Parameter Identification in Multibody System Dynamics using the Adjoint Sensitivity Analysis

Publication, 2018


S. Oberpeilsteiner - Parameter Identification in Multibody System Dynamics using the Adjoint Sensitivity Analysis - Phd Thesis, Technische Universität Wien, Institut für Mechanik und Mechatronik, Austria, 2018


The usage of state-of-the-art software for analyzing the dynamics of multibody systems allows to reduce the number of prototypes in an product development process. Often, unknown parameters cause notable deviations of simulation results compared to measurements taken during experiments. Improving the quality of the virtual prototype may be achieved by matching the parameters used in the simulation model with the real ones. With an increasing number of parameters, adjusting their values manually in order to improve the accordance is hardly possible due to the complexity of the multibody system. Therefore, an automated and efficient strategy for parameter identification represents the only reasonable approach for gaining better simulation results. Another problem arises due to the fact, that the chosen excitations may not cause a sufficient reaction of the components under consideration. In such a case the result of the identification relies on insufficient data, and therefore the accuracy of the virtual prototype is not satisfactory. Automating the process of parameter identification requires a meaningful performance measure in order to quantify the deviation of experiment and simulation. This allows for using an iterative approach that aims at solving the optimization problem that minimizes a scalar performance measure. The gradient required by the optimization algorithm is computed by using the adjoint sensitivity analysis. Addressing the problem raised by insufficient information contained in the measurements is done by adjusting the system inputs in order to maximize the performance measure’s sensitivity onto parameter changes, usually denoted as optimal input design. For both special issues, the computation of the gradient and the optimization of system inputs, detailed derivations are done. Besides the description of procedures developed, comprehensible examples are presented for emphasizing the performance of the respective method.