Ultra-high-performance concrete (UHPC) has rapidly become a material of choice in modern construction owing to its favorable properties, such as superior strength, toughness, and durability. Yet, accurate quantification of damage states using appropriate engineering demand parameters (EDPs) remains a significant challenge in the performance-based seismic design (PBSD) of UHPC columns. This gap is especially evident in the scarcity of research focused on the PBSD of UHPC columns. Therefore, this study aims to develop an explainable machine learning (ML) powered predictive model for the drift ratio limit states of UHPC bridge columns across four damage states. To achieve this, a fiber-based numerical model was developed and then validated against large-scale experimental results of UHPC columns tested under reverse cyclic loading. The Latin hypercube sampling method is employed to generate a comprehensive database of UHPC bridge columns (10,000 UHPC columns), considering uncertainties across a spectrum of design factors, including column geometry, material characteristics, reinforcement ratio, and axial load ratio. The drift ratios at the initiation of four damage states are determined using the numerical model. Subsequently, the power of the ML algorithm is leveraged to introduce drift ratio limit states for UHPC columns that capture the onset of four different damage states. The efficacy of the developed model is assessed using a range of statistical metrics, including the composite fitness score, DX% metrics, coefficient of determination, root mean squared error (RMSE), normalized RMSE, and mean absolute error, which demonstrated high predictive accuracy of the drift ratio limit states on both the training and unseen or test datasets. Furthermore, a model-agnostic Shapley Additive exPlanation (SHAP) is used to explain the output of the model and examine the influence of various design factors on drift ratio across the damage states. Notably, the axial load ratio and aspect ratio are found to be major factors in determining the drift ratios across the damage states. The developed model is deployed into a software tool in order to enable its practical application. Additionally, the implementation of the developed model in the context of PBSD is illustrated through a design example. Furthermore, the study proposed capacity uncertainty parameters to aid the fragility assessment of UHPC columns. Finally, the proposed model, along with capacity uncertainty parameters, is used for the fragility assessment of UHPC columns reinforced with varying steel bar grades. The results demonstrate the less vulnerability of UHPC columns reinforced with higher bar grade, yet the effect of bar grade lessens at severe damage levels.
Read full abstract