Damage detection of steel frames under fire using time-series acceleration and machine learning = Chẩn đoán hư hỏng khung thép chịu lửa sử dụng chuỗi gia tốc và học máy

Abstract

In this paper, a damage detection methodology for steel frame structures under fire load using time-history acceleration and machine learning (ML) is proposed. A randomly created dataset by finite element analysis (FEA) is utilized to develop deep neural networks (DNNs). In which, the inputs of the model are the time-dependent acceleration at limited degrees of freedom (DOFs) of a steel frame structure, while the outputs are damage ratios of frame members. The damage ratio of damage elements is defined by the reduction of material Young’s modulus. The accuracy of DNNs is continuously upgraded by eradicating low-risk members after each iteration via a damage thresshold. A planar frame including damage detection scenarios with and without the fire effect programmed by Python are tested to confirm the validity of the proposed paradigm.