Reliability-Based Optimal Design of SMA and Yield Damper for Seismic Vibration Control of Connected Building: A Predictive Model Using Machine Learning

Past studies reported the stochastic structural optimal design of shape memory alloy (SMA) damper without considering the failure probability or reliability of structure. Further, none of the previous studies provided any formulation to quickly predict the optimal design under the simultaneous and wide range variation of critical design parameters. Thus, present study focuses on the optimal design of SMA damper (SMAD) and yield damper (YD) to minimize the failure probability of flexible building against random earthquakes. Machine learning (ML) regression methods is used to predict the optimal design of dampers and optimal responses of structure. Flexible and stiff building are modeled as two linear single degree of freedom (SDOF) systems and connected with SMA or yield damper. Stochastic linearization process is used to model the force–deformation behavior of dampers. Random earthquakes are modeled through Kanai-Tajimi power spectra. A sensitivity analysis is performed to identify the critical parameters affecting dampers’ performance. Subsequently, a wide range of these input parameters is considered, and input sample sets are generated using the Latin hypercube sampling (LHS) method. Employing stochastic optimization method, optimal design of SMA and yield dampers and optimum structure responses are estimated. ML regression methods such as multiple linear, Ridge, Lasso, and elastic net regression are utilized to fit the data and propose the best predictive equation for the optimal design of dampers and optimal responses of structures. Overall, the study utilizes stochastic optimization and ML techniques to predict the optimal damper design and optimum response of connected buildings, under random earthquakes.