Engineering Structures 29 (2007) 1064–1073

www.elsevier.com/locate/engstruct

Evaluation of dynamic loads on a skew box girder continuous bridge

Part II: Parametric study and dynamic load factor
Demeke B. Ashebo a , Tommy H.T. Chan a,∗ , Ling Yu a,b
a Department of Civil and Structural Engineering, The Hong Kong Polytechnic University, Kowloon, Hong Kong, China
b College of Science and Engineering, Jinan University, Guangzhou, 510632, PR China

Received 2 November 2005; received in revised form 23 June 2006; accepted 5 July 2006
Available online 25 September 2006

Abstract

Studies on dynamic loads are important for bridge engineering as well as pavement design. A large number of research studies have indicated
that bridge dynamic loads increase road surface damage by a factor of 2–4. Although the field test is the best available approach to understanding
actual vehicle-induced dynamic loads on bridges, according to pervious studies there is only a limited amount of field data available on skew box
girder continuous bridges. This paper presents an evaluation of vehicle-induced dynamic loads, based on a field test that was carried out on a
skew box girder continuous bridge as reported in a companion paper (Part I). The effects of different parameters such as the weight, speed, type,
number of axles and position of vehicles on dynamic loads are investigated. Based on the statistical analysis, the use of the dynamic load factor
(DLF) is proposed. The dynamic load factor obtained in this study is less than the values provided by most bridge design codes.
c 2006 Elsevier Ltd. All rights reserved.

Keywords: Skew bridge; Dynamic load factor; Dynamic load allowance; Normal traffic condition

1. Introduction estimated based on the static load, it has never been ascertained
as it varies from case to case.
Vehicles that are expected to cross a bridge are accounted for Most studies have used the analytical approach to study the
in the design or evaluation of the bridge through a static design bridge–vehicle interaction problem and estimated the dynamic
loading and a certain prescribed fraction of it that is referred to load factor (DLF). Even though the field test is the best
as an impact factor, a dynamic load allowance, or a dynamic available approach to understanding the actual bridge–vehicle
increment [1]. interaction to estimate the DLF, the amount of data available
As opposed to static design loading which is a tangible from the field on dynamic loads is limited [2]. Billings [3]
entity that can be formulated from the static weights of actual carried out field measurements of dynamic loads on 27 bridges
or foreseen vehicles, determining the dynamic load is not a of various kinds. He found that the dynamic load allowance
straightforward procedure because of the complex nature of the ranges from 0.05 to 0.10 and from 0.08 to 0.20 for prestressed
interaction between the bridge and moving vehicles. concrete and steel bridges, respectively. Tests carried out by
Numerous studies on bridge dynamics have been carried Cantieni [4] on 226 bridges in which most of the bridges
out. It has been discovered that the magnitude of the dynamic were loaded with the same vehicle, under the same load,
load depends on several factors, including bridge dynamic and with the same tire pressure showed that a dynamic load
behaviors, road roughness, vehicle dynamic characteristics, allowance as high as 0.7 could be obtained from the field.
vehicle speed, type, weight, number of axles, axle spacing, Chan and O’Connor [5,6] have found maximum value of 1.25
the position of vehicles on the bridge, and so forth. Although for a dynamic load allowance. Recently, on the basis of the
the exact magnitude of this dynamic load can be reasonably analytical simulations and field tests, Nowak et al. [7] pointed
out that the dynamic load allowance is considerably lower
∗ Corresponding author. Tel.: +852 2766 6061; fax: +852 2334 6389. than code-specified values. According to their findings, the
E-mail address: cetommy@polyu.edu.hk (T.H.T. Chan). maximum simulated and measured dynamic load allowances

0141-0296/$ - see front matter c 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2006.07.013
 D.B. Ashebo et al. / Engineering Structures 29 (2007) 1064–1073 1065

do not exceed 0.17 for a single heavy truck, and 0.10 for two 2.3. Design of the filter
trucks moving side-by-side. It can be understood that, given the
limited number of field tests, most of the studies on vehicle- As for the analysis of a large amount of field data on the
induced dynamic loads on bridges have ended up with different dynamic responses of the bridge owing to normal traffic, the
outcomes. This shows the need to conduct more field tests on fastest and the most convenient method to estimate the static
various bridges using unbiased random samples of vehicles. responses of a bridge due to a moving vehicle is to use an
In this study, a field test was conducted on a skew box appropriate filter. However, no universally accepted method yet
girder continuous bridge to collect vehicle-induced dynamic exists for designing a digital filter specifically for the study of
response data on the bridge and information from various kinds bridge dynamics. The method currently used, to choose the best
of vehicles under normal traffic conditions. The details about filter, is based on a trial-and-error approach.
the tested bridge such as the cross section, location of sensors, In selecting the low pass cutoff frequency to filter out
roadway width are presented in a companion paper [15]. A the dynamic portion of the responses and estimate the static
parametric study was carried out to determine the factors that response of the bridge due to moving vehicles, careful attention
affect the DLF. To determine the appropriate DLF value for should be taken to ensure that nearly all static frequency
design purposes, a statistical analysis was conducted. responses must occur below the cutoff frequency and nearly all
dynamic frequency responses should be filtered out above this
2. Dynamic load factor
cutoff frequency [14].
2.1. Definition of the dynamic load factor A low frequency “body bounce” may influence the
calculated dynamic load factor values if this dynamic mode is
Various studies have been carried out to determine and below the low pass threshold and remains among the presumed
evaluate the magnitude of the dynamic load factor. Past studies extracted responses. According to Billings [3] and Chan and
have used several definitions to represent this factor. Bakht and O’Conner [6], a typical vehicle “body bounce” occurs in the
Pinjarkar [8], in their state-of-the-art review of bridge dynamic range between 2 and 5 Hz frequencies. A vehicle body bounce,
tests, have summarized eight definitions of dynamic load factor. if present, will result in higher calculated static responses,
After comparing and analyzing the eight definitions, they which in turn will cause the calculated dynamic load factor to
suggested the following equation for computing the dynamic decrease.
load factor: The typical frequency spectrum graphs from both accelera-
DLF = 1 + DLA (1) tion and strain responses of the bridge under moving vehicles
are shown in Fig. 10 of the companion paper [15]. During the
where DLF is the dynamic load factor, and DLA is the dynamic selection of appropriate filter parameters, the cut off frequen-
load allowance given by cies greater than the frequency range that affects the static strain
Rdyn − Rstat and less than the first natural frequency of the bridge were con-
DLA = (2) sidered. In addition, in order to avoid the body bounce effect
Rstat
of a vehicle, cutoff frequencies in between 2 and 5 Hz were
where, Rdyn is the maximum dynamic response and Rstat is the avoided. A low pass Butterworth digital filter was used in this
maximum static response. study. This type of filter ideally retains all frequency responses
Various researchers (such as Cantieni [9], Chan and below the cutoff frequency and omits all frequency responses
O’Conner [5,6], Kim and Nowak [10], Laman et al. [11]) have above the cutoff frequency [14]. The other important parameter
used these equations to compute the dynamic load factor, and which needs consideration in filter design is the order of the fil-
the equation has also be used for this purpose in design codes ter. As indicated by Johnson et al. [16], by keeping the low pass
(Ministry of Transportation of Ontario [12], AASHTO [13]). cutoff frequency constant, the measured response improves as
2.2. Dynamic load factor from the responses of a bridge the filter order increases.
The dynamic responses of the bridge induced by the control
Most of the previous studies on the dynamic load factor vehicle (calibration truck) as reported in [15] were filtered
used the measured responses of a bridge as their input. The by applying different low pass cutoff frequencies and filter
dynamic responses of a bridge are mainly measured in terms orders. The results were compared with the known measured
of the bending moment or the deflection at selected locations static strain responses of the bridge caused by the calibration
of the bridge structure. Then, the dynamic load factor can truck as well as with theoretically computed bending moment
be computed by dividing the measured maximum dynamic responses.
response to the known static response by using Eqs. (1) and (2). The comparison of the measured dynamic and the
According to Eq. (2), Rstat , the static response can be acquired corresponding filtered responses with the measured and the
from a theoretical analysis, by measuring or filtering out theoretically computed responses was carried out on the
the dynamic component of the measured dynamic responses, selected strain gauge location of the bridge. The selected strain
whereas Rdyn can be directly obtained from the peak measured gauges were strain gauges 7, 8, 9, and 22 (refer to Figure 2
dynamic response. The next section will present the method of in the companion paper [15] for the locations of the strain
estimating the static response Rstat from the measured dynamic gauges). The first three strain gauges were selected from the
responses. middle of each web of the first span so that the maximum
1066 D.B. Ashebo et al. / Engineering Structures 29 (2007) 1064–1073

(a) Maximum measured static, and dynamic and filtered strain responses: (b) Maximum theoretically obtained static, and dynamic and filtered
fast lane. measured bending moment responses: fast lane.

(c) Maximum measured static, and dynamic and filtered strain responses: (d) Maximum theoretically obtained static, and dynamic and filtered
slow lane. measured bending moment responses: slow lane.

Fig. 1. Comparison of responses before and after filtering.

response could be obtained from those locations. As to negative

bending moment responses, strain gauge 22 was selected to
study the responses due to negative bending moment responses
at the bridge deck near the support. In addition to the criteria
presented in the above paragraphs, the selection of the filter
parameters was based on the good agreement between the
filtered responses and the measured responses from the known
loads. It was found that a Butterworth filter with a low pass
cutoff frequency of 1.4 Hz and filter order 10 was fit for the
given criteria. Some of the results of the filtered responses with (a) Dynamic response.
measured and theoretical values are shown in Fig. 1. From the
four figures in Fig. 1, it can be noticed that, in general, the
filtered responses agree well with the measured and theoretical
responses.
Typical dynamic and static responses for computing the
dynamic load factor obtained from strain gauge 9 (refer to
Figure 2 in [15] for the location of the sensor) for five axle
trucks moving at highway speed are given in Fig. 2. After
estimating the static responses (in this case the static strain
responses) by filtering, it is a straightforward procedure to
compute the dynamic load factors using Eqs. (1) and (2). (b) Static response.

3. Factors affecting the dynamic load factor Fig. 2. Typical responses for computing dynamic load factor.

As mentioned in the preceding sections, many factors affect 3.1. Weight of the vehicle
the magnitude of the dynamic load factor. When carrying
out dynamic measurements in the field, in addition to strain It was indicated in previous studies (Chan and O’Conner [5],
and acceleration responses, information on the speed, weight, Nassif and Nowak [17], Laman et al. [11], Nowak et al. [7])
number of axles, axle spacing, and position of the vehicles on that the dynamic load factor is dependent on the weight of
the bridge was acquired. The influence of these parameters to the vehicles. In this study, the strain responses of the bridge
the dynamic load factor was studied and is presented in the were recorded for each case having a single vehicle traversing
following sections: on the bridge. According to the recorded data, in 5 days,
 D.B. Ashebo et al. / Engineering Structures 29 (2007) 1064–1073 1067

gauge locations. The maximum static strains are found to be

recorded in the strain gauges 9 and 19, which are located on
the first web of the bridge deck under the slow lane [15]. The
mean calculated DLA for each strain gauge location of 7–9, 18,
19, and 22 are 0.233, 0.208, 0.193, 0.182, 0.165, and 0.149,
respectively. Although the strain recorded in the strain gauge
at location 22 is found to be the smallest, the mean computed
DLA is not the largest of all. On the contrary, the mean DLA
computed from this location is the smallest of all. In contrast
to the other strain gauge locations mentioned here, strain gauge
22 recorded negative bending moments near the support.
(a) At strain gauge 9.
3.2. Number of axles

Some bridge design codes like OHBDC [12] and

CHBDC [18] relate the dynamic load factor with the number
of axles of the design vehicle. In this study, an investigation has
been carried out to determine if there is a correlation between
the dynamic load factor and the number of axles of the vehicles.
From the data samples that were collected of single vehicles,
it was found that the number of axles of the vehicles ranged
from two to six. Vehicles with five axles make up the largest
proportion in the collected database followed by vehicles with
three, four, two and six axles.
(b) At strain gauge 19.
The DLA versus the number of axles for some of the
strain gauge locations are given in Fig. 4. The statistical values
(means, standard deviations, maximum values, and minimum
values) of the computed DLA versus the number of axles
are given in Table 1. The computed means of the DLAs are
summarized and given in Fig. 5. From Fig. 5 and Table 1, it
is noticed that the maximum mean DLA obtained from strain
gauges 7–9, 18 and 19 are all from vehicles with two axles,
whereas the minimum DLA for the same strain gauge locations
are from six-axle vehicles. The minimum mean DLA for strain
gauge 22 was obtained from a three-axle vehicle. In general,
the mean DLAs obtained from strain gauge locations 7–9, 18
(c) At strain gauge 22. and 19, for the three-axle vehicles, are lower than that of other
vehicles except for the six-axle vehicles. The reason for this,
Fig. 3. DLA versus static live load strain. according to the video data collected from the field, is that
most of the single three-axle vehicles recorded in the three-
309 good cases (a single vehicle at a time) were selected. axle vehicle category were double-decker buses (which are very
From the collected data, it was noticed that most of the single common in Hong Kong), carrying a more or less similar load
vehicles (more than 95%) during the test choose Lane 2 (the on them. In other words, the samples of the three-axle vehicles
slow lane) to cross the bridge. As for the selection of the collected are more uniform in type than the others.
locations of the strain response measurements, those locations The standard deviation given in Table 1 shows which
with the expected maximum response measurement points were samples of the data are more uniform than others. Excluding
included in the analysis. Therefore, the strain gauge sensors the six-axle vehicles (since the sample of these are few in
along the lines m3 and d2 as shown in [15] (strain gauges 7–9, number), the data sample of the three-axle vehicles has the
18 and 19, refer to Figure 2 in [15] for the locations of the strain lowest standard deviations for all strain gauge locations except
gauges) were expected to have their maximum positive bending that of strain gauge 7, while the standard deviations obtained
moment in the first span, and strain gauge 22 was expected to from the data sample of the five-axle vehicle category are
have its maximum negative moment at the support which is near the largest for all strain gauge locations. Therefore, the data
the first pier. samples in the three-axle category are the most uniform, while
The computed DLA versus the corresponding static live load the data samples from the five-axle vehicles are the most diverse
strain at strain gauge locations of 9, 19, and 22 are given in terms of the types and weights of the vehicles.
in Fig. 3. It is clear from the figure that the dynamic load Again, after inspecting the video data of the five-axle and
allowance decreases as the static load increases in all strain four-axle vehicle sample categories, it was visually identified
1068 D.B. Ashebo et al. / Engineering Structures 29 (2007) 1064–1073

Table 1
Summary of statistical results of DLA versus number of axles

Strain Axle DLA

gauge no. no.
Mean Std. dev. Max Min
2 0.289241 0.073942 0.452593 0.176001
3 0.161441 0.082628 0.573417 0.07903
7 4 0.262988 0.114585 0.55448 0.09321
5 0.248076 0.133971 0.651799 0.04201
6 0.147119 0.045657 0.197093 0.107589
2 0.267037 0.084624 0.473362 0.138211
(a) At strain gauge 9. 3 0.196875 0.055161 0.430297 0.079114
8 4 0.202642 0.091372 0.441486 0.030014
5 0.20401 0.104979 0.625103 0.0439
6 0.12408 0.028023 0.156433 0.107361
2 0.229799 0.088268 0.414607 0.015755
3 0.178874 0.066278 0.624146 0.054791
9 4 0.176979 0.101345 0.503527 0.003317
5 0.198215 0.112807 0.570945 0.0205
6 0.12252 0.067875 0.194421 0.059556
2 0.225658 0.073324 0.452552 0.110732
3 0.118234 0.07255 0.515012 0.048401
18 4 0.18941 0.092925 0.465682 0.014411
5 0.200829 0.129912 0.774827 0.023184
6 0.110482 0.019553 0.12633 0.088632
(b) At strain gauge 19. 2 0.186908 0.059795 0.369169 0.110036
3 0.154918 0.056567 0.401061 0.018597
19 4 0.163951 0.089192 0.393354 −0.00326
5 0.167935 0.099998 0.51417 0.02535
6 0.096808 0.020875 0.11596 0.074557
2 0.200712 0.119993 0.577771 0.024848
3 0.041231 0.056498 0.303156 −0.02028
22 4 0.176328 0.104393 0.441144 0.029277
5 0.181741 0.172069 0.906661 −0.03882
6 0.136383 0.19762 0.363467 0.003389

the visually identified unloaded trucks from the samples. A

total of 52 vehicles were identified as unloaded trucks. Forty-
(c) At strain gauge 22. three of them were five-axle vehicles, eight of them were four-
axle vehicles and one of them was a two-axle vehicle. The
Fig. 4. DLA versus number of axles.
computed mean DLA values are plotted in Fig. 6. The figure
shows a decrease in the mean DLA for the four- and five-
axle vehicles. The decrease in the mean DLA is especially
significant for the five-axle vehicles. This is because a large
number of unloaded trucks that were excluded from the sample
were five-axle trucks. Thus, while studying the relationship
between the number of axles and the DLA it is important
to consider the actual load condition of the vehicles being
examined.
From the information presented in Figs. 4–6, it is not
apparent whether there is any relationship between the number
of axles and the DLA. Therefore a correlation coefficient study
has been carried out on the data of the number of axles and
the corresponding DLA values to determine what kind of
relationship exists between these two variables.
Fig. 5. Number of axles versus mean DLA. In probabilty theory and statisitcs, correlation is a numeric
measure of the strength of the linear relationship between
that a significant number of unloaded trucks were included in two random variables. A correlation coefficient is a number
the samples. In order to study the effects of these unloaded between −1 and 1 that measures the degree to which two
trucks on the samples, an analysis was carried out by excluding variables are linearly related. A correlation coefficient of 1 or
 D.B. Ashebo et al. / Engineering Structures 29 (2007) 1064–1073 1069

Table 2
Summary of correlation coefficients for number of axles versus DLA

St gauge 7 St gauge 8 St gauge 9 St gauge 18 St gauge 19 St gauge 22

Corr. coef. 0.08173 −0.1336 −0.0295 0.1125 −0.0229 0.1869

(a) Strain gauge 9.

Fig. 6. Number of axles versus mean DLA after excluding unloaded trucks.

−1 means that there is strong linear relationship between the

two variables, whereas a correlation coefficient of 0 means that
there is no linear relationship between the two variables. Based
on this statistical theory, the correlation coefficients between the
number of axles and the DLA are computed for selected strain
gauge locations and presented in Table 2. From the table, it can
be noticed that the absolute value of the correlation coefficients
is much less than 1 and closer to zero. This shows that the (b) Strain gauge 19.
dependency of the dynamic load allowance on the number of
axles in a vehicle is very small or non-existent.

3.3. Speed of vehicles

The effect of the speed of vehicles on the dynamic load

factor has been studied both analytically and experimentally.
The study carried out by Hwang and Nowak [19] reported
that the lightest vehicles induced an increase in DLA with
an increase of speed, however, the heavier vehicles induced a
smaller DLA value as speed increased. Whereas, based on data
obtained from field measurements, Laman et al. [11] concluded (c) Strain gauge 22.
that there is no correlation between vehicle speed and DLA
values, while Schwarz and Laman [20], in their investigation Fig. 7. Speed of vehicles versus DLA.
involving a limited number of samples collected from the field,
indicated that DLA increases with speed. Chan et al. [21]
concluded that, for the same axle spacing parameter (ASP = around 55 km/h. Beyond this speed up to the largest speed,
1.0), the dynamic load allowance increases with vehicle speed. the relationship between the speed and the DLA is not clear.
Thus, different conclusions about the relationship between the In order to study the relationship between these two variables,
speed of vehicles and DLA were reached in previous studies. the correlation coefficients were computed for each strain gauge
Vehicles with different speeds were randomly selected from location and the results are tabulated in Table 3. The fact that
the database and analyzed to compute the DLA in different the values of the correlation coefficients are less than 1 by a big
strain gauge locations. The results of the DLA versus the speed margin and significantly greater than 0 in all sensor locations
of vehicles are plotted and presented for some of the strain shows that the speed and the DLA are weakly correlated. Even
gauge locations in Fig. 7. The figures show that the DLA though the correlation between DLA and speed is not strong,
increases with the speed of vehicles until the speed of the a positive sign in the correlation coefficients indicates that the
vehicles reaches around 55 km/h. The largest DLA value for two variables have an increasing linear relationship (i.e., as the
all strain gauge locations is also noticed at a vehicle speed of speed of vehicles increases the DLA will also increase).
1070 D.B. Ashebo et al. / Engineering Structures 29 (2007) 1064–1073

Table 3
Summary of correlation coefficients between vehicle speed and DLA

St gauge 7 St gauge 8 St gauge 9 St gauge 18 St gauge 19 St gauge 22

Corr. coef. 0.2184 0.3511 0.3868 0.2537 0.4373 0.3471

Fig. 8. Typical responses for vehicles moving on a bridge side by side or in a

group. (a) At strain gauge 9.

Fig. 9. Typical axle signal output for vehicles moving on a bridge side by side
or in a group.
(b) At strain gauge 19.

3.4. Side by side vehicles or groups of vehicles

In order to study the effect of vehicles that are moving side

by side or in a group on dynamic load factor, samples of the
two types of vehicles have been extracted from the data base
and analyzed. A typical response graph of trucks moving side
by side on the bridge with the corresponding axle sensor output
for one of the heaviest cases is given in Figs. 8 and 9. The
calculated DLAs for different strain gauge locations are plotted
for some of the strain gauge locations as a function of the
corresponding static micro strain, and are shown in Fig. 10.
For a given maximum static stress, differences in the measured (c) At strain gauge 22.
DLA can be observed in Fig. 10 among strain gauge locations.
However, a correlation between the maximum static stress and Fig. 10. Static live load strains versus DLA.
the DLA is evident for each strain gauge location. Lower DLA
values consistently correspond to larger static strains. The same Table 4
relationship between the DLA and the maximum static live load Statistical results for DLA of vehicles moving side by side or in a group
strain has been observed in the previous sections for the static Strain DLA
live load strain of single vehicles versus the DLA. gauge no.
Mean Std. Dev. Max Min
The mean, variance, maximum and minimum values of DLA
for each strain gauge location are tabulated in Table 4. From 7 0.209085 0.08383 0.421886 0.047536
the table, it can be seen that the computed DLAs in all strain 8 0.165128 0.069273 0.434519 0.027063
9 0.164231 0.094062 0.481776 0.020425
gauge locations are lower than the corresponding mean DLAs 18 0.169556 0.075642 0.475827 0.025449
computed for single vehicles. It can be concluded that the DLA 19 0.142914 0.073156 0.475951 0.025449
obtained from vehicles moving side by side or in a group is 22 0.115249 0.074274 0.365941 −0.0097
less than the DLA obtained from a single vehicle moving on
 D.B. Ashebo et al. / Engineering Structures 29 (2007) 1064–1073 1071

Table 5
Statistical parameters of DLA at strain gauges 9, 19, 22, and 9 and 19 combined

Strain gauge Factors Mean Std. dev. Variance Max Min

9 DLA 0.191311 0.098226 0.009648 0.624146 0.003317
Micro-strain 7.542015 3.395904 11.53216 17.79328 1.474164
19 DLA 0.16023 0.081369 0.006621 0.51417 −0.00326
Micro-strain 8.079886 3.591397 12.89813 19.13313 1.474585
22 DLA 0.14425 0.148982 0.022196 0.906661 −0.03882
Micro-strain 4.463992 2.01605 4.064459 10.1301 0.698893
9 and 19 combined DLA 0.17577 0.091448 0.008363 0.624146 −0.00326
Micro-strain 7.810951 3.502417 12.26693 19.13313 1.474164

the bridge. The same conclusion was reached by Hwang and

Nowak [19] and Nowak et al. [7].

4. Statistical analysis of the dynamic load factor

A large scatter of the values of dynamic load factors can be

obtained from field measurements, even when the bridge and
the vehicle are the same. Bakht et al. [1] concluded that the
dynamic load factor is not a deterministic quantity. To obtain a
single value of this factor for design purposes it is necessary to
know the statistical properties of the scatter data. Nassif and
Nowak [17], and Kim and Nowak [10] have used statistical Fig. 11. DLA with the corresponding static strain for data of strain gauges 9
and 19 combined.
approaches to derive the cumulative distribution function and
other statistical parameters of the values of dynamic load differences in the measured DLA can be observed in Fig. 11
factors collected in the field. among strain gauge locations; however, a correlation between
In this study, a large amount of data on bridge responses the maximum static strain and DLA is evident for each strain
and vehicle information was acquired through a field test. gauge location. The statistical parameters for the computed
Therefore, it is important to introduce a statistical analysis to DLA and static strain are tabulated in Table 5. Although the
obtain the appropriate value of DLA. In the previous sections distance between strain gauges 9 and 19 is not very large, the
it was found that DLA is dependent on the static weight of a mean DLA value for strain gauge 9 is greater than that of strain
vehicle. It was also found that the speed of vehicles can affect gauge 19. This may be attributed to the difference in their mean
the DLA. strain.
From the identified parameters, it was determined that DLA In probability theory and statistics, the variance in the
is strongly affected by the static weight of the vehicles. It was random variables is a measure of their statistical dispersion,
also shown that the DLA from a single vehicle at a time is indicating how far from the expected value (mean) their values
larger than that of vehicles moving side by side on the bridge. typically are. The variance in DLA from strain gauge 19 is less
Therefore, it was decided that the DLAs obtained from a single than that of 9, which shows that the DLA values in strain gauge
vehicle moving on the bridge would be the governing DLA 19 are less dispersed. The DLA from strain gauge 22 has the
for the statistical analysis. However, more than 95% of the largest variance which indicates that the distribution of DLA is
single vehicles crossed the bridge using Lane 2 (slow lane, refer widely dispersed at this location.
to [15]). The remaining 5%, of which most were light vehicles, The maximum value of the DLA of the positive bending
crossed the bridge using Lane 1 (fast lane, refer to [15]). Thus, moment responses in the mid-point of the first span was 0.62,
the locations of maximum response were found near strain obtained at strain gauge 9. Meanwhile, the maximum value
gauges 9 and 19. For these reasons, the DLA values obtained of 0.91 DLA was obtained for the negative bending moment
from these two strain gauge locations will be included in the responses near the support at strain gauge 22. However, the
statistical analysis for maximum responses in the middle span mean DLA at strain gauge 22 is the smallest of all, which is
of the first span, and the DLAs from strain gauge 22 will be used equal to 0.14. Since strain gauges 9 and 19 are very close to
in the statistical analysis to determine the appropriate DLA for one another and located at the mid-point of the span on the
negative bending moment responses near the support. first web of the first span, the data obtained from them are
The live load static strain response was obtained in each combined to widen the data samples. The statistical results from
strain gauge location (strain gauges 9, 19, and 22) for each the combination of the two data sets of the strain gauges 9 and
vehicle crossing. Since strain gauges 9 and 19 are located on 19 are also listed in Table 5.
the first web underneath the first span some distance apart, the In order to decide the design value of DLA, the 90%
data obtained from these two locations are combined together confidence interval of DLA is computed. During the field test,
and presented in Fig. 11. For a given maximum static strain, a three-axle truck weighing 240 kN (which is less than the
1072 D.B. Ashebo et al. / Engineering Structures 29 (2007) 1064–1073

Table 6
Summary of DLF of bridges according to different bridge design codes and present study

AASHTO (Standard) AASHTO (LRFD) CHBDC BS5400 BD 37/01 SDM (Hong Kong) Present study
DLA 0.252 0.33 0.25 0.25 0.80 0.80 0.24
DLF 1.252 1.33 1.25 1.25 1.80 1.80 1.24

5.3. Eurocode

In the Eurocode [23], the dynamic amplification established

for a medium pavement quality and pneumatic vehicle
suspension has been included for traffic load models, which
depends on various parameters such as road roughness and
location of the load under consideration.

5.4. British standards institute

Fig. 12. The 90% confidence interval of DLA. The British Standard Institute’s BS 5400 Steel, Concrete
and Composite Bridges Part 2, Specification for Loads (BSI
weight of the AASHTO HS20-44 design truck) generated about 1978) [24] gives one type of highway bridge live load, HA
14.5 micro strains at strain gauge locations 9 and 19. To be loads, a 25% factor for impact or dynamic effect by multiplying
conservative enough in deciding the design values of DLA the live load by this factor. For an HB loading, no dynamic load
while computing the percentage confidence interval, strains factor is included.
of less than 14.5 micro strains have been included in the The amended Manual for Roads and Bridge Designs
analysis. Therefore, the 90% confidence interval of DLA for BD37/01 [25] shows that a dynamic factor of 80% was made in
those vehicles that generated strains greater than or equal to deriving the HA loading. The factor was applied to the highest
the mean strain, obtained from the data of strain gauges 9 and axle load and only included in the single vehicle loading case.
19 combined together, is computed, and the result is presented
in Fig. 12. The maximum value of DLA in this percentage
5.5. Hong Kong structures design manual
confidence interval was found to be about 0.24. Thus, for a
safe design, this maximum value is taken as the design value
According to the Structures Design Manual for Highways
of DLA.
and Railways of Hong Kong [26], highway structures and
5. Comparing different bridge design codes railway bridges shall be designed on the basis of the limit
state philosophy contained in BS 5400 Steel, Concrete, and
5.1. American associations of states highway and transport Composite Bridges.
officials All undated references to BS 5400 refer to the current
edition of the Structures Design Manual for Highways and
In the standard AASHTO code [22], to allow for dynamic, Railways of Hong Kong [26] with the exception of loading
vibratory and impact effects, the live load stress in the (originally given in BS 5400: Part 2 [24]), which shall refer to
superstructure of the bridge due to the H and HS loadings the current edition of the United Kingdom Highways Agency’s
should be multiplied by a DLA as defined by, Departmental Standard BD 37/01 [25].
15.4 From Table 6 it can be noticed that the dynamic load factor
DLA = (3) obtained from the present study is less than the values obtained
L + 38.1
from all of the design codes. BD 37/01 and the Structures
where L is the loaded length in meters, and the DLA does not
Design Manual for Highways and Railways [26] were found to
exceed 0.3.
be the most conservative in deriving the value of the dynamic
In the new LRFD AASHTO code [13], a live load is
load factor for design purposes.
specified as a combination of a HS20 truck and a uniformly
distributed load of 9.3 kN/m. A DLA value of 0.33 for the truck
effect is taken with no dynamic load applied to the uniform 6. Conclusions
loading.
In this study an evaluation of vehicle-induced dynamic load
5.2. Canadian highway bridge design codes that was carried out on a skew continuous bridge is presented.
The bridge response data acquired in the field test were mainly
In the newly introduced Canadian Highway Bridge Design measured strains from strain gauge sensors. As the strains are
Code [18], the DLA is practically equal to 0.25 for all heavy proportional to the bending moment of the bridge, therefore, the
trucks. Larger DLAs of 0.4 and 0.3 are applied to single and computed DLFs values obtained in this study are based on the
dual axle vehicles, respectively. bending moment of the bridge.
 D.B. Ashebo et al. / Engineering Structures 29 (2007) 1064–1073 1073

On the basis of the results obtained from this particular field [5] Chan THT, O’Conner C. Wheel loads from highway bridge strain. Journal
study the following conclusions, which may be applicable for of Structural Engineering Division ASCE 1990;116(7):1751–71.
[6] Chan THT, O’Conner C. Vehicle model for highway bridge impact.
similar kinds of bridges, are drawn:
Journal of Structural Engineering Division ASCE 1990;116(7):1772–93.
(1) The dynamic load factor was found to depend on the weight [7] Nowak AS, Szerszen MM, Eom J. Dynamic loads on bridges. In: Ko JM,
of the vehicles, as the weight of the vehicles increases the Xu YL, editors. Advance in structural mechanics, vol. 1. 2000. p. 407–14.
[8] Bakht B, Pinjarkar SG. Dynamic testing of highway bridges—a review.
dynamic load factors (DLFs) decrease. Transportation Research Record 1989;1223:93–100.
(2) Weak correlations were found between the speed of the [9] Cantieni R. Dynamic load testing of highway bridges. In: Proceedings
vehicles and the DLF. IABSE, 1984; p. 75–84.
(3) No relationship can be found between the number of axles [10] Kim S, Nowak AS. Load distribution and impact factors for I-girder
and the DLF. bridges. Journal of Bridge Engineering ASCE 1997;2&3:97–104.
[11] Laman JA, Pechar JS, Boothby TE. Dynamic load factor for through-truss
(4) The mean DLF obtained from vehicles moving side by side
bridges. Journal of Bridge Engineering ASCE 1999;4(4):231–41.
in a given sensor location was found to be less than the [12] Ontario highway bridge design code. Ministry of transportation and
mean DLF obtained from a single vehicle. For example the communication. Ontario (Canada): Highway Engineering Division; 1991.
mean DLF values of 1.164 and 1.193 were obtained for [13] American association of state highway and transportation officials
vehicles side by side and a single vehicle, respectively. (AASHTO). LRFD bridge design specifications. Washington (DC);
1998.
(5) The DLF obtained based on the statistical analysis in the [14] Thater G, Chang P, Schelling DR, Fu CC. Estimation of bridge static
study was found to be less than the DLF values provided by responses and vehicle weights by frequency responses analysis. Canadian
most bridge design codes. Journal of Civil Engineering 1998;25:631–9.
[15] Ashebo DB, Chan THT, Yu L. Evaluation of dynamic loads on a skew box
Acknowledgements girder continuous bridge: Part I field test and modal analysis. Engineering
Structures 2007;29(6):1052–63.
[16] Johnson DE, Johnson JR, Moore HP. A handbook of active filters.
The work described in this paper was supported by Englewood Cliffs (NJ): Prentice-Hall; 1980.
grants from the Research Grants Council of the Hong [17] Nassif HH, Nowak AS. Dynamic load spectra for girder bridge.
Kong Special Administrative Region (PolyU 5033/02E), the Transportation Research Records 1995;1476:69–83.
Hong Kong Polytechnic University Postdoctoral Fellowship [18] Canadian Highway Bridge Design Code CSA International (Toronto).
Research Grants (G-YX25) and the National Natural Sciences Ontario, Canada; 2000.
[19] Hwang ES, Nowak AS. Simulation of dynamic load for bridges. Journal
Foundation of China (50378009); the support provided of Structural Engineering ASCE 1991;117(5):1413–34.
by the Highways Department of the Hong Kong Special [20] Schwarz M, Laman JA. Response of prestressed concrete I-girder bridges
Administrative Region is also gratefully acknowledged. to live load. Journal of Bridge Engineering ASCE 2001;6(1):1–8.
[21] Chan THT, Yu L, Yung TH, Chan JHF. A new bridge–vehicle system
References part II: Parametric study. Structural Engineering and Mechanics an
International Journal 2003;15(1):21–38.
[1] Bakht B, Jaeger GL, Mufti AA. Bridge engineering recent innovation. [22] Association of state highway and transportation officials (AASHTO).
Vashi, New Bombay: R&D Center; 1994. Standard specifications for highway bridges. Washington (DC); 1996.
[2] Paultre P, Chaallal O, Proulx J. Bridge dynamics and dynamic [23] Eurocode 1: Actions on structures—Part 2: Traffic loads on bridges. The
amplification factors: A review of analytical and experimental findings. European Standard EN 1991-2:2003.
Canadian Journal of Civil Engineering 1992;19:260–78. [24] BS 5400. Steel, concrete and composite bridges. Part 2. Specification for
[3] Billings JR. Dynamic loading and testing of bridges in Ontario. Canadian loads. London: British Standard Institution; 1978.
Journal of Civil Engineering 1984;11:833–43. [25] BD37/01. Loads for highway bridges. London: British Department of
[4] Cantieni R. Dynamic behaviour of highway bridges under the passage of Transport; 2001.
heavy vehicles. Report no. 220. Swiss Federal Laboratories for Materials [26] Structures design manual for highways and railways. The Hong Kong
Testing and Research (EMPA); 1992. p. 240. Government; 2002.