A novel data-driven analysis for sequentially formulated plastic hinges of steel frames

Abstract

Although many different nonlinear analysis techniques for steel frame structures, they causes the increase in total computational cost. A data-driven approach can then be used to substitute the classic finite element nonlinear analysis including second order methods and inelastic modelling. In this paper, a novel data-driven method using ensemble learning models for geometric and material nonlinear analyses of frame structures is proposed. For this approach, the data acquisition process and machine learning-based models are needed to train a large amount of dataset, which is collected from the nonlinear analysis and a yield surface equation. Ensemble learning algorithms are then used to build the data-driven models. After training, the sequence of plastic hinge formation in steel frame structures can be predicted without any nonlinear analysis process. In that situation, the data-driven model is more effective, especially in practical cases with the lack of experimental information due to limited sensors. The validity of the proposed method has been demonstrated by several numerical examples.