Machine learning-driven probabilistic residual displacement-based design method for improving post-earthquake repairability of steel moment-resisting frames using self-centering braces

Owing to their sufficient ductility, conventional steel moment-resisting frames (MRFs) are widely used as seismic-resistant systems in building structures. However, the ductile behavior introduces significant residual displacement after earthquakes, leading to massive economic losses caused by difficult or even unfeasible repairs. This paper intends to propose a machine learning-driven probabilistic residual displacement-based design method for retrofitting MRFs using self-centering braces (SCBs) and improving the post-earthquake repairability by reducing the residual displacement to a target level. The influence of the SCB's design parameters on the peak and residual displacements of the enhanced MRF are first investigated through parametric dynamic analysis of single-degree-of-freedom (SDOF) systems. The probabilistic residual displacement prediction models are developed using different machine learning algorithms. Analysis results indicate that the artificial neural network (ANN) models demonstrate the highest accuracy with the greatest coefficient of determination (R²) and lowest root mean squared error (RMSE). A software was developed based on the ANN models for predicting the probabilistic peak and residual displacement responses of the retrofitted MRF (RMRF). Then the proposed probabilistic residual displacement-based design steps are introduced. Finally, a three-story benchmark MRF is retrofitted using the proposed design method with two different performance objectives. Nonlinear static and dynamic analyses are conducted to investigate the seismic responses of the designed RMRFs. The analysis results show that the RMRFs can achieve the desired performance objective, confirming the effectiveness of the proposed design method. The SCB can significantly reduce the residual displacement of the RMRFs, where the RMRFs are fully recoverable (with a probability of 100%) without any need for repair after maximum considered earthquakes.