International Symposium “Steel Structures:Culture & Sustainability 2010”

21-23 September 2010, Istanbul, Turkey

Paper No: 14

GUIDELINES FOR MODELING THREE DIMENSIONAL STRUCTURAL

CONNECTION MODELS USING FINITE ELEMENT METHODS

Serdar SELAMET1, Maria GARLOCK2

1
Graduate Student, Department of Civil and Environmental Engineering, Princeton University
2
Assistant Professor, Department of Civil and Environmental Engineering, Princeton University

ABSTRACT
Simple shear (fin plate) connections, which are designed to resist shear loads only, are commonly used in the
US. However, as observed in the Cardington large-scale building experiment, these connections carry large
compressive forces during the heating phase of a fire that can lead to local buckling of the connecting members
(i.e. beam). Further, large tensile forces develop near the connection during the fire decay which can lead to the
failure of the connections. The objective of this research is to provide guidelines and address common problems
to researchers in modeling three dimensional connection details using commercial finite element software such
as ABAQUS. Modeling such FE models, which consists of several parts in contact, requires knowledge in
contact mechanics with friction, meshing techniques, matrix solver and stability and convergence algorithms. In
recent studies, researchers have made several attempts to model and run double angle (web cleat) or single plate
connection models under a given fire load. A general consensus has been difficulty in setting up a proper contact
surface configuration and overcoming rigid body motions and convergence problems related to contact and local
buckling. With some examples from previous and ongoing research on simple shear connections at Princeton
University, we aim to give suggestions on how to improve convergence characteristics of such models by
selecting an optimum meshing level near contact areas, using stability methods and matrix solver techniques.
Steel connections under fire events have been the least researched yet crucial area in the structural engineering
and fire practice. Due to the high cost of conducting experiments of connections in a furnace, the finite element
method is a cost-effective way to investigate the strength and behavior of connections under fire. We have
observed that contact surfaces with edges or corners create convergence difficulties. Although using an explicit
solver might look like a better alternative to an implicit solver for large models, the results from an explicitly
solved solution could be unreliable and hence this technique requires careful attention by the user during post-
processing. Since an implicit solver requires the balance of forces for each iteration, the results are inherently
stable.
Keywords: Finite Element, Contact, Implicit, Shear Connection, Steel, Guidelines.

INTRODUCTION
The behavior of connections has become an important research field in predicting performance-based
design for steel frames especially after many researchers realized a knowledge gap in estimating the
behavior of steel structures under fire conditions. Highly non-linear behavior of steel under elevated
temperatures coupled with large deflections, fire-induced forces and local instabilities in the fully or
partially restrained beams make the finite element method (FE) a favorable choice in estimating the
ultimate load capacity of steel subassemblies. Among the various types of connections, simple shear
1
connections are considered to be most vulnerable to fire scenarios because they are only designed for
shear (gravity) loads and cannot fully resist the large axial forces in the beams and the rotational
demand during fire.
The objective of this research is to provide guidelines and address common problems to researchers in
modeling three-dimensional connection details using commercial finite element software such as
ABAQUS. Modeling such FE models, which consists of several parts in contact, requires knowledge in
contact mechanics with friction, meshing techniques, matrix solver and stability and convergence
algorithms. In previous research (Garlock and Selamet 2010, Selamet and Garlock 2010), we validated
a single plate shear connection that is used in Cardington full-scale building tests (Lennon and Moore
2003). FE model of this connection provided us valuable insight in possible challenges that researchers
from the same research area might encounter. We will first discuss about the previous benchmarks on
three-dimensional finite element modeling of steel connections and then present our guidelines and
supply several examples from our lap joint models or the single plate connection FE model from the
Cardington test.

GUIDELINES IN CONNECTION MODELING

Previous Approach:
Bursi and Jaspart (1997, 1998) have done extensive research on FE modeling of bolted end plate steel
connections. Their research provided guidelines not only to end plate steel connections but also to all
bolted steel connections. They suggested that 3D FE models are superior in estimating the connection
capacity, which leads to a more realistic global behavior of steel members (i.e. beams, columns). Thick
or thin shells are unable to produce acceptable results especially for the bolt behavior since bolts could
have varying stresses through their thickness and shells cannot accurately capture such stress state. Due
to the limited computing power a decade ago, simulating contact conditions between bolts, plates and
beams in a connection was a grand challenge. Therefore, Bursi and Jaspart limited their focus on pre-
loaded and not pre-loaded bolted endplate connections, which are at ambient temperature and loaded
with monotonically increasing loading. Since the authors’ main concern was bending dominated
problems at ambient temperature, they compared FE results to the experiments through moment-
rotation diagrams. Overall, FE results compared well with the experiments in terms of the initial
stiffness and the ultimate load capacity but they did not accurately capture the onset of yield strength
on which Bursi and Jaspart commented as the lack of residual stress representation in the FE models.
The authors also investigated discretization of the connection geometry, use of element types and
analysis category (elastic or plastic), the effects of friction (due to contact) in tangential direction
between connection parts (i.e. bolts, plates) as well as bolt pre-loading and prying force effects.
Bursi and Jaspart recommended the use of 3D first-order (linear) hexahedron elements with
incompatible modes (C3D8I) in ABAQUS. Linear elements are better for hyperbolic (plasticity)
problems, in which the strain yield is discontinuous. Each of these elements has 13 additional degrees
of freedom (DOF) to the existing 24 DOF, which provides superior performance in bending dominated
problems without having shear locking behavior or zero energy modes. The authors also suggested
using at least 3 elements through thickness of a section if the section’s behavior is bending dominated.
They calibrated beam elements and used this assemblage to represent shear and bending behavior of
the bolts. Moreover, they experimented with gap elements (as opposed to surface contact method) to
simulate contact conditions. The beam assemblage method for the bolts and gap elements for contact
conditions were utilized for the purpose of reducing the computational expense. The findings suggested
that changing the Columb friction coefficient for tangential contact between bolts and end plate does
not affect the rotational response of the end plate connection.
2
Van der Vegte and Makino (2004) investigated solution techniques to the structural problems involving
bolted connections. He commented on the use of implicit and explicit solution schemes for such 3D FE
connection models. Explicit method is usually used for dynamic problems because it determines the
solution without iterating but by explicitly advancing the kinematic state from previous increment.
Explicit problems do not need to form a global stiffness matrix because the linear equations are not
solved simultaneously for the entire system (like in implicit method) but the stress wave propagates
element-to-element (local). Implicit method steps in time by assembling global stiffness matrix for the
entire structure and inverting it to find all nodal displacements. Hence, the user could experience high
computational expense and converging difficulties since bolted connection models involving contact
conditions are highly non-linear. Finally, the authors recommend using explicit solvers for large
models with several parts in contact.

New and Enhanced Guidelines:

Our paper expands and enhances this previous research in several directions. First, it investigates single
bolt lap joints (tensile connections composed of 1 bolt and two plates) and simple shear connections in
a subassembly. Previous research (Bursi and Jaspart 1998) focuses on isolated connections, which fail
under monotonically increasing load. In our FE analyses, axial force, moment and shear develop
simultaneously due to fire and imposed boundary conditions on the subassembly. Second, our paper
considers highly nonlinear steel material behavior at elevated temperatures. Third, it uses state of the
art solving techniques and contact configurations provided by the finite element software ABAQUS
(DS-Simulia 2008).
Element type and integration order:
Whether the structure is at ambient temperature or under fire conditions, the FE models must be
modeled for plastic problems if the goal is to estimate the ultimate load capacity (limit state). Hence,
first-order elements should be used for bolted connection type problems in order to capture strain
discontinuities (yield lines). Quadratic elements such as C3D20 are more accurate for elliptic (elastic)
problems but they create additional difficulties when contact surfaces exist in the structure, because the
shape function on element edges is not linear. However, it is inefficient to use only C3D8I elements for
the entire model because each element has 13 additional DOF when compared to the fully integrated
elements (C3D8). When the elements are not in the contact zone or not expected to have large stress
concentrations, C3D8R (reduced integration) elements can be used to decrease the computational
expense. They are not suited for contact zone because they are inherently rank deficient and can switch
to zero energy modes. When these modes are triggered, the element can deform without any resistance
to the load. C3D8 elements are generally more accurate than C3D8R elements, but they are subject to
shear locking behavior, which can lead to overestimation of the load capacity in bending dominated
problems. They can be used in parts of the structure where local stress concentration is high but no
large bending is expected. The user should adopt hexahedron elements (C3D8, C3D8R and C3D8I)
where possible. If the user needs to represent irregular geometry such as fillet radius of an I-Beam or a
welded region, wedge elements (C3D6) can be used with care because the mesh density should be very
fine in order to get acceptable stress results.
Figure 1 shows how closely the FE model with C3D8I elements can capture (C3D8 elements
overestimate the load capacity because of shear locking behavior) the load-displacement history of
L8x5/8x4 angles (Garlock et al. 2003). The displacement is measured from the heel of the angle (uplift)
and the force is measured from the thick plate that is connected to the angle through 4 bolts. For
brevity, the load is applied for one cycle only until the deformation in the angle heel displaced by 1
inch. Since the angle in the experiment does not show any reduction in ultimate strength for each cycle
until it fails by fatigue, the FE model is loaded only once.
3
(a) L8x5/8x44 Angle (b) Looad-Displacement Plot for
f the Experiment and FEF model
F
Figure 1. Coomparison between
b Expperimental Cyclic
C Loadinng Angle Teest (Garlock et al. 2003)) and FE
M
Model (C3DD8I Elementss).

Meshing an
M nd steel matterial properties:
The finite element
e theoory states thhat the nummerical modeel will converge to the true solutioon if the
n
number of elements
e in that
t region gets
g finer. For contact problems,
p thiis also holdss true. Howeever, the
cost of commputation is always
a a conncern, and inntelligent meshing
m decissions must be b made. Sinnce high
stress conceentrations ex xist near thee connectionn, bolts, plaate- and beaam bolt-holee regions shhould be
m
meshed withh 20 to 24 elements
e aroound the circcumference of a typicall bolt diameeter (7/8’’ too 1 ¼’’).
This number is found by y mesh convvergence stuudies, which also agree well w to the suggestions
s b Bursi
by
and Jaspart (1998). If the t connectiion is designned for tenssile loading (see Figure 2), we recoommend
u
using 2 C3D D8 elementss through thee plate thickkness. If the connection should alsoo resist comppression,
local buckliing is a posssible conceern and the user shouldd use at least 3 elemennts though the t plate
t
thickness. P
Post-buckling g strength of a plate depends on thhe bending behavior
b of the elementts; hence
C3D8I elem ments are suiitable for coonnections under
u compreession. A sim mple single plate conneection as
shown in Fiigure 3a-b iss subjected to shear, com mpression, teension and bending
b throoughout the analysis,
a
h
hence we reecommend using
u C3D8 elements onn the contact zones, C3D D8I elementts for the beeam web
and beam flaange, the sin ngle plate annd bolts in thhe connectionn region andd C3D8R eleements outsidde of the
connection region
r (for th
he rest of thee subassembbly).
F
Figure 2 shoows the sim mplest tensilee connectionn, which is used
u to validdate the FE model befoore more
complex moodels are buiilt. The connnection conssists of one bolt b (M20) connected
c too 2 plates with
w 3/8’’
t
thickness. T bolt-holee is 1/16’’ laarger than thhe bolt diam
The meter (7/8’’). This is com mmon in enggineering
p
practice andd will allow the bolt movve relative to t the two pllates. The model
m is fixed in one plaate and a
displacemennt-controlled d loading is applied
a to thhe other plate until either bolt-hole bearing
b or boolt shear
limit state iss reached (Garlock and Selamet
S 20100).

4
(a) Lap Joinnt (b) Lap Joint Detail Cuut (Bolt is Puurple Color))
F
Figure 2. Finnite Elementt Model of thhe Lap Jointt Assembly (Garlock
( andd Selamet 20010).

(a) Beam-Giirder Subasssembly (b)) Single Platee Connectionn Detail

F
Figure 3. Finnite Elementt Model of thhe Cardingtoon Subassem mbly (Selameet and Garloock 2010).

For all partss (including bolts)

F b in the FE models represented in this papeer, steel withh isotropic haardening
is used for which
w the yielding is deefined usingg ‘von Misess’ yield criteerion. At eleevated tempeeratures,
w adopted the Eurocod
we de steel mateerial propertties (Eurocode 3, 2001).. For the single plate connnection
m
model, the single platee is welded to the girdeer (see Figuure 3b). How wever, we recommend
r that the
w
welded parts are modeleed using *TIIE Constrainnts option inn ABAQUS. This methood combiness the two
p
parts in all DOF at thee connected (welded) reegion and thhe region is assumed too fail only when w the
y
yielding strress level is reached. Such sim mplification significantlly reduces additional contact
configuratioon and conveerging challeenges due too a more com mplex weld material
m behhavior. Howeever, the
u
user is responsible to check
c if the total strain stays withinn the weld rupture limiits (~20% sttrain for
Eurocode 3)). Further, we w recommeend using trrue strain annd stress inpput for steell material prroperties
b
because the FE connecttion models experience large l plasticc deformatioons. ABAQU US and severral other
sophisticatedd commerciaal FE softwaare require trrue strain annd stress inpuut. Eurocodee steel propeerties are
given as enggineering strress-strain innput, which should
s be coonverted to true
t stress annd true (logaarithmic)
strain using Equations (1)( and (2) where
w eng and eng are the engineerring (nomina al) strain annd stress,
r
respectively y whereas truue and true are
a the true sttrain and stress, respectiively.
truue ln 1 eng e (1)

trrue eng 1 eng (2)

The importaance of such h formulationn is shown inn Figures 4aa-c. Two squuare bars, onne with true material
p
property andd the other with
w eng matterial properrty, are loade
ed in tensionn until they are
a fully plasstic. The
steel materiial has the (engineering
( g) yield streess of 303M MPa with isootropic straiin hardeningg up the
(engineeringg) ultimate stress of 470 MPa. Iff the engineeering stresss is used as a a direct input in
5
ABAQUS, the bar stretches near the fixed end, because the plastic stress ( eng) stays constant (Figure
4b) from 4% to 15% of plastic strain. Hence, the engineering stress/strain formulation causes a
reduction in the bar load capacity although no descending branch (ductile damage) is implemented for
the steel material. Further, it allows for an unrealistic ductile capacity of the steel material. As
expected, the true stress formulation gives realistic results with an ultimate (flat) load capacity until the
bar cross section becomes fully plastic.

(a) (b) (c)

Figure 4. Engineering versus True Stress/Strain Formulation in (a) 10 m Bars with 1x1 m Cross
Section in Tension with C3D8 Elements (Mises Contours Shown), (b) Plastic Strain/Stress and (c)
Load-Displacement Plots.

Contact configuration and some challenges:

The contact algorithm in ABAQUS/Standard checks for open or closed slave nodes on contact
surfaces. Open slave nodes are not in contact with the master node, thus they are unconstrained,
whereas closed slave nodes are constrained in the direction of the surface normal by the corresponding
master nodes. ABAQUS/Standard also determines if closed slave nodes are moving tangentially to
master surface (sliding or sticking). The tangential behavior is determined by the shear stress between
the master and slave surfaces. There are two methods to enforce contact constraints. The first method is
the traditional Lagrange multiplier method, which exactly enforces the contact constraints by adding
degrees of freedom to the global structure matrix. The second method is the penalty method, which
approximately enforces the contact constraints by use of springs without adding degrees of freedom to
the matrix structure. Using the penalty method, some penetration of contact surfaces is allowed, which
improves the convergence rate. We recommend using penalty method as both tangential and normal
contact surface enforcement. In our FE models, linear penalty stiffness formulation is used. The default
penalty stiffness is 10 times the underlying element stiffness; hence selecting the default value means
that the penalty scale factor is k=1.0. However, it is possible to scale the penalty stiffness in order to
decrease the computational time and avoid ill-conditioning of the matrix (DS-Simulia 2008). We
investigated the effect of the penalty stiffness by observing the load-deflection curve of the single-bolt
lap joint model from k=1.0 to k=0.001. As seen in Figure 5, the load-deflection curves with scale
factors k=0.1 and k=1.0 follow very closely meaning that the forces from one plate to another are
successfully transmitted through the contact surface in the bolt region. Hence, we recommend using the
penalty scale factor k=0.1.
Contact discretization is another important configuration. For surface-to-surface contact, contact
conditions are enforced in an average sense, rather than at discrete points such as node-to-node

6
discretizatioon. Such aveeraging techhnique proviides more acccurate and smooth conntact state trransition
and hence, wew recommeend surface-tto-surface foor all contactt surfaces.
Contact traccking algoriithm must be b defined in i ABAQUS S/Standard. A general but b computaationally
expensive algorithm
a is the finite slliding, which takes accoount for largge relative movements
m between
contact pairs compared to their elem ment sizes and
a updates their contacct tracking sttate for eachh contact
iteration. Smmall sliding algorithm
a dooes not allow
w large movvement betweeen surfacess and its calcculations
are done onnly once in the beginniing of the simulation.
s W recomm
We mend the finnite sliding tracking
algorithm because it is well suited for connecttion models especially under u fire coonditions wiith large
p
plastic deforrmations.

Figure 5. Looad-Displaceement Plot for

F fo Single Boolt Lap Jointt Model at am
mbient tempperature for different
d
p
penalty scale factors.

(a) (b) (
(c) (d)
F
Figure 6. Mises
M Stress Contours ono (a) Deforrmed Beam (b) Beam Bolt
B Regionn and Illustrration of
Contact Surfaces on (c) Bolts and (dd) Bolt-hole Regions in the
t Beam.

In static conntact analysees, care musst be taken to t ensure thee proper orieentation of the
t contact surfaces.
s
B
Bolts are ussually more rigid (quencch-tempered)) than hot-roolled steel members;
m hennce they aree usually
denoted as master
m surfaaces in contaact pairs andd meshed coarser than thhe slave surffaces to incrrease the
contact connvergence raate (DS-Sim mulia 2008)). As seen in Figure 6c, the bollt shank suurface is
p
perpendicula ar to the bollt head and bolt
b nut surfaace. Therefoore, these maaster surfacees will have different
d
surface norm mals, which define how w the slave nodes
n get inn contact witth the masteer nodes. Iff a slave
n
node (on thee bolt hole of the plate) gets
g in contaact with the master
m node that is at thee corner of bolt
b head
and bolt shaank, then two o different suurface norm
mals will imppose two connstraints on the
t same slavve node,
w
which will likely
l producce a numericcal singulariity (solution is not uniquue). Althouggh the analyssis could
7
continue, the slave node will experience unrealistic distortion and very large spikes in stresses. Figure
6a shows the deformed beam taken from our single plate connection model. The analysis stops
converging due to large contact force residuals at the node (shown in red circle in Figure 6b). In order
to avoid this problem, two master and two slave surfaces that are perpendicular to each other must be
separated as illustrated in Figures 6c and 6d, where the red color represents two master (or slave)
surfaces with different normals and the purple (Figure 6c) and the white (Figure 6d) color represents
the buffer zone which is not defined as a contact surface. This formulation avoids contact problems at
corners with a tradeoff in stress accuracy at those locations.

(b)

(a) (c)
Figure 7. (a) Triangular Loading Distribution for two Parts (red and grey) in Contact. Mises Contours
at the End of Analysis of (b) the Sharp Corner and (c) the Round Corner.

Another solution to convergence problems around the corners is to make the sharp corners round with a
given fillet radius. Such geometry is very common in hot-rolled steel members and it helps to define a
contact region with only one surface normal and increases the convergence rate as well as provides a
smoother stress transition from one contact element to another. Two FE models consisting of two parts
are created with a sharp corner and with a round corner with 10 mm fillet radius and as shown in
Figure 7a, the grey parts is loaded (displacement-controlled) with a triangular distribution where the
maximum displacement is 15 mm. The sharp corner produces element distortion and high stress levels
as seen in Figure 7b when compared to the round corner (Figure 7c). Moreover, the model with a sharp
corner completes the analysis with 134 iterations in 45 minutes (CPU time) whereas the model with a
round corner has only 107 iterations in 38 minutes. This computational efficiency plays a more
important role when the models are larger with more contact surfaces.
The initial step in contact analyses is usually the hardest step to establish convergence because the parts
are not fully in contact. Here, two parts are considered to be in contact when the contact pressure is
nonzero on the contact surfaces. For the analysis to complete the first step, rigid body motions due to
the lack of boundary conditions should be avoided. For instance, the bolts in our single plate
connection model will experience rigid body motions if they are not fixed in all three degrees of
freedom at some nodes in the first loading step. Such boundary conditions on the bolts need to be
imposed artificially to provide convergence in the initial phase. However, the user has to make sure
that the initial loading with these artificial boundary conditions is small enough not to affect the global
behavior of the connection in later steps. Once the loading ensures proper contact between parts
(nonzero contact pressure), the artificial boundary conditions are removed and the analysis is
8
continued. Furthermore, we recommend that the small loading in the initial step should be
displacement-controlled instead of force-controlled since they provide a higher numerical stability to
the system.
Solving techniques:
Solving large FE models with high nonlinearity involving contact is a crucial issue for most
researchers. Van der Wegte (2004) recommended explicit integration in time for connection models for
ease with contact configuration and convergence. However, he warned the users to carefully analyze
and interpret results. Explicit time integration is advantageous for large non-linear dynamic problems,
for which inertial effects are important and the time of simulation is measured in milliseconds.
However, 3D connection modeling at ambient or elevated temperatures is essentially a static problem.
Explicit methods could be used quasi-statically for connection modeling under ambient or elevated
temperatures, where the system produces kinetic energy (inertial forces) but this energy stays below a
certain threshold (~10%) when compared to the internal energy of the system throughout the
simulation. Explicit problems are also conditionally stable, because the stress wave propagation cannot
exceed the smallest element size (critical element characteristic length). Hence, the stable time
increment ( cr) is usually very small (refer Table 1) for finely meshed models. Using very small cr
makes it impossible to use a real time scale for quasi-static problems in Explicit method; instead a total
time (ttotal) 0.005 seconds is selected (usually 10-50 times larger than cr) such that the analysis
completes faster, but the noise due to contact and the kinetic energy in the system are kept minimal.
Implicit method does not have a time instability issue, but its computational expense grows almost
exponentially with larger degrees of freedom and it inherently creates convergence problems when
contact surfaces and high nonlinear materials are used in the model. We have included in Table 1 an
example of the single bolt lap joint model (shown in Figure 2a) to compare explicit and implicit
solution techniques.

Table 1. Implicit versus Explicit method of Single Bolt Lap Joint Model (see Figure 2a) using 64-bit
System with Intel Xeon E5345, 2.33GHz, Quad-core CPU (parallel processing), 32 GB of RAM.
Parameters IMPLICIT EXPLICIT
Number of DOF / number of iterations 104262 / 735 79158 / 207165
Contact Enforcement Penalty method Penalty method
Computational Expense (CPU time / Memory) 3.93 hrs / 565 MB 2.16 hrs / 136 MB
Stable time increment ( cr) N/A 1x10-7 sec

Total time (ttotal) 1.0 (time is irrelevant) 0.005 sec

Field equations do not The ratio of deformation speed to wave
Analysis End (Bearing of the Bolt-hole)
converge due to plastic failure. speed exceeds 1.0 for most elements.

Figure 8 shows the deformation of the bolt-hole in bearing for both implicit (Figure 8a) and explicit
(Figure 8b) methods at failure. Implicit method gives acceptable range of ductile deformation when the
bolt region becomes fully plastic and the field equations do not converge. Explicit method, however,
fails by unrealistic deformation of the bolt region, although the ultimate load capacity of the models is
approximately the same (not shown). The user should look for an acceptable deformation rate during
post-processing the results before the failure time is estimated. The implicit method does not require
such user intervention in the post-processing, since Mises contours (Figure 8a) show that the bolt
region has already become fully plastic at the end of the analysis. Investigating the pros and the cons of
two methods, we recommend using implicit solution scheme although it requires a more intelligent
contact configuration and greater computational expense.

9
 (a) (b)
Figure 8. Mises Contour and Deformation Plots of Single Bolt Lap Joint Models (ambient temperature)
with Bolt-hole Bearing Limit State using (a) Implicit Method and (b) Explicit Method.

CONCLUSION
Finite element software has evolved in the last decade to enable the researchers to implement 3D
topologies such as connections with user friendly graphical support. The researchers still need to
understand and overcome some of the difficulties that such models produce. In this paper, we discussed
about previous principles for FE modeling of connections and developed more recent and
comprehensive guidelines for connections under ambient or elevated temperatures in a subassembly.
We also supported our recommendations with original examples such as lap joints or single plate shear
connection from Cardington tests.

Acknowledgement
This research is supported by the National Science Foundation (NSF) under grant number CMMI
0756488. All opinions, findings and conclusions expressed in this paper are the authors' and do not
necessarily reflect the policies and views of NSF. The writers are grateful to Professor Panos
Papadopoulos (Department of Mechanical Engineering, University of California at Berkeley) for his
assistance with the ABAQUS and developing the finite element model.

REFERENCES
DS-Simulia, ABAQUS, version 6.8 (2008), Documentation, Providence, RI.
Bursi, O.S., Jaspart, J.P. (1997), Benchmarks for Finite Element Modelling of Bolted Steel
Connections, Journal of Constructional Steel Research, 43 (1-3), 17-42.
Bursi, O.S., Jaspart, J.P. (1997), Calibration of a Finite Element Model for Isolated Bolted End-plate
Steel Connections, Journal of Constructional Steel Research, 44 (3), 225-262.
Bursi, O.S., Jaspart, J.P. (1998), Basic Issues in the Finite Element Simulation of Extended End-plate
Connections, Computers and Structures, 69 (3), 361-382.
Eurocode 3 (2001), Design of steel structures Part 1.2: General rules structural fire design ENV 1993-
1- 2:2001, European Committee for Standardization, Brussels, Belgium.
Garlock, M.M., Ricles, J.M., and Sause, R. (2003), Cyclic Load Tests and Analysis of Bolted Top-and-
Seat Angle Connections, Journal of Structural Engineering, 129(12), 2003, 1615-1625.
Garlock, M.E., Selamet, S. (2010), Modeling and Behavior of Steel Plate Connections Subject to
Various Fire Scenarios, Journal of Structural Engineering, accepted for publication for July 2010.
Selamet, S., Garlock, M.E. (2010), Robust Fire Design of Single Plate Steel Connections, Engineering
Structures, accepted for publication.
Van der Vegte, G.J., Makino, Y. (2004), Numerical Simulation of Bolted Connections: The Implicit
versus the Explicit Approach, AISC-ECCS, Amsterdam, Netherlands.

10