Вы находитесь на странице: 1из 11

Journal of Constructional Steel Research 64 (2008) 515–525

www.elsevier.com/locate/jcsr

Numerical simulation of bolted steel connections in fire using explicit


dynamic analysis
Hongxia Yu a,∗ , I.W. Burgess a , J.B. Davison a , R.J. Plank b
a Department of Civil and Structural Engineering, University of Sheffield, Sheffield S1 3JD, UK
b School of Architecture, University of Sheffield, Sheffield S10 2TN, UK

Received 20 July 2007; accepted 16 October 2007

Abstract

Behaviour of steel connections in fire is a multi-dimensional problem involving parameters such as temperature, tying forces and large
deformations. Investigation of this behaviour will remain one of the main subjects for fire engineering research in the coming years. Finite
element simulation plays an important role in the study of connections because fire tests are expensive to perform. Unlike normal structural
analyses, finite element simulation of bolted steel connections is a challenging task, as large numbers of contacts exist in the model. This leads
to convergence difficulties in static solvers. This paper explores the use of an explicit dynamic solver to analyse bolted steel connections. By
comparing the results with those from static analysis and tests, it is shown that the explicit dynamic solver, with proper control, gives satisfactory
predictions of the responses of steel connections up to post-failure deformations.
c 2007 Elsevier Ltd. All rights reserved.

Keywords: Explicit dynamic analysis; Steel; Connections; Fire; Contact

1. Introduction of bolted steel connections is different from normal structural


analysis in two respects. First, 3D solid elements are generally
Structural connection failures were observed in the collapse used instead of structural beam and shell elements; second,
of the WTC buildings [1] and full-scale fire tests at many contact relationships are usually involved. Contacts
Cardington [2]. They caused concern over the ability of impose difficulty for the numerical solution procedure because
connections to maintain their structural integrity in fire and led the connection is not initially tied together, and does not provide
to subsequent studies of the high-temperature behaviour of steel stiffness until contacts on the load transfer route are closed.
connections. A general understanding of the moment–rotation Several simulations of steel connections have been
characteristics of steel connections at ambient temperature reported previously. Liu [3] modelled two extended endplate
is insufficient in fire engineering because material properties connections in fire using a self-developed program. Shell
and stress states can be affected by temperature, and ultimate elements were used to discretize the beam, column stiffeners
resistance, rather than elastic stiffness, becomes the major and endplate. A beam element having special characteristics
issue of concern. The University of Sheffield is performing an was used to simulate the behaviour of bolts and the contact
investigation of the robustness of common steel connections ‘link’ between endplate and column flange. This was a fairly
at elevated temperatures. As a large number fire tests will be simple treatment of the contact problem, but gave successful
performed in the course of the investigation, finite element predictions of the test results up to a maximum rotation of
modelling will also be widely employed, because it can extend 0.1 rad. Bursi and Jaspart [4] discussed the modelling of
the range of the study. By providing the complete stress state of bolts in bolted connections at ambient temperature using the
the structure during its whole loading history, FE modelling can general commercial program ABAQUS. A “spin” model, using
help understanding of the failure mechanisms. FE modelling an assembly of beam elements, was proposed to represent a
bolt. This was verified against 3D solid models for simple
∗ Corresponding author. Tel.: +44 114 222 5726; fax: +44 114 222 5793. problems such as T -stubs, and then used in more complicated
E-mail address: H.Yu@sheffield.ac.uk (H. Yu). endplate connections. Both models gave very close predictions

c 2007 Elsevier Ltd. All rights reserved.


0143-974X/$ - see front matter
doi:10.1016/j.jcsr.2007.10.009
516 H. Yu et al. / Journal of Constructional Steel Research 64 (2008) 515–525

to the test results. Swanson et al. [5] modelled a test on For simulations of steel connections in fire it is proposed
a T -stub using ABAQUS, incorporating contact relationships to use an explicit dynamic solver because this allows the
between the bolts and the base materials, and between different definition of very general contact conditions for complicated
connection plates. However, details of the contact techniques contact problems, without generating numerical convergence
employed are not given. Spyrou [6] performed two-dimensional difficulties. Also, an explicit dynamic solver uses a consistent
modelling of T -stub tests at elevated temperatures in ANSYS large-deformation theory that can model large rotations and
using 2D plane stress elements. The contacts were simulated large deformations. This is beneficial for the analysis of
by point-to-surface contact elements. However, the results were structures in fire because they are generally subjected to large
not very satisfactory, which seems to indicate that 2D models deformations.
are not appropriate. Maggi et al. [7] performed a numerical There are two major advantages of explicit dynamic
study of the behaviour of bolted endplate connections using analysis:
ANSYS. They concentrated on the structural behaviour, and • An explicit dynamic solver allows easy preprocessing.
gave little information on the modelling techniques employed. Structural bolted connections usually involve bolts in
Coelho [8] gave a very detailed description of the numerical clearance holes, which are at least 2 mm larger than
simulation of T -stubs using the commercial package LUSAS the diameter of the bolts. Depending on the direction of
with eight-node full-integration brick elements. The contact loads, a bolt can move freely in the hole until it finds
was simulated using node-to-node nonlinear contact friction the bearing surface. In static analysis, the bolt surface
elements, basically a kind of spring element that connects and the inner edge of the hole must be carefully brought
two adjacent nodes, whose properties for the behaviour of into initial contact, otherwise, the model will encounter
various degrees of freedom can be defined by the contact zero stiffness, and thus numerical singularity. For simple
properties of the surfaces. Al-Jabri [9] simulated a series of contacts with obvious load-transfer routes this is easy to do.
fire tests on flush endplate connections in ABAQUS/standard However, for connections with multiple bolts and subjected
using eight-node brick elements with reduced integration. to more complex loading, the pattern of initial slip might
For the contact areas, surface-to-surface contact was used. not be so obvious. Using an explicit solver, this is not of
Efforts to help numerical convergence included using a small- concern, because the contact surfaces can be left in the
sliding option and providing restraints to prevent rigid body noncontact state, and the free components can have rigid
movements. The numerical analysis results agreed well with body movements under the applied load until they come into
test results, even though the meshing was surprisingly coarse contact with the target surface, at which point the structure
compared to general practice. Sarraj et al. [10] performed automatically starts to pick up the stiffness.
modelling of steel fin-plate connections in fire using eight-node • The explicit dynamic solver also provides an easy solution
hexahedral brick elements in ABAQUS/standard. Surface-to- procedure. In a static analysis, for each load increment, the
surface contact with a small sliding option was used to define continuity of the contact states is checked first. If any of the
the contact relationships. Good agreement was achieved when contacts in the model changes state, the stiffness matrix is
the simulation algorithm was validated against test results for adjusted on the basis of the changed state of contacts, and
both ambient temperature and elevated temperatures. Again this increment is reperformed. It is not until the continuity
it was mentioned that significant efforts were required to of all contacts is satisfied that the force equilibrium can
make the numerical procedure converge. Lou and Li [11] be checked. This analysis algorithm imposes huge difficulty
performed simulations of one ambient-temperature test and two when many contact pairs are defined in the model, especially
elevated-temperature tests on extended endplate connections at the beginning of each analysis because it is difficult
using ANSYS. They used SOLID45 elements with at least two for the solver to find a stable state for all the contact
layers of elements through the thickness of the plates. Contact pairs at the same instant. In a bolted connection, one bolt
elements TARGET170 and CONTA173 were used to simulate generally requires the definition of four contact pairs; the
the contact relationships. The simulated results were generally contact between the head and nut and the plate surfaces,
good, with discrepancies observed for only one test. and the contact between the bolt shank and the holes of
Contact is a nonlinear and discontinuous phenomenon. In an the two connected plates. To analyse a bolted structural
idealized hard-contact, when the contact is open the pressure steel connection could easily involve over 20 contact pairs
between the surfaces is zero. When the two surfaces come into in the model, making it almost impossible for the static
contact, the overpressure can reach very high values, but no analysis solver to converge. In an explicit dynamic solver,
penetration of one surface into the other is permitted. Accurate a kinetic contact algorithm is used, in which the model
simulation of this behaviour is practically impossible due to the moves into a predicted configuration without considering the
discontinuity of the stiffness matrix. In the simpler treatments, contact conditions. The equilibrium state is then determined
the contact relationships are established between relevant by compensation to the forces and displacements of the
nodes using spring elements. This method loses accuracy if nodes on the contact surfaces according to the amount of
large deformations develop. A more general approximation is penetration. Several corrections may need to be performed to
surface-to-surface contact, but the contact property is softened. remove the residual penetration, but no convergence check
The softer the contact pressure-gap over-closure relationship, is required. Stability is maintained by limiting the time
the easier it is for the solver to converge. increment for each step to a very small value. The stability
H. Yu et al. / Journal of Constructional Steel Research 64 (2008) 515–525 517

Fig. 2. Comparison of analytical results with classical solutions.

3. Effect of loading speed on response

Explicit dynamic analysis gives the transient response of the


problem analysed. It is computationally inefficient for quasi-
static problems if real time is used, because the time needed
to finish an analysis is proportional to its duration. However,
it is often possible to scale the real time to a very small
time period if the response of the structure remains basically
Fig. 1. Finite element model for the Hertz contact problem. static. According to classical dynamic theory, when a dynamic
system is subjected to a linearly rising load, its response can
limit for the explicit time integration is approximately be approximately treated as static if the duration of the loading
estimated as the smallest transit time of a dilatational wave stage is large compared to the natural period of the system. For
across any of the elements in the mesh. It is proportional to complex structures, this statement gives no direct advice on the
the smallest element dimension in the mesh. exact time duration to use, but provides a general implication
that a time duration which has been shown to produce a quasi-
Explicit dynamic analysis is less often used than static
static response for one problem will be suitable for another
analysis for structural problems. This paper discusses some
with similar natural frequencies. In the following sections, two
issues concerned with simulating steel bolted connections using
simple connection problems are studied using a static solver
the explicit solver provided by the general commercial program
and an explicit dynamic solver with different loading durations,
ABAQUS. Methods for justification of the analytical results
to discover the appropriate time range for the complete loading
are proposed. The analysis method is validated by extensive
process.
comparisons with static solvers and test results.
3.1. Effect of loading speed on a simple bolt-contact problem
2. Validation of the contact algorithm
Bolts transfer forces through the connected plates by contact
The contact algorithm in ABAQUS/Explicit is validated with the bolt-hole edges or bearing of the head (or nuts) against
against a classical contact problem in which an infinitely long, the steel plates. In most connections, the stress state for the bolts
solid elastic cylinder rests on a rigid plane. It starts to bear is a combination of shear, tension and bending. In this section,
against the rigid plane when it is subjected to a downward two simple bolt contact problems are studied using both static
concentrated load P at the peak point. The Hertz contact analysis and explicit dynamic analysis. In the first example, the
theory gives the normal contact pressure between the elastic bolt is subjected to pure tension and transmits forces by bearing
cylinder and the rigid plane. The finite element model for of the head against the plate. Fig. 3(a) shows the geometry.
this problem is shown in Fig. 1. The infinitely long cylinder The plate is supported at both ends and the bolt is pulled at
is simulated by plane-strain 2D solid elements. Half of the the shank. In the other example, the bolt is subjected to shear,
cylinder (represented by a disk in the plane-strain setup) was by bearing of the shank against the hole circumference. This is
analysed in the finite element model due to the inherent shown in Fig. 3(b). The lower surfaces of both the plate and the
symmetry of the problem. A refined mesh was used for the shank are restrained against normal movement to create pure
contact region. The normal contact pressures taken from the shear at the plate interface.
output for each node in the contact region are plotted in Fig. 2 To make it easier for the static analysis to converge,
as points. The theoretical solutions for this problem are given displacement-controlled analysis was employed. The cross-
by Ju and Rowlands [12], and are reproduced in Fig. 2 using section of the shank was moved downward in the tension
solid lines. The results agree extremely well except at the edge case and sideways in the pure shear case. The resultant
of the contact region. However, the aggregate effect of this local reaction forces at the supports are plotted against the applied
discrepancy is very small. displacements in Figs. 4 and 5 for the tension and shear cases
518 H. Yu et al. / Journal of Constructional Steel Research 64 (2008) 515–525

(a) Model 1: The bolt in pure tension. (b) Model 2: The bolt in pure shear. (c) Upper view of the geometrical
model.

Fig. 3. Model to study simple contacts between bolt and plate.

Fig. 4. Comparison of results from static analysis and explicit dynamic analysis
for bolts subjected to tension.
Fig. 6. Comparison of results from static analysis and explicit dynamic analysis
for a lap joint test.

3.2. Effect of loading speed in a lap joint test

In this section, a lap joint test [13] is simulated in explicit


dynamic analysis with various loading speeds. The test involved
two plates, 9.525 mm thick and 12.7 mm thick, connected by
one bolt of 19 mm diameter. The bolt’s grade was ASTM A325
and the steel plates were A36. The two plates were clamped to
a testing rig and pulled against each other. Sarraj et al. [10]
simulated this test using ABAQUS/Static. A comparison of
different simulation results against the test results is shown in
Fig. 5. Comparison of results from static analysis and explicit dynamic analysis
Fig. 6. For the explicit analysis, if the loading is applied in
for bolts subjected to shear. 1 s or 0.1 s, the result is smooth and very close to the static
analysis result by Sarraj. When the loading is applied in 0.01 s,
respectively. Three loading durations, of 1, 0.1 and 0.01 s, were the response shows minor vibration in the initial linear phase,
used to assess the maximum viable loading speed. It can be but becomes close to the static analysis in the elasto-plastic
observed that, for both cases, loading times of 1 and 0.1 s give response phase.
smooth results which lie very close to those from static analysis.
The results from t = 0.01 s exhibited vibration, especially in 3.3. Discussion
the initial range when the structure was elastic and stiff.
These two examples are very simple contact problems. In The previous two sections have shown that, by controlling
the tension case, only one contact between the base material the loading speed, explicit dynamic analysis can be used to
and the head of the bolt was defined. In the shear case, two simulate quasi-static problems. There are two potential pitfalls:
contacts were defined; the interaction between the steel plate one is when too rapid an increase of load is applied, the
and the head of the bolt and the interaction between the hole structural response will basically be dynamic; the other is
circumference and the shank of the bolt. To make the results that the post-failure resistance may be overestimated because
comparable, hard-contact was used for both static and dynamic inertial force helps to balance the applied load. Neither problem
explicit analyses. However, convergence difficulties were found can be resolved simply, and a pragmatic approach needs to
in the static analysis, as a result of which the increment size had be taken. For the first problem, it may be concluded from
to be reduced to 1.0E−7 mm to achieve convergence. the previous two studies that, for normal steel connections, a
H. Yu et al. / Journal of Constructional Steel Research 64 (2008) 515–525 519

Fig. 8. T -stub for parametric study on mesh size.


Fig. 7. Parameters to validate a quasi-static analysis.
consequence will reduce the required time increment so that
loading time between 0.1 and 1 s should be appropriate for more steps are needed to finish the analysis. In this section, a
most applications. The softer the structure analysed, the longer parametric study will be performed on the effect of mesh size
should be the loading time. for bolted connections. In steel connections, the deformation
For the second problem, two methods can be used to is generally concentrated into three parts: the bolt deformation,
check if the structure has failed and the analysis is continuing the inter surface between the bolt and the connected plate holes,
simply as dynamic motion. The first is to check the ratio of and the yield lines of the connected plates. A T -stub is a typical
kinetic to internal energy. It is stated in the ABAQUS/Explicit component used in bolted endplate connections. It can be used
manual [14] that the procedure is quasi-static if the ratio of as an example, and parametric studies will be performed on the
the kinetic energy to the internal energy is less than 10%. Any mesh sizes of the bolt, the contact area around the hole and
responses which have an energy ratio larger than this should through the thickness of the bending plate.
be treated as dynamic and removed from the results. Another Fig. 8 shows the model for the parametric study. Due to
method is to check the equilibrium of the applied load and the symmetry, only half of the T -stub is shown. The symmetry
reaction force. When the response is static, the reaction force plane, the middle plane of the web of the T -stub, is restrained
should be approximately equal to the applied load. When the by symmetry about the x axis (designated by S–x in Fig. 8). A
structure reaches failure, the reaction force will drop because rigid plate, fixed in space, is put below the T -stub to simulate
the structure has become too weak to transfer the applied load the existence of another T -stub. A bolt is used to connect these
to the boundary. A plot of these parameters for the analysis of two T -stubs. In this model, only half of the bolt is modelled,
Fig. 13 mesh 1 (the details of the analysis are given later in with the middle plane of the shank restrained by symmetry
Section 4.1) is shown in Fig. 7. The estimated ultimate load P0 about the z axis (S–z in Fig. 8). A Grade 8.8 M20 bolt within
of 50 kN is applied in 0.1 s. The horizontal axis shows the step a 22 mm diameter hole is used. Three contact interactions are
time. The analysis stopped at 0.0759 s due to structural failure. defined: the bearing of the back of the T section against the
The dashed line shows the ratio of kinetic energy to internal rigid surface, interaction between the hole and the bolt shank,
energy. Some disturbance is noted before 0.04 s, probably due and the interaction between the bolt head and the surface.
to contact noise. According to the criterion that the analysis Interaction between the washer and head is ignored, and the
can be classified as quasi-static if the ratio of kinetic energy to diameter of the washer (36 mm) is used as the head size. The
internal energy is less than 10%, the analysis result is valid up to T -stub is pulled by its web until failure at a rate which would
0.0745 s. The ratios of the applied load and the reaction force to achieve 50 kN in 0.1 s. Element type C3D8R is used for all the
the ultimate load P0 are also plotted in the figure. Comparison elements.
of these two curves gives similar results to the energy ratio
curve. The reaction force, after some disturbance, starts to drop 4.1. Element numbers through the thickness of the T -stub
at 0.0745 s. The deformed shape of the structure at step time flange
0.074 s is inset in the figure. At 0.075 s, the structure has been
torn apart at the root of the T -stub. This study is designed to test the number of elements
through the thickness of the plate required to simulate
4. Effect of mesh size on the calculation time and structural accurately the bending of plates and capture the formation
response of yield lines. Fig. 9 shows the mesh for the T -stub when
four layers of elements are used through the thickness of the
Another factor that significantly affects the calculation time flange. For different cases, the number of layers of elements
and accuracy is the mesh size. An explicit dynamic solver is was changed from one to six. Meshing for the bolt was the same
more sensitive to the mesh size than other methods because as Mesh 2 in Fig. 13. Comparison of the T -stub deflections and
refined mesh not only increases the number of degrees of the time needed to finish each analysis are shown in Fig. 10.
freedom, but also decreases the size of elements, which in The response of the T -stub with increase of load can be divided
520 H. Yu et al. / Journal of Constructional Steel Research 64 (2008) 515–525

Fig. 11. Different meshes for the contact zone.

Fig. 9. Meshes in study of the effect of number of elements through the plate
thickness.

Fig. 12. Comparison of the deflections with various meshes for the contact
zone.

size in this area would affect both the formulation of the yield
line adjacent to the bolt hole and the contact between the plate
and the bolt head. Fig. 12 shows that all four results are very
close. Meshes 4 × 3 and 6 × 4 give about 5% higher ultimate
resistance than 8 × 6 and 12 × 8. While 8 × 6 will generally
give good enough results, 4 × 3 will give a satisfactory solution
if the requirement for accuracy is not very high.
Fig. 10. Comparison of the deflections of the T -stub when different layers of
elements are used through the flange thickness. 4.3. Mesh in the bolt

approximately into three phases. After the first linear phase, the In previous studies, this T -stub failed by plasticity of the
second phase starts when two yield lines have formed. In the flange under bending, and the bolt was not critical. Hence, two
third phase the yield lines have reached ultimate stress. One models were used to study the effect of mesh size on bolt
layer of elements obviously gives inaccurate results, whereas performance. In one model the thickness of the flange was
n = 2 to n = 6 give nearly the same results for the linear phase t p = 9.6 mm and the thickness of the web was tw = 3.3 mm
and very close results in the third phase. In the second phase, as shown in Fig. 8; this model would fail by forming yield lines
n = 2 and n = 3 give slightly lower resistance than n = 4 and in the T -stub flange. In the other model, the thickness of the
n = 6. Considering the time need to finish each analysis, it may flange was t p = 20 mm and the thickness of the web was
be concluded that n = 2 is the minimum requirement for the tw = 6.6 mm, so that failure of the bolt would happen. For
number of elements through the plate thickness, but n = 3 or each study, three mesh sizes were used for the bolt, as shown in
n = 4 is the optimum choice, and n > 4 will be unnecessarily Fig. 13. For these two models, comparisons of the deflections
time-consuming. given by different meshes are shown in Fig. 14. For Model 2, it
is clear that Mesh 1 and Mesh 2 give very close results, but
4.2. Mesh in the contact zone Mesh 3 is far from being accurate. For Model 1 the failure
is controlled by the T -stub, and differences in effect between
The contact zone is the 50 mm × 50 mm area around the different bolt meshes are not so obvious. Mesh 2 is very close
bolt hole. It is subdivided into four equal trapezoidal areas by to Mesh 1 until the last phase, in which it gave slightly larger
four lines extending from the centre of the bolt hole to the four resistance. The time needed to finish each analysis is also shown
corners of the square. In this study the mesh for each trapezium in the figure. Considering both accuracy and efficiency, Mesh 2
was changed as shown in Fig. 11. Each mesh is designated by would be the optimum choice.
two numbers, the first being the number of elements along the
side of the outer square, and the second being the number of 5. Verifications
elements from the inner circle to the outer square. The bolt
mesh remains as Mesh 2 in Fig. 13. Three layers of elements In this section, the validity of using an explicit dynamic
are used through the thickness of the T -stub flange. The mesh solver to analyse the behaviour of steel connections is verified
H. Yu et al. / Journal of Constructional Steel Research 64 (2008) 515–525 521

within a contact pair, setting both contact surfaces as master


surfaces can cause “contact noise”. Therefore, the surfaces with
coarser meshes and less expected deformations are generally set
as the absolute master surfaces.

5.1. Verification against T -stub tests at elevated temperatures

A T -stub is a simplified model which is usually used to


represent the tension zone of an endplate connection. In a
typical T -stub model, two T -sections are placed back to back
and connected by several bolts. When the tees of the two
T sections are pulled apart, the bolts are pulled. There are
interactions between the contact surfaces of the two T -sections
and between the bolts and the T -sections. An extensive number
of tests have been performed on T -stubs, both at normal
Fig. 13. Three mesh sizes for the bolt, and the mesh of the T -stub when temperature and in fire. Here, a group of tests with the same
t p = 9.6 (when t p = 20, four layers of elements are used through the thickness geometry, but tested at various temperatures, performed by
of the flange). Spyrou [6] are simulated.
The test setup and specimen geometries are shown in Fig. 15.
Due to symmetry, only one-eighth of the whole T -stub model,
as shown by the dashed lines, was subjected to finite-element
analysis. The two cutting surfaces were given symmetrical
boundary conditions. A rigid plane was placed at the back of
the T -stub to simulate the bearing from the other. The bolt
was cut across the middle of the shank, and the cut surface
was restrained by symmetry about the y axis. Three contact
interactions were defined: the bearing of the back of the T
section against the rigid surface, interaction between the hole
and the bolt shank, and the interaction between the bolt head
and the surface. Interaction between washer and head was
ignored and the diameter of the washer was used as the head
size in the finite element model. For all four analyses, the total
(a) Model 1, t f = 9.6 mm. loading time was 0.1 s.
The properties of the structural steel at normal temperature
are given in Fig. 15. The stiffness and strength decrease at
elevated temperatures. However, the reduction rate is very
complicated, depending on the temperature, heating rate, strain
rate, thermal creep, etc. Basically, there are two kinds of
tests which are used to determine the properties of structural
steel in fire. One is to load the specimen and then to heat
it at a prescribed rate; this is named a transient test. The
other is to heat the specimen to a specified temperature and
then to load it; this is called a steady-state or isothermal
test. The stress–strain relationships obtained from these two
types of testing could be significantly different. Within these
test categories, the stress–strain relationships may also differ,
depending respectively on either the heating rate or the loading
(b) Model 2, t f = 20 mm. rate. The material model generally used for structural steel
in fire is according to EC3: Part 1.2 [15] which is based on
Fig. 14. Comparison of deflections of the T -stub for various bolt mesh sizes. transient tests. The series of tests to be modelled are steady-
state tests, in that the specimens were heated to a constant
by simulating several types of connection tests. In all the temperature, and then loaded to failure. A trial series of
models, the solid element C3D8R has been used. The contacts simulations using the EC3: Part 1.2 material model found that
are defined as surface-to-surface contact with a finite sliding they did not agree with the test results. A series of steady-
option. ‘Hard contact’ is used for the normal contact behaviour. state test results for S275 steel performed at the University
In the tangential direction, a friction coefficient of 0.3 is used. of Sheffield [16] was used. The stress–strain curves at various
No prestress to the bolts is considered. It has been found that, temperatures, normalized with respect to the yield strength at
522 H. Yu et al. / Journal of Constructional Steel Research 64 (2008) 515–525

Fig. 17. Comparison of the analytical results with T -stub test results.

section, only the tests with one bolt row (Test A1) and three
bolt rows (Test A3) are simulated. The column section was
UC253 × 73, and the beam was UB203 × 25 for Test A1 and
UB305 × 25 for A3. The angle cleats were L100 × 8. In A1,
Fig. 15. Geometries and material properties of the T -stub test.
two rows of bolts were used to connect the angle to the beam
web. Descriptions of the test setup and the specimen geometries
can be found from Owens and Moore [18]. The finite element
models for the two tests are shown in Fig. 18. The steel
beam length was 650 mm, but only 200 mm is simulated. The
upper and lower flanges are neglected in the model, because
in regions close to the connection they are basically stress-free.
The distributed tensile force is applied to the beam web with the
resultant force in line with the centroid of the bolt group. The
column section is represented by one flange only. This column
flange is restrained along its vertical centre-line, simulating the
restraint provided by the column web. It was observed from the
Fig. 16. Stress–strain relationships for structural steel at elevated temperatures. test photos that washers were not used on the side of the bolt
head that connected the angle to the column flange. Therefore,
ambient temperature are shown in Fig. 16. Properties of bolts the hexagon shape of the bolt head is modelled explicitly.
at elevated temperatures were defined according to tests on Comparison of the finite element analysis results with the test
Grade 8.8 bolts reported by Theodorou [17]. The four tests were results is shown in Fig. 19. Two identical tests were performed
performed at 200 ◦ C, 420 ◦ C, 570 ◦ C, and 745 ◦ C respectively. within each category, and the test results were very close. Here,
The results from the finite element model are compared to the finite element analysis results are compared to A1/2 and
the test results in Fig. 17. The simulation and the test results A3/2. Except for the initial phase, in which the amount of
are generally in good agreement. The major reason for the prestress in the bolts and the properties of contact friction had
difference observed at 200 ◦ C is probably due to uncertainty important effects on the behaviour, the predicted responses are
in the material properties of steel at low, elevated temperatures. very close to the test results. Fig. 18 shows the deformations of
these two connections. In both cases, the web cleat underwent
significant bending deformations. In A1, the connection failed
5.2. Verification against web cleat connection tests at ambient by the bolt head on the column side punching into the web
temperature cleat. For A3, the connection failed by bearing of the beam web,
which is also shown in an exploded view in Fig. 18. These are
Double-angle web cleat connections use two angle sections in agreement with the test observations.
to connect a beam to a column. Bolts are generally used to
connect the angles to both the beam web and the column 5.3. Verification against flush endplate connection test at
flange/web. 11 tests were performed on this kind of connection elevated temperature
by Owens and Moore [18]. Various sizes of beam and column
sections were tested with between one and seven rows of bolts. Al-Jabri [19] performed three groups of tests on steel flush
These connections were subjected to pure tension forces that endplate connections. The test setup was the same for all three
were in line with the centroids of the bolt groups. Web cleat groups, but each group used a different combination of beam
connections use a large number of bolts, and finite element and column sections. The test setup, beam and column sizes
models become huge when more rows of bolts are used. In this and connection details for Al-Jabri’s Group 1 tests are shown
H. Yu et al. / Journal of Constructional Steel Research 64 (2008) 515–525 523

Fig. 18. Finite element models and failure modes for two web cleat connections.

to the connection as a whole, it was decided that only the


fire-exposed parts, including the connection region and the
adjacent 100 mm of each of the two beams and the column
would be constructed in the finite-element model. Because of its
inherent symmetry, the finite-element model included only one
quarter of the connection and connected region, and the meshed
model is shown in Fig. 21. Boundary conditions and loads were
carefully applied to ensure that the connections were subjected
to the same actions as in the tests. The bottom of the column
was vertically supported. The middle of the column web was
restrained against displacement U1 in the direction shown as
“1” and the symmetry plane normal to “2” was restrained
Fig. 19. Comparison of the analytical results to the test results.
against U2 . The test loads, which were applied 1.5 m from the
in Fig. 20. In the tests, the connections were loaded to a axis of the column, were represented as three components at the
certain load level and then heated until failure. The moment beam end, a tension force applied to the upper flange, an equal
capacity of this connection was calculated to be 20.9 kNm. A compression force applied to the lower flange and a shear force,
total of four tests were performed, in which the connections equal to the applied load, applied to the web. To ensure that the
were respectively loaded to 20%, 40%, 60% and 80% of end cross-section of the beam remained plane, a thick plate was
this calculated moment capacity. They were then heated at a attached to the end.
constant rate of 10 ◦ C/min. During the heating procedure, the All steels used in the tests were Grade 43. The Young’s
temperatures of the column web and flange, beam web and modulus, yield stress and ultimate stress are given in Fig. 20.
flange, endplate and bolts were monitored. Temperatures of the Since these were transient tests, the stress–strain relationships
parts of the connection were fairly uniform, lying within a range recommended by EC3: Part 1.2 were used for the material
between that of the beam lower flange and 14% higher. properties at elevated temperatures. The material properties
To build the finite element model for all the members and of the bolts at elevated temperatures were those defined by
connections in the test setup needs a large number of elements. Theodorou [17]. For Tests FB11 and FB12, 0.02 s was used
It can be seen from Fig. 20 that most parts of the steel to apply the initial static loading and 0.1 s was then used to heat
beams and columns were fire-protected, and the temperatures the model to 800 ◦ C (or 700 ◦ C for FB12). For FB13 and FB14,
of these protected members were recorded as around 37% of the initial static loading was applied in 0.04 s. A comparison
the temperature of the fire-exposed connection. As the fire- of the connection rotation–temperature relationships is shown
protected parts remained relatively cool and strong compared in Fig. 22. In the tests, the rotations of the connections
524 H. Yu et al. / Journal of Constructional Steel Research 64 (2008) 515–525

Fig. 20. Setup and details for steel flush endplate-connection test Group 1.

This could be due to some uncertainty about the exact material


properties at elevated temperatures. Using a thick endplate
instead of simulating the whole beam length may also make
the model appear stronger.

6. Conclusions

Understanding the behaviour of connections in fire is now


becoming very important, because connections are increasingly
subjected to complicated loadings in fire, as fire protection
schemes are optimized, and their failure can conceivably lead
to progressive collapse of the whole structure. Finite-element
analysis is an efficient and cost-effective tool to supplement
experimental investigations of connection behaviour. However,
finite element simulation of bolted steel connections is
challenging because large numbers of contacts exist in the
models, and these cause discontinuities to the stiffnesses, which
lead to numerical convergence problems for normal structural-
analysis solvers.
This paper has explored the use of an explicit dynamic
solver to analyse the behaviour of bolted steel connections.
Fig. 21. Indicative drawing of the finite-element model and 3D view of the
It has been suggested that, for typical steel connections, a
meshes. loading duration of 0.1–1s in the explicit dynamic analysis
should ensure a quasi-static response. The effects of varying
were measured both by displacements of the beams and by mesh sizes representing the bolts and the connected plates have
clinometers. At the same temperature, the difference between been studied. The responses were generally not very sensitive
the different measurements and the difference between the to the mesh size unless extremely coarse meshes were used.
rotations of the left and right beams both reached a maximum Finally explicit dynamic analysis was used to simulate three
of 5 millirads in the linear response phase and 20 millirads in sets of connection tests at ambient and elevated temperatures.
the failure phase. An average of the different measurements The analytical results were found to be in close agreement with
is used for the test results in Fig. 22. It can be seen that the test results. Given that use of an explicit dynamic solver
the finite-element model gives a good prediction of the initial allows the definition of very general surface contacts with large
deformation under static load, and the linear response. At deformations and avoids numerical convergence difficulties, its
higher temperatures, the finite-element simulation seems to application to analyse the behaviour of bolted steel connections
provide slightly higher resistance than the tests in all four cases. is recommended.
H. Yu et al. / Journal of Constructional Steel Research 64 (2008) 515–525 525

(a) FB11. (b) FB12.

(c) FB13. (d) FB14.

Fig. 22. Comparison of the numerical-+analysis results to test results.

Acknowledgement J Construct Steel Res 2005;61:689–708.


[8] Coelho Girao AM. Characterization of the ductility of bolted endplate
The authors gratefully acknowledge the support of the beam-to-column steel connections. Ph.D. Thesis. Universidade de
Coimbra; 2004.
Engineering and Physics Sciences Research Council of Great [9] Al-Jabri KS, Seibi A, Karrech A. Modelling of unstiffened flush endplate
Britain under Grant EP/C510984/1. bolted connections in fire. J Construct Steel Res 2006;62:151–9.
[10] Sarraj M, Burgess IW, Davison JB, Plank RJ. Finite element modelling
References of fin plate steel connections in fire. In: Fourth international workshop
structures in fire. 2006. p. 315–26.
[1] FEMA. World Trade Center. Building performance study: Data collection, [11] Lou GB, Li GQ. Nonlinear finite element modelling of behaviour
preliminary observations, and recommendations. Washington, DC & New of extended endplate bolted moment connections in fire. In: Fourth
York (USA): Federal Emergency Management Agency; 2002. international workshop “structures in fire”. 2006. p. 327–44.
[2] Newman GM, Robinson JT, Bailey CG. Fire safety design: A new [12] Ju SH, Rowlands REA. Three dimensional frictional contact element
approach to multi-storey steel-framed buildings. The Steel Construction whose stiffness matrix is symmetric. J Appl Mech 1999;66:460–7.
Institute; 2006. [13] Richard RM, Gillett PE, Kriegh JD, Lewis BA. The analysis and design
[3] Liu TCH. Finite element modelling of behaviour of steel beams and of single plate framing connections. Eng J AISC 1980;17(2):38–52.
connections in fire. J Construct Steel Res 1996;45(1):181–99. [14] ABAQUS analysis user’s manual version 6.5. ABAQUS Inc; 2004.
[4] Bursi OS, Jaspart JP. Benchmarks for finite element modelling of bolted [15] BS EN1993-1-2. Eurocode 3: Design of steel structures, part 1.2: General
steel connections. J Construct Steel Res 1997;43:17–42. rules — structural fire design. British Standards Institution; 2005.
[5] Swanson JA, Kokan DS, Leon RT. Advanced finite element modeling [16] Renner A. The effect of strain-rate on the elevated-temperature behaviour
of bolted T -stub connection components. J Construct Steel Res 2002;58: of structural steel. Research dissertation. University of Sheffield; 2005.
1015–31. [17] Theodorou Y. Mechanical properties of Grade 8.8 Bolts at elevated
[6] Spyrou S. Development of a component-based model of steel beam- temperatures. Master’s dissertation. University of Sheffield; 2003.
to-column joints at elevated temperatures. Ph.D. thesis. University of [18] Owens GW, Moore DB. The robustness of simple connections. Struct Eng
Sheffield; 2002. 1992;70(3):37–46.
[7] Maggi YI, Goncalves RM, Leon RT, Ribeiro LFL. Parametric analysis [19] Al-Jabri KS. The behaviour of steel and composite beam-to-column
of steel bolted endplate connections using finite element modeling. connections in fire. Ph.D. thesis. University of Sheffield; 1999.

Вам также может понравиться