Probabilistic fatigue life estimates for riveted railway bridges

A large percentage of the railway bridges in the UK rail network and around Europe are of riveted construction exceeding in many cases 100 years of age. The remaining fatigue life of these bridges is difficult to estimate due to the uncertainties regarding the fatigue behaviour of wrought-iron and older steel material which were used for their construction. The problem is further compounded by the uncertainties associated with the loading both past and future. Previous global finite element analyses of a typical wrought-iron riveted railway bridge have shown that the fatigue critical details are the inner stringer-to-cross-girder connections (Imam et al. 2005). The analyses were carried out under a historical load model (Imam et al. 2005), developed to represent rail traffic in the period 1900-1970, and present day traffic (BS5400 1980) for the period 1970 onwards. Deterministic remaining fatigue life estimates of the connections were found to be sensitive to the level of dynamic amplification as well as the fatigue classification of the details. Following this work, this paper presents probabilistic fatigue life estimates for the most highly damaged stringer-to-cross-girder connection, as identified by the global analysis of the riveted bridge. On the loading side, the problem is randomised through the frequency of train traffic, dynamic amplification and uncertainties regarding the difference between actual and calculated stresses. On the response side, different assumed S-N curves used for detail classification and the Miner sum are also treated as random. The probabilistic analysis, which is carried out using Monte Carlo simulation, shows that the most heavily fatigue-loaded stringer-to-cross-girder connection has considerable fatigue life reserve. Through a sensitivity study, it is found that for a 2.3% probability of failure, the remaining fatigue life of the investigated connection is equal to 68 years for a pessimistic scenario. Under a more realistic combination of variables (base model), the 2.3% characteristic remaining fatigue life is found to be 480 years. Figure 1 shows the effect of different variables on the time to attainment of a 2.3% probability of failure assuming a base model. It can be seen that fatigue life estimates exhibit the highest sensitivity to detail classification, in other words the constant amplitude fatigue behaviour of the detail, and the factor α, which takes into account the difference between measured and calculated stresses. In particular, by changing the detail classification from Class WI (base model), which is used to represent wrought-iron riveted details, to Class D (BS5400 1980), which is used to represent bolted or riveted steel connections with a high clamping force, there is a 180% increase in the time required to attain a 2.3% probability of fatigue failure. On the other hand, a change to a modified Class B detail, which may be thought of being representative of a riveted connection with no or low clamping force in the rivets, is found to result in a 40% decrease in the same time. The use of a mean value of 0.70 for the factor α is found to result in an increase in the fatigue life at the 2.3% failure probability by about 110%. However, a mean value of α of 0.90 is found to result in a decrease in this fatigue life by about 50%. The mean values of the cumulative damage model (Δ) and the dynamic amplification factor (DAF) may be seen in Figure 1 to be of less importance. (Graph Presented). © 2006 Taylor & Francis Group