Machine learning-based multi-objective optimal design of superelastic resilient friction base isolated building under stochastic earthquake

This study proposes optimal design framework for a superelastic resilient friction base isolator (SR-FBI) system, using multi-objective optimisation, to mitigate stochastic seismic excitations in the isolated buildings. This optimisation approach aims to minimise both RMS top floor acceleration (RTFA) and RMS isolator displacement (RID) through a tailored weight factor strategy. Behaviour of the SR-FBI is compared with the R-FBI by evaluating the performance across varying building, isolator and earthquake parameters. By integrating SMA wires with the R-FBI system, the SR-FBI offers enhanced vibration control performance, significantly reducing isolator displacement while preserving isolation efficiency. Results reveal, two key design parameters, i.e. friction coefficient and strength of SMA wire as critical for maximising isolator control efficiency. Compared to R-FBI, the SR-FBI system requires 20 % to 70 % lower friction coefficients owing to the contribution of 10 % to 45 % SMA wire strength. Consequently, the optimally designed SR-FBI reduces RTFA by 20 % to 35 % and RID by 25 % to 60 % than the R-FBI. Finally, this study proposes prediction models with exponential and power terms to estimate the optimal design parameters and the structural responses under real GMs. Among the various ML regression methods, the lasso and elastic net regression models achieved the highest prediction accuracy in deriving the closed-form predictive equations. These models provide a reliable and efficient tool for the rapid optimal design of R-FBI and SR-FBI systems, and prediction of the seismic structural responses, thereby supporting the practical engineering applications without need for extensive simulations.