Statistical evaluation of monitored data intended for fatigue reliability assessment

For steel railway bridges, fatigue is the main reason for a limited service life. An initial prediction of the fatigue life is usually performed using loads and conservative assumptions stated in the governing standards. The result is used for decisions of further actions. Even if the result is an insufficient residual life, some bridges are not easy to strengthen or replace without causing large traffic disturbance. In this licentiate thesis, the real load effect is studied by monitoring a bridge during service. The aim is to improve the fatigue life prediction by reducing the uncertainties in the traffic load and finally attain a longer theoretical service life of the bridge. A case study of the Soderstrom Bridge in central Stockholm, one of Sweden’s most important railway bridges, is incorporated in the studies. Previously performed theoretical assessments have shown alarming results regarding the remaining fatigue life for some of the structural members in the bridge. To verify the result and hopefully receive more favorable stress ranges, the bridge was instrumented with a monitoring system in 2008. The bridge and the monitoring campaign are thoroughly presented. Owing to the large data volume produced by the monitoring system, robust routines and procedures for quality assurance have been required. A deterministic fatigue assessment is presented based on 43 days of continuous measurements comprising more than 17 thousand train passages. Unfortunately, also this fatigue assessment indicates high fatigue damages. A large monitoring system with many gauges enables a statistical evaluation of the quality of the measurements. By the principle of Analysis of Variance, a comparison of the stress range spectra can be performed with the aim of finding deviant spectra and corrupt gauges. An additional aspect is the length of the monitoring period, meaning the required duration for obtaining a stable result. A model for the convergence progress is proposed which could be incorporated in a monitoring system for use during service. The extensive monitoring program of the Soderstrom Bridge has resulted in reliable stress range spectra. By curve fitting and goodness-of-fit tests measured spectra have been compared with known statistical distributions. These distributions can be used for future fatigue reliability assessments.