Rapid dynamic analysis for structures with variable system parameters through multi-fidelity model

Abstract

Dynamic analysis for structures with variable parameters, such as structural dynamic reliability analysis, optimization and uncertainty quantification, usually requires a large number of samples, while high-fidelity experiments or simulations are usually too time-consuming to repeat many times. To address the above issue, this paper proposes a new approach called multi-fidelity based dynamic analysis (MF-DA) methodology for efficiently approximating dynamic responses with available multiple computational models. The responses from dynamic analyses with different fidelities can be integrated concurrently through MF-DA, thus improving the modeling efficiency and approximation accuracy simultaneously. The MF-DA method begins with the design of experiment procedure through optimized and inherit Latin hypercube sampling. Then, the obtained high-dimensional dynamic responses are transformed into several low-dimensional projections according to principal component analysis. Finally, the relationship between projections from two dynamic analyses with different fidelities is constructed based on radial basis function. Two numerical examples are provided to demonstrate the accuracy and robustness of MF-DA. Throughout the modeling process, only a small number of HF dynamic analyses are performed, thus the proposed method provides an efficient tool for dynamic analysis with variable system parameters under insufficient high-fidelity samples.