Contribution
Efficient model order reduction of a generic vibroacoustic model via multi-moment matching
* Presenting author
Abstract:
Finite-element-based modelling of vibroacoustic problems, e.g. in virtual prototyping, can yield large-scale models with millions of degrees of freedom which require unfeasibly large computational resources for the numerical solution. Especially in the background of parameter studies, replacing the large-scale model with a reduced order model decreases the computational costs tremendously - this can be realised by multi-moment matching, a projection-based model order reduction approach. Since the efficiency of this method strongly depends on the number of cross-moments (multi-moments) defining the reduced order model’s size, this contribution discusses the realisation of cross-moment reduction in multi-moment matching model order reduction. To reduce the number of multi-moment vectors, which is connected to the number of parameters, non-influential parameters are identified via sensitivity analysis. This yields two working principles: general and specific multi-moment vector reduction. However, specific multi-moment vector reduction can only be applied to additive models. As the vibroacoustic system is non-additive, only general multi-moment vector reduction is examined. For the application to a damped, generic vibroacoustic model, the parametric reduced order models derived by multi-moment matching based on general-multi-moment vector reduction are superior to these computed with standard multi-moment matching regarding the accuracy and computation time in the online phase.