Escolar Documentos
Profissional Documentos
Cultura Documentos
Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct
a r t i c l e
i n f o
Article history:
Received 13 June 2012
Revised 4 November 2013
Accepted 19 January 2014
Keywords:
Bridges
Connections
Finite-element analysis
Gusset plate
Nonlinear springs
Steel truss
a b s t r a c t
The investigation of the 2007 collapse of the I-35 W Highway Bridge in Minneapolis, Minnesota, used
very detailed nonlinear nite-element (FE) analysis. On the other hand, the Federal Highway Administration (FHWA) provided simple guidelines for the load rating of gusset plates, but load rating was never
intended to capture the actual behavior of gusset plates. The approach proposed here combines the accuracy of the rst method with the simplicity of the second. From the detailed FE analysis of a single joint,
the stiffness matrix of semi-rigid equivalent springs (linear in a simple model, nonlinear in a more
advanced model) was derived by applying forces and moments to the free end of each portion of member
(hereafter called stub member) that framed into the joint, one action at a time, while keeping the ends of
the other stub members xed. The equivalent springs were then placed in a global model, which was in
turn veried against a global, detailed FE analysis of the I-35 W Highway Bridge. The nonlinear equivalent
spring model was able to predict the correct failure mode. The approach was applied to a Howe truss
bridge as an example of performance prediction of bridges with semi-rigid connections, most of them
of one type. As the simplied spring model was developed from a detailed FE analysis of the joint considered, this approach would not be justied if all joints had to be modeled in detail. Examples where
the approach can be used include: structures where only specic joints need to be investigated (e.g.,
joints subjected to concentrated loads), and structures where the same joint model can be used repeatedly at multiple locations. In some cases, the effort required in performing detailed FE analyses of many
joints in order to develop simplied models can be justied if the simplied models can be used in subsequent multiple load cases, thus leading to overall computational savings. Under these circumstances,
the nonlinear connection model proposed here provides a simple and affordable way to account for connection performance in global analysis.
Published by Elsevier Ltd.
1. Introduction
The 2007 catastrophic collapse of the I-35 W Highway Bridge in
Minneapolis, Minnesota (I-35 W Bridge for short), under ordinary
trafc and construction loads, was triggered by the buckling of
an undersized gusset plate [1]. Gusset plates are complicated
structural components used to connect frame members such as
beams, columns and braces. Their use in buildings and bridges goes
back many decades and certainly predates the use of computers in
structural analysis and design. Practical design methods ensure
safety by providing a load path that satises equilibrium, boundary
conditions and does not exceed material yield limits. The resulting
stress eld is by denition a statically possible yield state of stress.
Safety against plastic failure is assured because, according to the
Corresponding author. Tel.: +1 301 975 4357; fax: +1 301 869 6275.
E-mail addresses: chiara.crosti@uniroma1.it (C. Crosti), dduthinh@nist.gov (D.
Duthinh).
http://dx.doi.org/10.1016/j.engstruct.2014.01.026
0141-0296/Published by Elsevier Ltd.
136
design analysis, which typically is linear and assumes rigid connections. Semi-rigid here means that the connection has a nite, dened rotational stiffness, as opposed to zero (hinged) or innite
(rigid) stiffness. The paper starts with a literature survey that focuses on approximate design and analysis methods, then follows
with a brief account of the I-35 W Bridge collapse and its investigation. Next, a simple linear equivalent spring model and a more detailed nonlinear one are developed from detailed FE analyses of a
joint, and used in a global model that, in the nonlinear case, correctly predicts the failure mode of the I-35 W Bridge. Finally, the
spring model is applied to a Howe truss bridge to show how it
can be used to predict the performance of bridges with connections
mostly of the same type.
eB
b
P; V B P; V C P;
r
r
r
q
2
2
a eC b eB
HB
HC
eC
P;
r
r
1
s
ry c 3
tb 1:5
EL2
where c = the shorter of the distances from the corner bolt or rivet
to the adjacent beam or column, E = modulus of elasticity, L2 = equivalent column length from the middle of the Whitmore width
(Fig. 1) and ry = yield stress.
The gusset plate is compact if its thickness t P tb and non-compact if t < tb. Dowswell and Barber [10] compared their theoretical
buckling capacities Pth with experimental and FE calculations in
the literature Plit. They used the average of the lengths from the
middle and the ends of the Whitmore width for the equivalent column length. For compact gusset plates, using an effective length
factor of 0.5, they found the ratio Plit/Pth = 1.47; and for non-compact gusset plates, using an effective length factor of 1.0, Plit/
Pth = 3.08. Thus, the separation of compact from non-compact gusset plates and the subsequent different effective length factors resulted in inconsistent and problematic factors of safety for design
against buckling.
Brown [11,12] developed analytical expressions for the edge
buckling of gusset plates, based on the elastic buckling stress of a
plate supported on its loaded edges, but otherwise unrestrained.
For edge buckling, the critical section bisected the long free edge
of length a and was perpendicular to the brace (Fig. 3). Its width
b was different from the Whitmore width. Only a fraction of the total brace load contributed to the edge buckling of the gusset, and
the rest was transferred directly to the steel frame. Brown
[11,12] compared her predictions with 18 experiments and produced 16 ratios of experimental to predicted values ranging from
1.01 to 1.38, and 2 values below 1 (0.99 and 0.81).
Yamamoto et al. [13] tested eight gusset plates (each connecting two horizontal chord members and two diagonal braces) and
performed FE analysis to establish stability design criteria for the
joints of the Warren truss designed to stiffen the deck of the HonshuShikoku suspension bridge. They focused on the development
of plastic zones, local buckling and ultimate strength, and found
that local yielding and local buckling preceded global buckling of
the gusset plates. The load at which local buckling started depended on the extent of yielding, which covered the inner portions
of the gusset plate, whose in-plane stiffness was constrained by the
surrounding elastic region. Under the assumption that buckling occurred when the stress reached the allowable stress in the material, ra = 0.58 ry, and with l1 = length of the vertical free edge,
Yamamoto et al. [13] proposed the following design thickness for
local buckling:
r
tcr 1:10l1
ra
E
137
particular, they focused on the effects of the splice plate and the
rotational restraint provided by the bracing member, but did not
formulate design equations or derive equivalent springs.
A widely used equivalent spring for beamcolumn connections
of braced frames was formulated by Richard and Abbott [15]. Their
nonlinear spring had one degree of freedom, and its momentrotation stiffness was represented by four parameters that modeled a
linear elastic part, a linear post-yielding, strain hardening part,
and a curved transitional part (Fig. 4 and Eq. (4)).
6
7
K Kp
M6
n 1=n K p 7
4
5h K CONN h
KK h
1 M0p
138
3. The length of the column is the average of three lengths extending in the direction of the framing member from the middle and
the ends of the last row of bolts to the edges of the gusset plate
or adjacent groups of bolts (L1, L2 and L3 in Fig. 1). This approximation is adapted from Thornton [17], who used the middle
length L2, but then went on to propose the average of L1, L2
and L3 as a more reasonable approximation of the critical
length of the column strip, with an effective length factor of
0.65, corresponding to a column with both ends xed. The
examples in the Guidance ignore any lateral constraint to the
gusset and use an effective length factor of 1.2, which corresponds to a column with one end xed and the other restrained
against rotation but free to translate. Brown [11,12] had previously proposed the factor of 1.2 based on experimental observation and calculations.
4. Detailed nite-element investigation
From the literature review, it is seen that there are simple design methods based on equilibrium and elastic behavior and proven safe by experiments. There is, however, no simple way of
calculating the actual behavior of a gusset plate, even in the elastic
range. Designers ensure that the connections are stronger than the
members, then proceed with a structural analysis that assumes rigid connections. Such a structural analysis is incapable of predicting connection failure, or account for the exibility of the
connection in the global behavior of the structure.
On the other hand, there exist detailed models such as the ones
analyzed by the NTSB. As part of the investigation of the collapse of
the I-35 W Bridge, the NTSB commissioned the FHWA, the State
University of New York (SUNY) at Stony Brook and the software
company Simulia to develop an FE model of the bridge. FHWA constructed a three-dimensional global model of the entire deck truss
portion of the bridge with (two-node, linear or cubic) beam and
(four-node) shell elements. The model, including the boundary
conditions at the piers, was calibrated with strain gage data from
a fatigue assessment conducted by the University of Minnesota
in 1999 [18]. In addition, as eld evidence pointed to gusset plates
as the trigger of the collapse, SUNY/Simulia developed detailed
models of the U10 (Fig. 6) and L11 nodes and incorporated them
into the global model. More details about the highway accident
and its analysis can be found in [1,19,20]. In [19] Hao provided a
possible explanation for why some of the gusset plates were under-designed: the main frame gusset plates and the upper chords
may have been designed from a one-dimensional model of the
bridge, i.e., a uniformly loaded beam supported at four points.
Each gusset plate of the detailed model was composed of four
layers of eight-node, linear, solid brick elements, whose largest
dimension in the plane of the gusset plate was 5 mm in highly
stressed regions and less than 15 mm elsewhere [20]. The
289,000 solid elements used reduced integration and hourglass
control to alleviate spurious displacement modes. The connection
model also included the ve main truss members, cut at 2/5 of
139
their lengths in the bridge (the so-called stub members), and modeled with solid and shell elements (Fig. 6). In their largest dimension, the shell elements ranged from 25 mm to 50 mm, and the
solid elements, which were in the transition zones between the
truss members and the gusset plates, from 8 mm to 25 mm. Transition between shell and solid elements was performed by surfacebased coupling constraints. All members in the connection model
other than the main truss members and gusset plate had maximum element dimension of about 15 mm. The sophistication of
these models, compared with Williamss [16] for example, reects
the extraordinary advance of computer and nite-element technologies in the last quarter of a century.
Even with such a detailed connection model, the rivets could
not be modeled. The shanks of the rivets had a radius of 13 mm
and the hemispherical heads a radius of 20 mm. The rivets were
modeled by coupling nodes of the fastened components normal
to the rivet axis within a radius of inuence of 13 mm. Contact
pairs, with a Coulomb friction coefcient of 0.1, were also used
with the fastener elements. Such a detailed analysis required advanced skills and powerful computers and was clearly beyond routine design.
Table 1
Computer models run.
Model
Software
Analysis
Purpose
1.
2.
3.
4.
5.
6.
7.
8.
9.
ABAQUS
STRAND
ABAQUS
STRAND
STRAND
ABAQUS
STRAND
STRAND
STRAND
Linear
Linear
Linear
Linear
Linear
Nonlinear
Nonlinear
Nonlinear
Nonlinear
Stiffness of 5 members
Stiffness of 5 springs
Stiffness of 5 members
Validity of rigid connection assumption
Behavior with various connection models
Stiffness of 5 members
Stiffness of 5 springs
Behavior with various connection models
Behavior with various connection models
Detailed connection
Spring connection
Detailed connection, various gusset thickness
Rigid connection
West main truss (2D)
Detailed connection
Spring connection
I-35 W (3D)
Howe truss (2D)
140
Table 2
Element 1, node 1, displacements d and rotations q obtained in ABAQUS for applied forces and moments (detailed model 1).
Node 1
dx (mm)
dy (mm)
dz (mm)
qx (mrad)
qy (mrad)
qz (mrad)
Fx = 444.8 (kN)
Fy = 444.8 (kN)
Fz = 444.8 (kN)
Mx = 228.8 (kN m)
My = 113.0 (kN m)
Mz = 161.0 (kN m)
1.84E1
8.19E2
1.88E5
4.19E6
7.83E7
7.43E3
8.19E2
3.50E0
3.39E4
1.22E5
1.94E5
7.09E1
1.88E5
3.42E4
2.10E+1
9.30E1
1.29E0
5.10E5
6.72E6
4.51E5
1.51E0
5.15E0
3.71E2
4.41E6
7.72E5
7.66E5
5.08E0
8.98E2
7.04E1
1.11E5
2.05E2
1.96E0
1.37E4
8.80E6
8.01E6
6.16E1
141
dx (mm)
dy (mm)
dz (mm)
qx (mrad)
qy (mrad)
qz (mrad)
Fx = 444.8 (kN)
Fy = 444.8 (kN)
Fz = 444.8 (kN)
Mx = 228.8 (kN m)
My = 113.0 (kN m)
Mz = 161.0 (kN m)
1.85E1
8.27E2
2.51E4
1.60E5
3.63E5
7.60E3
8.27E2
3.50E0
3.29E4
2.83E5
1.86E5
7.07E1
2.51E4
3.29E4
2.10E + 1
9.30E1
1.29E0
4.75E5
2.61E5
4.60E5
1.51E0
5.15E0
3.71E2
4.66E6
1.43E1
7.31E5
5.09E0
9.00E2
7.06E1
1.02E5
2.10E2
1.95E0
1.31E4
7.92E6
7.14E6
6.15E1
Table 4
Comparison of displacements d and rotations q of STRAND7/STRAUS7 model 2 with semi-rigid connection, model 4 with rigid connection and ABAQUS FE model 3 with various
gusset plate thicknesses [numbers in brackets are ratios to base case with thickness of 12.7 mm].
STRAND mod. 4 rigid joints
dx (mm)
dy (mm)
dz (mm)
qx (mrad)
qy (mrad)
qz (mrad)
0.184 [1.00]
7.42 [2.12]
12.8 [0.61]
3.12 [0.61]
0.759 [1.08]
0.802 [1.30]
0.185 [1.00]
3.50 [1.00]
21.0 [1.00]
5.15 [1.00]
0.706 [1.00]
0.615 [1.00]
15.87 mm
19.05 mm
25.4 mm
0.184 [1]
3.50 [1]
21.0 [1]
5.15 [1]
0.704 [1]
0.616 [1]
0.1748 [0.95]
3.276 [0.94]
18.54 [0.88]
4.688 [0.91]
0.6618 [0.94]
0.5998 [0.97]
0.1677 [0.91]
3.105 [0.89]
16.90 [0.80]
4.393 [0.85]
0.6302 [0.90]
0.5872 [0.95]
0.1575 [0.86]
2.860 [0.82]
14.83 [0.71]
4.021 [0.78]
0.5862 [0.83]
0.5682 [0.92]
Fig. 9. 2D model of I-35 W Bridge showing loads and support conditions. Circles and squares indicate support conditions (restrained displacements d and rotations q) for both
main trusses.
142
9. Results
143
Table 5
Ratio of calculated force to axial capacity for connection U10-W at load factor 0.92.
Connection element
Load ratio
Tension or compression
1
2
3
4
5
0.28
0.56
1.00
0.02
0.41
Compression
Tension
Compression
Tension
Tension
model. Various attempts to circumvent this difculty were unsuccessful and the literature review conrmed that previous simplied models were very limited in accuracy and use. This
simplied connection model is then most useful for structures that
use the same connection repeatedly, or only with changes in plate
thickness or material properties that can be accommodated easily
in the same detailed FE model; or for loadings that would cause the
analyst to focus on a few specic connections only; or for multiple
load cases, where a detailed model can be used efciently multiple
times. An example follows.
144
Fig. 16. (a) ABAQUS model showing failure of U10 under axial compression of diagonal member 3 and (b) inspection photograph of U10 [1].
Four models with large displacement formulations, elasto-plastic members and different connection models were considered:
9A: All joints rigid;
9B: one critical joint (at the concentrated loads) modeled by the
nonlinear, ve-element connection previously described, all
other joints rigid;
9C: all ve-element joints modeled by the nonlinear connection
elements previously described; and
9D: all ve-element joints hinged.
Nonlinear analyses (that accounted for both material and geometric nonlinearity) were run by applying an increasing load factor
to the concentrated loads only, with the uniform loads held constant. Fig. 18 presents the load factor versus midspan vertical displacement for the four different models considered. As expected,
model 9D (pinned joints), had the lowest load capacity, model 9A
(rigid joints) had the highest, and the others (semi-rigid joints) fell
in between.
10.1. Model 9A: Rigid joints
Model 9A reached the ultimate load factor of 3.26. Fig. 19 presents the ratio of bending moment to exural capacity in the members at the ultimate load factor. Since the joints were rigid, collapse
occurred because of exural yielding of the members.
145
the critical joint under the concentrated loads, and the other joints
remained elastic as the critical joint failed by exural yielding. Section 6 showed that the assumption of rigid behavior was a good
approximation for elastic joints, and the difference between models 9B and 9C was in the treatment of joints that turned out to remain elastic. If there were instead several heavily loaded joints,
then model 9C (with all joints semi-rigid) would predict a different
load distribution than model 9B (with only one semi-rigid joint) as
some of the heavily loaded joints begin to plastify. In this case, the
failure loads predicted by models 9B and 9C would be expected to
be different.
10.4. Model 9D: All ve-member joints pinned
Fig. 22 presents the ratio of bending moment to exural capacity in the members at the ultimate load factor of 1.46. For this model, failure was reached as soon as a plastic hinge was developed in
member 1 of the joint under concentrated loads.
Table 7 summarizes the results. Models that did not account for
realistic connection behavior could be quite erroneous, especially
when concentrated loads made a joint critical.
When structural members were assumed rigidly connected, the
predicted capacity could be quite erroneous if the assumption that
joints were stronger than members turned out to be incorrect. The
methodology proposed here made global analysis with proper
accounting of connection performance efcient and affordable.
11. Conclusion
Conclusion 15 of the NTSB investigation of the I-35 W Bridge
collapse stated [1]: Because bridge owners generally consider
gusset plates to be designed more conservatively than the other
Fig. 19. Model 9A, rigid joints: bending moment ratio at load factor of 3.26.
146
Fig. 21. Model 9B, axial forces in the ve-element connection at concentrated load.
Table 6
Ratio of calculated force to axial capacity for critical connection at load factor 1.98.
Connection member
Load ratio
Tension or compression
1
2
3
4
5
0.71
1.00
0.24
0.11
0.38
Compression
Compression
Compression
Compression
Compression
Fig. 22. Model 9D, pinned joints. Bending moment ratio at load factor of 1.46.
Table 7
Analysis of Howe truss with various connection models.
Model
Description
Failure mode
9A
9B
9C
9D
3.26
1.98
1.98
1.46
147
[8] Thornton WA. On the analysis and design of bracing connections. In: Proc.
National Steel Construction Conf. American Institute of Steel Construction,
Washington D.C.; June, 1991. p. 2633.
[9] American Institute of Steel Construction. Manual of Steel Construction. 14 th
ed. AISC, Chicago, IL; 2011.
[10] Dowswell B, Barber S. Buckling of gusset plates: a comparison of design
equations to test data. In: Proc. Annual Stability Conf. Structural Stability
Research Council, Rolla, MO; 2004.
[11] Brown VL. Stability of gusseted connections in steel structures. Ph.D. thesis.
University of Delaware, Newark, DE; 1988.
[12] Brown VL. Stability design criteria for gusseted connections in steel-framed
structures. In: Proc. An. Tech. Sess. on Stability of Bridges. Structural Stability
Research Council, St. Louis, MO; 1990. p. 35764.
[13] Yamamoto K, Akiyama N, Okumura T. Buckling strength of gusseted truss
joints. ASCE JSE 1988;114(3):57590.
[14] Cheng JJR, Yam MCH, Hu SZ. Elastic buckling strength of gusset plate
connections. ASCE JSE 1994;120(2):53859.
[15] Richard RM, Abbott BJ. Versatile elastic-plastic stress-strain formula. ASCE J
Eng Mech Div 1975;101(4):5115.
[16] Williams GC. Steel connection designs based on inelastic nite-element
analysis. Ph.D. dissertation, Civil Engineering and Engineering Mechanics,
University of Arizona, Tucson, AZ 85721; 1986.
[17] Thornton WA. Bracing connections for heavy construction. Eng J AISC
1984;21(3):13948.
[18] OConnell HM, Dexter RJ, Bergson P. Fatigue evaluation of the deck truss of
bridge 9340. Minnesota Dept. of Transportation Final Report; 2001.
[19] Hao S. I-35W bridge collapse. ASCE J Bridge Eng 2010;15(5):60814.
[20] National Transportation Safety Board. Structural and local failure study of
gusset plate in Minneapolis bridge collapse. Modeling group chairman nal
report 08119, NTSBC070010, Nov. Washington D.C. 20594; 2008.
[21] ABAQUS Analysis Users Manual v. 6.5. West Lafayette, IN; 2004. <http://
www.simulia.com/>.
[22] Crosti C. Improving the safety of steel bridges through more accurate and
affordable modeling of connections. Ph.D. dissertation, Department of
Structural and Geotechnical Engineering, Sapienza University of Rome,
Rome, Italy; 2011.
[23] Crosti C, Duthinh D. Instability of steel gusset plates in compression. J Struct
Infrastruct Eng: Mainten ManageLife-Cycle Des Perform; 2013. http://
dx.doi.org/10.1080/15732479.2013.786102 [on line 4.11.2013].
[24] Petrini F, Li H, Bontempi F. Basis of design and numerical modeling of offshore
wind turbines. Struct Eng Mech 2010;36(5):599624. ISSN:12254568, eISSN:
15986217.
[25] STRAND7/STRAUS7, Sydney, NSW, Australia; 2010. <http://www.hsh.info/>.
[26] National Transportation Safety Board (NTSB 2007). Loads on the bridge at the
time of the accident. Modeling Group Study Report No. 07-115. <http://
www.ntsb.gov/dockets/highway/hwy07mh024/385259.pdf>.
[27] Liao M, Okazaki T, Ballarini R, Schultz A, Galambos T. Nonlinear nite-element
analysis of critical gusset plates in the I-35W Bridge in Minnesota. ASCE JSE
2011;137(1):5968.
[28] Ocel JM, Zobel RS, White D. Prediction of gusset plate limit states. In: 29th Int.
Bridge Conf., Pittsburgh, PA; June 2012.
[29] Bontempi F, Gkoumas K, Arangio S. Systemic approach for the maintenance of
complex structural systems. Struct Infrastruct Eng 2008;4:7794. http://
dx.doi.org/10.1080/1573247060115 523. ISSN: 1573-2479.