BEHAVIOR AND ANALYSIS OF A CURVED AND SKEWED I-GIRDER BRIDGE

BIOGRAPHY
Ozgur Cagri is a Graduate Research Assistant at the Georgia Institute of Technology. His research interests include structural stability, finite element methods, and the design and behavior of steel and composite steel-concrete bridge structures. His doctoral research is focused on behavior and design of horizontally curved and skewed Igirder bridges. Donald White is Professor at the Georgia Institute of Technology. Dr. Whites research covers a broad area of design and behavior of steel and composite steelconcrete structures as well as computational mechanics, methods of nonlinear analysis and applications to design. Professor White has been an active participant in the background research and in the development of updated provisions for design of steel girder bridges in the 3rd and 4th Editions of the AASHTO LRFD Specifications.

SUMMARY
Curved and/or skewed steel girder bridges can experience significant three-dimensional deflections and rotations. Various methods and techniques are available for the design of these types of structures. In certain cases, simpler methods such as the V-load method or 2D grid analysis can suffice. In other cases, e.g., longer spans, more severe curvature and more severe skew, the use of more sophisticated 3D analysis methods may be prudent. This paper investigates the overall design behavior of an example curved and highly skewed I-girder bridge predicted by 3D FEA and 3D Grid models. The 3D FEA modeling approach applied in this study has been shown to accurately predict experimentally measured responses in prior research by the senior author. Hence, the solutions provided in this study can serve as useful benchmarks for evaluation of other simpler methods of analysis. The out-of-plumbness of the girder webs at the completion of the construction is investigated. These results are compared to predicttions from simplified hand calculations. The potential importance of geometric nonlinear (secondorder) effects during construction is discussed. The strength behavior of the bridge is investigated using a refined 3D inelastic analysis. The results from refined inelastic analyses are compared to elastic analysis and design calculations using the 4th Edition of the AASHTO LRFD Specifications.

Cagri Ozgur

Don White

BEHAVIOR AND ANALYSIS OF A CURVED AND SKEWED I-GIRDER BRIDGE

Cagri Ozgur and Donald W. White

Introduction
Tight geometric requirements are often placed on highway structures due to right-of-way restrictions in congested urban areas. Skewed and/or horizontally curved steel I-girder bridges are among the most economical options for satisfying these demands. Increasingly strict and complex site constraints are leading to bridge projects with longer spans, more severe curvature and more complex geometries. These characteristics exacerbate the inherent three-dimensional (3D) response of curved and skewed bridge structures. As a result, the behavior of these types of bridges needs to be better understood. The ability of various levels of analysis to capture (or account for) the 3D bridge responses needs to be studied in more depth, and the implications of various analysis and design approximations on the safety, constructability and economy of curved and skewed bridges need to be defined more clearly. 3D FEA methods generally are accepted as providing the most rigorous and theoretically the most accurate solutions. The prediction accuracy of these methods is demonstrated in previous research studies, e.g. (5). However, the sophistication of 3D FEA models can make them potentially more prone to inadvertent errors in design. Furthermore, there are numerous approximating assumptions that can influence the accuracy of the sophisticated analysis models. Simpler methods of analysis often can exhibit fewer of these difficulties, although the simpler methods can require special considerations of their own. Recent research studies have developed and demonstrated 3D grid analysis capabilities that also provide highly accurate predictions of curved I-girder bridge elastic responses (3; 6). It is desirable to compare the results from these more sophisticated types of grid analysis to refined 3D FEA solutions using the types of modeling approaches developed in prior studies (5). Carefully presented results from both of these approaches also can serve as useful benchmark data for various other more simplified analysis solutions. Possibly one of the most important attributes of curved and/or skewed I-girder bridges is the behavior of these systems during construction. For instance, the NHI Design Manual, LRFD for Highway Bridge Superstructures (8) provides a simple equation for estimating the layover of I-girders at skewed bearing lines. It is useful to understand the origins of this equation and to check it against the results from refined analysis solutions. Furthermore, composite I-girder bridge systems tend to be more flexible in their noncomposite construction condition compared to their final composite condition. These flexibilities can make the systems sensitive to second-order effects during construction. It is desirable to understand the significance of these potential sensitivities on representative curved and skewed I-girder bridges. One bridge attribute that can be important in the context of both of the above considerations is the influence of the method of detailing of the cross-frames on the bridge response. One of the types of detailing that is often preferred is called no-load fit detailing, or detailing for the girder webs to be plumb in the theoretical no-load condition (1; 3). With this type of detailing, the cross-frames are fabricated to connect perfectly to the girders with the webs plumb in their theoretical cambered no-load geometry. As a result, no additional stresses are locked into the system due to lack of fit. However, due to the torsional deformations of the bridge under its self weight, the girder webs will not be plumb in the final constructed geometry neither after erection of the steel nor after placement of the deck slab. For curved and/or skewed I-girder bridge systems, the prediction of the system displacements can be very important. Girder rotations must be checked against maximums that can be tolerated by the bearings, including additional rotations due to the subsequent bridge live loads. Also, the deflected slab positions must be checked against roadway alignment tolerances, tolerances at closure pours, etc. Furthermore, engineers and owners sometimes question whether web out-ofplumbness can have any adverse effect on the strength limit states.

Page 1 of 16

In addition to the above construction considerations, there still are areas where further design economy can be realized by improved characterization of the maximum resistances. One of these areas is the characterization of the resistance of curved composite I-girders in positive bending. Research studies have indicated that the true strength of curved composite I-girders in positive bending is close to the full plastic moment capacity of the cross-section (Mp) with some minor reduction for flange lateral bending effects (5; 9; 10; 11). However, the research and development teams involved with the creation of the unified AASHTO provisions opted to restrict the design of all curved composite I-section members to a yield moment (My) or flange yield stress (Fy) based resistance with some reduction for flange lateral bending effects. This restriction was implemented because of limited information about the effect of girder inelastic deformations on the overall behavior of horizontally curved I-girder bridge systems. The primary concerns involved the influence of inelastic redistribution from more heavily loaded curved I-girders on the validity of elastic analysis results. Of particular concern was the potential underestimation of the forces in cross-frames, since these members serve an essential role in curved bridge construction but may not respond in as ductile of a fashion as the I-girders if subjected to larger than predicted loads. The more recent studies (5; 9) have shown that the influence of the girder inelastic deformations on the system responses is minor up to a flexural resistance characterized by the equation. 1 Mu + f Sx f M n (Eq. 1) 3 where: f = flange lateral bending stress, M u = member major-axis bending moment, S x = elastic section modulus about the major-axis of the section to the flange under consideration taken as Myf /Fyf, and f M n = factored flexural resistance in terms of member major-axis bending moment. This paper summarizes the results from the studies in reference (9).

Description of the Study Bridge

The study bridge shown in Figures 1 and 2 was recommended by Mr. Dann Hall of BSDI Inc. This bridge has the following key attributes: One simply-supported span. Six horizontally curved I-girders with 8.5 ft girder spacing. Deck slab thickness of 7.5 inches, resulting in tributary widths larger than the slab effective widths based on the AASHTO (2007) rules. Staggered cross-frames. Significant skew angles with maximum values of 60.46 at the left bearing and 64.64 at the right bearing on the inside Figure 1. Plan view of the noncomposite structure. girder G6. Section transitions in the bottom flanges of Girder 1 (G1) and Girder 2 (G2) but otherwise all the girders are prismatic. Cross-frames detailed for the no-load fit condition. The yield stress of the steel is 50 ksi for the steel girders and 60 ksi for the slab reinforcing steel. The strength of the slab concrete is taken as 4.913 ksi. This is the average measured strength of the concrete from the experimental tests documented in reference (5).

Page 2 of 16

Figure 2. Composite bridge geometry. The girder flange and web dimensions are shown in Figure 2. Girder 1 changes from Section G1-1 to G1-2 at 39.76 ft from the left bearing line and it changes back to section G1-1 at 128.32 ft from the left bearing line. Additionally, Girder 2 changes from Section G2-1 to G2-2 at 39.62 ft from the left bearing line and it changes back to section G2-1 at 79.68 ft from the left bearing line. The radius of curvature of each of the girders, R, is provided in Figure 1. The arc span length ranges from 159.88 ft for G1 to 162.48 ft for G6. The maximum subtended angle between the cross-frames is 0.03, which satisfies the AASHTO (2007) limit of Lb / R 0.1. The slab is cast in place with a thickness of 7.5 inches except for the haunch over each of the girder flanges and the slab overhangs. Wood forms are used for the concrete casting. This simplifies the interpretation of the slab behavior from the analysis. The haunch thickness is 3.5 inches from the bottom surface of the slab to the bottom face of the top flanges. As shown in Figure 2, the total width of the deck is 50.88 ft. The roadway is composed of 3 lanes and has a width of 46 ft. The overhang width is 4.19 ft. Also, future wearing surface and parapet loads are considered. The parapets are 2.440 ft high with an average width of 1.583 ft. Regarding the displacement boundary conditions for the bridge, a guided bearing providing restraint in the radial direction is assumed at one end of girder G3 while a fixed bearing is assumed at the other end of G3. All the other bearings are assumed to allow free movement within the horizontal plan. All the bearings are assumed to allow rotation about any axis.

3D FEA Model
A finite element model of the bridge is constructed by using ABAQUS 6.5.1 (4). Figure 3 shows a perspective view illustrating the overall geometry of the bridge and the specified boundary conditions of the

Page 3 of 16

FEA model used throughout this study. Since the system is curved, a global cylindrical coordinate system is used to describe the boundary conditions. In the FEA model, the girder webs and concrete deck are modeled with the S4R element, which is 4-node quadrilateral displacement-based shell element with reduced integration and a large-strain formulation, and the S3R element, which is a compatible 3-node triangular shell element used for modeling some parts of the deck. The S3R is a degenerated version of S4R. Twelve shell elements are used through the depth of the web in all the girders. The number of elements along the girder length is selected such that the web elements have an aspect ratio close to one

Figure 3. Perspective view of the bridge FEA model and its boundary conditions.

A total of 53 shell elements are used across the width and an average of 161 shell elements is used along the length of the deck. Since the bearing lines are skewed, the number of the shell elements varies along the length of the deck. A coarser grid is used for modeling the concrete deck and the top flange of the steel girders compared to grid for the webs. The mesh density for the slab and the top flange of the girders in the length direction is one half of the girder web mesh density. The ABAQUS linear multi-point constraint is used for the mesh refinement. This command constrains each degree of freedom at middle nodes in the denser mesh to be interpolated linearly from the corresponding degrees of freedom at the adjacent nodes in the coarser mesh. Five integration points for the webs of the steel I-girders and nine integration points for the concrete deck are used through the thickness of the shell elements. The number of integration points is increased in the concrete deck in order to capture the progressive failure of the concrete deck more effectively in the subsequent nonlinear analysis (4). The EQUATION command is used to model the ideal full composite action between the top flanges and the concrete deck. Specifically, the EQUATION command is used to implement a constraint similar to the ABAQUS beam-type multi-point constraint. The composite slab is introduced in the noncomposite dead load configuration of the structure without inducing any deformations and internal stresses due to the noncomposite system displacements by using the EQUATION command. The flanges of the girders and the transverse stiffeners (bearing stiffeners, intermediate transverse stiffeners and cross-frame connection plates) are modeled using the B31 element, which is a two-node beam element compatible with the S4R and S3R shell elements. At the section transition points on G1 and G2, prismatic beam elements with a thickness equal to the average of the flange thickness on each side of the transition are used for two element lengths along the flanges. The bottom chord and diagonal members of the cross-frames are modeled with T31 truss elements while the top chords are represented by B31 beam elements to maintain the stability of the cross-frames in the direction normal to the plane of the cross-frames. The cross-frames in this study bridge were oriented with the middle vertex of the K in the top chord. Generally, it is more efficient in the final constructed geometry to locate the middle vertex in the bottom chord.

3D Grid Model
A 3D grid model is created using the structural analysis program GT-SABRE (2). All the girders are modeled with the B14DGLW element which is a displacement-based element based on thin-walled open-section beam theory. The B14DGLW element has seven degrees of freedom per node to capture the warping effects on the Page 4 of 16

member. Moreover, the diagonals of the cross-frames are modeled using T6 truss elements, which have three degrees of freedom per node. The top and bottom chords of the cross-frames are modeled with conventional beam elements, which have 6 degrees of freedom per node. Furthermore, the slab is modeled by a grid of conventional beam elements. The tributary width of the slab over each girder is represented by a longitudinal grid member, and the transverse actions of the slab are represented by Figure 4. 3D grid model view of the study bridge in the transverse grid members at each of the crossframe locations. The open-section thin-walled GT-SABRE Viewer beam element model of the girders and the slab grid model are connected by rigid offsets with idealized hinges at the girder top flange. The torsional and lateral bending rotations are released at these hinges. The displacements of the top flange, including lateral bending and warping, are constrained to be compatible with the slab grid model displacements throughout the girder lengths. Figure 4 shows a view of the study bridge in the GT-SABRE Viewer.

Material Properties
Two different sets of material properties are used in this study. The first one is used for the elastic bridge FEA and design checks. The second one is used for the full nonlinear FEA.

Material Properties for Elastic Model

The material properties for the elastic bridge FEA and design checks are as follows. A steel modulus of elasticity of Es = 29,000 ksi and a concrete modulus of elasticity of Ec= 3320 ksi are assumed. The material strengths are reported in the previous section.

Material Properties for Inelastic Model

The concrete damaged plasticity model formulated by Lee at al. (7) is used for the concrete constitutive description. This model is based on damage theory and the corresponding stiffness degradation. Tensile and compressive damage variables are established separately to account for different damage states. The multilinear representation of the concrete compression and tension stress-strain responses are shown in Figures 5 and 6 respectively.

Figure 5. Multi-linear representation of concrete compression stress-strain response (5).

Figure 6. Multi-linear representation of concrete tension stress-strain response (5).

Page 5 of 16

For the steel, J2-plasticity (von Mises yield criteria) and multi-linear isotropic hardening are assumed in the full nonlinear analysis. A multilinear stress-strain curve is fit to the true stressstrain results by defining four anchor points (9). Figure 7 shows the resulting multi-linear stressstrain curve. Table 2 gives the specific stressstrain data. For the slab reinforcing steel, an elastic-perfectly plastic stress-strain response is assumed. Table 2. Data points for multi-linear stressstrain response for steel members Figure 7. True stress-strain responses for the structural steel (Grade 50).
Point True Strain (%) Yielding 0.17 Onset of strain Hardening 1.12 Intermediate strain hardening 3.84 Ultimate Strength 11.65 True Stress (ksi) 50 50.38 66.31 79.56

Noncomposite Construction Condition

The noncomposite construction condition is defined as the stage where the steel section alone must resist the permanent component loads applied before the concrete deck is hardened. These permanent component loads include the weight of the steel components, wet concrete, slab reinforcing steel, forms and construction equipment. The study bridge is small enough such that it may be assumed that the slab is cast in one continuous stage. Therefore, staged casting does not need to be considered for this bridge. First, the importance of the analysis type is investigated by using the 3D FEA elastic noncomposite model of the bridge. The GRAV command is used in the FEA model to calculate the steel self weight. The weight of the formwork, slab reinforcing steel, wet concrete is applied to the top flanges as uniformly distributed line loads based on the tributary width of each girder across the cross-section of the bridge. The overhang bracket loads are modeled as equal and opposite radial forces that create a uniformly distributed torque on the exterior girders. The exterior girder (G1) flange lateral bending responses are increased because of the additional torque due to overhang bracket loads. To sum up, the total applied vertical load due to the weight of steel components is 465.07 kips and the total applied vertical load from the wet concrete, slab reinforcing steel and formwork is 939.56 kips. Second, the noncomposite construction results of the 3D FEA model are compared with the results of the 3D grid model. The steel dead weight is applied to the girders of the 3D grid model as uniformly distributed loads in GT-SABRE. The weight of formwork, reinforcement, wet concrete and construction equipment is applied in the same fashion as in the 3D FEA model. Also, the weights of the cross-frames are applied to the girders as point loads. Furthermore, eccentric overhang bracket loadings and parapet loadings are treated as uniform vertical loads and torques on the fascia girders. The twist angle of the outside girder at the completion of the construction is investigated, assuming fabrication of the crossframes such that they fit-up with the girders with their webs plumb in the no-load condition. These results are compared to the results from simplified hand calculations. Figure 8. Girder top flange deflections at a fixed bearing location. Web out-of-plumbness is caused by the skewed bearing lines as well as the horizontal curvature of the bridge. The cross-

Page 6 of 16

frames at the skewed bearing lines tend to rotate about their own skewed axis and warp (twist) out of their plane due to the girder rotations. Representative girder top flange deflections at a fixed bearing location are illustrated in Figure 8. The skewed cross-frame causes the major-axis bending and torsional rotations to be coupled at the girder end, since the in-plane cross-frame deformations tend to be small relative to the other displacements. Therefore, the layover of the girders at the skewed bearing lines can be calculated based on the compatibility between the girders and cross-frames, assuming negligible cross-frame in-plane deformations, as x = z Tan( ) (Eq. 2) where, x is the layover, z is the deflection of the top flange at the bearing due to the major-axis bending rotation and is the skew angle of the considered bearing. Correspondingly, the twist angle or out-ofplumbness of the girders can be calculated as

z = x Tan( )

(Eq. 3)

where, x is the girder major-axis bending end rotation and z is the girder major-axis bending end rotation. In the case of a non-fixed bearing, Equation 2 gives the relative displacement of the top and bottom flanges.

First-Order versus Second-Order Responses

It is worthwhile to compare the results from first- and second-order analysis of the noncomposite bridge under the construction loadings since the noncomposite system can be more prone to second-order effects compared to composite structure. Figures 9 and 10 show the bottom and top flange stresses obtained from geometrically nonlinear (second-order) and linear elastic (first-order) analyses under noncomposite dead and construction loads (DC1) for G1. The cross-frame locations are denoted by the lightercolor vertical lines in the plots. It is noted that the flange lateral bending stresses are largest at the cross frame locations and close to the mid-span of the unbraced lengths. Additionally, the major-axis bending stresses are not affected significantly by the geometric nonlinearity whereas the influence of Figure 9. Top flange major-axis bending and flange geometric nonlinearity is noticeably high for the lateral bending stresses along the length of G1 flange lateral bending stresses. The top flange lateral under DC1, comparison of linear elastic analysis bending stresses tend to decrease whereas the and geometric nonlinear analysis (Load factor = 1). bottom flange lateral bending stresses tend to increase when the second order effects are considered. Figure 11 presents the vertical deflections obtained from geometrically nonlinear and linear elastic analyses of the outside girder (G1) under un-factored noncomposite dead and construction loads. The maximum vertical deflection at G1 is 8.47 inches when the influence of geometric nonlinearity is considered and 8.33 inches when linear elastic analysis is conducted, an increase of only 1.7 percent. Figure 12 shows the radial displacement of the top and bottom flanges Figure 10. Bottom flange major-axis bending and along the length of the G1 as well as the radial flange lateral bending stresses along the length of displacements of the top flange relative to the G1 under DC1, comparison of linear elastic analysis bottom flange. Although, the both flanges show different behavior when the second order effects are and geometric nonlinear analysis (Load factor = 1) Page 7 of 16

Figure 11. Vertical displacement along G1 obtained from geometric nonlinear and linear elastic analyses under DC1 (Load factor = 1).

Figure 12. Top and bottom out-of-plumbness along the length of the G1 and their relative movement each other under DC1 (Load factor = 1).

considered, their relative movement is similar in both analysis types. It should be noted that the geometric nonlinearity is having a very minor effect on the G1 vertical deflections and out-of-plumbness whereas Figure 12 shows that the radial deflections of the top flange are substantially amplified (1.9 times larger at the normalized length of 0.36). Since the influence of geometric nonlinearity is important for the noncomposite dead and construction loadings in this bridge, the responses from geometric nonlinear (second-order) analysis should be combined with the responses from the subsequent linear elastic (first-order) analyses for the composite loadings in the subsequent design checks.

3D FEA VS 3D Grid Model Results

The responses from the 3D FEA and 3D Grid models are compared at 1.25 DC1 in this section. Figure 13 shows the vertical deflections of the noncomposite structure along the outside girder under 1.25 DC1. Also, the web out-of-plumbness for the noncomposite structure under 1.25 DC1 is shown in Figure 14. Moreover, the top and bottom flange stress distributions under 1.25 DC1 are presented in Figures 15 and 16 respectively. Figures 13 through 16 show that the variation in the girder responses between the 3D FEA and the 3D grid models is very small. The overall behavior of the cross-frames is discussed in the subsequent sections.

Figure 13. Vertical displacement along G1 which is obtained from different models (Load factor =

Layover of the Outside Girder

Table 3 compares the twist angle and the layover of the outside girder from Equations 2 and 3 and the refined FEA under DC1. It should be noted that the layover and twist angle predictions for the outside girder from Equations 2 and 3 are very close to the FEA results. The difference between results of the equations and FEA predictions is 1.2 % for the left bearing and 13.7 % for the right bearing. This difference is due to the cross-frame in-plane deflections.

Figure 14. Web plumbness along the length of G1 due to DC1 (Load Factor =1.25).

Page 8 of 16

Figure 15. Top flange major-axis bending Figure 16. Top flange major-axis bending and flange lateral bending stresses along G1 and flange lateral bending stresses along G1 due to DC1 from both models (Load due to DC1 from both models (Load factor=1.25). factor=1.25). Table 3. Comparison of the twist angle (z) and of the outside girder from the approximated equations and refined FEA under DC1.
Skew angle
Left Bearing Right Bearing 54.7 58.2 x 0.01555 -0.01555 z(Approx.) 0.0220 -0.0250 z(FEA) 0.0217 -0.0220

% Diff.
1.2 13.7

Layover(Approx.) Layover(FEA)
1.63 -1.85 1.61 -1.63

% Diff.
1.2 13.7

Strength Behavior of the Study Bridge

The strength behavior of the outside girder is investigated by conducting full nonlinear analyses using the 3D FEA model of the study bridge. These full nonlinear results are presented and compared with the elastic analysis responses at different load levels. Based on the assumption that the system is elastic and that the response under composite loadings is geometrically linear, superposition is valid. Therefore, separate analyses are conducted for noncomposite dead and construction loads (DC1), weight of parapets (DC2), wearing surface dead load (DW) and design live loads (LL). The resulting stresses and deflections are superimposed for the design checks. For checking the flexural resistance of curved composite I-girders in positive bending, the AASHTO (2007) provisions are restricted to the use of the form
fb + f 3 Rh Fyt

(Eq. 4)

where; = Tension flange major-axis bending stress. fb = Tension flange lateral bending stress. f = Specified minimum yield strength of the tension flange. Fyt = Hybrid factor Rh This equation is referred to below as the Fy-based 1/3 rule. However, for checking the flexural resistance of compact composite sections in positive bending, the AASHTO (2007) provisions use Equation 1. In this study, Equation 1 is used with Mn = Mp in order to assess the impact of the potential most liberal format of the AASHTO (2007) resistance equations. Equation 1 with Mn = Mp is referred to below as the Mp-based 1/3 rule. The ratio of the left and right-hand sides in Equations 1 and 4 can be used to check the adequacy of the girders relative to the AASHTO required loadings at a given STRENGTH load level. The critical section on the outside girder is determined by comparing the unity checks at various sections for the strength limit states under the STRENGTH I load combinations defined in AASHTO (2007). Figure 18 shows the critical section on G1 and the corresponding AASHTO live load configuration. This figure also indicates the most heavily loaded cross-frame. The critical AASHTO live load configuration for this cross-frame is slightly different

Page 9 of 16

than the one shown in Figure 18. However, the AASHTO live load configuration which is shown in Figure 18 is used to obtain the cross-frame forces. Table 4 compares the strength unity checks for the critical section based on the Fy-based 1/3 rule as well as the Mp-based 1/3 rule. The STRENGTH I unity check is reduced 36% for the critical section on G1 by use of the Mp-based 1/3 rule versus the Fy-based 1/3 rule. Table 4. Comparison of the strength unity checks for girder G1 based on the Fy-based 1/3 rule and the Mp-based 1/3 rule
Bottom flange STRENGTH I Strength
[f b +(1/3) f ] / f F n

Strength
[M u +(1/3) f S xt ] / f M p

Ratio of Strengths 1.355

1.172

0.881

Also, the AASHTO live load can be scaled to the level at which either Equation 1 or Equation 4 is satisfied. These load levels are referred to below as Fy-based 1/3 rule and Mp-based 1/3 rule load levels. Although the noncomposite structure is analyzed under DC1 to predict the initial camber, the camber is factored by 1.25 so that the girder top flanges are ideally located in a horizontal plane under this loading. This allows the slab to be instantiated in the finite element model in an initially flat position for the subsequent full nonlinear analysis studies. Furthermore, this exaggerates the influence of any girder twists from the initiallyplumb no-load fit condition due to the noncomposite dead load (since the factored dead load 1.25 DC1 is used rather than the actual 1.0 DC1). Figure 17 shows the cambers specified for each of the bridge girders.

Figure 17. Initial cambers of the bridge Figure 18. Critical section on G1, critical cross-frame girders along the normalized length (1.25 and corresponding live load. DC1). Figure 19 shows the vertical deflections versus the fraction of factored live load at the critical section on G1 at the different load levels. The vertical deflection of the critical section at the Mp-based 1/3 rule load level is

Figure 19. Vertical deflection at the critical section on G1 under different fractions of factored live load

Figure 20. Internal moments at the critical section on G1 under different fractions of factored live load.

Page 10 of 16

23.8 inches when the full nonlinear analysis is considered whereas it is 21.1 inches based on the elastic analysis. The internal moment at the critical section is determined by obtaining the nodal forces in the deformed FEA model and summing their contributions to the cross-section moment using the undeformed orientation of the cross-section. The full tributary slab width is used for these sections. Prior research studies (5) showed that the error associated summing the nodal forces over the undeformed cross sections was negligible. Figure 20 illustrates the internal moments at the critical section for the different load levels. Internal moments at the critical section start to decrease relative to the linear elastic results at 78 % of the factored live load due to inelastic redistribution. The total internal moment at the critical section on G1 deviates from the elastic predictions by 1.6% at the Mp-based 1/3 rule load level. Although, the moment redistribution increases with the increasing load, the nonlinearity is small up to the Mp-based 1/3 rule load level. Figures 21 and 22 present the normalized bottom flange major axis bending and flange lateral bending strains along the length of G1 at the Mp-based 1/3 rule load level. Since the bottom flanges of G1 are yielded, the strain results are shown for the bottom flanges. The maximum bottom flange major-axis bending strain is 2.95 times the yield strain. The maximum flange lateral bending strain is 2.1 times the yield strain. Significant yielding is observed near the mid-length of G1 at Mp-based 1/3 rule load level. Also, the major-axis bending and flange lateral bending strains peak at the cross-frame locations and mid cross-frame locations. Figure 23 shows the web out-of-plumbness of G1 at the Mp-based 1/3 rule load level. The skewed bearing lines and horizontally curved girders introduce torsional deflections and differential displacements in the girders.

Figure 21. Normalized major-axis bending strain along G1 bottom flange at Mp-based 1/3 rule load level

Figure 22. Normalized flange lateral bending strain along G1 at Mp-based 1/3 rule load level

Figure 23. Out-of-plumbness along G1 at Mpbased 1/3 rule load level

Figure 24. Longitudinal strains at the top of the slab along the width of the critical section on G1 at Mp-based 1/3 rule load level.

Page 11 of 16

Figure 23 illustrates these 3D effects on the girders. Moreover, Figure 23 shows that the web out-ofplumbness of G1 which are obtained from the full nonlinear and linear analyses match with each other. Most importantly, although the deflections become slightly nonlinear at the Mp-based 1/3 rule load level, still there is a good match between the full nonlinear and linear analyses. Furthermore, the slab top surface strains are plotted along a radial line across the width of the bridge at the critical section corresponding to the G1 Mpbased 1/3 rule load level in Figure 24. It should be noted that the accumulated strains are within the elastic limit of the defined concrete stress-strain response and are significantly less than the nominal concrete crushing strain of 0.003. In addition, the strains at the top of the slab vary in an approximately linear fashion across the slab width. Although the above major-axis and lateral bending strain plots give useful assessment of the regions subjected to higher strains, they do not show where the girder strains exceed the yield strain. Therefore, the normalized equivalent plastic strain variation along G1 at the Mp-based 1/3 rule load level is shown in Figure 25 to illustrate the spread of yielding in the steel sections. Equivalent plastic strains are monitored at the outside, inside tip and middle of the bottom flanges. The maximum equivalent plastic strain is 4.4 times the yield strain at the outside tip, 2.1 times the yield strain at the inside tip and 2.3 times the yield strain at the Figure 25. Normalized equivalent plastic strain along middle. All maximum strain values are observed at G1 at Mp-based 1/3 rule load level. the bracing locations. It is clear that there is significant yielding near the mid-length of the bridge and the yielding is greatest at several cross-frame locations on the outside tips of the G1 bottom flanges at the Mp-based 1/3 rule load level. Additionally, it should be noted that the entire bottom flange is yielded near the mid-length. Figure 26 presents the normalized major-axis bending strain at the bottom flange of the critical section under different fractions of the factored live load. Figure 27 shows the normalized bottom flange lateral bending strain at the critical section on G1 under different fractions of the factored live load. The major-axis bending strain for G1-S1 at the Mp-based 1/3 rule load level is 2.1 times the yield strain when the full nonlinear analysis is considered whereas it is 1.3 times the yield strain when the elastic analysis is considered. The flange lateral bending strain for G1-S1 at the Mp-based 1/3 rule load level is 0.5 times the yield strain when the full nonlinear analysis is considered whereas it is 0.08 times the yield strain when the elastic analysis is considered. It should be noted that the major-axis and flange lateral bending strains at the critical section which are illustrated in Figures 26 and 27 deviate from the linear predictions at the Mp-based 1/3 rule load

Figure 26. Normalized major-axis bending strain at bottom flange of the critical section

Figure 27. Normalized bottom flange lateral bending strain at the critical section on G1.

Page 12 of 16

level. However, this nonlinear variation is not significant relative to the linear predictions in the context of the overall response of the bridge. Figures 28 and 29 show the vertical reactions under different fractions of the factored live load. The vertical reactions show slight nonlinearity close to the Mp-based 1/3 rule load level (maximum of 5.7 % deviation). Nonetheless, it is clear that the variation in the girder reactions are predominantly linear up to the Mp-based 1/3 rule load level.

Figure 28. Vertical reactions at left bearing. under different load levels.

Figure 29. Vertical reactions at right bearing under different load levels.

The cross-frame highlighted in Figure 18 has the maximum bottom chord force among all cross-frames under STRENGTH I load combination. Figure 30 shows the axial forces in this critical cross-frame obtained from the linear elastic analysis, geometric nonlinear analysis and full nonlinear analysis using the 3D FEA model. It is observed from the Figure 30 that the axial force in the bottom chord is significantly larger in the full nonlinear analysis. The bottom chord axial force is 12.4% larger than in the geometric nonlinear analysis and 19.6% larger than in the linear elastic analysis at the Mp-based 1/3 rule load level. Figure 31 compares the axial forces in the critical cross-frame predicted by geometrically nonlinear 3D FEA and 3D grid models to the full nonlinear analysis 3D FEA results. It can be seen in Figure 31 that the bottom chord force in the critical cross-frame is 10% smaller in the full nonlinear analysis compared to the 3D grid model results at the Mp-based 1/3 rule load level. The bottom chord force predicted by the 3D grid model is 24.9% higher than that predicted by the geometrically nonlinear 3D FEA model at the Mp-based 1/3 rule load level. The difference between the geometric nonlinear analysis results of the two different models is believed to be due to the web distortional flexibility in the beam-shell FEA model. It appears from Figure 31 that the prediction of the Figure 30. Axial force variation at critical magnitude of the cross-frame forces can be influenced cross-frame from 3D FEA model. significantly by using the 3D grid or the 3D FEA models. Since the 3D grid model overestimates the bottom chord force, the use of the 3D grid model may be sufficient to estimate the cross-frame forces in general for this bridge. However, in other cases the results might be underestimated by using the 3D grid model. Furthermore, it is apparent from Figure 31 that the geometrically nonlinear 3D FEA model can significantly underestimate the true cross-frame forces at the Mp-based load level. Therefore, careful studies should be done to better understand the prediction of the cross-frame forces by Figure 31. Axial force variation at critical different types of analysis. If bridge designs are based on cross-frame from 3D FEA and 3D grid plastic moment based resistance equations, possibly Page 13 of 16

analysis factors similar to overstrength factors in seismic design should be introduced to guard against the underestimation of cross-frame forces by the different types of analysis. Nevertheless, the current AASHTO (2007) rules require the girders to satisfy the Fy-based 1/3 rule strength checks. One can observe from the various results discussed in the above that the study bridge is fully elastic for all practical purposes at the Fybased 1/3 rule load level.

Summary and Conclusions

In this study, various approximating assumptions that can influence the accuracy of 3D analysis models are considered. A detailed 3D beam-shell FEA model of a highly skewed and horizontally curved bridge is prepared to illustrate and investigate analysis and design considerations using the 4th edition of the AASHTO LRFD Specifications (1). The important considerations and key findings about the noncomposite construction stage are illustrated. Both linear (first-order) and geometric nonlinear (second-order) analyses are conducted on the noncomposite structure during construction. The ABAQUS 6.5.1 software (4) is used for these analyses. Stresses, moments and deflections at various sections are obtained throughout the bridge by superposing the DC1, DW, DC2 and LL analysis results. Additionally, the elastic FEA results are compared with the elastic results from refined 3D grid models implemented in the structural analysis program GT-SABRE (2). Layover and twist angle predictions from 3D FEA model are compared with simplified hand calculations. Refined full nonlinear analyses are conducted using the same finite element discretization employed in the above elastic FEA solutions. These models are used for investigating the strength behavior of the bridge. The full nonlinear analyses include the simulation of final construction dead load effects on the noncomposite structure followed by the effects of applied loads on the completed composite structure. The results from these FEA studies are considered with the results of prior studies to arrive at recommendations for the design of horizontally curved and skewed I-girder bridges. Key observations and findings from this study are as follows: It is observed that major-axis bending stresses and deflections are not affected significantly by the geometric nonlinearity whereas the influence of geometric nonlinearity is noticeably high for the flange lateral bending stresses and radial deflections. Since the influence of geometric nonlinearity is obviously important for the noncomposite dead and construction loadings in this bridge, the responses from the geometric nonlinear (second-order) analysis for these loadings are combined with the responses from the geometrically linear (first-order) analysis for the composite loadings in the design checks. It is noted that the variation in the girder responses between the elastic 3D FEA and the 3D grid models is quite small. However, the cross-frame forces show significant differences between two models. This is believed to be due to the web distortional flexibility in the beam-shell FEA model. The layover of the girders at the bearing lines can be approximated as:

x = z Tan( )
where;

(Eq. 8)

x z

= layover = deflection of the top flange at the bearing due to the major-axis bending rotation = skew angle of the considered bearing

If the girder end rotation can be determined accurately by a given type of analysis, the layover can be easily predicted by Equation 8. Both stress and moment-based flexural resistance equations are used for checking the strength limit states. The reduction in the strength ratio by using the AASHTO moment-based flexural resistance equation is determined for a number of critical sections for the subject bridge. The reduction in the strength ratio is 36% for the critical section on G1. This section has the maximum unity check among the critical sections.

Page 14 of 16

The load level based on Mn = Mp is referred to as the Mp-based 1/3 rule load level. Equation 1 is used with Mn = Mp in order to assess the impact of the potential most liberal format of the AASHTO (2007) resistance equations. The vertical displacements are found to be predominantly linear up to the Mp-based 1/3 rule load level. The results of linear elastic analysis can be used to obtain good estimates of deflections at this load level. Although, inelastic moment redistribution increases with increasing load, the total internal moments at the critical section on G1 deviates from the elastic predictions by only 1.6 % at the Mp-based 1/3 rule load level. At the Mp-based 1/3 rule load level significant yielding is observed near the mid-length of the bridge. The yielding is greatest at several cross-frame locations on the outside tips of the G1 bottom flanges. Also, the yielding is progressed on the inner tips and middle of the G1 bottom flanges. The top surface slab longitudinal strains across the width of the bridge at the critical section on G1 is within the elastic limit of the defined concrete stress-strain response and are significantly less than the nominal concrete crushing strain of 0.003 at the Mp-based 1/3 rule load level. In addition, these strains vary approximately in a linear fashion across the bridge width. The vertical reactions deviate slightly from the elastic predictions (maximum of 5.7 % deviation). The variation in the girder reactions is predominantly linear up to the Mp-based 1/3 rule load level. The cross-frames are critical load carrying elements in horizontally curved bridges. It is noted that the axial forces in the critical cross-frame are increased by as much as 12.4 % due to the girder inelasticity in the subject bridge. The 3D grid model overestimates the critical cross-frame bottom chord force from the refined inelastic analysis by 10 % whereas the elastic 3D FEA model underestimates the critical crossframe bottom chord force by 12.4% at the Mp-based 1/3 rule load level. The effect of using 3D grid versus the 3D FEA models should be studied more thoroughly in further research. The previous studies (5) conclude from various parametric studies that the nonlinearity is minor on the system responses up to the Mp-based 1/3 rule resistance level. However, the results of this study show that there is some significant deviation of the cross-frame forces from the linear analysis results at the Mpbased 1/3 rule resistance level on the subject study bridge. Thus, further study should be done with various parameters to further understand the degree of the nonlinearity on the results, particularly on the internal cross-frame forces, for loadings up to the Mp-based 1/3 rule resistance level. The current AASHTO (2007) approach of limiting the girder resistances to the Fy-based 1/3 rule appears to ensure that the overall response of the bridge is fully elastic for all practical purposes under the STRENGTH loadings.

Acknowledgements
The authors would like to express their sincere thanks to Dann Hall of Bridge Software Development International, Ltd. for providing this interesting bridge for study. Special thanks are extended to Michael Grubb of Bridge Software Development International, Ltd. for his valuable contributions and comments. This research was funded by Professional Services Industries, Inc. and the Federal Highway Administration. The financial support from these organizations is gratefully acknowledged. The opinions, findings and conclusions expressed in this paper are the authors and do not necessarily reflect the views of the above individuals, groups and organizations.

References
1) AASHTO (2007). AASHTO LRFD Bridge Design Specifications, 4th Edition, American Association of State and Highway Transportation Officials, Washington D.C. 2) Chang, C.-J. (2006). GT-SABRE Manual. Structural Engineering, Mechanics and Materials, School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, GA.

Page 15 of 16

3) Chang C.-J. (2006) Construction Simulation of Curved Steel I-Girder Bridges, PhD Thesis, School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta 4) HKS (2004), ABAQUS/Standard Manual Version 6.5-1, Hibbitt, Karlsson & Sorensen, Inc., Pawtucket, RI. 5) Jung S.K. (2006) Inelastic Strength Behavior of Horizontally Curved Composite I-Girder Bridge Structural Systems, PhD Thesis, School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta 6) Krzmarzick, D. P. and Hajjar, J. F. (2006). Load Rating of Composite Steel Curved I-Girder Bridges Through Load Testing with Heavy Trucks, Report No. MN/RC-2006-40, Minnesota Department of Transportation, St. Paul, Minnesota 7) Lee, J. and Fenves, G.L. (1998). Plastic-Damage Model for Cyclic Loading of Concrete Structures, Journal of Engineering Mechanics, ASCE, 124(8),892-900. 8) NHI (2007), Load and Resistance Factor Design (LRFD) For Highway Bridge Superstructures, NHI Course No. 130081A,U.S. Department of Transportation Federal Highway Administration, Publication No. FHWA-NHI-07-035. 9) Ozgur C. (2007) Behavior and Analysis of a Horizontally Curved and Skewed I-Girder Bridge, MS Thesis, School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta 10) White, D.W. and Grubb, M.A. (2005). Unified Resistance for Design of Curved and Tangent Steel Bridge I-Girders, TRB 6th International Bridge Engineering Conference, Transportation Research Board, 20pp 11) White, D. W., Zureick, A. H., Phoawanich, N. P., and Jung, S. K. (2001). Development of Unified Equations for Design of Curved and Straight Steel Bridge I Girders, Final Report to AISI, PSI, Inc. and FHWA, October.

Page 16 of 16