A Sensitivity Study of Error Estimation in Reduced Elastic Multibody Systems

Abstract Error estimation is important for the further acceptance and usage of reduced order models to speed up simulations. We focus in this work on an a-posteriori error estimator for second-order mechanical systems which is valid for all model order reduction techniques based on Galerkin reduction. We analyze and improve this estimator in the following ways: We conduct a sensitivity analysis of the error estimation on a beam model. It is shown that the estimator is sensitive to the reduction methods, the input functions, and the model itself. It is also shown that the overestimation can be arbitrarily small. Matrix norm inequalities are used to prevent inversion of the matrix. This results in an overall speedup of the error estimation routine as well as allows the scaling of the error estimator to larger systems, which was not possible before. Additionally, we present how the error estimator itself is plugged in a non-intrusive way to the Elastic Multibody simulation software Neweul-M2 with least possible effort.