Parametric model order reduction for efficient frequency response evaluation

In many mechanical applications, a design optimization is performed on the frequency response functions of the mechanical system. In this work, a novel model reduction technique is explored for the parametric analysis of the frequency response functions of dynamical mechanical systems. This approach entails two novel aspects. First, a super-local reduced order basis is introduced which allows models of much lower order compared to models obtained with single, frequency independent reduced order basis approaches. Secondly, a novel hyper-reduction scheme is introduced which selects a subset of the considered, discrete frequency domain and computes the corresponding interpolation matrix. During the online phase, the full frequency domain response is reconstructed by interpolating between these sampled frequency lines. The proposed approach is validated numerically on an academic mass-spring-damper system and shows promising results.