In recent years, the design of modern mechanical and mechatronic system is becoming an increasingly complex task. This trend is due to the continuously increasing functional requirements and to the vast amount of different designs that have to reach the market within tight schedules. For these reasons it becomes infeasible to perform extensive testing on each new prototype. Simulation tools offer a faster and cost-effective alternative to physical testing provided that complex systems can be simulated accurately and within reasonable time frames. In order to match these requirements novel simulation methodologies and tools are needed and are subject of constant research effort. Flexible multibody dynamics is one of these approaches, developed in recent years, to deal with complex systems including high performance and light-weight machines. Within this approach the single components of a mechanism can be modeled to include the effect of deformation on the global system motion. Despite the constant research effort in this field, flexible multibody simulation of complex machines remains computationally very demanding, especially if accurate prediction of local quantities such as dynamic stresses and strains is of relevance. So-called component and system-level model order reduction strategies have been developed over the years to reduce the computational burden to an acceptable level. Nonetheless some key issues remain unsolved. If the flexible components of a mechanism present a large number of connections with other components or experience loading in multiple locations, standard model order reduction schemes fail to give an appropriate solution due to the large number of degrees of freedom required unless accuracy is largely sacrificed.In many industrial applications though, a component can be loaded at many different locations in time but only a few of those are actually loaded at a given time instant. A typical example is a pair of gears, meshing with each other in which all the teeth can be loaded during a full revolution but only a limited amount is in contact at a certain moment in time. Other examples are e.g. bearings and sliders. The topological variation of these components makes it challenging to perform reliable and efficient simulations within a flexible multibody framework.The main focus of this dissertation is on providing solutions that allow to simulate flexible multibody systems with variable topology efficiently with minimal losses in accuracy. Component-level model order reduction strategies should be adapted to reach this goal. This dissertation proposes two alternative solutions to the problem. At first a methodology named static modes switching is developed and further extended. The core idea behind this strategy is to discontinuously adapt the reduction space used to model the displacement field of the flexible components on-line during simulation. In this way, all the degrees of freedom that show a negligible contribution to the system response can bediscarded to improve efficiency. Several application examples of increasing complexity are used to analyze this strategy both from a performance and an accuracy point of view. A second method named static modes sliding, is also proposed and validated. This method is developed to mitigate a few limits of static modes switching related to its discontinuous character. In this approach, the reduction space is continuously adapted by interpolating a pre-computed set of shape vectors during simulation. The influence of the time variability of the reduction space is retained in the equations of motion so that the scheme remains energy consistent.Both methods developed are solutions to the complex problem of flexible mechanisms with variable topology and lead to an efficient and accurate description of global and local dynamic phenomena. The research performed in this PhD project has also paved the way for improvements and further developments that are suggested as future research tracks.