| The topic of the thesis is the design sensitivity analysis for multibody dynamic systems. Work that was performed to date in this field used equations of motion that are expressed in a centroidal body-fixed reference frame. This simplifying assumption can yield inaccurate results when, due to design changes, the location of the centroid changes with respect to a body-fixed reference frame. Moreover, under this simplifying assumption, the equations of motion contain terms that are treated as identical zero, therefore being ignored in sensitivity analyses. Since the location of the centroid can depend on model parameters, the centroidal coordinate assumption is a potential source of errors in sensitivity analysis.; A unified approach to design optimization for multibody dynamic systems is presented, using the most general formulation of the Newton-Euler equations of motion; i.e., relative to non-centroidal body reference frames. First and second order dynamic sensitivity equations for model parameters are generated using two methods; direct differentiation and adjoint variables.; CAD-based optimization requires sensitivity information with respect to design variables. Thus, methods of obtaining derivative information for model parameters with respect to design parameters are analyzed, ensuring that the results from the dynamic sensitivity analysis can be transferred back in the solid model. A library of functionals that can be used in the optimization process is presented. The library includes functionals that can be used for optimization of general-purpose mechanical systems, as well as functionals that are suitable for the case of vehicle systems; i.e., functionals that can be used for optimization of vehicle systems and subsystems.; Algorithms are presented for direct differentiation and adjoint variables and an extensive library of derivatives with respect to generalized coordinates, generalized velocities, and model parameters is developed. First and second order sensitivity results for several multibody dynamic systems and model parameters are compared to finite difference approximations of different orders. Convergence of the finite difference approximations to the calculated sensitivities validates the algorithms and the sensitivity equations. |
