Effects of material properties uncertainty on seismic fragility of reinforced-concrete frames using machine learning approach

Material properties play a crucial role in governing the seismic response of reinforced-concrete buildings. Investigating the effects of material properties uncertainty using nonlinear dynamic analysis can be computationally expensive to obtain the seismic fragility curve parameters. Machine learning models are used to address this challenge, which takes material properties, the fundamental period, and the number of stories in the building as input and predicts their seismic fragility curve parameters. Reinforced-concrete buildings with 4, 8, and 12 storeys are investigated considering the uncertainty in material properties associated with concrete and steel rebars. Latin hypercube sampling is used to obtain 200 sets of variations of the material properties, and hence a total of 600 buildings are analyzed using nonlinear dynamic analyses, considering three different intensity measures, namely, PGA, Sa(T1), and Saavg. The seismic fragility curve parameters are obtained corresponding to three limit states, i.e., 'Immediate Occupancy', 'Life Safety', and 'Collapse Prevention', for all the investigated buildings, and intensity measures. It is found that the material properties uncertainty significantly affects both the median and standard deviation parameters of seismic fragility curves, for both Sa(T1) and Saavg, as compared to PGA. The predictions of trained machine learning models are explained using SHAP analysis, which is then used to identify the effects of input material and building-specific features on seismic fragility curve parameters and the subsequent changes in their rankings due to change in the buildings' limit states. The SHAP analysis revealed that the fundamental period of the building, and the unconfined concrete material properties are the important input features that govern the seismic fragility curve parameters. An interactive dashboard is developed as a part of this work that can be used to predict both the fundamental period and seismic fragility curve parameters for reinforced-concrete buildings at the intensity measures and limit states of interest.