06fsj Ls Dyna Sfe
06fsj Ls Dyna Sfe
06fsj Ls Dyna Sfe
net/publication/316533037
CITATIONS READS
4 273
3 authors:
Guillermo Rein
Imperial College London
209 PUBLICATIONS 3,075 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Panagiotis Kotsovinos on 30 April 2017.
A R T I C L E I N F O A BS T RAC T
Keywords: Due to the complex nature of structural response in fire, computational tools are often necessary for the safe
Dynamics design of structures under fire conditions. In recent years, use of the finite element code LS-DYNA has grown
LS-DYNA considerably in research and industry for structural fire analysis, but there is no benchmarking of the code
Benchmarking available in the fire science literature for such applications. Moreover, due to the quasi-static nature of structural
Structures
response in fire, the majority of the computational structural fire studies in the literature are based on the use of
Fire
static solvers. Thus, this paper aims at benchmarking the explicit dynamic solver of LS-DYNA for structural fire
Computational
analysis against other static numerical codes and experiments. A parameter sensitivity study is carried out to
study the effects of various numerical parameters on the convergence to quasi-static solutions. Four canonical
problems that encompass a range of thermal and mechanical behaviours in fire are simulated. In addition, two
different modelling approaches of composite action between the concrete slab and the steel beams are
investigated. In general, the results confirm that when numerical parameters are carefully considered such as to
not induce excessive inertia forces in the system, explicit dynamic analyses using LS-DYNA provide good
predictions of the key variables of structural response during fire.
1. Introduction tests provide unrealistic results [5]. They do not represent real fire
conditions in the compartment and base fire resistance on the
Modern building designs and innovative architectural solutions performance of the individual elements ignoring the effects of the
pose a challenge to structural engineers. This is particularly the case for surrounding structure. Conversely, full-scale testing of real structures
structural fire engineers due to the complex interactions of modern is complex, expensive and time consuming. In addition, the limited
structural systems in fire. The performance of even generic structures number of full-scale experiments carried out worldwide (e.g.
exposed to fire is not straightforward and cannot be easily predicted. As Cardington tests [5]) has been on buildings of generic rectangular
a result, there is often the engineering need to be able to assess geometry. Thus, they cannot be generalised to predict the performance
structural behaviour under fire conditions from first principles and not of all structures, especially where more innovative irregular structural
rely on blanket prescriptive guidance. arrangements are used. As a results, designers use computational tools
The behaviour of isolated structural elements under standard fire to predict and assess the performance of complex structures under fire
conditions through furnace testing has been extensively studied over conditions.
the past decade, and can now be predicted with some degree of With increasing computational capabilities, the fire resistance
accuracy using analytical and computational means [1]. However, it assessment of various structural arrangements under different fire
has been shown in the past [1–3] that structural fire performance of scenarios is becoming more and more used in practice. However, these
isolated elements does not resemble the performance of a whole models have to be benchmarked against experimental data or known
structure. The whole structure performance in a real fire depends on solutions to make sure that they produce accurate and physically
a number of factors. They include restraint, stress redistribution, correct results. Most commonly used numerical models for structural
composite action, and continuity within the structure [4]. The involve- fire analysis include commercial general finite element analysis
ment of the many variables makes the analysis and prediction of the packages (Ansys, Abaqus) and purpose-based finite element models
fire performance of realistic structures a difficult process. Standard fire developed or extended specifically for structural fire analysis (Vulcan,
⁎
Corresponding author.
E-mail address: g.rein@imperial.ac.uk (G. Rein).
http://dx.doi.org/10.1016/j.firesaf.2017.03.002
Received 16 August 2016; Received in revised form 2 March 2017; Accepted 14 March 2017
0379-7112/ © 2017 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/BY/4.0/).
E. Rackauskaite et al. Fire Safety Journal 90 (2017) 123–138
SAFIR and OpenSees). All of these models have been widely used for correctness of the results [15,16]. The LS-DYNA model is benchmarked
structural fire analysis in the recent years and have been validated against fire tests on loaded steel framework results [28–30], two
against various small-scale and full-scale tests (e.g. Cardington) [6– benchmarks published by Gillie [31] and results published by
17]. Rackauskaite and El-Rimawi [32] on the numerical study of 2D steel
More recently, researchers and designers [18–20] have adopted LS- frames subject to localised fires.
DYNA for structural fire analysis. Kilic and Selamet [19], and Selamet
[18] used LS-DYNA to investigate the effect of fire location on the 2. LS-DYNA benchmarking models
collapse of a 49 storey high-rise steel structure. The whole 3D building
model was used for the analysis. Law et al. [20] studied the structural For the benchmarking of LS-DYNA for structural fire analysis, we
response of structural arrangements with bi-linear columns to fire. A use the double precision LS-DYNA (Release 7.1.1) version. Each
slice of the 3D model of a generic substructure was used that included a benchmarking case chosen for this paper encompasses different
4-storey high bi-linear column with adjacent composite beams and mechanisms of structural response in fire, which are required to get
concrete slabs. In all of these studies [18–20] the authors used beam a realistic response. In total four benchmark cases are considered. The
and shell elements to represent steel beams and concrete slab first benchmark is based on the natural fire test of the 2D steel frame
respectively. However, neither of them included the benchmarking of carried out in 1987 [28–30]. It allows the benchmarking of LS-DYNA
the adopted LS-DYNA model. Both et al. [21] published a benchmark against experimental results and assessing whether the model correctly
for predicting the structural fire response of a centrally loaded steel- captures the effects of non-uniform heating, material non-linearity,
concrete column. The column was modelled using solid elements. The restraint, and stress redistribution.
results of the coupled thermal-mechanical analysis were presented for The remaining 3 benchmarks are based on and allow benchmarking
ANSYS, LS-DYNA, and ABAQUS. Temperature and displacement of LS-DYNA against the results of numerical analysis published in the
development in the column showed a good agreement between those literature [31,32]. Gillie [31] has published results for 2 problems
software packages. The maximum differences between the peak values which provide benchmark solutions for structural fire analysis. They
in relation to LS-DYNA results were approximately 23 °C (5%) and allow users to check whether their models capture the required
0.23 mm (23%), respectively. Kwaśniewski et al. [22] carried out a phenomena, which occur when structures are heated. The first bench-
coupled thermal-mechanical analysis of a restrained steel column mark [31] is on a uniformly heated steel beam with 75% support
subjected to fire using LS-DYNA and validated against experimental stiffness. This benchmark allows to confirm whether the model
results. The detailed 3D model (as in the benchmark by Both et al. [21]) captures the effects of material non-linearity, geometric non-linearity
adopted solid elements. and restraint conditions [31]. The second benchmark [31] is on a
LS-DYNA is a commercial general purpose finite element software heated composite concrete floor. This benchmark assesses whether
originally developed for highly nonlinear and transient dynamic phenomena such as stress redistribution, localised heating and com-
analysis [23]. It is robust in the analysis of problems involving posite action effects can be captured. These two benchmarks provide a
transient effects, contact and large deformations and has high compu- computational challenge against most of the fundamental mechanisms
tational efficiency. As a result, LS-DYNA is one of the most commonly that occur when structures are heated and, thus, were chosen for the
used numerical explicit integration simulation programs. Common benchmarking of LS-DYNA. In addition to the above, a third bench-
applications of LS-DYNA include automotive, aerospace, metalforming, mark has been chosen based on the study by Rackauskaite and El-
and multi-physics problems. In structural engineering, common appli- Rimawi [32] on the heating effects on a 2D steel frame. In the latter
cations include earthquake, blast impact, and progressive collapse study non-uniform heating of the beams was assumed. Therefore, the
analysis. LS-DYNA has been used for the aircraft impact and progres- adoption of this study as a benchmark allows to assess whether thermal
sive collapse analysis of the World Trade Centre (WTC) towers by NIST bowing, restraint from the surrounding structure and stress redistribu-
[24,25]. However, most likely due to the lack of available scientific tion are appropriately captured in the analysis. In the following
work using LS-DYNA, for structural fire response analysis in the same sections these benchmarks are described and the results from the LS-
WTC study a different numerical program was used. LS-DYNA as a DYNA analyses are presented.
software has been fully validated and verified by its developers
(Livermore) for its generic applications. Even though, in recent years, 2.1. Benchmark #1 (BM1): fire test on a loaded steel framework
the explicit dynamic solver of LS-DYNA has grown considerably in
research and industry uses for the analysis of structures in fire [18–22], Benchmark #1 (BM1) represents a natural fire test on a 2D steel
to the best of the authors knowledge, there is still no benchmarking of frame carried out in 1987 [28]. The test frame comprised of a 4.55 m
the code available in the literature specifically with regards to the long steel beam and two 2.53 m high columns. It was connected to
structural fire performance. Available research is limited to the internal secondary framework to prevent lateral instability. Details of the frame
benchmarking work carried out in Arup [26] and detailed 3D solid and loading are shown in Fig. 1. This test has already been successfully
based individual structural member models [22]. Solid elements are modelled in CEFICOSS [29], Abaqus and SAFIR [30]. Therefore, a
not frequently used for global models or for design purposes. In similar modelling approach was used in LS-DYNA for comparative
addition, due to the quasi-static nature of the structural response in purposes.
fire, the majority of the structural fire analyses available in literature Beam web temperature measurement from the test is only available
are carried out using static solvers. at one instant during the whole fire exposure. Thus, member tempera-
Thus, this paper aims at benchmarking the explicit dynamic solver tures, which show a good agreement with the measured test data for
of LS-DYNA for the structural fire analysis of 3D composite structures beam flanges and column (see Fig. 1), were taken from [29]. The beam
and 2D steel frames against experiments and other numerical codes. was subjected to non-uniform gas temperatures along its length. To
An extensive parameter sensitivity study is carried out to study the account for this, a temperature reduction function following a sinusoi-
effects of various modelling parameters on the kinetic energy and dal shape was applied along the beam as in [29,30] with the
convergence to quasi-static solution. Four canonical problems that temperature at the connection being 0.9 of the beam temperature at
encompass a range of thermal and mechanical behaviours in fire are mid-span. Column temperatures along the height were assumed to be
simulated. The term benchmarking is used in this paper to refer to constant.
verification and validation of computational models, based on the In LS-DYNA, the steel frame was modelled using Hughes-Liu beam
definitions adopted by the ASTM [27]. That is, evaluating the software elements with user defined cross-section integration. This element has
for correct application to known benchmark problems and for physical a single integration point along its length in the middle of the element,
124
E. Rackauskaite et al. Fire Safety Journal 90 (2017) 123–138
Fig. 1. Details of Benchmark #1 used for benchmarking of LS-DYNA for fire analysis: illustration of the tested frame (left), and comparison of beam temperatures at mid-span and
column temperatures at mid-height measured during the test [28] with CEFICOSS numerical simulation results [29,30] that were used as an input in LS-DYNA analysis (right).
where sectional plasticity is monitored. Hughes-Liu beam elements within the beam is assumed to be uniform. Different levels of restraint
allow for the treatment of finite strains and are simple, computationally are considered and include 0% (i.e. simply supported), 5%, 10%, 25%,
efficient and robust. It is the first beam element implemented in LS- 50%, 75%, and 100% (i.e. pinned) of support stiffness.
DYNA and thus it is used as the default element type in the program In LS-DYNA, the steel beam was modelled using Hughes-Liu beam
[33]. The same beam formulation was used in [18–20]. A thermally- elements as for BM1. The elastic-plastic material (MAT 004) formula-
sensitive steel material type MAT 202 formulation based on Eurocode 3 tion was used to represent the steel. This material model allows the
(EN 1993-1-2:2005) was used for the analysis. Thermal loading to the user to define temperature dependent material properties as provided
beam and columns was applied using the formulation (i.e. “load in the original benchmark [31]. One end of the beam was set as pinned
thermal variable beam”) that allows the definition of variable and on the other end a discrete beam element was used with a linear
through-thickness temperature distribution. Using this formulation, a elastic material (MAT 66) formulation. This material model allows
known temperature-time history at different cross-section coordinates defining 6 springs acting about the local degrees of freedom [23]. The
of beam elements can be defined. Along the length of the beam element discrete beam element was used to represent varying support restraint
only a single uniform temperature can be defined, due to the Hughes- conditions of the beam. To model the 75% support restraint, the
Liu beam element formulation. Therefore, for this benchmark a uni- translational stiffness in the discrete beam material model was set to
form average temperature value over the beam element length was 1.902×108 N/m (75% of the beam stiffness, EA/L) as suggested in [31].
calculated and applied. Out of plane translational restraint was added to all nodes to ensure a
Due to symmetry, only half of the frame was modelled with the two dimensional behaviour. Illustration of this benchmark (BM2) is
support conditions set as shown in Fig. 1. To represent the support shown in Fig. 2.
from the secondary framework a discrete beam element was used at the
column height of 3.2101 m with nonlinear elastic material (MAT 67)
formulation. This material model allows defining 6 springs acting about 2.3. Benchmark #3 (BM3): composite steel-concrete floor
the local degrees of freedom and nonlinear spring stiffness [23]. Spring
behaviour was taken from [29,30]. LS-DYNA has a 3D model space and Between 1994 and 1996 a series of full-scale fire tests were
as a result, out of plane translational restraint to the nodes was added conducted on an 8-storey steel frame within the UK Building
to simplify the benchmark into a two dimensional space. Research Establishment at Cardington to gain a better understanding
of global structural performance in fire [5]. Benchmark #3 (BM3) [31]
represents a simplified idealisation of the first Cardington test. In this
2.2. Benchmark #2 (BM2): uniformly heated beam
benchmark part of the 4.5 m x 4.5 m x 0.13 m composite steel-concrete
floor is exposed to linearly increasing heating and cooling. The heated
Benchmark #2 (BM2) [31] represents a 1 m long restrained
area and the geometry of the benchmark are shown in Fig. 3. The
rectangular steel beam with cross-section dimensions of 35 mm x
concrete floor slab is connected to primary and secondary steel beams.
35 mm. A uniform load of 4250 N/m is applied to the beam (see
The heated part of the secondary beams has a uniform temperature
Fig. 2). The beam is linearly heated up to a temperature of 800 °C and
distribution and reaches a peak temperature of 800 °C. The heated part
then cooled down to room temperature. The temperature distribution
of the floor slab is assumed to have a linear through thickness thermal
gradient with the absolute peak temperature values of 600 °C at the
bottom of the slab and 0 °C at the top. The uniform gravity loading
applied on the slab is 5.48 kN/m2.
For this benchmark two LS-DYNA models were developed to
compare two different approaches used to represent the composite
action between steel beams and the concrete slab. In Model A (BM3-A),
the beam elements representing the steel beams and the shell elements
Fig. 2. Details of Benchmark #2 [31] used for the benchmarking of LS-DYNA for fire representing the concrete slab were defined using separate nodes. To
analysis. represent composite action, a tied contact between beam and shell
125
E. Rackauskaite et al. Fire Safety Journal 90 (2017) 123–138
Fig. 3. Illustration of Benchmark #3 from [31] used for benchmarking of LS-DYNA for fire analysis (top); Visualisation of Model A and Model B of Benchmark #3 where different
formulations were used to simulate composite action between steel beams and the concrete slab (bottom).
elements was used. This contact uses a penalty based formulation and benchmarking study two materials were created using this formulation
force and moment resultants are transferred by discrete spring for plain concrete and plain reinforcement. These were then used to
elements [23]. In Model B (BM3-B), the beam and shell elements represent different through-thickness layers of the concrete slab
share the same nodes in order to account for composite action effects section. The slab was divided into a chosen number of layers, each
between steel beams and concrete slab. Beam and shell element representing either plain concrete or plain reinforcement, with the
formulations with offset option were used to offset the central axis of middle layer characterising steel reinforcement. The equivalent thick-
the elements to geometrically represent the position of the elements in ness of the reinforcement layer (0.283 mm) was calculated based on
relation to one another. The illustration of the geometry and the two the total volume of steel rebars in the concrete slab.
models are shown in Fig. 3. Both modelling approaches assume that The thermal load to the beams and shell elements was applied using
there is a perfect contact between the steel beams and the concrete the formulations that allow the definition of varying through-thickness
slabs, and that shear stud failure does not occur during the fire. temperatures. It was identified by Gillie [31] that gravity loading was
The remaining parameters and configuration (e.g. mesh size, included in the uniformly distributed load on the concrete slab and
material properties, loading) were the same in both Model A and thus it was not explicitly applied in the current model. The concrete
Model B. For primary and secondary steel beams, the Hughes-Liu slab was modelled as continuous with appropriate lateral and rota-
beam element formulation was used. The concrete slab was modelled tional constraints around the boundary.
using the Belytschko-Tsay shell element formulation, which is compu-
tationally efficient. It is based on a combined co-rotational and velocity- 2.4. Benchmark #4 (BM4): 2D steel frame
strain formulation [33]. Material type MAT 172 and MAT 202
formulations were used for concrete and steel respectively. The steel The fourth LS-DYNA benchmark model used for benchmarking,
material model is based on Eurocode 3 (EN 1993-1-2:2005) and the Benchmark #4 (BM4), is based on the numerical study by Rackauskaite
concrete model is based on Eurocode 2 (EN 1992-1-2:2004). Both and El-Rimawi [32] on the structural response of 2D frames subject to
material models are thermally-sensitive and allow the users to define localised fires. The latter study [32] was conducted using an in-house
temperature dependent properties by overriding the values from the specialist finite element software based on the secant stiffness approach
Eurocodes. MAT 172 can be used to represent plain concrete, plain that was developed specifically for structural fire modelling purposes
steel or a smeared combination of concrete and steel [23]. In this [32,34–36]. A three-storey three-bay steel frame was subject to six
126
E. Rackauskaite et al. Fire Safety Journal 90 (2017) 123–138
Table 1
Initial model parameters used for parameter sensitivity study.
Model Preload duration, Thermal load application Element length Beam integration refinement Number of shell through thickness
t pre duration, th & tc factor, k integration points
*
Refers to beam element length. Column element length for BM1 and BM4 is ~ 0.39 m and 0.875 m, respectively.
127
E. Rackauskaite et al. Fire Safety Journal 90 (2017) 123–138
One of the key model parameters that can influence the accuracy of
the numerical solution is mesh density. Thus, for all the benchmark
cases, a mesh convergence study is carried out. For cases BM1, BM2,
BM3, and BM4, the beam element length is varied between 0.03 and
0.476 m, 0.00625 and 0.5 m, 0.0625 and 0.5 m, and 0.0625 and 4 m,
respectively. Column element length for cases BM1 and BM4 is varied
between 0.048 and 0.386 m and between 0.0625 and 0.875 m,
respectively. The beam and shell element lengths for case BM3 are
the same. For the mesh density study of BM1, a uniform temperature
along the beam length is assumed. This is to avoid any secondary
effects due to the local increase of temperatures with a finer mesh size.
Hughes-Liu beam elements have one integration point in the middle of
Fig. 5. The sequence of load application used in the LS-DYNA modelling approach for
the element and therefore a uniform average temperature value has
structural response to fire analysis. tpre is the preload duration, tss is the steady-state, th is
been assumed along the length of each element as identified previously
the heating duration, and tc is the cooling duration.
in Section 2.1. With finer mesh, i.e. a higher number of elements with a
shorter beam element length, the average temperature over the
element length applied to beam elements would increase in case of
on the convergence of the output and kinetic to internal energy ratios non-uniform heating along the length of the member. This would
for different models. The preload is applied first and then kept constant therefore have an effect on the thermal expansion and stresses within
for 1 s. Following that once the steady-state solution has been reached, the member. Thus, the outcome of the sensitivity study would not be
the thermal load is applied. In the cases where cooling is considering, only mesh dependent but also temperature dependent. Uniform
cooling duration to ambient temperature, tc , is the same as the heating heating was only considered as part of the mesh sensitivity study.
duration, i.e. tc = th . During the thermal load application period, the
static load is unchanged and considered to be constant. In the cases 3.3. Beam element integration
where damping is applied during preload only, the critical damping
factor is reduced to 0 before the application of the thermal load. The As noted earlier, in LS-DYNA, the Hughes-Liu beam element
order of load application is shown schematically in Fig. 5. formulation has a single integration point in the middle of the element
In addition to varying the preload and thermal load duration, the and user defined cross-section integration [33]. Integration points for a
application of global damping was considered. For each benchmark, an rectangular cross-section (BM2) and an I-section (BM1, BM3, and
LS-DYNA implicit eigenvalue analysis was run to determine the lowest BM4) are defined as shown in Fig. 6. The number of integration points
fundamental frequency of the structural system. Material type MAT can vary depending on the desired accuracy required on the through-
202 formulation is not currently available on the implicit solver of LS- depth plasticity. Each fibre has a prescribed uniaxial stress-strain
DYNA R7.1. Thus, to carry out the eigenvalue analysis for BM1, BM3 relationship. A greater number of integration points can also more
and BM4 an elastic material (MAT 1) formulation for steel parts was accurately represent the thermal gradients within a structural member.
used. The resulting frequencies, f , were 62.4, 77.06, 7.70, 7.71, and In this study, the beam cross-section integration refinement factor, k
4.28 Hz for BM1, BM2, BM3-A, BM3-B, and BM4, respectively. These (see Fig. 6), for all cases was varied between 0 and 11. That is, the
frequencies were used to determine the critical damping factor, dcr , for number of cross-sectional integration points was varied for the beam
the structural system using Eq. (2) as suggested in [23]: elements between 9 and 169 for the rectangular cross-section (BM2),
and between 6 and 50 for the I-section (BM1, BM2, and BM3).
dcr = 4πf (2)
The static load was applied using a half sine function shape over the 3.4. Shell element integration
duration 2 times the normal period (1/f ) of the system in the damped
cases. After preload, the load was kept constant for 1 s. It should be In LS-DYNA, based on the modelling approach presented for BM3
noted that there may be cases where a dynamic analysis would be best in this paper (see Section 2.3), the concrete slab is modelled as a
suited for structural fire analyses such as in the case of fire induced composite section. Different shell through-thickness integration points
progressive collapse [37,42] and therefore selecting the appropriate are defined for steel rebars and slab with appropriate material models.
analysis parameters that would minimise the introduction of pseudo- A smeared cracking approach is adopted and therefore rebar rupture is
inertia effects is crucial. not explicitly considered and would need to be assessed by the modeller
independently. The rebar layer is modelled at the centre of the shell
Fig. 6. Rectangular and I-section beam element cross-section refinement used in LS-DYNA. Number of integration points for integration refinement factor k = 1 [23].
128
E. Rackauskaite et al. Fire Safety Journal 90 (2017) 123–138
the highest preload time of 64 s is similar for both cases BM3-A and
BM3-B and is around 3.1 mm. However, the difference in the axial
forces is quite high, approx. 10 kN, considering that both cases
represent the same benchmark problem.
When damping is applied, the kinetic energy in all cases is close to
0%. Results do not indicate any oscillations and show a nearly perfect
steady-state solution. This solution in cases BM1, BM2 and BM4 is
almost identical to results where damping was not applied.
Nevertheless, this is not the case for BM3. The damping solution
Fig. 7. Representation of shell through thickness integration layers as modelled in LS-
DYNA BM2. t is the total thickness of the concrete slab. The image represents a shell differs by up to 5 kN and 2 kN when compared to the solution with no
element with 7 integration layers. damping for cases BM3-A and BM3-B, respectively.
129
E. Rackauskaite et al. Fire Safety Journal 90 (2017) 123–138
Fig. 8. Variation of the kinetic to internal energy ratio with preload application time (top), and the effect of preload application time on displacement (middle), and axial force (bottom)
for all benchmark cases. Displacements and axial forces shown are during 1 s (tss ) after the preload was applied and kept constant. Plotted data is of the following structural members:
BM2 – beam mid-span displacement and axial force; BM3 – heated beam mid-span displacement and axial force; and BM4 – mid-span displacement and axial force of the ground floor
edge beam.
high localised stresses because Hughes-Liu beam elements have only the development of beam mid-span displacement (Fig. 11). This
one integration point along the length. Therefore, with a decreasing indicates that increasing the number of beam integration points for
element size the constant stress along the length would increase. There these models (BM1, BM3, and BM4) does not improve the solution
is no clear relationship between the kinetic energy in the system and significantly. However, this is not the case for BM2. The development
the change in element length comparing the different cases. Due to the of beam mid-span displacement and the resultant values are affected by
instabilities present in the solutions for the finest mesh densities, the beam integration refinement. Convergence is only reached for a
mesh with the beam element length of 0.25 m is selected as the refinement factor, k, of 5 which gives a more refined solution in
“converged” solution for BM3. It should be noted that these analyses comparison to the case when k is 0. In the cases with k=1 and k=3
are based on capturing the global structural fire behaviour. The relative error reaches up to 20%. Thus, in some cases the integration
selected mesh densities for these problems would likely need to be factor has to be carefully chosen in order to obtain conservative results.
different if localised responses such as concrete cracking are desired to The kinetic energy in all cases shows negligible variation with the
be captured. change in the number of beam integration points. The maximum
difference is less than 1.5%. This indicates that the number of beam
cross-section integration points is unlikely to affect the inertia of the
4.1.4. Effect of the number of beam element integration points
system. It should be noted that if there is a scenario such that the
The variation of the relative errors of the beam peak mid-span
response is significantly underestimated with a chosen number of
displacement and axial force, kinetic energy and heated beam mid-span
integration points, this assumption may need to be reassessed.
displacement development with the change of the integration refine-
ment factor is shown in Fig. 11. The errors were calculated based on
the numerical results for an integration refinement factor of 11, which 4.1.5. Effect of the number of shell element integration layers
includes the maximum number of fibres. For all cases except for case The variation of the key structural variables with the number of
BM3-A, a convergence of the resultant values with an increasing shell element integration layers is shown in Fig. 12. The relative errors,
integration factor can be observed. Yet, the relative errors for BM1, as in the previous sections, have been calculated based on the outcome
BM3-A, BM3-B and BM4 are negligible ( < 3.5%) as it can be seen from for the model with the largest number of shell element through-
130
E. Rackauskaite et al. Fire Safety Journal 90 (2017) 123–138
Fig. 9. Variation of relative error of peak displacement and axial force, 95th percentile of kinetic to internal energy ratio, displacement (preload 32 s), and axial force (preload 32 s) with
preload and thermal load application duration for all benchmark cases. Plotted data is of the following structural members: BM2 – mid-span displacement and axial force; BM3 – heated
beam mid-span displacement and axial force; and BM4 – mid-span displacement and axial force of the ground floor edge beam. DH refers to the cases where damping was continued
during the heating.
131
E. Rackauskaite et al. Fire Safety Journal 90 (2017) 123–138
Fig. 10. Variation of relative error of peak displacement and axial force; 95th percentile of kinetic to internal energy ratio; and axial force with mesh size for all benchmark cases. Plotted
data is of the following structural members: BM2 – mid-span displacement and axial force; BM3 – heated beam mid-span displacement and axial force; and BM4 – mid-span
displacement and axial force of the ground floor edge beam.
hand, the axial force development for the 3 integration layers indicates at 16 min. Predictions from LS-DYNA on the deflections also compare
some instabilities between temperatures of 300 and 400 °C. Thus, for well with other software packages (CEFICOSS [29], SAFIR and Abaqus
the purposes of this study the convergence is assumed to be reached for [30]) and indicate a conservative solution. However, in comparison to
7 shell element integration layers. the latter software packages, LS-DYNA underestimates peak axial
forces by approximately 30 kN (25%). The sensitivity study presented
4.2. Benchmarking of LS-DYNA in the previous sections indicates that there is very little or no influence
from the investigated parameters on the development of axial forces.
In this section, the final results for each benchmark case obtained Therefore, LS-DYNA errors in comparison with other software
with LS-DYNA are presented and compared with the results published packages could be associated with the inherent assumptions in the
by Cooke and Latham [28], Santiago et al. [30], Gillie [31], and programs and the element types used to represent the support from the
Rackauskaite and El-Rimawi [32]. Based on the time scaling and secondary framework. In the case without the lateral support (i.e. that
parameter sensitivity studies, the final model parameters were chosen discrete elements are not used), LS-DYNA underestimates peak axial
and are shown in Table 3. The parameters were determined based on forces by only approximately 2.9–4.3 kN (6.7–11.7%) in comparison
the convergence of results and computational time efficiency. The with CEFICOSS, SAFIR and Abaqus [30]. It should be noted, however,
results for BM1, BM2, BM3, and BM4 are shown in Fig. 13, Fig. 14, that LS-DYNA gives a conservative prediction of the failure time by up
Fig. 17, and Fig. 18, respectively. The LS-DYNA final solution errors in to 1.4 min.
respect to other software packages and/or experimental results are In benchmark BM2 a single rectangular steel beam with 75%
summarised in Table 4. restraint was heated and then cooled. For the BM2 model as seen in
Benchmark BM1 represents the experiment on the 2D steel frame Fig. 14 LS-DYNA predicts well the development of heated beam mid-
heated until failure. The results shown in Fig. 13 show a good span displacements and axial forces with temperature in comparison to
agreement between the LS-DYNA solution and the experimental results other software packages (Abaqus, Vulcan, and Ansys [31]). Unlike in
[28] for the beam mid-span deflection and deflected shape of the frame BM1, in BM2 the peak axial force predicted with LS-DYNA is within
132
E. Rackauskaite et al. Fire Safety Journal 90 (2017) 123–138
Fig. 11. Variation of relative error of peak displacement and axial force (N), 95th percentile of kinetic to internal energy ratio, and displacement development with beam integration
refinement factor for all benchmark cases. Plotted data is of the following structural members: BM2 – mid-span displacement; BM3 – heated beam mid-span displacement; and BM4–
mid-span displacement of the ground floor edge beam.
the same range as predicted by other software packages with the LS- DYNA prediction in comparison to the prediction by Abaqus [31] is
DYNA solution being closest to the Abaqus explicit-dynamic solution. A worse than for BM1 and BM2 (see Fig. 17). While LS-DYNA is able to
comparison of the results from LS-DYNA and Abaqus for varying predict the development of axial forces during the heating for BM3, it
support stiffness is shown in Fig. 15. The results indicate that with the underestimates the values during cooling below the temperature of
final chosen model parameters, LS-DYNA is able to predict well the 400 °C. LS-DYNA is unable to capture the peak axial force of
development of deflections and axial forces for all restraint levels approximately 1.5 MN at the end of cooling. The predicted values are
except for the 0–10%. For the latter cases high dynamic oscillations 293 kN (19%) and 394 kN (26%) lower for BM3-A and BM3-B,
occur indicating large inertia effects for very low restraint levels. An respectively. Interestingly, this value has been captured by the model
interesting observation is that for these cases inertia effects could only with the lower value of shell through-thickness integration points (i.e.
be dissipated to achieve a quasi-static solution as in Abaqus by applying 3) (see Fig. 12). Comparing models BM3-A and BM3-B, the latter one
damping during the heating duration (see Fig. 15). The comparison of gives a closer prediction of axial force development in the beam during
the 95th percentile of the kinetic to internal energy ratios (see Fig. 16) heating in comparison to Abaqus. In BM3-B a shared node between
shows a significant reduction (a factor up to 30,000) in kinetic energy beam and shell elements was used to represent the composite action.
in the system when damping is applied for cases with low restraint Both BM3 models (A and B) show significantly higher beam mid-span
levels (0–10%). Results shown in Fig. 16 also indicate that the deflection development values than Abaqus. Displacement results
application of damping during the heating duration may not always using BM3-A and BM3-B models are by up to approximately 1.4 and
be the most effective or efficient solution. For restraint levels higher 1.2 times higher, respectively, compared to Gillie's [31] results. Thus,
than 10% solutions with damping have very similar level of kinetic in the terms of deflections LS-DYNA solution can be considered as
energy in the system as solutions without damping. The maximum conservative.
difference in kinetic to internal energy ratio is up to 0.00006%. The differences in LS-DYNA and Abaqus solutions could be due to
In benchmark BM3, part of the composite concrete floor slab and different inherent assumptions in the analysis programs or input
the supporting beams was heated and then cooled. For BM3, the LS- parameters used. No details of the input file for Abaqus [31] for BM3
133
E. Rackauskaite et al. Fire Safety Journal 90 (2017) 123–138
Fig. 12. Variation of relative error of peak displacement and axial force, 95th percentile of kinetic to internal energy ratio, and heated beam axial force development with number of
shell element integration layers for BM3-A (top) and BM3-B (bottom).
Table 3
Final model parameters used for the LS-DYNA simulations.
Model Preload duration, Thermal load application Element length Beam integration refinement Number of shell through thickness
t pre duration, th & tc factor, k integration points
*
Element length used in [32].
Fig. 13. Comparison of the experimental BM1 results [28] and finite element analysis results from CEFICOSS, SAFIR and Abaqus [29,30] with results obtained using LS-DYNA.
Variation of heated beam mid-span deflection (left) and axial forces (middle) with temperature and the deflected shape of the frame at 16 min. Deflected shape scale factor is 10.
134
E. Rackauskaite et al. Fire Safety Journal 90 (2017) 123–138
Fig. 14. Comparison of BM2 results published by Gillie [31] with results obtained using LS-DYNA. Variation of heated beam mid-span displacement (left) and axial forces (right) with
temperature.
Table 4
Comparison of the final LS-DYNA solution with other software packages and/or experimental results for all benchmark cases.
have been published. Moreover, some details on the benchmark were published the deflected shape of the whole model for benchmarking.
not clearly defined in [31], for example, the location of steel rebar in In benchmark BM4 various bays of the 3 storey × 3 bay 2D steel
the concrete slab and whether the self-weight of the members is to be frame [32] were subjected to heating. The comparison of the results
included. Considering the uncertainty in the benchmark itself, Abaqus from [32] and the LS-DYNA solution for the different cases is shown in
parameters used for the analysis, and significant sensitivity of the LS- Fig. 18. For most localised fire cases the results show a good agreement.
DYNA solution to various parameters investigated in this paper, it is The development of the beam mid-span displacements and bending
difficult to identify more specific reasons for the differences in LS- moments is almost identical in LS-DYNA and [32]. For axial forces LS-
DYNA and Abaqus solutions. In addition, LS-DYNA models BM3-A DYNA produces slightly higher axial forces (up to approximately 15 kN
and BM3-B at the end of heating give considerably different deflected or 15% for cases A to D), but they can be considered to be conservative.
shapes, especially in the bays adjacent to the heated slab. Deflections in Also, the trend of axial force development is the same in the results
these unheated bays in BM3-B are almost as high as in the heated part produced by LS-DYNA and [32]. Nevertheless, for the BM4 cases where
of the slab and thus they do not seem to be realistic. Gillie [31] has not the top floor and middle floors were heated, even though LS-DYNA
135
E. Rackauskaite et al. Fire Safety Journal 90 (2017) 123–138
Fig. 15. Comparison of BM2 results published by Gillie [31] with results obtained using LS-DYNA for varying levels of support restraint stiffness. Variation of heated beam mid-span
displacement and axial forces with temperature for the models without (left) and with (right) damping included during heating.
5. Conclusions
136
E. Rackauskaite et al. Fire Safety Journal 90 (2017) 123–138
Fig. 17. Comparison of BM3 results published by Gillie [31] with results obtained using LS-DYNA models BM3-A and BM3-B. Variation of the heated beam mid-span displacement and
axial forces with temperature (top), deflected shape at the end of heating from LS-DYNA model (bottom). Displacement scale factor is 5.
Fig. 18. Comparison of BM4 results published by Rackauskaite and El-Rimawi [32] with results obtained using LS-DYNA. Variation of heated beam mid-span displacement (left), axial
forces (middle) and bending moments (right) with temperature. For the references of beams in the frame refer to [32].
a more significant influence on the final result of the LS-DYNA explicit the time scaling study shows that the generally applied ad-hoc rule in
solution than preloading due to static loads. The high variation in the explicit dynamic analysis that the kinetic to internal energy ratio for the
results and kinetic energy with different parameters, especially for the most of the time has to be less than 5% to achieve a quasi-static
composite concrete floor slab case, highlights that an extensive para- solution might be too high for structural analysis in case of fire.
meter sensitivity study has to be carried out for every model to ensure In this study LS-DYNA (Release 7.1.1) version was used. It should
that the LS-DYNA solution converges and is quasi-static. In addition, be noted that results obtained using a different LS-DYNA version might
137
E. Rackauskaite et al. Fire Safety Journal 90 (2017) 123–138
be to some extent different. Thus, it should be used cautiously and the [19] S.A. Kilic, S. Selamet, Symmetric and Asymmetric Collapse Mechanisms of a Multi-
Story Steel Structure subjected to Gravity and Fire, in: Struct. Congr. 2013 Bridg.
model should be benchmarked before carrying out the analysis of Your Passion with Your Prof. Pittsburgh, Pennsylvania, May 2-4, 2013, American
interest. Additionally, further research is required to assess other Society of Civil Engineers, Reston, VA: pp. 2545–2554. doi:http://dx.doi.org/10.
element types such as solid elements or formulations of beams or shell 1061/9780784412848.222, 2013.
[20] A. Law, P. Kotsovinos, N. Butterworth, Engineering geometrically bi-linear columns
elements different to those examined in this paper. to deliver fire resistance: standard heating, Eng. Struct. 100 (2015) 590–598.
http://dx.doi.org/10.1016/j.engstruct.2015.06.046.
Acknowledgements [21] I. Both, K. Ostapska, L. Kwasniewski, R. Zaharia, D. Dubina, Composite column
subjected to compression and fire, in: F. Wald, I.W. Burgess, L. Kwasniewski,
K. Horová, E. Caldova (Eds.), Benchmark Stud. Verif. Numer. Model. Fire Eng.,
The research has been funded by the Engineering and Physical CTU Publishing House, Czech Technical University, Prague, 2014, pp. 196–201.
Sciences Research Council (EPSRC, UK) with grant number EP/ [22] L. Kwaśniewski, F. Ali, M. Balcerzak, Coupled Structural-thermal Calculations for
Restrained Steel Columns in Fire, J. Struct. Fire Eng. 4 (2013) 59–70. http://
K502856/1, Ove Arup and Partners Limited (UK), Centre d′Études et
dx.doi.org/10.1260/2040-2317.4.1.59.
de Recherches de l′Industrie du Béton (CERIB, France) and [23] LSTC, LS-DYNA Keyword User’s Manual, Volumes I to III (Version R7.1),
Educational & Scientific Foundation of the Society of Fire Protection Livermore Software Technology Corporation (LSTC), Liwermore, 2014.
Engineers (SFPE, USA). The authors appreciate the valuable technical [24] A. Irfanoglu, Using numerical simulations and engineering reasoning under
uncertainty: studying the collapse of WTC-1, Comput. Civ. Infrastruct. Eng. 27
discussions held with Professor Ann Jeffers (University of Michigan) (2012) 65–76. http://dx.doi.org/10.1111/j.1467-8667.2010.00700.x.
and Susan Deeny and Graeme Flint (Arup). [25] T.P. McAllister, F. Sadek, J.L. Gross, S. Kirkpatrick, R. . MacNeill, R.T. Bocchieri,
et al., Structural analysis of impact damage to world trade center buildings 1, 2, and
7, Fire Technol. 49 (2013) 615–642. http://dx.doi.org/10.1007/s10694-012-0286-
References 5.
[26] Arup, Benchmarking and validation for LS-DYNA. Internal report, 2013.
[1] V.K.R. Kodur, M. Garlock, N. Iwankiw, Structures in fire: state-of-the-art, research [27] ASTM, ASTM E176-15ae1 Standard Terminology of Fire Standards, West
and training needs, Fire Technol. 48 (2012) 825–839. http://dx.doi.org/10.1007/ Conshohocken, PA, 2015. doi:http://dx.doi.org/10.1520/E0176-15AE01.
s10694-011-0247-4. [28] G.M.E. Cooke, D.J. Latham, The inherent fire resistance of a loaded steel frame-
[2] A. Haksever, Fire response of total systems in a localfire, Fire Saf. J. 4 (1981) work, Steel Constr. Today 1 (1987) 49–58.
141–146. http://dx.doi.org/10.1016/0379-7112(81)90012-6. [29] J.M. Franssen, G.M.E. Cooke, D.J. Latham, Numerical simulation of a full scale fire
[3] J.A. Purkiss, Developments in the fire safety design of structural steelwork, J. test on a loaded steel framework, J. Constr. Steel Res. 35 (1995) 377–408. http://
Constr. Steel Res. 11 (1988) 149–173. http://dx.doi.org/10.1016/0143-974X(88) dx.doi.org/10.1016/0143-974X(95)00010-S.
90040-5. [30] A. Santiago, C. Haremza, F. Lopes, J.-M. Franssen, Numerical behaviour of steel
[4] J.C.M. Forrest, The whole structure, in: R.D. Anchor, H.L. Malhotra, J.A. Purkiss columns under localized fire loading, in: F. Wald, I.W. Burgess, L. Kwasniewski,
(Eds.), Int. Conf. Des. Struct. Against Fire, Elsevier Applied Science Ltd., K. Horová, E. Caldova (Eds.), Benchmark Stud. Exp. Valid. Numer. Model. Fire
Birmingham, 1986: pp. 111–125. Eng., CTU Publishing House, Czech Technical University, Prague, 2014, pp. 60–72.
[5] British Steel, The Behaviour of Multi-Storey Steel Framed Buildings in Fire, [31] M. Gillie, Analysis of heated structures: nature and modelling benchmarks, Fire
Rotherham, 1999. Saf. J. 44 (2009) 673–680. http://dx.doi.org/10.1016/j.firesaf.2009.01.003.
[6] Vulcan-Solutions, Comparisons against experimental and clasical results, (n.d.) [32] E. Rackauskaite, J. El-Rimawi, A study on the effect of compartment fires on the
〈http://www.vulcan-solutions.com/*Page_id=40. http://www.vulcan-solutions. behaviour of multi-Storey steel framed structures, Fire Technol. 51 (2015)
com/*Page_id=40〉 (accessed 4 December 2015). 867–886. http://dx.doi.org/10.1007/s10694-014-0419-0.
[7] J.-M. Franssen, SAFIR: a thermal-structural program for modelling structures [33] J.O. Hallquist, LS-DYNA theory manual, Livermore Software Technology
under fire, Eng. J. Q3 (2005) 143–158. Corporation. 〈http://www.dynasupport.com/manuals/additional/ls-dyna-theory-
[8] R. Sun, Z. Huang, I.W. Burgess, The collapse behaviour of braced steel frames manual-2005-beta/at_download/file〉, 2006.
exposed to fire, J. Constr. Steel Res. 72 (2012) 130–142. http://dx.doi.org/ [34] I.W. Burgess, J.A. El-Rimawi, R.J. Plank, A secant stiffness approach to the fire
10.1016/j.jcsr.2011.11.008. analysis of steel beams, J. Constr. Steel Res. 11 (1988) 105–120. http://dx.doi.org/
[9] J. Jiang, L. Jiang, P. Kotsovinos, J. Zhang, A. Usmani, OpenSees software 10.1016/0143-974X(88)90046-6.
architecture for the analysis of structures in fire, J. Comput. Civ. Eng. 29 (2015) [35] J.A. El-Rimawi, I.W. Burgess, R.J. Plank, The analysis of semi-rigid frames in fire–
1–13. http://dx.doi.org/10.1061/(ASCE)CP.1943-5487.0000305. a secant approach, J. Constr. Steel Res. 33 (1995) 125–146. http://dx.doi.org/
[10] J. Jiang, A. Usmani, Modeling of steel frame structures in fire using OpenSees, 10.1016/0143-974X(94)00018-D.
Comput, Struct 118 (2013) 90–99. http://dx.doi.org/10.1016/j.comp- [36] J.A. El-Rimawi, I.W. Burgess, R.J. Plank, Studies of the behaviour of steel
struc.2012.07.013. subframes with semi-rigid connections in fire, J. Constr. Steel Res. 49 (1999)
[11] V.K.R. Kodur, M.Z. Naser, Effect of shear on fire response of steel beams, J. Constr. 83–98. http://dx.doi.org/10.1016/S0143-974X(98)00002-9.
Steel Res. 97 (2014) 48–58. http://dx.doi.org/10.1016/j.jcsr.2014.01.018. [37] R. Sun, I.W. Burgess, Z. Huang, G. Dong, Progressive failure modelling and
[12] W.-Y. Wang, G.-Q. Li, Fire-resistance study of restrained steel columns with partial ductility demand of steel beam-to-column connections in fire, Eng. Struct. 89
damage to fire protection, Fire Saf. J. 44 (2009) 1088–1094. http://dx.doi.org/ (2015) 66–78. http://dx.doi.org/10.1016/j.engstruct.2015.01.053.
10.1016/j.firesaf.2009.08.002. [38] Abaqus, ABAQUS Analysis User’s Manual. Version 6.5, 2004.
[13] M. Gillie, A.S. Usmani, J.M. Rotter, A structural analysis of the first Cardington [39] J.T. Wang, T. Chen, D.W. Sleight, A. Tessler, Simulating Nonlinear Deformations of
test, J. Constr. Steel Res. 57 (2001) 581–601. http://dx.doi.org/10.1016/S0143- Solar Sail Membranes Using Explicit Time Integration, in: 45th AIAA/ASME/
974X(01)00004-9. ASCE/AHS/ASC Struct. Struct. Dyn. Mater. Conference, PalmSprings, CA, USA:
[14] H. Yang, F. Liu, S. Zhang, X. Lv, Experimental investigation of concrete-filled pp. 1–15, 2004.
square hollow section columns subjected to non-uniform exposure, Eng. Struct. 48 [40] F. Pan, J. Zhu, A.O. Helminen, R. Vatanparast, Three point bending analysis of a
(2013) 292–312. http://dx.doi.org/10.1016/j.engstruct.2012.09.011. mobile phone using LS-DYNA explicit integration method, in: W.L. Mindle (Ed.),
[15] F. Wald, I.W. Burgess, L. Kwasniewski, K. Horová, E. Caldova (Eds.), Benchmark 9th Int. LS-DYNA Users Conf., LSTC, Dearbron, Michigan USA, 2006: pp. 13–31.
studies. Experimental validation of numerical models in fire engineering., CTU [41] H. Yu, I.W. Burgess, J.B. Davison, R.J. Plank, Numerical simulation of bolted steel
Publishing House, Czech Technical University, Prague, 2014. connections in fire using explicit dynamic analysis, J. Constr. Steel Res. 64 (2008)
[16] F. Wald, I.W. Burgess, L. Kwasniewski, K. Horová, E. Caldova (Eds.), Benchmark 515–525. http://dx.doi.org/10.1016/j.jcsr.2007.10.009.
studies. Verification of numerical models in fire engineering, CTU Publishing [42] P. Kotsovinos, A. Usmani, The world trade center 9/11 disaster and progressive
House, Czech Technical University, Prague, 2014. collapse of tall buildings, Fire Technol. 49 (2013) 741–765. http://dx.doi.org/
[17] G. Flint, Fire Induced Collapse of Tall Buildings, University of Edinburgh, 2005 10.1007/s10694-012-0283-8.
〈http://hdl.handle.net/1842/1172〉. [43] M. Gillie, The Behaviour of Steel-Framed Composite Structures in Fire Conditions,
[18] S. Selamet, The impact of fire scenario to the collapse of a tall structure, in: SEMC University of Edinburgh, 2000 http://hdl.handle.net/1842/1466.
Proceedings of the Fifth International Conference Struct. Eng. Mech. Comput. 2−4 [44] C. Röben, The Effect of Cooling and Non-uniform Fires on Structural Behaviour,
Sept. 2013, Cape Town, South Africa: pp. 711–716, 2013. The University of Edinburgh, 2009 http://hdl.handle.net/1842/14292.
138