Machine learning based optimal design of superelastic friction pendulum for controlling underground blast-induced vibration of building

This study shows the machine learning (ML) based optimal design and response controllability of superelastic shape memory alloy (SMA) integrated friction pendulum (SMA-FP) isolator under the blast induced ground motion (BIGM), and compared with the FP isolator. An elastic steel shear building isolated with the FP or SMA-FP system is analysed through nonlinear time-history analysis (NLTHA). Design of base isolators (BIs) are obtained through optimizing two conflicting objectives, i.e. top floor peak acceleration (TFPA) and peak isolator displacement (PID). A multi-objective optimization (MOO) algorithm is used to estimate the optimal combination of friction coefficient and SMA wire strength. Comparison of the pareto optimal front clearly reveals a better trade-off between two objective functions for SMA-FP BI than FP BI. Robustness of optimal design and control effectiveness is studied through extensive parametric studies, for various parameters of isolator, building, and BIGM. Study results reveal that, SMA-FP can substantially reduce the TFPA (up to 30 %) in conjunction with the PID and residual isolator displacement (RID), reduction up to 42 % and 60 %, respectively, than FP BI. Finally, employing different ML based regression methods (multilinear, ridge, lasso, elastic-net regression), predictive models have been proposed for optimal design and optimum responses of structure and isolator.