Gaussian Process–Based Optimal Design of a Negative Stiffness Device under Stochastic Seismic Excitation

Abstract

The protection of buildings under seismic events necessitates the development of efficient vibration mitigation devices. The recently proposed negative stiffness vibration (NegSV) device materializes such a seismic protection mechanism, which leverages the concept of negative stiffness. The NegSV introduces a negative stiffness mechanism to a specified story, thus modifying the dynamics of a system. Therefore, the top part of the building, in reference to the modified story, is treated as a resonator with respect to the lower part. Despite the demonstrated benefits of the device, a drawback lies in the introduction of larger interstory drifts at the level of the modification. To counteract this behavior, the geometrically nonlinear nature of the device is exploited here for optimization of its properties toward maximizing its effectiveness. A parameterized seismic motion model incorporating probabilistic inputs is used for identifying the optimal device parameters based on specific objectives. A Gaussian process regression model, trained with comprehensive input–output data, is then adopted for optimizing the configuration of the device through coupling with active learning. The efficacy of the NegSV device is assessed on the basis of the achieved reduction in the level of vibration at critical locations within the structure. The investigation reveals a consistent set of optimal solutions, ensuring the integrity of the protected structures under seismic excitation.