Explanation of Excessive Long-Time Deflections of Collapsed Record-Span Box Girder Bridge in Palau

Explanation of the excessive deflections of the Koror-Babeldaob (KB) Bridge in Palau is presented. This bridge was built in 1977 by the cantilever method and collapsed 3 months after remedial prestressing in 1996. It was a segmental prestressed concrete girder having the worldrecord span of 241 m (790 ft.) and maximum girder depth of 14.17 m (46.5 ft). The final mid-span deflection was in design expected to be 0.55 to 0.67 m (21.6 to 26.4 in), but after 18 years it reached 1.39 m (54.6 in.) and was still growing. Presented is a comprehensive analysis using 5906 three-dimensional (3D) finite elements and step-by-step integration in time. For the concrete creep and shrinkage properties, the B3, GL, ACI and CEB (or CEB-FIP, fib) models are considered and predictions compared. Model B3, in contrast to the others, does not give an unambiguous prediction because, in addition to concrete design strength, it necessitates further input parameters which are unknown. These are three mix parameters which can be set to default values but can also be varied over a realistic range to ascertain the range of realistic predictions. The findings are as follows: 1) For model B3, one can find plausible values of input parameters for which all the measured deflections (as well as Troxell et al.’s 23-year creep tests) can be closely matched, while for the other models they cannot even be approached and the observed shape of creep curve cannot be reproduced. 2) The 19-year deflections calculated by 3D finite elements according the ACI, CEB and GL models are about 67%, 62% and 54% less, respectively, than those measured and calculated according to model B3, and their deflection curves have shapes rather different from the those measured. 3) The shear lag is important since it increases the downward deflection due to self-weight much more than the upward deflection due to prestress. 4) The deflection is highly sensitive to prestress loss because it represents a small difference of two large numbers (deflections due to self-weight, and to prestress). According to the ACI, CEB and GL models, the prestress loss obtained by the same finite element code is, respectively, about 54%, 44% and 34% smaller, and according to the classical lump estimate used in design about 54% smaller. 5) Model B3 is in agreement with the measurements if 3D finite elements with step-by-step time integration are used to calculate both the deflections and the prestress losses, and if the differences in shrinkage and drying creep properties caused by differences in slab thickness and temperature are taken into account.