Accelerating parametric studies in computational dynamics: Selective modal re-orthogonalization versus model order reduction methods

- Selective reorthogonalization is proposed to accelerate dynamic analysis in large structures.
- The computational performance is compared against several model order reduction methods.
- Accuracy and numerical efficiency of all methods is compared in detail.
- The proposed method is more efficient than condensations and preserves a high accuracy in the results.

In the dynamic analysis of a structure, it is frequent the use of parametric studies to consider several design configurations or possible modifications of the structure. These changes modify the physical properties of the structure, and therefore, finite element models need updates in order to compute the response of the modified structure. A wide variety of model order reduction methods which may be suitable for this task has been developed, either static or dynamic, which also consider non-classical damping, which is especially relevant in the design of vibration absorption devices. In this paper, we compare the use of selective reorthogonalization with other model order reduction techniques, both in terms of computational time and in accuracy, using three computer architectures. The proposed reorthogonalization method allows for parametric structural modifications and evaluates the solution using a modified complex modal domain only along a selection of a few degrees of freedom that are relevant for the dynamic analysis of the system. This acceleration method does not result in any significative decrease of the quality of the results of interest due to approximations, whereas remains very competitive when compared to usual model order reduction techniques.