Engineering Structures 173 (2018) 251–268

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Performance and sensitivity analysis of UHPFRC-strengthened bridge T

columns subjected to vehicle collisions
⁎
Wei Fana,b, , Xin Xua, Zhiyong Zhanga, Xudong Shaoa
a
Key Laboratory for Wind and Bridge Engineering of Hunan Province, College of Civil Engineering, Hunan University, Changsha 410082, China
b
Department of Civil Engineering, University of Toronto, ON M5S 1A4, Canada

A R T I C LE I N FO A B S T R A C T

Keywords: Bridge columns made of normal concrete are evidenced to be susceptible to vehicle collisions. Particularly in the
Vehicle collision United States, vehicle collision has become one of the primary causes of bridge failures. This is largely due to the
Crashworthiness low crashworthiness of a conventional reinforced concrete (RC) column. Ultra-high-performance fiber-re-
FE modeling inforced concrete (UHPFRC) as one of advanced concrete materials has been experimentally demonstrated to
Ultra-high-performance fiber reinforced
possess excellent strength, durability, impact resistance and energy-absorbing capacity. Accordingly, one type of
concrete (UHPFRC)
UHPFRC-strengthened columns was proposed in this study as an alternative to RC columns that may be at risk
Residual load-carrying capacity
Response surface model for vehicle collision incidents. High-resolution finite element (FE) models were developed to investigate the
performance of UHPFRC-strengthened columns subjected to vehicle collisions. In the high-resolution FE model, a
three-span simply-supported girder bridge (including girder, pier column, column cap, bearing, etc.) was
adopted and modelled. Material models regarding normal concrete and UHPFRC as well as the vehicle model
were carefully calibrated by experimental data. The influence of initial gravity loads on impact responses was
found to be pronounced, and a damping-based method was proposed to efficiently exert permanent loads on pier
columns prior to a collision. Three different simplified models, as published in current studies, were investigated
to replace the whole bridge model. Two single-column models with different boundaries were shown to have low
accuracy. The pier-bent model considering the superstructure gravity was demonstrated as capable of predicting
collision-induced responses that are in good agreement with the high-resolution FE model. The impact re-
sistances of both RC and UHPFRC-strengthened columns were extensively investigated using the appropriate
simplified model. The crashworthiness of UHPFRC-strengthened column was found to be considerably superior
to that of RC column. An extensive parametric study was performed using response surface methodology to
explore the influences of reinforcement ratios, thickness of UHPFRC jacket, UHPFRC strength and initial impact
speed. The impact-resistant performance is mostly sensitive to changes in the thickness of UHPFRC jacket when
the impact speed is not very high. On the contrary, the residual capacity of the bridge column is hardly increased
by thickening UHPFRC jacket. In addition, the developed response surface models provided easy estimation of
impact-induced responses of an UHPFRC-strengthened column, which have potential use as the surrogates of
time-consuming FE simulations to efficiently examine the reliability and optimization of bridge columns under
impact loadings.

1. Introduction Among these recent accidents, the damages of bridge structures

were often catastrophic when an aberrant vehicle hit one of circular
A great number of catastrophic accidents on vehicle-bridge colli- reinforced concrete (RC) columns in a multiple-column bent, such as
sions have been documented around the world [1–4]. From the bridge the Tancahua Street Bridge over IH-37 and the Bridge on 26½ Road
failure data recorded during the period 1951–2000 in the United States over IH-70 [4]. This is partially attributed to the fact that the multiple-
[1,2], ~ 17% failures were due to collisions [5], which was the second column bents have been widely used in overpass bridge structures.
causes of bridge failures. Also, Buth et al. [4] and Xu [3] reviewed lots More importantly, this indicates that multiple-column bent bridges are
of severe vehicle-bridge collision accidents happened in recent years susceptible to vehicle collisions. Although much research has been
and illustrated the seriousness of the subject. devoted to investigating the performance of infrastructure subjected to

⁎
Corresponding author at: Key Laboratory for Wind and Bridge Engineering of Hunan Province, College of Civil Engineering, Hunan University, Changsha 410082, China.
E-mail addresses: [email protected], [email protected] (W. Fan).

https://doi.org/10.1016/j.engstruct.2018.06.113
Received 26 January 2018; Received in revised form 9 May 2018; Accepted 28 June 2018
0141-0296/ © 2018 Elsevier Ltd. All rights reserved.
W. Fan et al. Engineering Structures 173 (2018) 251–268

vehicle collisions [6–9], few studies placed an emphasis on the vul- 2.1. Bridge FE model and parameters
nerable multiple-column bent under vehicle impact loading. Liu [10]
and Xu [3] numerically investigated the vehicle-impact-resistant per- According to the layout of the overpass bridge shown in Fig. 1, the
formance of the multiple-column bent, where the cross sections of the corresponding high-resolution FE model was developed, as presented in
columns were rectangular, as opposed to the circular one commonly- Fig. 2. Solid elements were employed to model pier columns, bent caps,
used in engineering practice. In addition, some researchers [11,12] footings, bridge girders and abutments as well as bearings. The long-
investigated the performance of infrastructure subjected to vehicle itudinal and transverse reinforcements embedded in pier columns were
collisions through simplified single column model, where the fixed or explicitly modelled using beam elements. An overpass bridge is often
simple-supported boundary was used to consider the influence of the constructed to cross a highway in the excavation section of mountai-
omitted structure. However, the applicability of the simplified single nous areas. The geological conditions (e.g., soil type) are usually good,
column model has not yet been clarified, particularly for multiple and the piles exhibits a very high lateral resistance under impact
column bents. loadings with a short duration. Hence, like the Refs. [3,9,10], the in-
On the other hand, the collapses of multiple-column bent bridges fluence of soil-pile interaction was not taken into account in this study.
are largely attributed to the low crashworthiness of conventional RC Fix boundary conditions were arranged on the bottoms of the abut-
columns. For this reason, some retrofit measures have been proposed to ments and the footings. In addition, no obvious damages in abutments
improve the impact-resistant performance. For example, steel and FRP were reported in vehicle-bridge collision accidents. Hence, similar to
(Fiber reinforced plastics) jackets were recommended by Liu [10] and the Refs. [3,9,10], the abutment modeling was simplified in this study
Sha and Hao [13] to strengthen bridge piers that were vulnerable to to improve the computational efficiency.
vehicle or vessel impacts. These studies pointed out that steel jackets In terms of the T-girder, the elastic material model was used to si-
and FRP wrapping were effective in reducing damages of bridge piers mulate the reinforced concrete. Similarly, the elastic material model
during vehicle or vessel impacts. However, primary disadvantages of was also employed for the abutments, the footings and the bent caps.
steel jackets are their susceptibility to corrosion and the possibility of This means that these members were assumed to be low damage under
metal-to-metal contact with vehicle carrying flammable cargo. With vehicle impact loading for simplification. Although moderate damages
regard to FRP wrapping, adhesive epoxy resin is usually used to secure were observed on bent caps in the small number of vehicle-collision
the FRP well bonded to a column. But, some disadvantages such as the accidents, dominated damages (or even collapse) usually occurred in
durability of epoxy resin adhesive limit its application into engineering pier columns for most accidents. This was a reason why only single
practices. To overcome the limitations of a FRP sheet jacket (e.g., the column models were often adopted in much research [11,12,21] to
slackness in the confinement), Choi et al. [14,15] proposed FRP wire investigate vehicle-impact resistances. Similar to Fan et al. [22], the
winding jackets and applied them into seismic retrofit of circular RC two-parametric material model (∗MAT_MOONEY-RIVLIN_RUBBER) in
columns. The need is still evident for developing an alternative measure LS-DYNA was used to describe the behavior of the rubber bearing.
to improve the impact resistance of a bridge column. Ultra-high per- Two different bridge models with various pier columns were de-
formance fiber-reinforced concrete (UHPFRC) developed in the mid- veloped to compare the vehicle-impact resistant performance of con-
1990’s has been experimentally demonstrated to have many excellent ventional RC columns with UHPFRC-strengthened columns. One bridge
properties such as superior strength (e.g., compressive strength over model only had the conventional RC pier columns that are shown in
150 MPa and tensile strength over 8 MPa), durability and energy ab- Fig. 2(b). An UHPFRC-strengthened column was used to replace the RC
sorption capacity [16–18]. More importantly, UHPFRC materials have column in the other bridge model, as shown in Fig. 2(a) and (c).
been regarded as one of promising engineering materials to resist severe Compared with the conventional RC column, a 100-mm thickness
loading conditions such as impact and shock loadings [19,20]. An al- UHPFRC jacket was arranged to improve the vehicle-impact resistance
ternative measure can be developed when UHPFRC jackets are used in and a corrugated steel duct (thickness = 1.6 mm) was used to improve
multiple column bents. Owing to the excellent properties of UHPFRC, the bond between the outer UHPFRC layer and the inner normal con-
the potential of UHPFRC-strengthened columns is worth exploration. crete core. Nguyen et al. [23] experimentally investigated seismic
In this paper, high-resolution finite element (FE) models were de- performance of the similar column and pointed out that the presence of
veloped to investigate the performance of UHPFRC-strengthened pier hybrid fiber-reinforced concrete (HyFRC) tube could improve seismic
columns subjected to vehicle impacts. A typical three-span overpass resistances and facilitate the application of accelerated bridge con-
bridge with multiple column bents was finely modelled, including su- struction in engineering practice. Hence, the impact resistance of an
perstructure, bearing, column bent, abutment and so on. To ensure the UHPFRC-strengthened column is definitely worth further investigation.
rationality of FE models, material models regarding UHPFRC and All pier columns including the concrete and the reinforcements were
normal concrete as well as the vehicle model were carefully discussed modelled using elastic-plastic material models to reasonably capture
and validated against experimental data. To improve computational their damages under vehicle impact loading. Similar to Fan and Yuan
efficiency, three different simplified models were examined to address [24] and Fan et al. [22], the elastic-plastic model
their accuracy and applicability. The impact resistance of the UHPFRC- (∗MAT_PLASTIC_KINEMATIC) that is able to model isotropic and ki-
strengthened pier column was demonstrated to be considerably su- nematic hardening plasticity with strain-rate effect was employed to
perior to those of the RC pier column. A detailed parametric study was simulate the longitudinal and transverse reinforcements embedded in
performed based on response surface methodology to quantify the in- the column concrete. Since the available reinforcing bars have enough
fluences of reinforcement ratios, thickness of UHPFRC jacket, UHPFRC long to avoid the unnecessary connections of longitudinal reinforce-
strength and impact speed on impact-resistant performances. ments in the 5-meter high columns, the effect of reinforcement con-
nections was not considered in this study. The material parameters used
2. Bridge and vehicle FE models and verifications in the bridge are presented in Table 1, which were mainly determined
based on the recommendations from China’s design specification [25].
Many bridge structures with multiple column bents were damaged With regard to normal concrete and UHPFRC modeling, the detailed
due to vehicle collisions [4]. For this reason, a typical three-span discussions are presented in the next section because they play an im-
overpass bridge was selected in this study, as illustrated in Fig. 1. This portant role in simulating vehicle collisions with bridge structures.
bridge is similar to the objects of Liu [10] and Xu [3], but cross sections
of the columns are different from each other. As mentioned earlier, 2.2. Concrete and UHPFRC material models and validations
most of the damaged columns in vehicle-collision accidents were cir-
cular. Thus, this study placed an emphasis on circular columns. The previous study [26] performed low-velocity impact tests on

252
W. Fan et al. Engineering Structures 173 (2018) 251–268

3×20=60
Guardrail

Girder
Deck Bent cap Deck
Abutment
Footing Pier column
(a) Elevation of bridge structure
8.0 #20 steel bar
Bent cap

1
#16 steel bar

0.6
Pier

5.0
column (c) Cross section of pier column
8.0

0.3
Footing
1

0.5 3.5 3.5 0.5 Stringers Diaphragms

8.0 1.6 1.6

(b) Layout of multiple column bents (d) Cross section of bridge girder
Fig. 1. Overpass bridge and layout (Unit: m).

axially-loaded RC columns (Fig. 3(a)), and indicated that the con- [29]. To validate the modeling method, the low-velocity impact tests on
tinuous or smooth surface cap (CSC) model in LS-DYNA has potential in the axially-loaded RC columns were simulated using the FE model (e.g.,
concrete simulation. Hence, the CSC model was employed in this study as shown in Fig. 3(b)) for all the impact cases in the previous study
to simulate the normal concrete. In the CSC model, the failure surface [26]. Generally, good agreements were achieved between the FE pre-
function is based on three stress invariants and the cap hardening dictions and the experimental results. For two typical impact cases with
parameter κ , which can be written as [27] different impact energy, Fig. 3(c) and (d) present the numerically-ob-
tained damage and impact force versus midspan displacement curves
f (I1, J2′, J3′, κ ) = J2′−R2F f2 Fc (1) along with the experimental results. In addition to the impact tests, the
where I1 is the first invariant of the stress tensor J2′ and J3′ are the second ‘compression after impact (CAI)’ tests on impact-damaged columns (see
and third invariants of the deviatoric stress tensor, respectively; R is the Fig. 3(e)) [30] were also used for verifications. As shown in Fig. 3(f),
Rubin three-invariant reduction factor; Ff is the shear failure surface, the numerically-obtained residual strengths were found to have a rea-
which is defined in the model along the compression meridian; and Fc is sonable agreement with the experimental data presented in [30], in-
the hardening cap, which is a two-part function that is either unity or an dicating the applicability of the FE modeling method of RC columns.
ellipse. The input parameters (e.g., α , β , λ and θ ) of the CSC model Certainly, some differences between the FE simulations and the impact
listed in Table 2 are relevant to determine the failure surface. tests can be observed, such as the residual displacement at the low-
Importantly, Liu et al. [26] pointed out that the CSC model with energy impact event. This is due to the influence of both the bond-slip
default parameters cannot reasonably capture the confinement effect and crack closure behaviors on the impact-induced responses. For the
induced by transverse reinforcements and axial loads. According to the tested RC columns, the improved FE modeling was developed by Liu
model theory [27,28], the softening behavior in compression depends et al. [26] to consider the influence of the bond-slip and crack closure
on the ductile damage parameter d that is defined as: behaviors. However, it is complicated to extend the modeling technique
into vehicle-bridge collision simulation because the tensile and com-
d max ⎡ 1+B
d (τd ) = −1⎤ pressive zones cannot readily be predicted for various collision sce-
B ⎢ ⎣ 1 + B exp
−A (τd − r 0d ) ⎥
⎦ (2) narios. Moreover, residual axial-carrying capacities of bridge columns
after a collision are usually conservative when the residual displace-
where dmax is the maximum damage allowed to accumulate; τd is the
ment tends to be larger, as shown in Fig. 3(f). For these reasons, the
instantaneous strain energy-type term for damage accumulation; r0d is
improvements on the bond-slip and crack closure behaviors of the CSC
the ductile damage threshold; B is the ductile softening parameter; A is
model, as suggested by Liu et al. [26], were not included here.
the compressive softening parameter, as determined by
In this study, the normal concrete with a strength grade of C40 was
A = A (d max + 0.001) pmod (3) used in bridge columns. This means that the compressive strength ( fcu )
tested from a 150 × 150 × 150 mm concrete cube is 40 MPa in China.
Compression simulations were iteratively conducted using a
Usually, the uniaxial compressive strength fc′ of a cylinder concrete is
150 × 300 mm cylinder model to determine the material parameters B
equal to 0.79 times fcu [26,31]. In addition, China’s design specification
and pmod for the confined concrete [26]. As evidenced in Fig. 4, the
[25] prescribed that a scale factor of 0.88 should be considered for the
stress-strain relationship obtained using the adjusted parameters are in
compressive strengths between the structural concrete and the tested
good agreement with the theoretical results suggested by Mander et al.

253
W. Fan et al. Engineering Structures 173 (2018) 251–268

RC column

UHPFRC-
strengthened column
(Impacted column)

(a) Entire model of overpass bridge

5×0.15 m

0.6 m
0.6 m
Cross section Cross section
Normal
5.0 m

5.0 m
concrete

6×0.15 m
Confined
concrete

UHPFRC
Concrete
cover

RC column Reinforcements Details Strengthened column Corrugated duct Details

(b) FE model of RC column (c) FE model of UHPFRC-strengthened column
Fig. 2. High-resolution FE model of overpass bridge.

small-size cube sample. Accordingly, fc′ = 40 × 0.79 × 0.88 = 27.8 MPa, proposed UHPFRC-strengthened column was not overestimated.
which was used to define the uniaxial compressive strength in the CSC Since the CSC model behaves very well in modeling concrete, the
model. For the concrete cover in the bridge columns, the default values previous study [20] extended it into simulating UHPFRC based on the
suggested by the CSC model [27,28] were employed for the given com- constitutive model theory and the UHPFRC properties. Specifically, the
pressive strength. The corresponding stress-strain curve is presented in equations to determine the bulk and shear moduli were presented based
Fig. 4. For the confined concrete used in bridge columns, only the material on the fitting equations given by Graybeal [33]. Yield surface equations
parameters B and pmod related to the confinement effect were calibrated including triaxial compression surface (TXC), triaxial extension (TXE)
based on the theoretical model, as illustrated in Fig. 4. Because these two and torsion surface (TOR) were developed through typical stress states
parameters need to be adjusted for the confined concrete, the keyword of UHPFRC [20]. To conveniently apply the developed UHPFRC model
∗MAT_CSCM with many input parameters as listed in Table 2 had to be at any given strength, quadratic equations of the yield surface para-
used for the confined concrete. meters were fit as a function of unconfined compression strength in
With regard to the proposed UHPFRC-strengthened column, the [20]. For example, the TXC parameters (e.g., α , β , λ and θ ) are ex-
core concrete shown in Fig. 2(c) was modeled as normal and as con- pressed by [20]
fined for comparisons, respectively. Different treatments yielded quite
similar impact responses because the influence of the UHPFRC jacket α = −2.381 × 10−5fc′ 2 + 8.064 × 10−1fc′ + 2.178 × 101 (MPa) (4)
was dominate. In addition, the UHPFRC jacket and the core concrete
need to be casted separately in the construction. The debonding se- λ = 8.333 × 10−5fc′ 2 + 7.168 × 10−1fc′ + 1.880 × 101 (MPa) (5)
paration might occur like concrete-filled steel tube column [32] during
long-term use, which may result in decreasing the confinement beha- β = 5.381 × 10−8fc′ 2 −3.187 × 10−5fc′ + 6.926 × 10−3 (MPa−1) (6)
vior of the core concrete. Hence, the core concrete shown in Fig. 2(c)
was modeled as normal in this study so that the impact resistance of the θ = −2.381 × 10−7fc′ 2 + 1.357 × 10−4fc′ + 1.306 × 10−1 (7)

254
W. Fan et al. Engineering Structures 173 (2018) 251–268

Table 1
Material parameters of the overpass bridge.
Member Material model Parameter Magnitude

Bridge girder MAT_ELASTIC (#01) Mass density 2.40 × 103 kg/m3

Poisson’s ratio 0.20
Young’s modulus 3.45 × 104 MPa

Rubber bearings MAT_MOONEY-RIVLIN_RUBBER(#27) Mass density 1150 kg/m3

Poisson’s ratio 0.4997
Parameter A 0.43 MPa
Parameter B 0.076 MPa

Longitudinal reinforcement MAT_PLASTIC_KINEMATIC(#3) Mass density 7.85 × 103 kg/m3

Young’s modulus 2.06 × 105 MPa
Poisson’s ratio 0.30
Yield stress 400 MPa
Failure strain 0.2
Strain rate parameter C 7.274 × 107
Strain rate parameter P 11.22

Transverse reinforcement MAT_PLASTIC_KINEMATIC(#3) Mass density 7.85 × 103 kg/m3

Young’s modulus 2.06 × 105 MPa
Poisson’s ratio 0.30
Yield stress 300 MPa
Failure strain 0.25
Strain rate parameter C 7.274 × 107
Strain rate parameter P 11.22

Corrugated steel duct MAT_PLASTIC_KINEMATIC(#3) Mass density 7.85 × 103 kg/m3

Young’s modulus 2.06 × 105 MPa
Poisson’s ratio 0.30
Yield stress 320 MPa
Failure strain 0.17
Strain rate parameter C 7.274 × 107
Strain rate parameter P 11.22

where fc′ is the uniaxial compressive strength of UHPFRC. The cap reported in [21] (Fig. 6(d)). The numerically-obtained peak impact
parameters were determined by fitting pressure-volumetric strain force of 1256 kN was in agreement with that (1170 kN) measured from
curves measured in hydrostatic compression and uniaxial strain tests. In the collision test. Therefore, the rationality of the SUT model was
addition, Guo et al. [20] modified the strain-rate parameters in tension confirmed again in this study.
and compression by means of the fitting formulas in [34,35], which
were developed based on the experimental data. The developed con-
stitutive model was shown to be capable of achieving the desired me- 3. Vehicle-bridge collision FE model and simplification
chanical behaviors such as strength and strain-hardening behavior [20].
More importantly, the impact-induced responses obtained from the High-resolution FE models for vehicle-bridge collision simulations
extended CSC model with the parameters listed in Table 2 matched well were generated by defining the nonlinear contact between the afore-
with the impact test data, as shown in Fig. 5. Hence, the extended CSC mentioned bridge and vehicle models, as shown in Fig. 7(a). To guar-
model developed by Guo et al. [20] was employed in FE simulations of antee the accuracy and efficiency of collision simulations, some mod-
UHPFRC-strengthened bridge columns subjected to vehicle collisions. eling issues such as mesh convergence, initialized gravity loading and
Additional details on the development and validations of the CSC model model simplification were discussed in this section.
for UHPFRC modeling can be found in Guo et al. [20].

3.1. Mesh convergence test

2.3. Vehicle FE model and verification
It is critical to select mesh size in FE simulations for achieving a fine
Fig. 6(a) shows the FE model (35, 400 elements) of a Ford F800
balance between computational accuracy and efficiency. First, mesh
Single Unit Truck (SUT) developed by the FHWA/NHTSA National
size need be sufficiently small to yield convergent FE results. On the
Crash Analysis Center (NCAC) at the George Washington University
other hand, it cannot be excessively small due to the constraints of
[36]. The SUT model has been widely used to investigate the infra-
computational time and resources. For this reason, the influence of the
structure performances under vehicle collisions [9,36,37]. Similar to
mesh size of the impacted column was investigated. Three different
these studies, the SUT model was employed as the aberrant vehicle in
mesh sizes (i.e., 25 mm, 35 mm and 50 mm) were used to perform mesh
this study. The truck is 8.5 m long with a wheel base of 5.3 m, and its
convergence tests. For different mesh sizes, the column displacements
total mass is 8 Ton, including the 0.64-Ton engine and the cargo. For
at 1.2 m high above the ground are presented in Fig. 8 when being
simplicity, elastic material models with various modulus were used to
subjected to the vehicle collision with an initial impact speed (v0) of
model the engine and the cargo.
60 km/h. The FE model with 50-mm mesh provided the underestimated
Although the SUT model had been validated in current studies [36],
results compared to those with the other mesh sizes, whereas the FE
the double check was still performed using the collision test [21] to
model with 35-mm mesh yielded similar results to using 25-mm mesh.
ensure the rationality of the vehicle FE model prior to vehicle-bridge
Hence, to keep the balance between computational accuracy and effi-
collision simulations. As shown in Fig. 6(b) and (c), the damage pattern
ciency, 35-mm mesh was adopted to model the impacted bridge column
predicted by the FE simulation was similar to the collision test. The
in this study.
impact-induced displacements of the bollard obtained from the FE si-
mulation were approximate to the test results and the numerical results

255
W. Fan et al. Engineering Structures 173 (2018) 251–268

Sliding bearing
Guide rod for
Confined Concrete
drop hammer cover Drop
concrete
hammer
Column
A-A A

Cable elements for

High-strength steel axial pre-load
bars for axial pre-load Sliding bearing

Displacement gauge

Steel supports

(a) Impact test setup [26] (b) FE model for impact simulations
700
Test FEA
600
Low-energy impact

Impact force (KN)

500
400 Low-energy impact
300
200
100
0
-100
0 5 10 15 20 25
1000
Midspan displacement (mm)
800
Impact force (KN)
High-energy impact

600

400 High-energy impact

200

0 10 20 30 40 50 60
Midspan displacement (mm)

(c) Impact-induced damage distributions (d) Impact force versus mid-span displacement
Pu_FEA/Pu_EXP= 991/1279 = 0.775
Low-energy impact

Load cell Damaged

Axial loading
column

Pu_FEA/Pu_EXP= 780/853 = 0.914

High-energy impact

Data acquisition system

(e) “Compression after impact” test [30] (f) Comparisons of residual strengths
Fig. 3. Low-velocity impact test on axially-loaded RC column and validation of concrete material model.

3.2. Initialization of permanent load axial loads play a negative role (often called P-delta effect) when the
deformation of the bridge column is very large during a high-energy
Axial forces induced by superstructure self-weight have a great in- impact. Thus, all vehicle-collision FE simulations should incorporate
fluence on the impact resistance of a bridge column, as evidenced by self-weight loads of structural members due to gravity into the structure
the impact tests on axially-loaded RC columns [26]. Axial loads usually as part of an impact analysis.
improve the impact resistance of a bridge column when the impact- Although the application of permanent loads is straightforward for a
induced deformation of the column is relatively small. On the contrary, static analysis, careful considerations need to be given for the

256
W. Fan et al. Engineering Structures 173 (2018) 251–268

Table 2 40 CSC model with

adjusted parameters
Material parameters of UHPFRC and NC used in the CSC model (Unit: MPa,
mm, sec). Mander model
CSC model with
Parameter Meaning Magnitude 30 default value
for confined concrete

Stress (MPa)
NC UHPFRC Confined concrete
Loading
RO (Mg/mm3) Mass density 2.4E-9 2.6E-9 20
Stirrup
NH Hardening initiation 1 1

300 mm
CH Hardening rate 0 0
G (MPa) Shear modulus 1.117E+04 1.958E+04 CSC model for
K (MPa) Bulk modulus 1.224E+04 2.611E+04 10 concrete cover
α (MPa) TXC surface constant term 1.419E+01 1.422E+02 (default value)
θ TXC surface linear term 2.897E-01 1.456E-01
λ (MPa) TXC surface nonlinear term 1.051E+01 1.282E+02 150 mm
β (MPa−1) TXC surface exponent 1.929E-02 3.358E-03 0
α1 TOR surface constant term 0.7473 1 0.000 0.005 0.010 0.015 0.020
θ1 (MPa−1) TOR surface linear term 1.206E-03 0 Strain (mm/mm)
λ1 TOR surface nonlinear term 0.1700 0.4226
β1 (MPa−1) TOR surface exponent 7.260E-02 1.148E-03 Fig. 4. Stress-strain relationships of unconfined and confined concrete using
α2 TXE surface constant term 0.6600 1 CSC model.
θ2 (MPa−1) TXE surface linear term 1.453E-03 0
λ2 TXE surface nonlinear term 0.1600 0.5
β2 (MPa−1) TXE surface exponent 7.260E-02 1.148E-3 weighted damping method was used in this study. With mass propor-
R Cap aspect ratio 5 2.65 tional system damping, the acceleration is computed as [28]:
X0 (MPa) Cap initial location 8.959E+01 3.791E+02
n
W Maximum plastic volume 0.0500 4.257E-03 α n = M−1 (P n−F n−Fdamp ) (8)
compaction
D1 (MPa−1) Linear shape parameter 2.500E-04 2.825E-10 where M is the diagonal mass matrix; Pn
is the external load vector; F n
D2 (MPa−2) Quadratic shape parameter 3.492E-07 3.352E-06 n
is the internal load vector; and Fdamp is the force vector due to system
B Ductile shape softening 4.000E+02 1.000E+02
damping, which can be defined as:
parameter
GFC (MPa * mm) Fracture energy in uniaxial stress 5.382 1.777E+01 n
Fdamp = Ds mv (9)
D Brittle shape softening parameter 0.1000 5.000E+03
GFT (MPa * mm) Fracture energy in uniaxial 5.382E-02 4.760
where Ds is the damping constant. Generally, the damping constant for
tension
GFS (MPa * mm) Fracture energy in pure shear 5.382E-02 4.760 the system to reach static equilibrium is based on the critical damping
stress factor for the lowest frequency mode of interest [39]:
pwrc Shear-to-compression transition 5.0 5.0
parameter Ds = 2ωmin (10)
pwrt Shear-to-tension transition 1.0 1.0
parameter
where ωmin is the lowest natural circular frequency of a structural
pmod Modify moderate pressure 2.9 0.0 system. The natural frequencies of the bridge structure can be derived
softening parameter from the fast Fourier transform (FFT) analysis of the axial force time
η0c Rate effects parameter for 1.000E-04 9.928E-04 history with no damping. For the axial force shown in Fig. 9(a), the
uniaxial compressive stress
natural frequencies (ω1, ω2 ) corresponding to the first and second vi-
Nc Rate effects power for uniaxial 0.7800 0.7817
compressive stress bration modes are 61.63 rad/s and 246.40 rad/s, respectively. Based on
η0t Rate effects parameter for 5.992E-05 1.336E-04 Eq. (10), Ds1 = 123.26 rad/s and Ds2 = 492.79 rad/s, respectively.
uniaxial tensile stress According to structural dynamics, at least the fundamental natural
Nt Rate effects power for uniaxial 0.4800 0.7087 vibration period (T1 = 0.10 s) of the bridge structure is required to
tensile stress
overc (MPa) Maximum overstress allowed in 2.065E+01 7.500E+01
reach the static equilibrium when the damping is applied with the
compression gravity loading together at beginning of the FE simulation, as shown in
overt (MPa) Maximum overstress allowed in 2.065E+01 1.587E+01 Fig. 9(b). From Fig. 9(a), the static equilibrium cannot be met when the
tension duration of applying the mass-weighted damping (e.g., 0.07 s) is shorter
Srate Ratio of effective shear stress to 1.0 1.0
than the fundamental period (T1). Because high-resolution FE collision
tensile stress fluidity parameters
repow Power that increases fracture 1.0 0.5 simulations need supercomputing resources and tens of hours of com-
energy with rate effects puting time, it is of great importance to quickly reach the self-weight
IRATE Rate effect options 0.0 1 equilibrium. To this end, an alternative method based on mass-
weighted damping was proposed in this study to shorten the time re-
quired for static equilibrium. The damping change was introduced to
application of non-transient loads during a dynamic analysis [38]. As implement the transitions between different vibration systems (e.g.,
shown in Fig. 9(a), axial forces in the pier columns were amplified undamped, underdamped, overdamped systems), as illustrated in
(almost two times the desired static force) due to the abrupt application Fig. 9(b). For clear, single degree of freedom (SDOF) systems were used
of self-weight loading in a transient dynamic analysis. To properly in- to interpret the essence of the proposed method. For SDOF systems with
corporate permanent loads into a transient dynamic analysis, LSTC [28] different amount of damping, the dynamic displacements can be solved
provided some alternative methods such as mass-weighted damping as follows:
method, dynamic relaxation method and implicit-static analysis
method. The dynamic relaxation method usually requires lots of com- v0
u (t ) = u 0cosωn t + sinωn t when Ds = 0 (undamped)
putational time to reach static equilibrium, particularly when a struc- ωn (11)
tural system has a great number of nonlinear solid elements. For the
v0 + ζωn u 0
implicit-static analysis method, the types of material models and ele- u (t ) = e−ζωn t ⎛u 0cosωd t +
⎜ sinωd t ⎞ when Ds < 2ωn
⎟

ments needed to be manually changed when the implicit-static analysis ⎝ ωd ⎠

is altered to the explicit dynamic analysis. For these reasons, the mass- (underdamped) (12)

257
W. Fan et al. Engineering Structures 173 (2018) 251–268

Test FEA
300

Impact load (kN)

Drop height = 0.8 m
Max disp .=14.1 mm
200

100
Drop height=0.8 m
Max disp .=13.6 mm
0

0.000 0.005 0.010 0.015 0.020 0.025 0.030

400 Time (s)

Impact load (kN)

Drop height = 1.6 m

Max disp .=36.1 mm 300

200
Drop height=1.6 m
100

Max disp .=36.6 mm 0

0.000 0.005 0.010 0.015 0.020 0.025 0.030

Time (s)

(a) Damage distribution (b) Impact-force time history

Fig. 5. Low-velocity impact test of UHPFRC beam and validation of UHPFRC material model [20].

u (t ) = u 0 e−ωn t + (v0 + ωn u 0) te−ωn t when Ds = 2ωn (critically-damped) Based on Eqs. (11)–(14), different schemes are presented in Fig. 10
(13) for obtaining the static equilibrium. Accordingly, the results shown in
Fig. 9(a) can be explained clearly. First of all, the time to attain the
v0 + ζωn u 0 desired displacement (static displacement) could be shortened when
u (t ) = e−ζωn t ⎛u 0coshωd t +
⎜ sinhωd t ⎞ when Ds
⎟

⎝ ωd ⎠ the damping was absent at the beginning phase. Then, the damping was
> 2ωn (overdamped) added into dissipating the kinetic energy of the system. Since the time
(14)
to reach the static equilibrium may be longed because of suddenly
where u 0 and v0 are the initial displacement and velocity for free vi- applying the critical damping or overdamping, a transition between the
bration, respectively; ωn is the natural circular frequency of vibration; ζ underdamped system and the critically-damped (or overdamped)
is the damping ratio; ωd = ωn |ζ 2−1| . system was arranged. In fact, the fastest system was undamped for

(a) FE model (b) Vehicle collision test [21]

50
Test
KCC model
40
CSC model
Displacement (mm)

30

20

10

0
0.00 0.05 0.10 0.15 0.20 0.25
Time (s)
(c) Damage extent (d) Displacement at top of the bollard
Fig. 6. Vehicle FE model and validation.

258
W. Fan et al. Engineering Structures 173 (2018) 251–268

Fig. 7. (a) High-resolution FE model for vehicle collision with bridge; (b) global damage of bridge.

30 To explore the influence of self-weight equilibrium prior to collision

25-mm mesh analysis, dynamic responses of the RC column under a vehicle collision
35-mm mesh are shown in Fig. 11. The self-weight equilibrium has little effect on the
25 50-mm mesh
impact force. This is attributed to the fact that the impact force is
mainly dependent on the resistance of the impacting vehicle, rather
Displacement (mm)

20 than the impacted structure. However, the self-weight equilibrium

played an important role in the prediction of the impact-induced dis-
15 placement, which highlights the importance of reaching the self-weight
equilibrium before collision simulations.

10
3.3. Simplified FE model and discussion

5 To improve the computational efficiency of vehicle-bridge collision

simulations, simplified FE models were investigated by comparing with
0 the developed high-resolution model. Fig. 7(b) presents the damage dis-
0.00 0.05 0.10 0.15 0.20 tributions in the bridge structure subjected to a relatively high-energy
Time (s) vehicle impact. Apparently, the damage was only localized at the impacted
column. For this reason, several simplified models such as single-column
Fig. 8. Mesh convergence test (impact speed = 60 km/h). model and pier-bent model had been used to simulate vehicle collisions
with bridge columns. However, the applicability of these simplified
attaining the desired displacement, e.g., the time t0 in Fig. 10. For models is not clarified yet, particularly for single-column model. There-
Fig. 9(a), t0 = 0.034 s and the time to attain the static equilibrium fore, three typical simplified FE models used in current studies were in-
(1.4t 0 = 0.05 s) was consistent with that estimated from the SDOF vestigated, including: (1) single-column model with a simple support at
system shown in Fig. 10. Also, Fig. 10 presents the reason why the column top (denoted by Model-A, Fig. 12(a)); (2) single-column model
overdamping is required for static equilibrium. The damping curve with a roller boundary (Model-B, Fig. 12(b)); (3) pier-bent model with an
defined in Fig. 9(b) is not the sole solution, but one of the best solutions. equivalent superstructure mass (Model-C, Fig. 12(c)).

1.5 300
With no damping (a) (b)
Before collision During collision
1.2
Mass damping (s )

200 Conventional method

-1
Axial force (MN)

0.9 Ds=123 rad/s

Desired result
Proposed method
0.6 (Ds = 180 rad/s)
Ds=180 rad/s 100
Ds = 123 rad/s
0.3
Conventional method
0.0 0

0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20
Time (s) Time (s)
Fig. 9. Permanent load initialization for vehicle-impact analysis: (a) Axial force in the impacted column; (b) Damping curve for permanent load initialization.

259
W. Fan et al. Engineering Structures 173 (2018) 251–268

Fig. 10. Different schemes for achieving static equilibrium.

4 50
(a) (b)
Before collision During collision
Before collision During collision 40
3
Impact force (MN)

Displacement (mm)
With no damping
30
With no damping
2
20
With mass damping
1
10
With mass damping
0 0
0.00 0.05 0.10 0.15 0.20 0.25 0.00 0.05 0.10 0.15 0.20 0.25
Time (s) Time (s)
Fig. 11. Influence of self-weight static equilibrium on impact-induced response.

These simplified FE models along with the high-resolution model models yielded similar results when the impact-induced displacement
were used to simulate several vehicle-collision scenarios. Fig. 13 pre- was relatively small (i.e., v0 = 40 km/h). For another collision events,
sents the impact-induced displacements of the RC and UHPFRC- however, the displacements obtained from both two single-column
strengthened columns obtained from these models. Self-weight analysis models were greatly larger than those determined from the high-re-
prior to a collision analysis was included in FE simulations, but the time solution model. This is partially attributed to the fact that the ideal
at the beginning of a collision was set to zero for convenience. All boundary may trigger the heavier damage around the column bottom

Axial load Axial load

Equivalent mass

Simple Roller
support Roller
support
support
5.0 m

5.0m

Fixed boundary

(a) (b) (c)

Fig. 12. Simplified FE models: (a) Model-A; (b) Model-B; (c) Model-C.

260
W. Fan et al. Engineering Structures 173 (2018) 251–268

12 12
(a) NC, 40 km/h (b) UHPFRC, 40 km/h

Displacement (mm)

Displacement (mm)
9 9

6 6

3 3

0 0
0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20
Time (s) Time (s)
50 50
(c) NC, 60 km/h (d) UHPFRC, 60 km/h
Displacement (mm)

Displacement (mm)
40 40

30 30

20 20

10 10

0 0
0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20
Time (s) Time (s)
200 200
(e) NC, 80 km/h (f) UHPFRC, 80 km/h
Displacement (mm)

Displacement (mm)

150 150

100 100

50 50

0 0
0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20
Time (s) Time (s)
400 400
Displacement (mm)

(g) NC, 100 km/h (h) UHPFRC, 100 km/h

Displacement (mm)

300 300

200 200

100 100

0 0
0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20
Time (s) Time (s)
Whole bridge Model-A Model-B Model-C
Fig. 13. Comparison of impact-induced displacement using different simplified FE models.

and top during a relatively high energy impact. In contrast, good RC columns for each collision scenario (i.e., v0 = 20–100 km/h).
agreements were found between the pier-bent model and the high-re- In general, the damage extents in the RC columns were greatly se-
solution model. Similar observations can be found in Liu [10]. Although verer than those of the UHPFRC-strengthened columns for the same
the computational efficiency of the pier-bent model was lower than that collision events. Specifically speaking, the slight damage occurred in
of the single-column model, it was considerably greater than that of the the RC column only when the initial impact speed was just 20 km/h,
high-resolution model. Therefore, the validated pier-bent model was whereas the UHPFRC-strengthened column exhibited the obvious da-
employed in the following comparative and parametric studies. mage when the impact speed exceeded 60 km/h. The severe damage
and the large displacement were observed for the RC column at the
impact speed of 70 km/h. On the contrary, only the slight or moderate
4. Comparison between UHPFRC-strengthen column and RC
damage exhibited in the UHPFRC-strengthened column. Undoubtedly,
column
the impact resistance of a bridge column was greatly improved due to
the presence of the UHPFRC jacket.
Using the simplified pier-bent model, the vehicle-impact-resistant
In addition, the peak impact forces and displacements of both the
performances of RC columns and UHPFRC-strengthened columns were
UHPFRC-strengthened columns and the conventional RC columns were
widely investigated for different collision scenarios. Fig. 14 compares
summarized in Fig. 15 for each collision scenarios. As shown in
the damages of UHPFRC-strengthened columns with the conventional

261
W. Fan et al. Engineering Structures 173 (2018) 251–268

V0=20 km/h V0=40 km/h V0=50 km/h V0=60 km/h V0=70 km/h V0=80 km/h V0=100 km/h

(a) Conventional pier

(b) UHPFRC-strengthened pier

Fig. 14. Comparison of impact-induced damage between RC column and UHPFRC-strengthened column.

Fig. 15(a), peak impact forces increased with increasing impact speeds were increased due to the presence of UHPFRC jacket. The nearly linear
for both RC columns and UHPFRC-strengthened columns. For a given decline in residual capacity with impact speed was observed for RC
collision event, the impact force of the UHPFRC-strengthened column columns. However, some step (nonlinear) downs were observed in the
was comparable to that of the RC column, as shown in Fig. 15(b). This is residual capacities of the UHPFRC-strengthened columns.
attributed to the fact that the impact force is mainly dependent upon According to [40], a damage index (D) based on residual strength is
the stiffness of the impacting vehicle, and the stiffness of the impacted defined as follows:
column plays a minor role. This means that vehicle damage and injury
are not observably worsened due to the addition of UHPFRC-strength- D = 1−Presi/ Pm (15)
ened jacket. For the impact-induced displacements (Fig. 15(c) and (d)),
the observations were consistent with the damage severity shown in where Presi is the residual axial load-carrying capacity of the impact-
Fig. 14. The impact-induced displacements were dramatically reduced damaged bridge column; Pm is the ultimate axial capacity of the un-
(at least 50% drops) due to the presence of the UHPFRC jacket. These damaged column. Using Eq. (15), the values of D are shown in
results demonstrate the effectiveness of applying UHPFRC to improve Fig. 16(b) for various collision event scenarios. Generally, the D values
the impact resistance of bridge structures with a multi-column bent. of the UHPFRC-strengthened column were often but not always smaller
Because bridge columns must support substantial axial loads, it is than those of the RC column. Accordingly, the evaluations based on the
important to evaluate the residual axial load-carrying capacity, the damage index may lead to different conclusions from those drawn from
damage severity and the risk of collapse when being subjected to lateral Fig. 16(a). It should be noted that UHPFRC-strengthened columns
vehicle impact loads [30]. Hence, such axial performance character- possess larger ultimate axial capacity (Pm) than RC columns. At impact
istics of bridge columns were quantitatively examined using ‘com- speed of 80 km/h, two types of bridge columns yielded the same D
pression after impact’ simulations. Residual capacities of the RC col- values. However, the residual capacity of the UHPFRC-strengthened
umns and the UHPFRC-strengthened columns after vehicle impacts are column after collision was approximated to the ultimate axial capacity
presented in Fig. 16(a) for each impact event. The residual axial load- of the undamaged RC column. Apparently, using damage index D to
carrying capacities of the UHPFRC-strengthened columns were almost compare impact-resistant performance is meaningful when the same
two times those of the conventional RC columns, because impact-in- ultimate axial capacity and axial load ratios are arranged between the
duced displacements were reduced and axial load-carrying capacities UHPFRC-strengthened column and the conventional RC column. Since
UHPFRC materials have a very high compressive strength, the sectional

262
W. Fan et al. Engineering Structures 173 (2018) 251–268

10 150
NC NC
(a) UHPFRC (b) UHPFRC

Ratio of impact force (%)

97
8 14% 15%
10% 9%

7.
6% 6%

91
Impact force (MN)
0%

6.
100
6

84
11
4.
5.
08
73
4

4.
3.
50

94
58
2.
2.
75

98
8
2
85
1.
1.

1.
1.
84
84
0.
0.

0 0
v=20 v=40 v=50 v=60 v=70 v=80 v=100 v=20 v=40 v=50 v=60 v=70 v=80 v=100
Velocity (km/h) Velocity (km/h)
120
NC NC

6
(c) (d)

.9
250 UHPFRC

3
UHPFRC

24
100

Ratio of displacement (%)

Displacement (mm)

200
80
50% 58% 52% 56%
69% 59% 57%
150
60
5
.1
6
10

100
40
1
.3
76

50
2

20
.4

9
.0
39
3

33
1
.0

.0
5. 3
24
.4

19
91
29

17
12
91
61
31

9.
9.

0 0
2.
2.
1.

v=20 v=40 v=50 v=60 v=70 v=80 v=100 v=20 v=40 v=50 v=60 v=70 v=80 v=100
Velocity (km/h) Velocity (km/h)
Fig. 15. Peak responses for different collision scenarios: (a) impact force; (b) comparison of peak force; (c) displacement; (d) comparison of peak displacement.

dimension (e.g., diameter) of the UHPFRC-strengthened column would residual axial load-carrying capacity rather than damage index D was
be greatly decreased to achieve the same ultimate axial capacity and used for further investigating the sensitivity of design parameters.
axial load ratio as the conventional RC column. However, it seems to be
less realistic for construction and another resistance (e.g., flexural ca-
5. Sensitivity analysis using response surface
pacity) in engineering practice because the sectional dimension is too
small. On the other hand, it is hard to compare the impact-resistant
To investigate the sensitivity of impact-resistant performance of
performance when the sectional dimension of the UHPFRC-strength-
UHPFRC-strengthened column to potentially varying design para-
ened column is greatly different from that of the conventional RC
meters, a parametric study was performed using the validated simpli-
column. This study mainly focused on comparing the impact resistances
fied FE model. Five parameters were considered in this study, including
when their sectional dimensions were assumed to be the same. Hence,
thickness of UHPFRC layer (t), UHPFRC strength grade (s), longitudinal

25 1.0
(a) NC (b) NC
UHPFRC UHPFRC
20 0.8
Residual capacity (MN)

Damage index

15 0.6

10 0.4

5 0.2

0 0.0
0 20 40 60 80 100 0 20 40 60 80 100
Velocity (km/h) Velocity (km/h)
Fig. 16. (a) Residual capacity and (b) damage index after vehicle collisions.

263
W. Fan et al. Engineering Structures 173 (2018) 251–268

Table 3 Table 4
Factors and levels used in the Box–Behnken design. Box-Behnken design and residual load-carrying capacity at v0 = 60 km/h.
Independent variable Symbol Range BBD level Run t (mm) s (MPa) ρL (%) ρT (%) Residual strength (MN) Relative
error (%)
Thickness (mm) t 100–300 100, 200, 300 FE result Predicted
UHPFRC strength (MPa) s 150–210 150, 180, 210 value
Longitudinal reinforcement ratio (%) ρL 0.89–1.78 0.89, 1.34, 1.78
Transverse reinforcement ratio (%) ρT 0.4–1.2 0.4, 0.8, 1.2 1 300 180 1.34 0.4 27.77 28.01 0.87
Initial impact speed (km/h) v0 20–60 20, 40, 60 2 200 180 1.34 0.8 18.37 18.47 0.56
60–100 60, 80, 100 3 100 180 1.34 1.2 16.7 15.37 7.94
4 200 180 0.89 1.2 19.28 19.38 0.53
5 200 150 1.34 1.2 17.8 18.61 4.57
6 200 180 1.78 1.2 19.24 20.34 5.72
reinforcement ratio ( ρL ), transverse reinforcement ratio ( ρT ) and initial 7 200 150 1.34 0.4 18.98 18.00 5.16
impact speed (v0 ). Three levels were considered for all varying para- 8 100 180 0.89 0.8 15.99 17.24 7.80
meters, as listed in Table 3. If a full factorial design is employed for five 9 200 150 0.89 0.8 19.12 17.21 9.99
parameters with three levels, a total of 35 = 243 models have to be 10 300 180 1.78 0.8 31.71 30.85 2.73
11 300 180 1.34 1.2 32.98 30.92 6.23
involved. For this reason, the response surface methodology combined
12 200 180 1.34 0.8 18.63 18.47 0.84
with the Box-Behnken design (BBD) was employed to reduce the 13 100 210 1.34 0.8 18.58 18.30 1.52
number of required FE simulations and to improve the computational 14 200 180 1.34 0.8 18.52 18.47 0.25
efficiency of the sensitivity analysis. 15 300 210 1.34 0.8 35.55 35.18 1.05
16 100 180 1.78 0.8 15.87 14.19 10.61
17 200 180 1.34 0.8 18.16 18.47 1.73
5.1. Response surface method and factorial design 18 200 150 1.78 0.8 18.53 18.55 0.12
19 200 210 1.34 1.2 22.9 24.26 5.95
Response surface method (RSM) has been widely used in sensitivity 20 200 180 1.34 0.8 18.69 18.47 1.16
21 300 150 1.34 0.8 25.08 26.06 3.91
analysis and design optimization [41–43] as surrogate model to im-
22 100 180 1.34 0.4 15.93 16.90 6.10
prove computational efficiency. This method assumes that the struc- 23 300 180 0.89 0.8 25.17 27.24 8.21
tural response can be approximated by 24 200 180 1.78 0.4 18.37 18.97 3.25
25 200 180 0.89 0.4 19.77 19.37 2.03
N
∼ 26 200 210 1.78 0.8 22.23 23.06 3.72
y (x ) = ∑ aj φj (x )
27 100 150 1.34 0.8 15.21 16.28 7.04
j=1 (16) 28 200 210 1.34 0.4 23.92 23.49 1.80

where ∼
29 200 210 0.89 0.8 24.95 23.84 4.44
y (x ) is the approximation to structural response; φj (x ) is the
basis function, which is usually the polynomials; N is the number of
basis functions; aj is the coefficient of response surface function, which
Table 5
can be estimated based on the least-square method and the M sampling
Box-behnken design and residual load-carrying capacity for v0 = 20–60 km/h.
points determined from FE simulations (M > N ).
Run t (mm) s (MPa) v0 (km/ Residual strength (MN) Relative error
a = (ΦTΦ)−1 (ΦTy) (17) h) (%)
FE result Predicted
where a = [a1, a2 , ⋯, aN ]; Φ is the matrix consisting of basis functions value
evaluated at the sampling points as
1 200 180 40 22.04 21.59 2.02
(1) · ·· φN (x (1) ) ⎤ 2 200 210 20 28.78 29.36 2.03
⎡ φ1 (x )
⎢ ⎥ 3 200 210 60 22.9 23.87 4.23
⋮ ⋮ ⋮
⎢ ⎥ 4 300 180 60 29.71 30.71 3.36
(i) · ·· φN (x (i) ) ⎥
Φ = ⎢ φ1 (x ) 5 100 180 60 15.99 14.61 8.64
⎢ ⋮ ⋮ ⋮ ⎥ 6 200 180 40 21.96 21.59 2.02
⎢ (M )
⎥ 7 100 180 20 20.67 19.67 4.82
⎢ φ1 (x ) · ·· φN (x (M ) ) ⎥
⎣ ⎦ (18) 8 200 180 40 21.50 21.59 2.02
9 200 180 40 21.43 21.59 2.02
With regard to design of experiments (DOE), Box and Behnken [44] 10 200 180 40 21.04 21.59 2.02
proposed some three-level designs for fitting response surfaces. These 11 300 150 40 31.03 30.62 1.33
12 300 180 20 33.29 34.67 4.15
three-level designs were implemented by combining 2 k factorials with 13 100 210 40 21.56 21.97 1.91
incomplete block designs. Generally, the designs are very efficient in 14 300 210 40 42.59 40.63 4.61
terms of the number of required runs (e.g., the number of FE simula- 15 200 150 20 22.45 21.48 4.31
tions). Table 4–6 present the Box-Behnken designs in this study. Two 16 200 150 60 18.53 17.95 3.15
17 100 150 40 16.21 18.18 12.12
types of response surface models were considered for the sensitivity
analysis. The influences of varying parameters t, s, ρL and ρT were first
investigated when the impact speed was fixed to be constant 60 km/h.
R (t , s, ρL , ρT ) = 103.9230−0.2558t −0.7288s−5.6603ρL −15.1675ρT
Accordingly, the significant design parameters can be identified. Sub-
sequently, the influences of the identified significant parameters were + 5.9167 × 10−4ts + 3.7416 × 10−2tρL + 2.7750 × 10−2tρT
further investigated when the UHPFRC-strengthened columns were −3.9888 × 10−2sρL + 3.3333 × 10−3sρT + 1.9101ρL ρT
subjected to different energy collisions.
+ 3.5959 × 10−4t 2 + 2.0941 × 10−3s 2 + 1.5512ρL2
5.2. Response surface models and discussions + 4.5839ρT2 (19)

Based on Eqs. (16)–(18) and the sampling data shown in Table 4, The 3D plots of the response variable (i.e., residual strength) with
the response surface model of residual load-carrying capacity was es- respect to pairs of varying design parameters (other parameters are set
tablished for the UHPFRC-strengthened column when the initial impact to their center point values) are presented in Fig. 17. The influences of
speed of 60 km/h was given. varying design parameters on the impact-induced response can be

264
W. Fan et al. Engineering Structures 173 (2018) 251–268

Table 6 longitudinal and transverse reinforcement ratios. It can be concluded

Box-behnken design and residual load-carrying capacity for v0 = 60–100 km/h. from all the six 3D plots in Fig. 17 that the thickness of UHPFRC jacket
Run t (mm) s (MPa) v0 (km/ Residual strength (MN) Relative error was the most significant parameter, followed by UHPFRC strength and
h) (%) then reinforcement ratio, namely t > s > ρL ≈ ρT . Both the longitudinal
FE result Predicted and transverse reinforcement ratios had a slight influence on the re-
value sidual load-carrying capacity of UHPFRC-strengthened column when
1 200 180 80 9.69 9.12 5.89 they were in the typical ranges for structural designs. For this reason,
2 200 210 60 22.9 23.98 4.72 the changes in reinforcement ratios were ignored in developing the
3 200 210 100 9.39 9.31 0.85 response surface models that include the variable of initial impact speed
4 300 180 100 8.79 9.01 2.49 v0 to further improve the computational efficiency.
5 100 180 100 7.72 8.67 12.26
Fig. 16(a) clearly indicates that the relationships between the re-
6 200 180 80 9.25 9.12 1.42
7 100 180 60 15.99 15.77 1.36 sidual capacity and the impact speed as v0 ⩽ 60 km/h are different from
8 200 180 80 8.29 9.12 10.00 those as v0 > 60 km/h. If only one polynomial function without speed
9 200 180 80 8.78 9.12 3.86 division was used, the developed response surface model would have a
10 200 180 80 9.58 9.12 4.81
low accuracy. For this reason, similar to the treatments in [41,45], the
11 300 150 80 10.15 11.01 8.52
12 300 180 60 29.71 28.77 3.18 response surface functions (i.e., Eq. (20)) were developed piecewise to
13 100 210 80 9.34 8.48 9.24 improve the accuracy. Based on the sampling data presented in Tables 5
14 300 210 80 20.2 20.06 0.67 and 6, the response surface functions can be expressed by
15 200 150 60 18.53 18.61 0.44
16 200 150 100 7.5 6.42 14.40 R (t , s, v0)
17 100 150 80 9.13 9.27 1.50
⎧ 98.0510−0.1810t −0.8565s + 0.1428v0 + 5.1750 × 10−4ts
⎪
⎪ + 1.3750 × 10−4tv0−8.1667 × 10−4sv0 + 4.0018 × 10−4t 2
clearly identified from these 3D plots. As shown in Fig. 17(a), the re- ⎪ + 2.5019 × 10−3s 2−1.7019 × 10−3v02 (20 ⩽ v0 ⩽ 60 km/h)
sidual load-carrying capacity increased with increasing the thickness of ⎪
UHPFRC jacket and UHPFRC strength. More specifically, specifying a = 122.9215−0.0689t −0.4354s−1.5955v0 + 8.2000e−4ts
⎨
higher UHPFRC jacket thickness was more beneficial for improving the ⎪ −1.5812e−3tv0−1.0333 × 10−3sv0
residual load-carrying capacity. This is attributed to the fact that the ⎪
⎪ + 2.0297 × 10−4t 2 + 1.1747 × 10−3s 2
increase in the UHPFRC thickness not only reduces the impact-induced ⎪
+ 0.0110v02 (60 ⩽ v0 ⩽ 100 km/h)
damage, but also greatly enhances the axial loading-carry capacity. ⎩
Similarly, it is observed from Fig. 17(b) and (c) that the influence of the (20)
UHPFRC thickness was considerably greater than those of both the
Figs. 18 and 19 illustrate the 3D plots of the response variable

Fig. 17. Influences of thickness of UHPFRC layer, UHPFRC strength, longitudinal and transverse reinforcement ratios on residual strength of UHPFRC-strengthened
column (v0 = 60 km/h).

265
W. Fan et al. Engineering Structures 173 (2018) 251–268

Fig. 18. Influences of thickness of UHPFRC layer, UHPFRC strength and impact speed on residual strength of UHPFRC-strengthened column (20 ⩽ v0 ⩽ 60 km/h).

(residual capacity) with respect to pairs of varying input parameters. It obviously increased with the thickness of UHPFRC jacket.
is apparent that the response surfaces for 20 ⩽ v0 ⩽ 60 km/h were dif-
ferent from those for 60 ⩽ v0 ⩽ 100 km/h. Fig. 18(b) indicates that the 5.3. Verifications of predictions from RSM
initial impact speed had less influence on the residual capacity than the
thickness of UHPFRC jacket when the impact speed ranged from 20 km/ In addition to the above sensitivity analysis, it was deemed im-
h to 60 km/h. For 60 ⩽ v0 ⩽ 100 km/h, however, the impact speed was portant to examine the accuracy of the developed response surface
shown to have a greater influence on the residual capacity than the models. Besides the sampling points (cases) presented in Tables 4–6,
thickness of UHPFRC jacket (Fig. 19(b)). These observations can be some additional cases listed in Tables 7–9 were simulated by the FE
further confirmed by the damage change with the impact speed shown model to verify the developed response surface models. Relative error
in Fig. 14. Similar to Fig. 17(a), Fig. 18(a) and 19(a) indicate that the (RE), and the coefficient of multiple determination R2 were used to
residual capacity of the UHPFRC-strengthened column increases with evaluate the modeling accuracy.
the thickness of UHPFRC jacket and the UHPFRC strength. Compared to Tables 4–9 present the impact-induced responses predicted using
the impact speed and the thickness of UHPFRC jacket, the UHPFRC the developed responses surface model along with the corresponding
strength had a limited influence on the impact-induced response, as actual response obtained from FE simulations. Fig. 20 compares the
shown in Figs. 18(c) and 19(c). In addition, the influence of the residual resistances obtained from the response surface models with the
UHPFRC thickness on residual capacity was dependent upon initial corresponding FE results. For both two response surface models in Eqs.
impact speed, as shown in Figs. 18(b) and 19(b). The residual capacities (19) and (20), the values of R2 were very close to 1.0 and the average
increased with increasing the thickness of UHPFRC jacket when the RE values were 4.16% and 5.40%, respectively. Accordingly, the de-
initial impact speed was not very high. This is because the impact da- veloped response surface models were demonstrated as capable of
mage is slight and most of the UHPFRC at the bottom retains the high predicting the residual load-carrying capacity of UHPFRC-strengthened
axial load-carrying capacity after a vehicle impact (see Fig. 14(b)). The bridge column subjected to vehicle collisions. Therefore, in addition to
increase in the UHPFRC area due to the increase in the UHPFRC study the sensitivity of varying design parameters, the reliability and
thickness is significantly beneficial for improving the residual axial optimization of UHPFRC-strengthened bridge columns are expected to
capacity. On the contrary, such as v0 = 100 km/h, the residual capacity be efficiently performed when the validated response surface models
cannot be effectively enhanced by thickening UHPFRC jacket. In this are taken as the surrogates of the time-consuming FE simulations. These
case, sever shear damages (see Fig. 14(b)) always occurred at the works will be presented in the future due to page limitation.
bottom for all thicknesses investigated in this study, because the impact
energy was very high compared with the energy-dissipating energy of
6. Conclusions
the UHPFRC-strengthened column. Accordingly, the effective UHPFRC
area, which is good for improving residual axial capacity, cannot be
An UHPFRC-strengthened bridge column was presented in this

Fig. 19. Influences of thickness of UHPFRC layer, UHPFRC strength and impact speed on residual strength of UHPFRC-strengthened column (60 ⩽ v0 ⩽ 100 km/h).

266
W. Fan et al. Engineering Structures 173 (2018) 251–268

Table 7 study to improve the impact resistance, because the conventional RC

Sample point sets used for the RSM verification at v0 = 60 km/h. column was shown to be vulnerable to vehicle collisions. FE modeling
Run t (mm) s (MPa) ρL (%) ρT (%) Residual strength Relative error techniques were carefully discussed for reasonably simulating vehicle
(MN) (%) collisions with bridge structures with a multi-column bent, and the
impact-resistant performance of UHPFRC-strengthened bridge column
FE result RSM was investigated in detail. Based on the detailed numerical studies, the
30 100 150 1.34 0.78 14.94 16.31 9.2 following conclusions can be drawn:
31 100 170 1.34 0.78 15.96 15.26 4.4
32 100 190 1.34 0.78 17.05 15.92 6.6 (1) Numerical results indicate the importance of reaching the self-
33 100 210 1.34 0.78 18.58 18.31 1.4 weight equilibrium of a bridge structure prior to vehicle collision.
34 200 150 1.34 0.78 18.53 17.52 5.5
More importantly, to shorten the time required for reaching the
35 300 180 1.34 0.78 29.71 28.63 3.6
static equilibrium (or the stable axial force in bridge columns), a
new initialization method was proposed based on varying mass-
Table 8 weighted damping with time. Specifically, compared to the con-
Sample point sets used for the RSM verification when v0 = 20 –60 km/h. ventional damping-based method, at least 50% time can be saved
by using the proposed initialization method. The reason that the
Run t (mm) s (MPa) v0 (km/ Residual strength (MN) Relative error
h) (%)
proposed initialization method can shorten the time was clearly
FE result RSM explained based on the classical dynamics of structures.
(2) Three different simplified models, as published in current studies,
18 100 150 50 15.27 16.93 10.9 were investigated to replace the whole bridge model for efficiency
19 100 150 60 14.94 15.23 1.9
20 200 150 40 22.23 20.61 7.2
consideration. It was found that two single-column models with
21 300 150 60 29.75 28.22 5.1 different boundaries exhibited low accuracy and good agreement
22 100 170 60 15.96 14.36 10.0 was achieved between the pier-bent model considering the influ-
23 100 190 60 17.05 15.30 10.1 ence of the superstructure gravity and the high-resolution FE
24 100 210 60 18.58 18.05 2.9
model.
25 100 180 40 19.44 18.04 7.2
26 300 180 40 32.58 33.59 3.1 (3) Compared to the conventional RC column, the impact-induced
27 200 210 20 28.78 29.36 2.0 displacements and damage severity can be dramatically reduced (at
least 50% reductions in displacement) due to the presence of the
UHPFRC jacket. On the other hand, the impact force was not ob-
Table 9 viously increased when the UHPFRC-strengthened column was
Sample point sets used for the RSM verification when v0 = 60 –100 km/h. used. This is attributed to the fact the impact force is mainly de-
Run t (mm) s (MPa) v0 (km/ Residual strength (MN) Relative error pendent upon the resistance of the colliding vehicle rather than the
h) (%) collided column. Therefore, the UHPFRC-based enhancement was
FE result RSM demonstrated as being an effective measure for improving the im-
pact-resistant performance of bridge structures with a multi-column
18 100 150 60 14.94 16.31 9.2
19 100 150 70 10.56 11.76 11.4 bent.
20 100 170 60 15.96 15.78 1.1 (4) A sensitivity analysis was conducted based on the validated sim-
21 100 190 60 17.05 15.93 6.5 plified FE model and the response surface method to investigate the
22 100 210 60 18.58 16.77 9.7 influences of thickness of UHPFRC jacket, UHPFRC strength, long-
23 200 150 60 18.53 18.61 0.4
24 300 180 60 29.71 28.77 3.2
itudinal and transverse reinforcement ratios and initial impact
25 100 180 80 9.20 8.10 11.9 speed on the residual capacity of UHPFRC-strengthened column
26 200 150 80 9.76 8.40 13.9 after vehicle collisions. The thickness of UHPFRC jacket was iden-
27 200 180 100 6.32 6.81 7.2 tified as the most significant factor affecting the impact-resistant
performance when the initial impact speed was not very high.

40 50

(a) (b)
35
40 R-Squared=0.9844
R-Squared=0.9627
Average RE=4.16% Average RE=5.40%
30
Predicted (MN)

Predicted (MN)

30

25

20
20

10
15 Sampling case Samping case
Additional case Additional case

10 0
10 15 20 25 30 35 40 0 10 20 30 40 50
Actual (MN) Actual (MN)
Fig. 20. Verifications of response surface models.

267
W. Fan et al. Engineering Structures 173 (2018) 251–268

However, the residual capacity of UHPFRC-strengthened column [13] Sha Y, Hao H. Laboratory tests and numerical simulations of CFRP strengthened RC
cannot be effectively enhanced by thickening UHPFRC jacket when pier subjected to barge impact load. Int J Struct Stab Dyn 2015;15:1450037.
[14] Choi E, Jeon J-S, Cho B-S, Park K. External jacket of FRP wire for confining concrete
the initial impact speed is very high. and its advantages. Eng Struct 2013;56:555–66.
(5) Response surface models were developed to predict vehicle impact- [15] Choi E, Cho B-S, Lee S. Seismic retrofit of circular RC columns through using ten-
induced responses of an UHPFRC-strengthened column. Generally, sioned GFRP wires winding. Compos B Eng 2015;83:216–25.
[16] Russell HG, Graybeal BA. Ultra-high performance concrete: a state-of-the-art report
the predictions obtained from the developed response surface for the bridge community. Federal Highway Administration, No. FHWA-HRT-13-
models were in good agreement with the results derived from the 060; 2013.
FE simulations that were complicated and time-consuming. Hence, [17] Habel K, Viviani M, Denarié E, Brühwiler E. Development of the mechanical
properties of an Ultra-High Performance Fiber Reinforced Concrete (UHPFRC). Cem
in addition to study the sensitivity of varying design parameters, Concr Res 2006;36:1362–70.
the developed response surface models as the surrogates of FE si- [18] Yoo D-Y, Banthia N. Mechanical properties of ultra-high-performance fiber-re-
mulations can facilitate a more efficient evaluation on the relia- inforced concrete: a review. Cem Concr Compos 2016;73:267–80.
[19] Yoo D-Y, Banthia N. Mechanical and structural behaviors of ultra-high-performance
bility and optimization of UHPFRC-strengthened bridge columns
fiber-reinforced concrete subjected to impact and blast. Constr Build Mater
subjected to lateral impact loading. 2017;149:416–31.
[20] Guo W, Fan W, Shao X, Shen D, Chen B. Constitutive model of ultra-high-perfor-
Some simplifications (e.g., interface/joint behavior) were assumed mance fiber-reinforced concrete for low-velocity impact simulations. Compos Struct
2018;185:307–26.
in this study because the related test data were not available. These [21] Chen L. Research on bridge piers subjected to vehicle collisions. Hunan, China:
simplifications might have an influence on the performance of Hunan University; 2015.
UHPFRC-strengthened column under vehicle impacts. Hence, physical [22] Fan W, Yuan WC, Chen BS. Steel fender limitations and improvements for bridge
protection in ship collisions. J Bridge Eng 2015;20:06015004.
experiments should be conducted in the future to further investigate the [23] Nguyen W, Trono W, Panagiotou M, Ostertag CP. Seismic response of a rocking
performance of the proposed UHPFRC-strengthened column under im- bridge column using a precast hybrid fiber-reinforced concrete (HyFRC) tube.
pact loadings and to refine the FE modeling. Compos Struct 2017;174:252–62.
[24] Fan W, Yuan WC. Numerical simulation and analytical modeling of pile-supported
structures subjected to ship collisions including soil–structure interaction. Ocean
Acknowledgements Eng 2014;91:11–27.
[25] Ministry of Communications of China. General code for design of highway bridges
and culverts (JTG D60). Beijing: China Communications Press; 2015.
This research is supported by the Major Program of Science and [26] Liu B, Fan W, Guo W, Chen B, Liu R. Experimental investigation and improved FE
Technology of Hunan Province (Grant Number: 2017SK1010), the Key modeling of axially-loaded circular RC columns under lateral impact loading. Eng
Research and Development Project of Hunan Province (Grant Number: Struct 2017;152:619–42.
[27] Murray YD. Users manual for LS-DYNA concrete material model 159. Federal
2016GK2025), the National Natural Science Foundation of China
Highway Administration, No. FHWA-HRT-05-062; 2007.
(Grant Number: 51308202) and the National Natural Science [28] LSTC. LS-DYNA Keyword User's Manual. CA: Livermore Software Technology
Foundation of China (Grant Number: 51408208). The authors also Corporation (LSTC); 2014.
thank Dr. Isojeh Benard and Dr. Bo Chen for their constructive sug- [29] Mander JB, Priestley MJN, Park R. Theoretical stress-strain model for confined
concrete. J Struct Eng 1988;114:1804–26.
gestions to improve the quality of the article. [30] Fan W, Liu B, Consolazio G. Residual capacity of axially-loaded circular RC columns
after lateral impact. J Struct Eng; 2018. (submitted for publication).
Appendix A. Supplementary material [31] Cheng WX, Wang TC. Concrete structure design principles. 5th ed. China
Architecture & Building Press; 2011.
[32] Shiming C, Huifen Z. Numerical analysis of the axially loaded concrete filled steel
Supplementary data associated with this article can be found, in the tube columns with debonding separation at the steel-concrete interface. Steel
online version, at http://dx.doi.org/10.1016/j.engstruct.2018.06.113. Compos Struct 2012;13. 000-.
[33] Graybeal BA. Material property characterization of ultra-high performance concrete
(FHWA-HRT-06-103). McLean, United States: Federal Highway Administration;
References 2006.
[34] Fujikake K, Senga T, Ueda N, Ohno T, Katagiri M. Effects of strain rate on tensile
behavior of reactive powder concrete. J Adv Concr Technol 2006;4:79–84.
[1] Harik I, Shaaban A, Gesund H, Valli G, Wang S. United States bridge failures,
[35] Fujikake K, Uebayashi K, Ohno T, Shimoyama Y, Katagiri M. Dynamic properties of
1951–1988. J Perform Constr Facil 1990;4:272–7.
steel fiber reinforced mortar under high-rates of loadings and triaxial stress states.
[2] Wardhana K, Hadipriono FC. Analysis of recent bridge failures in the United States.
In: Proceedings of the 7th international conference on structures under shock and
J Perform Constr Facil 2003;17:144–50.
impact, Montreal, 2008 of Conference. p. 437–446.
[3] Xu X. Performance based approach for loading and design of bridge piers impacted
[36] Mohan P, Marzougui D, Kan CD. Validation of a single unit truck model for roadside
by medium weight trucks. New York: The City College of New York; 2017.
hardware impact. Int J Veh Syst Model Test 2006;2:1–15.
[4] Buth CE, Williams WF, Brackin MS, Lord D, Geedipally SR, Abu-Odeh AY. Analysis
[37] Sharma H, Hurlebaus S, Gardoni P. Performance-based response evaluation of re-
of large truck collisions with bridge piers: phase 1, report of guidelines for designing
inforced concrete columns subject to vehicle impact. Int J Impact Eng
bridge piers and abutments for vehicle collisions. Texas Transportation Institute:
2012;43:52–62.
Texas A & M University System, College Station, Texas; 2010.
[38] Consolazio GR, Davidson MT, Getter DJ. BDK75 FCN. vessel crushing and structural
[5] Fan W. Dynamic demand of bridge structures and capacity of pile-supported pro-
collapse relationships for bridge design. University of Florida, Department of Civil &
tection structures under vessel impacts: Ph.D. dissertation. Shanghai, China: Tongji
Coastal Engineering; 2010.
University; 2012. (in Chinese).
[39] Hallquist JO. LS-DYNA Theoretical Manual: Livermore Software Technology
[6] Al-Thairy H, Wang YC. Simplified FE vehicle model for assessing the vulnerability of
Corporation; 2006.
axially compressed steel columns against vehicle frontal impact. J Constr Steel Res
[40] Shi Y, Hao H, Li Z-X. Numerical derivation of pressure–impulse diagrams for pre-
2014;102:190–203.
diction of RC column damage to blast loads. Int J Impact Eng 2008;35:1213–27.
[7] El-Tawil S, Severino E, Fonseca P. Vehicle collision with bridge piers. J Bridge Eng
[41] Myers RH, Montgomery DC. Anderson-Cook CM. Response surface methodology:
2005;10:345–53.
process and product optimization using designed experiments. John Wiley & Sons;
[8] Abdelkarim OI, ElGawady MA. Performance of bridge piers under vehicle collision.
2016.
Eng Struct 2017;140:337–52.
[42] Fang JG, Sun GY, Qiu N, Kim NH, Li Q. On design optimization for structural
[9] Do TV, Pham TM, Hao H. Dynamic responses and failure modes of bridge columns
crashworthiness and its state of the art. Struct Multidiscip Optim
under vehicle collision. Eng Struct 2018;156:243–59.
2017;55:1091–119.
[10] Liu G. Behavior of bridge piers during vehicular impacts. New York, USA: The City
[43] Hou SJ, Li Q, Long SY, Yang XJ, Li W. Design optimization of regular hexagonal
University of New York; 2012.
thin-walled columns with crashworthiness criteria. Finite Elem Anal Des
[11] Sharma H, Gardoni P, Hurlebaus S. Performance-based probabilistic capacity
2007;43:555–65.
models and fragility estimates for RC columns subject to vehicle collision.
[44] Box GE, Behnken DW. Some new three level designs for the study of quantitative
Computer-Aided Civ Infrastruct Eng 2015:n/a-n/a.
variables. Technometrics 1960;2:455–75.
[12] Thilakarathna HMI, Thambiratnam DP, Dhanasekar M, Perera N. Numerical simu-
[45] Hou SJ, Liu TQ, Zhang ZD, Han X, Li Q. How does negative Poisson's ratio of foam
lation of axially loaded concrete columns under transverse impact and vulnerability
filler affect crashworthiness? Mater Des 2015;82:247–59.
assessment. Int J Impact Eng 2010;37:1100–12.

268