Fatigue Reliability Assessment Of Steel Bridges Using Long-term Strain Monitoring Data

Bridge structures exposed to aggressive environments will be gradually deteriorated or damaged under long-term external cyclic loading conditions. This will result in reduction of bearing capacity and life-cycle reliability. For cross-sea bridges and other offshore steel structures, the combined action of cyclic loading and aggressive environment often results in significant reduction in the fatigue performance compared with that obtained under cyclic loading in inert environments. Therefore, it is important to consider the effect of steel corrosion on fatigue failure reliability of steel bridges. In addition, due to the limitation of sensor implementation techniques and specific in-situ conditions, the sensors for strain monitoring usually are not deployed at the most critical locations and thus only the nominal strain/stress is obtained. The nominal stress is far less than the hot spot stress in welded toe where fatigue crack are most likely to occur. To handle such a critical problem, the hot spot stress approach is necessary to developed for fatigue analysis of complicated welded steel details.Consequently, the main contents and conclusions achieved in this thesis are summarized as follows:1. In this thesis, a novel method for fatigue reliability assessment of welded joints in steel bridges is proposed by considering multiple effects in terms of the stress concentration phenomenon at the welded joint as well as the reduction of the effective cross-sectional area and the degradation of the fatigue behavior due to steel corrosion. By exploring the stochastic characteristics of the hot spot stress range, the reduction model of the cross-sectional area, and the degradation function of the fatigue behavior of the welded joint, a fatigue reliability model considering multiple effects is developed in the framework of a continuous probabilistic formulation of Miner’s damage cumulative rule.2. This thesis proposes a genetic algorithm-based (GA-based) mixture parameter estimation approach and applied for multi-modeling of stress spectrum. It not only is suitable for multi-modeling of stress range data, but also could fit the joint probability density function (PDF) of stress range and mean stress. The application range is wider than the traditional expectation maximization (EM) algorithm.3. The field monitoring data of dynamic strain from the instrumented Tsing Ma Bridge was employed to illustrate the GA-based mixture parameter estimation approach. Firstly, a wavelet-based filtering technique is employed to eliminate the temperature effect inherent in the measured strain data. The rainflow counting algorithm is used to transfer strain stress history to daily stress spectrum. A standard daily stress spectrum accounting for highway traffic, railway traffic, and typhoon effects is derived. Then, a multi-modal probability density function (PDF) of stress range is modelled by use of three kind of finite mixed distributions (normal, lognormal, and Weibull), with the parameters of finite mixture distributions are determined by GA and expectation maximum (EM) algorithm, respectively. A comparative study is conducted to assess the performance of GA-based and EM algorithm-based mixture parameter estimation approach with the aid of the Akaike’s information criterion (AIC) and distance measure (Δc). Finally, the joint PDFs of stress spectrum are also obtained by means of a mixture of bivariate distributions and GA-based mixture parameter approach.4. The proposed fatigue reliability assessment method is applied for fatigue assessment of the Tsing Ma Bridge installed with a long-term structural health monitoring system. A standard daily stress spectrum accounting for highway traffic, railway traffic, and typhoon effects is derived by statistically analyzing the long-term strain monitoring data. A multi-modal probability density function (PDF) of the standard daily stress spectrum is obtained by use of the method of finite mixture distribution in conjunction with a genetic algorithm-based (GA-based) mixture parameter estimation approach. A comparative study is conducted to investigate the effects of stress concentration and steel corrosion (i.e., reduction of the effective cross-sectional area and degradation of the fatigue behavior) on the failure probability and reliability index versus the predicted fatigue life of the welded joint. The results show that the fatigue reliability of the welded joint is significantly affected by the effects of stress concentration and steel corrosion.