Reconstructing hysteresis behavior for magnetorheological elastomer base isolator using bi-fidelity transferring evolution

This study proposes a novel methodology to characterize and predict nonlinear hysteresis behavior of a magnetorheological elastomer (MRE) base isolator. A Maxwell nonlinear model is utilized to capture the force-displacement and force-velocity loops under varying input currents. To enable efficient and rapid reconstruction of nonlinear responses, a surrogate-assisted evolutionary algorithm (EA) is developed, incorporating a clustering-driven online learning model management mechanism for bi-fidelity optimization. Specifically, an incremental Kriging model is constructed to approximate the high-fidelity objective function, serving as a low-fidelity evaluation. Representative solutions, determined through K-means clustering and the Kriging model, are selectively transferred to high-fidelity evaluations, guiding the search for the global optimum. Experimental data of displacement, velocity, and force obtained from the MRE isolator are used to validate the proposed algorithm, demonstrating highly accurate predictions with exceptional computational efficiency.