Orthotropic Steel Decks (OSDs) are widely used in various types of steel bridges due to their benefits of light weight, high load bearing capacity and speedy construction. Although many improvements in aspects of design, fabrication, inspection, and maintenance have been achieved over the years for such bridge decks, fatigue remains a predominant problem, mostly because of the complexity of prediction methods. Many researchers have tried to investigate this component through experiments. However, performing only experiments may not lead to a cost-effective solution. Therefore, it is necessary to combine the experimental data with the numerical approaches. Particularly Linear Elastic Fracture Mechanics (LEFM) allows to model and analyse the crack propagation until subsequent failure, and significantly reduces the requirement of experiment. ABAQUS® provides an enriched feature, commonly referred to as the Extended Finite Element Method (XFEM) which incorporates two enrichment function namely the discontinuity function which represents the gap between the crack surface and asymptotic function which captures the singularity and thus can be used to model discontinuity independent to the finite element mesh. To evaluate the modelling efficiency and validate the simulation methodology, two XFEM-model based on LEFM and Virtual Crack Closure Technique (VCCT) are developed and the simulated results are compared with the experimental data. The first phase of the thesis deals with the numerical simulation to investigate the crack propagation rate in Compact-Tension (CT) specimen for different stress ratios. The results of two-dimensional (2D) model are found to be in good agreement (within 1.48%) with the fatigue coupon test results. As most of the work concentrates on 2D shell model, the extension to three-dimensional (3D) solid requires the investigation of related parameters to consider through-thickness effects. Nevertheless, the mechanism of 3D model is studied, and the simulated results match with the 2D results for fatigue crack growth (a, N). Moreover, a reliable technique of computing Stress Intensity Factor (SIF) is obtained by comparing with the ISO 12108 standard formulation. However, when the SIF and fatigue crack growth are combined, the crack propagation rate in 3D is overestimated (about 26%) when compared to the experimental data possibly because of the imperfection in the application of boundary conditions. The second phase deals with the numerical simulation in welded connection of OSD to determine the Paris law constants (C, m) by correlating the numerical result of fatigue crack growth with the beach mark measurements obtained in the fatigue experiments. Prior to automated XFEM simulation, a set of finite element analyses are performed to determine the vertical deformation, longitudinal stain distribution and hotspot stresses to validate the numerical model as per the test setup. The results of numerical analyses showed a good correlation (within 18%) with test data and Paris law constant C is predicted to be lower than the recommended value by IIW standard. The validated methodology is then applied on large scale to an existing bridge (Suurhoff bridge) structure which was built in 1971. In this case study, a crack length of 230 mm was detected in the deck plate originating from the root of the stiffener-to-deck plate welded connection between the cross-beams using TOFD measurements. To verify the problem, a numerical model is developed based on the dimension of the bridge to evaluate the crack initiation period and the crack propagation period. The crack initiation period is predicted using hotspot stress method and the crack propagation period is evaluated using automated XFEM simulation. Overall, the total fatigue load cycles are predicted to be 7.86 million which is equivalent to 48 years. A similar crack length was however detected after a service life of 44 years. This overestimation can be possibly explained as the model did not take residual stresses and other welding defects into account. The numerical model showed a good correlation with the real scenario and is therefore used to predict the permissible limit of deck plate crack length of 500 mm. The model predicted 8.02 million load cycles for a crack length of 500mm, which is equivalent to 34 years after the crack initiation period. Nevertheless, the fracture mechanics approach showed improvements in the assessment of fatigue life.