IABMAS2010, The Fifth International Conference on Bridge Maintenance, Safety and Management, July 11-15, 2010, Philadelphia, USA

Cable-loss analyses and collapse behavior of cable-stayed bridges

M. Wolff
Grassl Engineering Consultans, Hamburg, Germany

U. Starossek
Structural Analysis and Steel Structures Institute, Hamburg University of Technology, Hamburg, Germany

ABSTRACT: The general aim in designing structures, where the consequences of a collapse are high, must be collapse resistance. This means that no structural damage should develop that is disproportionate to the triggering event. Generally, structures can be made collapse resistant by ensuring a high level of safety against local failure or by designing for the failure of elements and thus increasing the robustness. Increasing the robustness of cable-stayed bridges is achieved by means of designing for the loss of cables. For this, quasi-static analyses using a dynamic amplification factor are recommended by guidelines. This paper shows the possibilities and limits of such an approach for cable-stayed bridges. Furthermore, collapse analyses of a cable-stayed bridge are conducted. With this, structural properties are identified which are responsible for collapse propagation. The prevailing collapse type is described and recommendations for the design of robust cable-stayed bridges are given. 1 INTRODUCTION The failure of one structural element can lead to the failure of further structural elements and thus to the collapse of large structural sections or the entire structure. In many cases, the initial triggering event and the resulting damage are disproportionate. Such collapses have frequently been discussed and investigated in recent years and are generally summarized under the term progressive collapse. But work in this field refers mainly to buildings. Collapse resistance means insensitivity to accidental circumstances. This can be achieved by ensuring a high level of safety against local failure or by using a design which allows for local failure. The structure's property of being insensitive to local failure is termed robustness (Starossek 2009). For cable-stayed bridges, collapse resistance is primarily achieved by increasing the robustness. The loss of cables must be considered as a possible local failure since the cross sections of cables have usually a low resistance against accidental lateral loads stemming from vehicle impact or malicious actions. The loss of cables can lead to overloading and rupture of adjacent cables. A collapse progressing in such a way is called a zipper-type collapse (Starossek 2007). Because the bridge girder is in compression, the loss of cables, which leads to a reduction of bracing, increases the risk of buckling. To create robust structures, it is necessary to know the collapse behavior of a structure. With this knowledge, structural properties can be identified which are responsible for collapse propagation. This paper examines the dynamic response of a cable-stayed bridge to the loss of one or more cables. Such analyses require a great amount of expertise and modeling effort. When designing for the loss of a cable, only the maximum responses are of interest. Therefore, quasi-static analyses using dynamic amplification factors to account for the dynamic effects can be conducted instead (PTI 2007). In the first part of this paper, dynamic amplification factors are determined, information is given on how to determine these factors and the limits of the quasi-static approach for cable-stayed bridges are outlined. In the second part of this paper, the collapse behavior of a cable-stayed bridge after the loss of cables is analyses. To trace the collapse progression following the rupture of one or more cables, geometric and material nonlinear dynamic analyses in the time domain are conducted. Hereby, critical elements are identified and the prevailing collapse type is described. Finally, recommendations for the design of robust cable-stayed bridges are given. 2 INVESTIGATED BRIDGE SYSTEM AND ITS MODELING The cable-stayed bridge being considered and which was the basis for a number of parameter variations is

shown in Figure 1. Two cable planes are placed in a modified fan arrangement with 80 cables in each vertical plane and a cable spacing of 15 m at deck level, apart from the closely spaced outermost backstay cables. The pylons consist of reinforced concrete. The bridge girder consists of a 21.60 m wide orthotropic steel deck, two 2.6 m deep longitudinal steel girders and cross girders spaced at 3.75 m apart. In the longitudinal direction, the bridge girder is only restrained by the cables. The numeric investigation is conducted using a three-dimensional model of the bridge. The pylon and bridge girder (longitudinal and cross girders) are modeled with beam elements. For investigations with nonlinear material behavior, a combination of shell and beam elements are used for the deck. The cables are modeled with a series of truss elements with distributed mass and self weight. The influence of cable sag and transverse cable vibration on the dynamic response could therefore be investigated. It shows that for the loss of one cable, this detailed modeling of cables leads to smaller dynamic responses (Wolff & Starossek 2009). The loss of a cable is investigated by nonlinear dynamic analyses in the time domain, taking into account large deformations. Firstly, the static initial state of the structure with the considered static load cases is calculated. The cable to be considered for failure is eliminated from the structural model and the corresponding cable forces are applied to the anchorage nodes of that cable as static loads. The time-history analysis is begun on this modified and loaded system at rest. To model the sudden loss of the cable, a step loading of the same size as the static cable force but acting in opposite directions is applied to both anchorage nodes. The numeric calculations are done using the finite element analysis program ANSYS. 3 DYNAMIC AMPLIFICATION FACTORS In the design, only the maximum dynamic responses to cable loss are of interest. Therefore, guidelines suggest a quasi-static approach which accounts for the dynamic effects by a dynamic amplification factor (DAF). For single-degree-of-freedom

systems, this factor is 2.0. According to the PTI Recommendations (2007), a force which is the cable force multiplied by this factor of 2.0 acting in the opposite direction must be applied to calculate the maximum responses due to cable loss. The EC 3 (2006) stipulates that the bending moments and forces due to static removal of a cable be multiplied by a factor of 1.5. This smaller value might consider the fact that a sudden failure of a whole cable this causes a higher response than a gradual reduction is unlikely. But the failure of a cable anchorage at the Cycle Arc bridge in Glasgow, the rupture of the main cable of a cable car in Cavalese due to a jet plane impact or the rupture of a cable at the Rion-Antirion Bridge due to a lightning strike tell a different story. The PTI Recommendations additionally allow the determination of a dynamic amplification factor in a nonlinear dynamic analysis, because it is assumed that, in general, smaller factors can be chosen for cable stayed-bridges. But how this factor is to be determined or which assumptions are to be made are not described. Investigations as to realistic ranges of dynamic amplification factors are rare. Single values are calculated for an arch bridge in Zoli & Woodward (2005), in a simplified manner in Hyttinen et al. (1994), and in Park et al. (2007). In the following, dynamic amplification factors are calculated separately for all state variables in all structural elements of the described bridge. The aim is to give advice on how to determine this factor, to prove if the use of a uniform amplification factor is valid and if reductions are generally possible. The method by which the DAF is determined is described in (Wolff & Starossek 2008). It was concluded that a unique dynamic amplification factor cannot be specified. Instead, the value is dependent on the location of the ruptured cable as well as the type and location of the state variable being considered. Very different dynamic amplification factors result if the rupture of one cable is considered (Wolff & Starossek 2009). Amplification factors at locations further away from the ruptured cable are high. At these locations, static responses are small. While the static removal of a cable mainly causes local deflections and bending

Figure 1. Structural system.

moments, the sudden removal of a cable excites natural modes with deflections and moments over the whole girder length. Thus, the mainly excited natural modes are not affine to the static deflection curve. The dynamic responses at locations further away from the ruptured cable are, however, irrelevant when considering all cable loss load cases, because only responses in the vicinity of the ruptured cable are design governing. Concerning the positive vertical deflections, a dynamic amplification factor of between 1.5 and 1.8 results at these locations. The amplification factor for the positive bending moments in proximity to the ruptured cable lies between 1.3 and 1.6, while that for the negative bending moments between 1.4 and 2.7. In Figure 2, the envelopes of extreme bending moments from all load cases are shown, together with the corresponding amplification factors. The reason for amplification factors smaller than 2.0 is that the maxima of all superposed eigenmodes do not occur at the same time in the considered time period. Figure 3 shows the contribution of the first 300 eigenmodes in time-history to the total bending moment at the location of a ruptured cable. For the sake of clarity, the contributions are not sketched individually but grouped. The sum of the maxima of these groups (encircled) is My = 18.3 MN which is higher than the maximum of the total response in this time range. The theoretical maximum develops at t = 300 s provided no damping is present. The dynamic amplification factor for the design governing dynamic axial forces in the bridge girder can be high (Wolff & Starossek 2009). But the increase in the girders total stresses due to the normal forces resulting from cable loss is small compared to the increase due to bending moments:

For the investigated bridge, the value is 2 %. The dynamic amplification factors for the design relevant cable forces which develop in the cables adjacent to the lost cable are between 1.35 and 2.0, depending on the lost cable being considered. Special attention is necessary for the bending moments in the pylons. The dynamic amplification factor for the bending moments over the whole pylon height and for all cable losses is significantly higher than 2.0. At the pylon base, values of about 30 for negative and about 8 for positive moments occur. The static moments are small. However, the dynamic bending moments in the pylon are significant (Fig. 4). Here too, higher modes which are not affine to the static deflection curve are excited. Furthermore, the pylon is not only excited by the step loading of the failing cable, but each of the cables induces irregular forces which are composed of the redistributed loads from the failed cable plus the inertia forces from the vibrating bridge deck. (The anchorage points are evenly distributed over a length of 26 m at the pylon head.) These forces cause a complex structural response which cannot be simplified as described above. The results show that in the present case, dynamic amplification factors smaller than 2.0 are only possible for the bending moments in the bridge girder, since over wide parts, the values are smaller. Here, an explicit calculation of the DAF can be beneficial. For the safe design of the cables, a dynamic amplification factor of 2.0 is necessary. For the bending moments in the pylons, high dynamic forces occur which cannot be safely accounted for by a quasi-static analysis using amplification factors. In particular, when the static removal causes a decrease in responses while the dynamic removal causes an increase, quasi-static analyses cannot yield

Figure 2. a) Dynamic amplification factor (DAF) for positive () and negative () bending moments; (b) extreme bending moments in longitudinal girder in the plane of cable rupture due to permanent loads and cable losses (envelope)

Figure 3. Contribution of first 300 eigenmodes in time-history to total bending moment at the location of a ruptured cable (= maximum bending moment) Figure 4. Extreme bending moments in one pylon leg due to

amplification is nearly independent of the type of calculation. Therefore, it is also independent of the type of additional loading. Thus, different live load positions need not be considered. In (Wolff & Starossek 2008), the influence of damping on the structural response after cable loss is described. The effect of damping depends on the occurrence of the maximum responses in timehistory. As stated before, the theoretical maximum responses occur very late in time history when even a very small damping has stopped any vibration. The considered time-range for calculating the dynamic amplification factor was chosen in such a way that damping reduces the amplitudes by at least 25 % by the end of this time range. Design relevant maximum deflections and bending moments of the bridge girder and maximum cable forces develop early in this time range. Therefore, a damping has only a small effect on these maximum responses and the impact on reducing the dynamic amplification factor is small. The bending moments in the pylons, however, are significantly reduced, even by a small damping ratio of = 0.2 %. However, the dynamic amplification factors are still much higher than 2.0. 4 IMPORTANT AND CRITICAL ELEMENTS

permanent loads and cable losses (envelope)

correct results. Dynamic analyses seem vital here. The results presented in this paragraph are for the undamped system under self-weight without live loads. The dynamic amplification factors are determined in nonlinear analyses. However, it turns out that even if nonlinearities cannot be neglected in calculating the maximum responses, the dynamic

The effort of determining maximum responses due to cable loss can be reduced by minimizing the number of investigated cable-loss load cases. This is possible if the important elements of the structure are known, which are defined as those elements whose failure cause the highest responses. For the investigated bridge, the increase in cable stresses is highest when a short cable fails. The design governing stresses in the bridge girder are

due to the loss of a long cable near the center of the bridge. These results might be applicable to other cable-stayed bridges with closely spaced cables because the reason for this is the bridge girders elastic support by the cables: in the range of short cables, the support is stiffer because of the larger inclination and stiffer force-deformation behavior of short stay-cables. Thus, the loads are transferred to few cables. The soft support in the range of long cables causes a load distribution to more cables but at the expense of higher bending moments in the bridge girder. Details are given in (Wolff 2010) For the maximum bending moments in the pylons, the loss of long cables has to be analyzed, because the induced horizontal forces are highest. But this cannot be generalized, because resonance effects also play a role. In addition to the important elements, critical elements are defined which are elements whose failure leads to a disproportionate damage. Thus, they are dependent on the definition of the term disproportionate. Furthermore, the additional loading is important. Depending on their location in the structure, the number or size of critical elements can be different. For the investigated bridge, the loss of three adjacent short cables is critical, as explained in Section 5. 5 COLLAPSE BEHAVIOR To trace the collapse progression after an initial failure of one or more cables, geometrically and materially nonlinear dynamic analyses are necessary. The material behavior of the bridge girder and the cables is assumed to be elasto-plastic. In the collapse state, cable stresses can be significantly higher than in the initial state but also a slacking is possible. For these high stress variations, the cables forcedeformation behavior is highly nonlinear. Taking into account cable sag and transverse cable vibration therefore becomes crucial. To account for the exact stress distribution in the bridge girder, a combination of shell and beam elements is used. The element size is adapted according to the strain gradients. For the investigated bridge, the failure of one single cable does not lead to collapse progression. (Unfactored live loads are placed in the most unfavorable position.) Only local plastifications develop and deflections are not significant. Also, the cable tensions remain comparatively small. Live loads can be increased by a factor of three until ultimate state is reached. The minimum number of cables which have to fail for total collapse of the bridge are three adjacent short cables. In Figure 5, the initial state with the intact critical cables is depicted. After the sudden loss of these cables, vertical deformations with

plastic regions first begin to develop in the longitudinal girder of the damaged cable plane (in the front part of the figure). Thereby, the normal forces acting on the whole section of the bridge girder are transferred to the longitudinal girder of the intact cable plane (at the rear of the figure) where vertical deflections are small. Although this girder is continuously supported by the cables, it cannot resist these high normal forces and begins to buckle in the vertical direction. From this moment on, vertical deflections grow strongly and cannot be arrested since the bridge deck is not restrained by fix supports in the longitudinal direction. Ultimate stresses in the bridge girder are exceeded. During this process, the upward deflections and the missing restraint in the longitudinal direction cause a slacking of some cables which can lead to them disengaging from their anchors. The downward deflection of the longitudinal girder of the intact cable plane finally causes the rupture of the cable at this location. This state of the bridge deck is shown in Figure 6. The collapse state is sketched in Figure 7: due to the longitudinal motion of the bridge deck towards the damaged region, normal forces are transferred to the other, intact half of the bridge with the second pylon (not shown in the figures). This leads to high unbalanced cable forces in both bridge parts which in turn results in high bending moments in both pylons. The continuity of the bridge girder thus causes the failure of both pylons which ultimately leads to the total collapse of the bridge. The collapse can only be arrested by providing an alternate load path for the normal forces in the bridge girder. Therefore, an alternative system was investigated where the bridge deck was horizontally fixed at the ends of the bridge or alternatively at the pylons. But the results show that high forces of 40 MN develop indicating that horizontal bearings are not a good measure for increasing the robustness.

Figure 5: Von Mises stresses in the bridge girder due to permanent and live loads prior to loss of cables (one half of the bridge)

Figure 6: Von Mises stresses in the bridge girder at collapse state due to loss of three cables, permanent and live loads (for a better illustration, the pylon is omitted), -- slack cables, ruptured cable

Figure 7: Final collapse state for the failure of critical elements, scheme

max w = 3220 cm

Figure 8: Deformed state after the loss of 10 long cables (von Mises stresses)

The structure is more robust regarding the loss of long cables. This is due to the fact that at the anchorage points of the longer cables, the bridge girder exhibits smaller normal forces and therefore the increase in bending moments due to second order effects is small. Furthermore, the support of the bridge girder at the bridge center is softer than at the pylons. Thus, the forces from the failing cables are distributed to more cables. Figure 8 shows the deformed state after the loss of five cables in each cable plane. The bridge will finally collapse due to high bending moments in both pylons, since additionally to the high vertical deformations, high horizontal deflections towards the bridge center occur. 6 COLLAPSE TYPES According to Starossek (2007), six types of structural collapse can be distinguished: five pure collapse types and one mixed-type collapse. Each collapse type can be characterized by a propagating action which, after the failure of one element, leads to the failure of the next element. The study of propagating actions can give insights into the structural properties which promote collapse propagation. The collapse propagation for the failure of the critical elements, described above, does exhibit features of at least two of the pure collapse types and must therefore be categorized as mixed-type collapse. At the beginning, there is the initial failure of elements which are responsible although not primarily - for the stabilization of the bridge girder in compression. Lack of bracing then leads to an increase in vertical deflections and high stresses due to second order effects, firstly in the longitudinal girder of the affected cable plane and then in the second longitudinal girder (a more detailed analysis is given in Wolff (2010)). These are features of the instability-type collapse although the failure is not a pure buckling. The compression force is thus responsible for the onset of collapse propagation and the failure of the bridge deck. After the failure of the bridge deck, the normal forces in the bridge deck, resulting from the horizontal cable forces, are redirected to the intact bridge part utilizing the decks tension resistance (Figure 7). The forces are transferred to the pylons by the main spans cables (Figure 7). The pylons are pulled towards the main span of the bridge and fail in bending. This process exhibits features of the domino-type collapse: the horizontal forces cause an overturning and thus failure of these two individual structures with mainly vertical load bearing capacity. They would not develop if the bridge deck was made of concrete or is not continuous.

A rupture of adjacent cables as a direct consequence of the initial failing cables which is the main example for the zipper-type collapse and is often connected to the collapse of cable-stayed bridges does not occur. 7 ROBUSTNESS From the observations made regarding the collapse analysis, conclusions can be drawn as to the robustness of the investigated cable-stayed bridge, which has a slenderness ratio of 1/230 and a cable spacing of 15 m. In general, robustness can be defined as insensitivity to local failure (Starossek 2009). Robustness is therefore always related to the size of the initial failure and to the accepted amount of damage to the remaining structure. Both items have to be quantified as design aims (Starossek & Wolff 2006). Recommendations for cable-stayed bridges require the analysis for the failure of one single cable; other authors propose the failure of cables within a 10 m range (Starossek 2009). If the design aim is that no or only small plastifications are allowed, the investigated bridge acts only robust to the loss of one cable. Here, alternate paths can develop. Loads are transferred through the bridge girder to adjacent cables. Only local plastifications develop in the bridge girder and the cables remain elastic. If only total collapse should be avoided, the bridge can be termed robust regarding the loss of two adjacent cables plus one adjacent cable in the second cable plane. The failure of more cables is not possible without serious damage or total collapse. The bridges robustness is therefore limited to the failure of one to three cables depending on the predefined design criteria. Although the quantitative description of robustness will be different, the qualitative description will be analogous to similar cable-stayed bridges. The reason for the relatively robust behavior of the bridge lies in some general features of cable-stayed bridges with close cable spacing. The bridge decks elastic support by the cables allows a load distribution to many cables. Together with the low stresses in the serviceability state and a high deformation capacity due to their length, a failure of cables due to overloading is not very likely. For the investigated bridge, the bridge girder is the critical structural element. Its resistance against instability is crucial for the robustness of the structure in case of the loss of short cables. The resistance can either be increased by increasing the stiffness or by reducing the unsupported length, which means closer cable spacing. In the whole context of avoiding disproportionate collapse, the former alternative should be chosen: if the cables are placed close together, the probability that more

cables fail due to the same event is higher (Shankar Nair 2006) and the consequence to the girder remains the same. Additionally to a high crosssectional stiffness, the resistance to lateral buckling of the longitudinal girders should be high. Here, close cross girders with the same height as the longitudinal girders are useful. In case of the loss of long cables, the bridge decks deformation capacity is important. Thus, a steel deck is preferable. Another advantage of a steel deck is that tension forces can be transferred to the intact bridge part. Although the loss of the critical elements leads to the progression of collapse to the intact bridge half, it increases the robustness for the loss of long cables. 8 CONCLUSION This paper investigates the loss of any one cable by nonlinear dynamic analysis of a threedimensional model of a cable-stayed bridge. Dynamic amplification factors for quasi-static analyses are determined. The aim of these analyses is to give advice on how to determine this dynamic amplification factor, to prove if the use of a uniform amplification factor to calculate the maximum responses to cable loss is valid and if reductions of this factor are generally possible. The results show that a unique dynamic amplification factor cannot be specified. Instead, the dynamic amplification factor depends on the location of the ruptured cable and on the type and location of the state variable being examined. Using a factor smaller than 2.0 is only possible for the bending moments in the bridge girder. Here, an explicit calculation of the DAF can be beneficial. A dynamic amplification factor of 2.0 is necessary for the safe design of the cables. Regarding the bending moments in the pylons, large dynamic amplification factors result because of large dynamic responses resulting from a complex excitation. Dynamic timehistory analyses are therefore recommended, at least for the cable loss cases which yield the highest responses. These are generally the longer cables but not necessarily the longest cable. The dynamic amplification factor can be determined in linear dynamic analyses. Different live load positions do not have to be considered. Additionally to having appropriate analysis tools for creating robust structures, it is important to know which structural properties are important to increase structural robustness. These properties are identified by investigating the collapse behavior of a cablestayed bridge after the loss of cables. The results show that the normal forces in the bridge girder is the collapse promoting attribute of a cable-stayed bridge. For this reason, self-anchored cable-stayed react less robust to the loss of short cables where the

normal force in the bridge deck is highest. For the investigated cable-stayed bridge, two adjacent short cables in one cable plane plus one cable in the second cable plane can fail without disproportionate collapse. In case of the failure of three adjacent cables, the bridge collapses due to instability failure of the bridge deck. The robustness of the bridge can be increased by preventing instability. This is possible by increasing the stiffness of the bridge girder or by reducing the unsupported length, which means closer cable spacing. The former is recommended here due to a higher failure probability of closely spaced cables. ACKNOWLEDGMENTS This research was funded by the Deutsche Forschungsgemeinschaft DFG (German Research Foundation) which is gratefully acknowledged. REFERENCES
Eurocode 3. 2006. Design of steel structures, Design of structures with tension components. EN 1993-1-11, 2006 Fib (International Federation for Structural Concrete) 2005. Acceptance of stay cable systems using prestressing steels. Lausanne: fib. Hyttinen, E. & Vlimki, J. & Jrvenp, E. 1994. Cablestayed bridges effect, of breaking of a cable. In Cablestayed and suspension bridges, Proceedings AFPC Conference, October 12-15, 1994. Park, Y.S., Starossek, U., Koh, H.M., Choo, J.F., Kim, H.K. & Lee, S.W. 2007. Effect of cable loss in cable stayed bridgesFocus on dynamic amplification. Improving infrastructure worldwide; Report IABSE symposium, Weimar, Germany, 19-21 September 2007. Zurich: IABSE. Post-Tensioning Institute (PTI) 2007. Recommendations for stay cable design, testing and installation. 5th edition, Cable-Stayed Bridges Committee. Phoenix: PTI. Shankar Nair, R. 2006. Preventing Dispoportionate Collapse. Jounal of Performance of Cnstructed Facilities, 20(4), 309314, November 2006. Starossek U. 2007. Typology of progressive collapse. Engineering Structures, 29(9), 2302-2307, September 2007. www.starossek.de. Starossek, U. (2009). Progressive Collapse of Structures. Thomas Telford Publishing, London, July 2009. www.starossek.de Wolff, M. & Starossek, U. 2008. Robustness assessment of a cable-stayed bridge. Proceedings, 4th International Conference on Bridge Maintenance, Safety, and Management (IABMAS08), Seoul, Korea, July 13-17, 2008. Wolff, M. & Starossek, U. 2009. Cable loss and progressive collapse in cable-stayed bridges. Bridge Structures, 5:1, 17 28, March 2009 Wolff, M. 2010. Seilausfall bei Schrgseilbrcken und progressiver Kollaps (Cable loss in cable-stayed bridges and progressive collapse). Phd-Thesis, Structural Analysis and Steel Struct. Inst.. Hamburg University of Technology. Zoli, T. & Woodward, R. 2005. Design of long span bridges for cable loss. Structures and Extreme Events Report IABSE Symposium, Lisbon, Portugal, September, Zurich: IABSE