This dissertation focuses on the efficient and accurate generation of sensitivity information for various optimization problems characterized by a large design space in the field of flexible multibody system dynamics. These various optimization problems, such as parameter identification, optimal input and topology optimization, enable the mechatronic industry to reach targets on energy-efficiency, performance, reliability, etc. Accurate sensitivity information is for many performant gradient-based optimizers a necessary ingredient for obtaining good convergence. However, because a sensitivity analysis often takes up a relatively significant amount of computational cost, it is important to obtain the gradient information efficiently, especially when dealing with a large amount of design parameters. The main focus of this thesis is the development of a novel methodology for generating sensitivity information for various optimization problems in flexible multibody system. The approach uses the Flexible Natural Coordinates Formulation (FNCF) as underlying small deformation flexible multibody formulation for reducing the complexity of the (continuous and discrete) adjoint variable methods. FNCF combines the advantages of the Floating Frame of Reference (FFR) and Generalized Component Mode Synthesis (GCMS) multibody formulations, resulting in a simple structure of the equations of motion (EOM), which is for FNCF characterized by a constant reduced mass and stiffness matrix and quadratic constraint equations. The simple structure ensures that the several Jacobians of the terms appearing in the EOM, which are needed in the adjoint sensitivity equations, are eliminated, readily available, or reduced in complexity. This feature ensures a reduced implementation complexity for the Adjoint Variable Method (AVM) compared to when using the traditional multibody formulations. Although the equations of motion and the adjoint equations need to be solved for an increased amount of degrees of freedom, the method can be computationally more efficient when the evaluation/calculation of these Jacobians take up a significant part in the total sensitivity computation. Furthermore, leveraging on the AVM-FNCF sensitivity methodology, an efficient and generally applicable lumped parameter identification methodology is presented for flexible multibody systems using experimentally recorded data. It is illustrated for the parameter identification of several model parameters in a quarter car multibody model including flexible bodies. The experimental reference signals are obtained from a corresponding quarter-car test rig. The sensitivity information for the optimization problem has been compared and verified using the finite difference sensitivity method. Moreover, the lumped parameter identification problem is extended towards input optimization for flexible multibody systems. This is illustrated with an input optimization case where more than 700 parameters that parametrize the input torque signal of a flexible slider-crank mechanism have been optimized. The objective was to retrieve the torque signal such that the resulting angular velocity of the motor shaft driving the crank matches a reference angular velocity signal as close as possible. The sensitivity information has been compared and verified using the finite difference sensitivity method. Because the reference angular velocity signal has been generated by applying a predefined torque signal to the multibody model of interest, the optimization method was able to retrieve the latter torque signal accurately. This example of an input-optimization case can be a stepping stone for further deployment in optimal-control. Lastly, a novel sensitivity approach for the fully coupled topology optimization of structural components in flexible multibody systems, leveraging on the AVM-FNCF sensitivity methodology is presented. A key advantage of the latter methodology for these fully coupled topology optimization problems, is that it results in constant Jacobians of the reduced mass and stiffness matrix with respect to the design parameters. Therefore, these only need to be computed once for each design iteration, instead of for each timestep during each design iteration when using other multibody formulations. The AVM-FNCF based fully coupled method has been illustrated for the topology optimization of multiple components in a slider-crank mechanism. Moreover, optimized topologies on a system-level are obtained that cannot be achieved using the classical 'self-adjoint' based sensitivity methods due to their underlying assumptions. Furthermore, the sensitivity information has been verified and compared using the finite difference and self-adjoint sensitivity methods. status: published