Probabilistic prediction model for failure mode of RC columns based on a two-stage method combining machine learning and Bayesian classifier

Reinforced concrete (RC) columns under seismic load may exhibit various failure modes (FMs), such as flexural, flexural-shear, and shear failures. Traditional FM prediction models generally yield deterministic results and fail to account for uncertainties due to variability in material properties and loading conditions. This paper innovatively develops a probabilistic two-stage machine learning model for predicting FM of RC columns. In the first stage, machine learning algorithms are applied to estimate the shear demand to shear capacity ratio (Vp/Vn) by using the critical structural design parameters. In the second stage, the Bayesian method combining with the machine learning, is employed to estimate the occurrence probability of each FM. It has been demonstrated that the proposed method can provide more accurate and robust prediction results as compared with the traditional FM prediction methods. It is found that the Artificial Neural Network (ANN)-based probabilistic model exhibits the highest predictive performance, which can be further used in seismic design optimization and safety assessment of RC columns and structural systems.