November 1999

Abstract

Bridge deterioration with time and ever increasing traffic loads raise concerns about reliability of aging bridges. One of the ways
to check reliability of aging bridges is proof load testing. A successful proof load test demonstrates immediately that the resistance
of a bridge is greater than the proof load. This reduces uncertainty in the bridge resistance and so increases the bridge reliability.
The paper considers a reliability-based calibration of intensities of proof loads for aging bridges to verify either an existing or
increased load rating taking into account possible bridge deterioration. Intensities of proof loads are calibrated based on a consistent
target reliability index. The influence of test risk, dead to live load ratio, and uncertainties associated with dead and live loads and
bridge resistance is considered. The results presented in the paper relate to short and medium span bridges.  2000 Elsevier Science
Ltd. All rights reserved.

Keywords: Bridges; Proof loads; Time dependent reliability

1. Introduction (i) confirm an existing load rating (if bridge deterioration


is observed or suspected), or (ii) increase a load rating.
It is quite evident that the bridge population in the Analytical or predictive approaches to determining load
US, Canada, Europe, Australia and elsewhere is aging. ratings are well documented and widely used. However,
This is due to aging local road bridges and the large such approaches may be overly conservative. For
expansion of highways built during the post-war boom example, the actual load carrying capacity of a bridge is
era of the 1950s to 1970s. For example, in Australia, often higher than the predicted capacity, this may be due
over 60% of bridges for local roads are over 50 yr old to system effects, load redistribution, contribution of
and approximately 55% of all highway bridges are over non-structural elements and bearings not behaving in an
20 yr old. These figures are typical of other countries ‘idealised’ manner [1,2]. It is desirable that the assess-
also. The incidence of structural deterioration increases ment of existing bridges not be overly conservative since
with bridge age due to corrosion, fatigue, wear and tear such an approach may result in the replacement of
and other forms of material property degradation. During bridges that are in fact ‘satisfactory’. As such, a diagnos-
this same period vehicle loads and legal load limits have tic or proof load test may be more appropriate if
been steadily increasing. Aging bridges subject to
increasing legal load limits mean that existing bridges 앫 analytical analyses produces an unsatisfactory load
often fail to satisfy structural requirements as specified rating; or
for new bridges. There is obviously a strong financial 앫 analytical analysis is difficult to conduct due to
incentive that existing bridges be conserved and their deterioration or lack of documentation.
remaining service life extended.
In general, bridge assessment is conducted to either: A diagnostic test may be used to verify or refine analyti-
cal or predictive structural models, whilst a proof load
test is used to assess the actual load carrying capacity
* Corresponding author. Current address: School of Engineering,
James Cook University, Townsville, QLD 4811, Australia. Tel.: + 61- of a bridge. A successful proof load test demonstrates
7-4781-4722; fax: + 61-7-4775-1184. immediately that the resistance of the bridge is greater
E-mail address: dimitry.val@jcu.edu.au (D.V. Val). than the proof load. As a result, this reduces uncertainty

0141-0296/00/$ - see front matter  2000 Elsevier Science Ltd. All rights reserved.
PII: S 0 1 4 1 - 0 2 9 6 ( 9 9 ) 0 0 1 1 1 - X
1678 M.H. Faber et al. / Engineering Structures 22 (2000) 1677–1689

associated with the resistance of the bridge and so reliability index. As such, bridge reliability (or safety)
increases its reliability [3]. Moreover, a proof load test is influenced by resistance and loading variability. How-
may provide additional useful information about bridge ever, the observation that a bridge survives a proof load
properties and performance. However, extensive proof test indicates only the minimum load-carrying capacity
load testing programs are very costly. For example, of the bridge—it does not reveal the actual bridge
proof load testing of bridges in New South Wales, Aus- capacity and also does not provide a meaningful measure
tralia, can cost up to A$110,000 per bridge or 6% of of the safety of the bridge. To be consistent with LRFD
the bridge replacement cost [4]. Given the large costs principles, an assessment procedure should show that the
involved it is surprising that proof load testing pro- post-test bridge reliability is acceptable. Consequently,
cedures are not well documented [5]. It should be recog- decisions related to bridge assessment are based on
nised also that there is a risk that the bridge will be dam- uncertain or incomplete information. Deterministic
aged or not survive a proof load test (referred to herein approaches are not efficient for decisions taken under
as test risk) and so proof load testing may not always uncertainty since such decisions tend to be conservative
be cost-effective [6]. and based on ‘prudent pessimism’ or ‘worst case’ scen-
The present paper presents a reliability-based method arios. Reliability (or probabilistic) analysis includes
(i.e. reliability is used as the measure of structural information from all resistance and loading variables
performance) to determine the target proof load. Pre- influencing the assessment process (not just point
vious reliability studies [5,7,8] in this area are quite lim- estimates) and so provides a rational criterion for the
ited and have (i) ignored the risks associated with failure comparison of the likely consequences of decisions
of a bridge during the load test—this is referred to herein taken under uncertainty.
as the ‘test risk’ (although Moses et al. [8] have con- The initial stage of a bridge assessment is to estimate
sidered test risk); (ii) assumed bridge reliabilities were the reliability of the existing bridge, such a reliability
not influenced by the timing of the proof load test; and analysis should consider the following uncertainties:
(iii) ignored the effects of structural deterioration.
Clearly, these are not realistic assumptions. Stewart and 앫 representation of real structures by idealised predic-
Val [6] improved this previous work by developing a tion models (model errors);
probabilistic framework that considers bridge age, 앫 inherent variability of material properties;
deterioration, magnitude of proof loads, test risk, 앫 variability in workmanship, element dimensions and
updated bridge reliability (for prior service loading and environmental conditions;
proof loading) and associated decision-making appli- 앫 spatial variations in material and other properties;
cations such as a risk–cost–benefit analysis. The present 앫 prior and current variability of service loading; and
paper considers a practical implementation of such a pro- 앫 assessment of current fatigue, corrosion or other
babilistic approach, in this case by using reliability-based deterioration processes.
methods to propose target proof loads to either verify an
existing load limit or assess a bridge for an increased Reliabilities may be calculated using updated probabilis-
load limit, for non-deteriorated or deteriorated bridges. tic load and resistance models developed from data
In this approach, target proof loads are calibrated from which are representative of known site characteristics.
a consistent target reliability index and the influence of This may occur if inspection data are available, leading
test risk, dead to live load ratio, dead and live loading to additional sources of uncertainty, such as:
variability, and resistance modelling uncertainty is con-
sidered. A risk–cost–benefit analysis is not considered 앫 inaccuracy of inspection and maintenance tech-
herein. The results presented herein relate to short and niques; and
medium span bridges. 앫 statistical uncertainty due to limited number of obser-
vations or sample tests.

2. Reliability considerations for proof load testing Information regarding the reliability of the existing
bridge is also needed to determine the target proof load.
The carrying out of a proof load test is a complicated After the application of a target proof load the bridge
and often bridge-specific procedure specified by high- reliability can again be updated. This occurs since the
way agency, state or national guidelines. It is beyond the lower tail of the distribution of structural resistance is
scope of the present paper to describe such procedure; truncated at the target proof load, resulting in a higher
however, it can be found elsewhere [2,9]. bridge reliability. Although this could be at the expense
The design of new bridges is increasingly being based of an unacceptable test risk. Note that an additional
on LRFD or limit state design principles [10,11] in source of uncertainty can arise since the proof load may
which all ‘typical’ bridges have relatively uniform not always be applied in a correct and controlled manner.
reliabilities and that these reliabilities exceed a target After a proof load test it is often of interest to deter-
M.H. Faber et al. / Engineering Structures 22 (2000) 1677–1689 1679

mine when the bridge will need to be re-assessed (e.g. by Eq. (1) is calculated herein using the time variant
frequency of inspections or maintenance). This provides component reliability analysis option in the STRUREL
an indication when funds will need to be allocated again [12] software package.
for this bridge. In such cases a time-dependent reliability
analysis can be used to determine when the bridge 3.2. Probabilistic models
reliability will fall below a target reliability index. The
uncertainties associated with this form of reliability 3.2.1. Resistance model including effects of
analysis include, in addition to those described above, deterioration
In order to take into account the time-dependent
앫 prediction of longer term deterioration processes; and deterioration of bridges their structural resistance, R(t),
앫 prediction of lifetime service loads. can be presented as the product of the time-independent
initial resistance, R0, and a time-dependent degradation
In other words, the life-cycle performance of the bridge function, g(t). The degradation function defines the pro-
can be predicted from a reliability-based approach. portion of the initial resistance remaining at time t
For the present study, and as described above, it is [13], hence,
proposed that bridge reliabilities be compared with
reliability-based acceptance criteria such as a target R(t) ⫽ R0·g(t) (2)
reliability index. A complementary approach may be to As has been mentioned above, there are a number of
use bridge reliabilities to estimate the cost-effectiveness uncertainties associated with the prediction of bridge
of bridge assessment decisions using a risk–cost–benefit resistance. Uncertainties in the initial resistance include,
analysis [6]. among others, variability and spatial variations of
Since many bridges are similar with respect to their material properties and element dimensions, and model
spans, type of construction, traffic loads and volumes, uncertainty related to representation of a real bridge by
etc. (steel stringer, RC slab, RC stringer, pre-stressed an idealised model (i.e. idealisation of support conditions
concrete tee, etc.) it is possible to develop bridge-type and joints between bridge members, not accounting for
specific assessment specifications for these ‘generic’ contribution of nonstructural elements, etc.). Generally,
bridges using a calibration approach. Such an approach these uncertainties depend on bridge type, quality of
has been used in the development of LRFD codes materials used, workmanship, environmental conditions
[10,11]. However, long-span bridges, historical bridges, during construction; all of which vary from bridge to
and other ‘unique’ bridges are likely to require an indi- bridge. Resistance statistics for slab-on-girder type of
vidual or bridge-specific assessment in which bridge bridges (i.e. steel girder, reinforced concrete T-beam and
reliabilities may need to be calculated for each bridge. prestressed concrete girder) were estimated by Nowak
The present paper will present a reliability-based cali- [14]. According to his results, the coefficient of variation
bration of target proof loads for ‘generic’ bridges. This of resistance (in flexure, shear, tension and compression)
is now described. of girder bridges varies from 0.075 for prestressed con-
crete girders in flexure to 0.17 for reinforced concrete
T-beams in shear without stirrups. In this paper R0 is
3. Time-dependent reliability analysis treated as a lognormal random variable in which two
values of its coefficient of variation, VR—0.10 and
3.1. Computation of time-dependent structural 0.15—are considered.
reliability The resistance degradation function g(t) is adopted in
the following form [15]
When assessing the reliability and residual service life
g(t) ⫽ 1 ⫺ k1t ⫹ k2t2 (3)
of existing bridges the effect of time variations in both
the strength (e.g. due to deterioration) and the load where t is the elapsed time, and k1 and k2 are coefficients
characteristics must be taken into account. If it is defining the deterioration rate. Based on parametric stud-
assumed that n independent load events Si occur within ies of reinforced concrete beams subjected to corrosion
the time interval (0,T) at times ti(i = 1, 2, %, n) then attack the following values of k1 and k2 and the time of
the cumulative probability of failure, Pf(T), of a bridge corrosion initiation, Ti, corresponding to low, medium
during this time interval is and high deterioration were suggested [15]
Pf(T) ⫽ 1 ⫺ Pr[R(t1) ⬎ S1傽R(t2) (1) Low deterioration k1 = 0.0005 k2 = 0 Ti = 10 yr
⬎ S2傽%傽R(tn) ⬎ Sn] Medium deterioration k1 = 0.005 k2 = 0 Ti = 5 yr
High deterioration k1 = 0.01 k2 = 0.00005 Ti = 2.5 yr
where R(ti) represents bridge resistance at the time ti and
t1 ⬍ t2 ⬍ %tn. The probability of failure represented (4)
1680 M.H. Faber et al. / Engineering Structures 22 (2000) 1677–1689

The values (4) were developed from limited analytical taneous reliability considering collapse, serviceability
data and must be used with caution. For example, based and fatigue failure modes. Thus it is necessary that these
on the results of the analysis by Stewart and Rosowsky probabilistic load models represent correctly both the
[16] of a reinforced concrete slab bridge deteriorating extreme load events within a reference period (i.e. period
due to reinforcement corrosion with the corrosion rate of time) and the random-point-in-time load events.
of 3 µA/cm2 (which represents a high corrosion rate) k1 Since in the present context the main emphasis is on
= 0.0035 and k2 = 0 were obtained. This indicates that the probabilistic modelling of the load for ultimate limit
the values k1 and k2 (4) for medium and high deterio- state analysis, the Gumbel distribution for the random
ration may be too conservative, and that the ‘medium’ maximum load effect L within a reference period of n
case actually represents very severe bridge deterioration. yr is chosen [20]. This distribution has the property that
It should be noted also that sufficiently accurate and if the extreme value within one reference period is Gum-
experimentally and analytically verified degradation bel distributed then the extreme value corresponding to
functions have not yet been developed. any reference period being a multiple of the first is also
Generally, there is significant uncertainty associated Gumbel distributed. Furthermore, the standard deviation
with a deterioration process. However, at present no suf- is independent of the reference period. If mL and sL are
ficient data are available to determine statistical para- the mean value and the standard deviation of the distri-
meters of k1 and k2 so in this study they are assumed to bution of the annual maximum load effect, the mean
be deterministic. value of the distribution of the maximum load effect in
Information on parameters controlling bridge resist- n yr is readily calculated as [21]
ance can be updated prior to a proof load test by carrying
out on-site inspections of the bridge. Inspections can be 冑6
mnL ⫽ mL ⫹ sL ⫻ ln(n) (5)
used to determine material properties (e.g. core testing, p
the rebound hammer, ultrasonic pulse velocity to esti-
mate on-site compressive strength of concrete), element
Based on analysis conducted by Stewart and Rosow-
dimensions (e.g. electromagnetic covermeters to locate
sky [16] using probabilistic load models developed by
and measure cover of reinforcing bars), and to detect
Nowak [14] it was found that the coefficient of variation
defects or deterioration (e.g. impact-echo to locate large
of the Gumbel distribution for the annual extreme load
voids in concrete, half-potential and resistivity measure-
effect is close to 10% for a two-lane bridge. This value
ments to predict the likelihood and rate of corrosion).
is thus used in the following.
Predictive probabilistic models for the bridge resistance
In addition to the live load a load component due to
can then be updated with collected site-specific data
dead load of the bridge is also included. It is assumed
using Bayes theorem [17], which is a general theorem
that this load may be modelled by a time invariant nor-
applicable to any situation in which existing probabilistic
mal distribution with mean of 1.05 times the nominal
knowledge is updated with new evidence [18].
dead load and a coefficient of variation of 10% [14].
In the following analysis, the live load is normalised
3.2.2. Load modelling
with respect to the dead load in terms of r the ratio
The safety and residual service life of road bridges is
between the characteristic values of the dead load Dk and
usually dominated by the variable loading due to passing
the live load Lk
vehicles. Design loads prescribed by design codes for
specific classes of bridges are normally calibrated such Dk
that they will induce load effects which are exceeded r⫽ (6)
Lk
only once within a certain sufficiently long return period,
typically 50 yr. In modern design codes such a cali- Dk is the 50th percentile value or the mean value of the
bration is based on a probabilistic model of the loading, distribution of the dead load, and Lk is taken as the 98th
broadly valid for all the bridges within the considered percentile of the distribution function of the annual
class. maximum load effect in accordance with the definition
For short span bridges and for secondary structural of characteristic loads in Eurocode [22].
elements, the statistical characteristics of the individual For a given r, the mean value of the Gumbel distri-
axle loads are important, whereas for long span bridges bution of the annual maximum load effect may thus be
the statistical characteristics of the integrated effect of determined as
several axles are more important [19]. Based on known rmD
traffic characteristics the statistical characteristics of the mL ⫽ (7)
load effects may readily be evaluated. − ln[ − ln(0.98)] × VL 冑 6+1−
0.5772
p
冑6VL
The probabilistic traffic load models must be formu-
lated such that they can be used both to assess the service where VL is the coefficient of variation of the live load-
life reliability of the bridge and to assess the instan- ing.
M.H. Faber et al. / Engineering Structures 22 (2000) 1677–1689 1681

3.3. Effect of proof loading on time-dependent prior to the proof load test and lp is the load effect
reliability induced by the proof load.

Based on the survival of proof load testing the 3.4. Reliability-based determination of proof loads
reliability of a bridge may be updated in accordance with
the pattern and intensity of the proof load applied. To determine an appropriate intensity of proof load
Depending on the way the reliability of the bridge is then the desired increase in reliability of a bridge and
evaluated, different approaches may be followed to the risk of damaging the bridge during the proof load
include the information of survival in a formal test must be considered. In the present study, the
reliability analysis. required intensity of the proof load is evaluated as a
As mentioned previously, bridge reliability is evalu- function of the desired bridge load rating. The bridge
ated using a time variant component reliability analysis. load rating is defined herein in terms of four different
In this context the post-test reliability can be updated by levels of characteristic live loads, namely 100%, 105%,
truncating the distribution function for the bridge resist- 110% and 115% of the characteristic live load used in
ance in its lower tail; namely, at the value which is equal the initial design of the bridge. These levels cover most
to the load effect induced by the applied proof load. The reassessment and upgrading situations in practice. Note
updated distribution function of the resistance F⬙R(r) after that the load rating is not a rating factor, but rather a
the proof load test may then be written as general measure representing the proportional increase
F⬘R(r) − F⬘R(lp) in live loads relative to the characteristic live load
F⬙R(r) ⫽ r ⱖ lp (8) (100%) used in design. In the reliability analysis the load
1 − F⬘R(lp)
rating influences the distribution of annual maximum
where F⬘R(r) is the distribution function of the resistance load effects since a change in Lk will effect r and conse-

Fig. 1. ‘Low deterioration’, VR = 0.15, r = 1, and different bridge ages: (a) required proof load intensities; (b) test reliability.
1682 M.H. Faber et al. / Engineering Structures 22 (2000) 1677–1689

Fig. 2. ‘Low deterioration’, VR = 0.15, r = 1/2, and different bridge ages: (a) required proof load intensities; (b) test reliability.

quently mL in Eq. (7). It is assumed that a proof load test of 3.5 for a 75 yr intended design life used by Nowak
is to be conducted only once during the service life of [14]. Based on this target reliability requirement and for
a bridge and the intensity of the proof load is determined a given r and load rating of 100%, the mean initial
on the basis that this single load test should ensure safe resistance, R̄n, is estimated such that the reliability of
service of the bridge for the remainder of its service life. this ‘new’ bridge is equal to bT. As occurs in normal
To take into account different spans and bridge types design practice, deterioration is not taken into account
reliability analyses are carried out for three different when deriving R̄n.
characteristic dead to live load ratios used in design (r
= 1/2, 1 and 2) which cover bridges of short span and
medium spans up to 40 m. Two cases of deterioration 4. Results
are considered, namely ‘low deterioration’ and ‘medium
deterioration’, which are described by the degradation The following section presents results obtained from
functions given in Eq. (4). Furthermore, the effect of reliability analyses as described in Section 3.4. It is
bridge age at the time of the proof load test is also exam- assumed that a coefficient of variation of bridge resist-
ined by considering proof load testing after 20, 40, 60 ance is VR = 0.15 (see Section 3.2.1).
and 80 yr for bridges with an intended design life of 100 Considering the cases where ‘low deterioration’ is
yr. Finally, for all the cases mentioned above the risks assumed, Figs. 1(a)–3(a) show the proof load intensities
of bridge collapse associated with the proof load test required to verify an existing bridge on proposed load
are evaluated. rating, for r = 1, 1/2 and 2 respectively. Figs. 1(b)–3(b)
It is assumed that the target reliability index, bT, for show the test reliability, bPL = ⫺ ⌽ ⫺ 1(PfPL) (where PfPL
bridges with an intended design life of 100 yr is 3.4. is test risk, i.e. the probability of bridge failure during the
This corresponds largely with the target reliability index test, and ⌽(·) the standard normal distribution function),
M.H. Faber et al. / Engineering Structures 22 (2000) 1677–1689 1683

Fig. 3. ‘Low deterioration’, VR = 0.15, r = 2, and different bridge ages: (a) required proof load intensities; (b) test reliability.

associated with each proof load intensity. For example, loads well above the characteristic live load used in
in the case of ‘low deterioration’, r = 2, desired load design are required for all desired load ratings. This is
rating of 105% and the age of the bridge 40 yr it can expected; loss of structural resistance due to deterio-
be found in Fig. 3(a) that a proof load intensity of 1.2 ration requires larger proof loads to maintain a constant
times the characteristic live load used at the design is target reliability index. Figs. 4(b)–6(b) show that the cor-
required. The corresponding test reliability found in Fig. responding test reliabilities are very low and in most
3(b) is approximately bPL = 3.1. situations are thus unacceptable.
From Figs. 1(a)–3(a) it is seen that the required proof
load intensity decreases as the time of the proof load
application increases. This is not surprising as the corre- 5. Further issues of proof load testing
sponding residual service life and failure probabilities
are reduced and so it is relatively easier to reach the 5.1. Effect of resistance uncertainty
target reliability index of 3.4. It is seen that in the cases
corresponding to 80% expired service life the curve To examine the effect of resistance uncertainty on the
starts at 105% of the characteristic live load used in efficiency of proof load testing a reduced coefficient of
design (i.e. load rating). This means that proof loading variation of resistance, VR = 0.10 (compared with pre-
is not needed if the desired load rating is less than 105%, viously used VR = 0.15), will now be considered for the
since in this case during the remaining service life (20 case of ‘low deterioration’ and r = 1. There are two
yr) the reliability index will not fall below 3.4. possibilities:
For the cases where ‘medium deterioration’ is
assumed, Figs. 4–6 show the proof load intensities and 앫 the reduced coefficient of variation is used in the
test risks for r = 1, 1/2 and 2, respectively. In compari- initial design of a bridge (e.g. steel girders [14]);
son to the case of ‘low deterioration’ it is seen that proof 앫 the initial design is based on VR = 0.15 and the coef-
1684 M.H. Faber et al. / Engineering Structures 22 (2000) 1677–1689

Fig. 4. ‘Medium deterioration’, VR = 0.15, r = 1, and different bridge ages: (a) required proof load intensities; (b) test reliability.

ficient of variation is reduced by carrying out an on- a lower variability of resistance is adopted in the initial
site inspection prior to the proof load test. design. This occurs because the lower tail of the distri-
bution of resistance is less sensitive to truncation at
When VR = 0.10 is used in the initial design this results lower proof loads.
in a lower mean initial resistance, R̄n/Dk = 3.175, to When the initial design is based on VR = 0.15, it is
ensure a target reliability index bT = 3.4 for a 100-yr assumed that data obtained by an on-site inspection will
intended design life (whereas for VR = 0.15, R̄n/Dk = reduce material property and dimensional uncertainty.
3.825). As such, results of the analysis (see Fig. 7 for As such, in the reliability analysis the coefficient of vari-
VR = 0.10, and Fig. 1 for VR = 0.15) indicate that higher ation of the bridge resistance may be, in some cases,
proof loads need to be applied to a bridge with VR = assumed to be reduced to 0.10 while the mean value of
0.10, compared to one with VR = 0.15, to provide the the resistance is not changed. Proof loading testing is
same reliability index for the same residual period of then not necessary since even for the worst case con-
intended design life and with the same load rating. This sidered (i.e., the inspection is carried out after 20 yr of
also leads to a higher test risk for a bridge with VR = the bridge service life and an increased load rating
0.10. For example, if a proof load test is carried out after 115%) the reliability index for the remaining service life
20 yr of bridge service to verify an increased load rating is 4.2 (i.e. much higher than the target value 3.4).
110% (i.e. 110% of characteristic live load used in Generally, these results illustrate an obvious fact that
design) for VR = 0.10 the required test load intensity is the higher uncertainty of the initial information the more
1.48Lk and the corresponding test reliability bPL = 2.2 reserves of bridge resistance are available and the more
(or test risk PfPL = 1.4 × 10 − 2), while for VR = 0.15 these efficient updating may be in reducing this uncertainty.
values are 1.24Lk and bPL = 3.2 (or PfPL = 6.9 × 10 − 4). Both proof load testing and an on-site inspection update
The test risk is 20 times lower in the latter case. These information on bridge resistance and, consequently,
results show that proof load testing is less efficient when improve the assessment of bridge reliability. However,
M.H. Faber et al. / Engineering Structures 22 (2000) 1677–1689 1685

Fig. 5. ‘Medium deterioration’, VR = 0.15, r = 1/2, and different bridge ages: (a) required proof load intensities; (b) test reliability.

there is risk of bridge failure during a proof load test, rating of the bridge needs to be increased by 5%. If the
while on-site inspections are ‘risk-free’ in this respect, bridge will not be re-assessed at a later stage the inten-
but may not be sufficient in themselves to assess an sity of proof load required to ensure that the reliability
existing bridge for a proposed load rating. index will not be less than 3.4 for the remaining 80 yr
of the bridge service life is 3.18Lk [see Fig. 4(a)] and
5.2. Reduction of test risk by reducing the reference the corresponding test reliability bPL = ⫺ 1.3 (test risk
period PfPL = 0.90) [see Fig. 4(b)]. However, if the next bridge
assessment is planned after 10 yr (i.e. the reference per-
Previously it has been assumed that a proof load test iod is 10 yr), then the intensity of proof load required
is to be conducted only once during the service life of a is only 1.01Lk and the corresponding test reliability bPL
bridge (i.e. there are no subsequent intended inspection, = 3.38 (PfPL = 3.6 × 10 − 4) (see Fig. 8). In other words,
maintenance or repairs). Essentially, this is ‘test and for- a proof load of 1.01Lk will ensure that the bridge is ‘safe’
get’ bridge management. In the case of severely deterior- for the next 10 years; however, after this period the
ating bridges (i.e. in this study bridges subjected to bridge will need to be re-assessed. The effect of the
‘medium deterioration’) this results in a very high test proof load test on bridge reliability during these 10 yr
risk [PfPL = 0.1 ⫺ 0.5 and higher, see Figs. 4(b)–6(b)]. is shown in Fig. 9. Without the proof load test the
Clearly, it is not particularly realistic to inspect/test a reliability index falls below its target value of 3.4 within
bridge, especially a deteriorating one, only once during 5 yr, while the application of the proof load with inten-
its entire service life. More frequent inspections may be sity of 1.01 ensures that the reliability index will stay
needed. A reduction of the time between assessments above the target value for the entire reference period.
(i.e. reduced reference period) allows the magnitude of Generally, a risk–cost–benefit analysis may be used
the proof load and test risk to be reduced significantly. for making decisions about optimal periods between
As an example, consider a bridge with r = 1 subjected inspection or tests and proof load intensities, as well as
to ‘medium deterioration’. After 20 yr of service the load about other measures (e.g. strengthening, repair,
1686 M.H. Faber et al. / Engineering Structures 22 (2000) 1677–1689

Fig. 6. ‘Medium deterioration’, VR = 0.15, r = 2, and different bridge ages: (a) required proof load intensities; (b) test reliability.

replacement) which can be employed to ensure continu- subjected to ‘low deterioration’ with Vr = 0.15 and r =
ing service of aging bridges. More detailed consideration 1 will now be considered. Fig. 10 shows reliability indi-
of the application of a risk–cost–benefit analysis for ces for the remaining service life of the bridge with and
management of aging bridges, including the use of proof without taking into account that the bridge has survived
load testing, can be found elsewhere [6]. prior service loads (i.e. with and without resistance
updating). It is assumed that prior to the year considered
5.3. Effect of prior service loads (i.e. survival age) the load rating of the bridge is 100%
and it is then increased to 110%. It is also assumed that
Prior service loads may be treated as an uncertain the intended design life of the bridge is 100 yr. Thus,
proof load, since the fact that a bridge has survived ns the results in Fig. 10 show, for example, that if the bridge
yr in service means that the bridge resistance is greater is 60 yr old the reliability indices for the remaining 40
than the maximum load effect over this period of time. yr of the bridge life with a load rating increased to 110%
Thus, the reliability of service-proven (i.e. older) bridges are 3.6 with updating (i.e. with taking into account that
would increase [16,23]. However, the influence of the bridge has survived prior service loads corresponding
deterioration may negate this expected increase [6]. to a load rating 100%) and 3.2 without updating.
To update the distribution function F⬘R(r) of the bridge According to the results, the influence of prior service
resistance taking into account prior service loads Eq. (8) loads significantly increases reliabilities of aging
is modified by replacing the deterministic proof load bridges, especially with no or minor deterioration. For
effect, lp, with a random variable. In accordance with the example, if the bridge considered in the example has
probabilistic live load model described above this ran- survived for about 35 yr its load rating may then be
dom variable is Gumbel distributed and its mean value increased to 110% without any proof load tests, since
is determined from Eq. (5) with n = ns. the bridge reliability index for the remaining service life
To illustrate the effect of prior service loads a bridge will still be higher than bT = 3.4 (see Fig. 10).
M.H. Faber et al. / Engineering Structures 22 (2000) 1677–1689 1687

Fig. 7. ‘Low deterioration’, VR = 0.10, r = 1, and different bridge ages: (a) required proof load intensities; (b) test reliability.

Fig. 8. Reliability indices to select proof load intensity for 10 yr reference period, ‘medium deterioration’, VR = 0.15 and r = 1.

6. Conclusions dead/live load ratios, degradation functions and bridge


ages. The proof load intensities have been calibrated
Proof load intensities for the load rating of aging using reliability methods to ensure that the bridges achi-
bridges have been investigated. Proof load intensities eve a desired load rating and at the same time maintain
have been derived for a range of different load ratings, a prescribed residual service life reliability. In order to
1688 M.H. Faber et al. / Engineering Structures 22 (2000) 1677–1689

Fig. 9. Bridge reliability for 10 yr reference period, ‘medium deterioration’, VR = 0.15 and r = 1.

Fig. 10. Effect of updating by taking into account prior service load, ‘low deterioration’, VR = 0.15 and r = 1.

assess the feasibility of the proof load tests the associated The University of Newcastle, Australia. The work has
test risks (i.e. risks of bridge collapse during tests) has been co-sponsored by research grants from COWIfonden
been also evaluated. and from the Faculty of Engineering, The University of
It has been found also that the risks associated with Newcastle, for which the authors are very thankful.
the tests are within reasonable limits, if deterioration is
not significant, i.e. loss of around 5% of the initial resist-
ance over the intended design life (‘low deterioration’). References
However, the risks and the benefits may be considered
in more detail by cost–benefit analyses. For bridges [1] Bakht B, Jaeger LG. Bridge testing—a surprise every time. J
Struct Engng, ASCE 1990;116(5):1370–83.
which are subject to significant deterioration, i.e. loss of [2] Saraf V, Nowak AS. Proof load testing of deteriorated steel girder
around 50% of their resistance over the intended design bridges. J Bridge Engng, ASCE 1998;3(2):82–9.
life (‘medium deterioration’), proof load tests may still [3] Fujino Y, Lind NC. Proof-load factors and reliability. J Struct
provide a very useful tool for ‘controlled aging’. In this Engng, ASCE 1977;103(4):853–70.
case proof loads may still be efficient to achieve a [4] Ariyaratne W. Case history of bridges tested in NSW since 1995.
In: Chirgwin GJ, editor. Proceedings of the AUSTROADS 1997
desired load rating of the bridge within reasonable test Bridge Conference ‘Bridging the Millennia’, Sydney: AUS-
risks but for a reduced reference period. TROADS Inc., 1997;3:283–90.
Finally, the residual service life reliability of bridges [5] Fu G, Tang J. Risk-based proof-load requirements for bridge
is evaluated, taking into account that the bridges have evaluation. J Struct Engng, ASCE 1995;121(3):542–56.
already survived a certain period of time under service [6] Stewart MG, Val DV. Role of load history in reliability-based
decision analysis of ageing bridges. J Struct Engng, ASCE
loading. It is found that this information may be useful 1999;125(7):776–83.
for updating the reliability of existing bridges. [7] Nowak AS, Tharmabala T. Bridge reliability evaluation using
load tests. J Struct Engng, ASCE 1988;114(10):2268–79.
[8] Moses F, Lebet JP, Bez R. Applications of field testing to bridge
Acknowledgements evaluation. J Struct Engng, ASCE 1994;120(6):1745–62.
[9] Ransom AL, Heywood RJ. Recommendation for proof load test-
ing in Australia. In: Chirgwin GJ, editor. Proceedings of the
The present paper presents part of the work conducted AUSTROADS 1997 Bridge Conference, Bridging the Millennia,
during a three months research leave for M.H. Faber, at Sydney: AUSTROADS Inc., 1997;1:232–44.
M.H. Faber et al. / Engineering Structures 22 (2000) 1677–1689 1689

[10] OHBDC. Ontario highway bridge design code. Ontario: Ministry structures: statistical and reliability issues. Proceedings of the 2nd
of Transportation, 1993. RILEM International Conference on Rehabilitation of Structures,
[11] AASHTO-LRFD. AASHTO-LRFD bridge design specifications. Melbourne, Australia, 1998:91–101.
Washington, DC: Association of State Highway and Transpor- [18] Box GEP, Tiao GC. Bayesian inference in statistical analysis.
tation Officials, 1994. Reading, MA: Addison-Wesley, 1973.
[12] STRUREL, RCP Consulting Software, technical and users man- [19] Faber MH, Hommel DL, Maglie R. Aspects of safety and oper-
uals, version 6.1. Munich: RCP GmbH, 1998. ation of bridges during rehabilitation. International Symposium
[13] Mori Y, Ellingwood BR. Maintaining reliability of concrete struc- on Advances in Bridge Aerodynamics, Ship Collision Analysis
tures. I: role of inspection/repair. J Struct Engng, ASCE and Operation and Maintenance, Copenhagen, Denmark, 1998.
1994;120(3):824–45. [20] Madsen HO, Krenk S, Lind NC. Methods of structural safety.
[14] Nowak AS. Calibration of LRFD bridge design code, NCHRP New York: Prentice-Hall, 1986.
Project 12-33, University of Michigan, Ann Arbor, MI, 1993. [21] Thoft-Christensen P, Baker M. Structural reliability theory and
[15] Enright MP, Frangopol DM. Service-life prediction of deteriorat- its applications. Berlin: Springer-Verlag, 1982.
ing concrete bridges. J Struct Engng, ASCE 1998;124(3):309–17. [22] Eurocode. Basis of design and actions on structures—part 1: basis
[16] Stewart MG, Rosowsky DV. Time-dependent reliability of deter- of design. ENV 1991-1, 1994.
iorating reinforced concrete bridge decks. Struct Safety [23] Stewart MG, Rosowsky DV. Structural safety and serviceability
1998;20(1):91–109. of concrete bridges subject to corrosion. J Infrastruct Sys ASCE
[17] Val DV, Stewart MG, Melchers RE. Assessment of existing RC 1998;4(4):146–55.

You might also like