선형 동적 시스템의 주파수 응답 해석을 위한 심층 연산자 네트워크 기반 예측 알고리즘 응용

초록

This paper proposes a deep operator network (DeepONet)-based model for predicting the frequency response of linear dynamic systems. The frequency response is a key quantity that directly reflects the dynamic characteristics of a structure, and its computation requires repeated solutions of the linear system equations at individual frequency points. This procedure incurs significant computational cost for large-scale finite element models and limits applications such as real-time vibration prediction and digital twin implementation. To address this problem, this paper proposes a frequency response prediction framework based on a DeepONet, in which the physical-coordinate response of a structure is transformed into the modal domain and the modal coordinates are predicted from the modal force input. Numerical experiments on a 16-degree-of-freedom mass-spring-damper system demonstrated that the proposed model achieves high prediction accuracy and can serve as an efficient surrogate model for repeated frequency response analyses. These results indicate that the proposed model can be effectively applied to real-time frequency response prediction of large-scale structures and digital twin-based vibration analysis.

키워드