| Title |
Development of an Optimal Prediction Model for Structural
Member Forces of Box Structures Using Artificial Neural Networks |
| Authors |
신서연(Shin, Seoyeon);윤누리(Yun, Nu-ri);박건(Park, Gun);홍기남(Hong, Kinam) |
| DOI |
https://doi.org/10.12652/Ksce.2025.45.3.0297 |
| Keywords |
박스 구조물, 딥러닝, 유한요소법, 부재력 Box structure, Deep learning, FEM, Member force |
| Abstract |
Box-type structures have been widely employed in civil engineering from ancient times to the present, serving essential roles in the
construction of various infrastructure systems such as water supply and drainage, urban gas lines, electricity and telecommunications,
highways, and railways. Their usage is expected to increase further in both frequency and scale. Currently, the FEM is the predominant
analytical technique used to evaluate the safety of these structures. While FEM offers precise and reliable results, it often demands
considerable time and effort due to the complexity of modeling and simulation, particularly when applied to large-scale structures. This
can lead to inefficiencies in the design and review process. To address these limitations, this study performed both static and seismic
analyses of box structures using FEM, and subsequently utilized the resulting data to train a deep learning model aimed at predicting
structural member forces. A total of 600 numerical models were developed using MIDAS software. The predictive performance of the
deep learning model was assessed using metrics such as Mean Squared Error (MSE), Mean Absolute Error (MAE), and the coefficient
of determination (R2). The results indicated that the model's performance under static loading conditions was superior to that under
seismic conditions, which may be attributed to the complexity introduced by dynamic responses. This study highlights the potential of
deep learning as a complementary approach to traditional FEM-based structural analysis. The proposed methodology offers a
promising avenue for enhancing the efficiency of structural safety assessments and design processes in future engineering applications |