A hyper-reduction method using adaptivity to cut the assembly costs of reduced order models

Slight variances in the manufacturing of rotating machinery can lead to significant changes in the structural dynamic behavior compared to the behavior of ideal cyclic periodic structures. Therefore it is necessary to consider deviations from the perfect cyclic periodic structures in the mechanical design process of rotating machinery. To minimize the effort of numerical calculations, the application of reduced order models is indispensable. The objective of this paper is the comparison of two reduction methods which are not widespread in application of rotationally periodic structures. For the validation of the implemented methods, a generic model of a thin plate meshed with shell elements and a representative large size FE model of a radial turbine wheel are used. The first reduction method is called Improved Reduced System and is based on the classical Guyan reduction. The second reduction method is called SEREP method and is from a theoretical point of view closely related to the first method although the procedure to obtain the reduction basis is quite different. The results show for both test cases an excellent agreement between the reduced order models and the unreduced finite element model. Both reduction methods are also able to capture the phenomena of mode localization. It is also found that through the application of the reduced order methods the computation time can be reduced by two orders of magnitude. Based on the first reduction method, the statistical mistuning behavior is studied using the accelerated Monte Carlo simulation.

Download Full-text

On the Interpolation of Reduced-Order Models

10.14293/p2199-8442.1.sop-math.nsoqwl.v1 ◽

2018 ◽

Author(s):

Yao Yue ◽

Lihong Feng ◽

Peter Benner

Keyword(s):

Model Order Reduction ◽

Reduction Method ◽

Order Reduction ◽

Reduced Order Models ◽

Matching Method ◽

Model Order ◽

Reduced Order ◽

Parametric Model Order Reduction ◽

Order Reduction Method ◽

Parameter Dependency

A parametric model-order reduction method based on interpolation of reduced-order models, namely the pole-matching method, is proposed for linear systems in the frequency domain. It captures the parametric dynamics of the system by interpolating the positions and amplitudes of the poles. The pole-matching method relies completely on the reduced-order models themselves, regardless of how they are built. It is able to deal with many parameters as well as complicated parameter dependency. Numerical results show that the proposed pole-matching method gives accurate results even when it interpolates two reduced-order models of completely different nature, one computed by a projection-based method and the other computed by a data-driven method.

Download Full-text