A machine learning approach used to predict the peak displacement, base shear and fundamental frequency of multi-storey steel structures under seismic excitation

This paper details the effectiveness of adopting Artificial Neural Networks (ANNs) for predicting the seismic response of multi-storey steel structures. A comprehensive framework is established to conduct a parametric study, generating input and output parameters for the ANNs. Fundamental periods for the multi-storey steel structures were first determined using modal analysis and adopted to select past earthquake ground accelerations that potentially trigger damage to the structural configuration. Response spectrum diagrams conforming to real-time data present a range of periods, that are analysed to correspond to the fundamental period generated by the modal analysis of the structure. Then, nonlinear time-history finite element analysis (FEA) models were developed to simulate the structural response using varying input parameters, such as geometric properties, member sizes and peak ground accelerations as well as output parameters, like the drift and the base shear. A parametric investigation of 3793 simulations were eventually conducted, adopting 18 input and 9 output parameters. The results obtained from the ANN models presented a fairly accurate solution yielding an R-value close to 1 and a mean square error (MSE) converging to zero. The accuracy of the results was validated by comparing results from the ANNs to those generated from FEA software (Ansys) on test datasets comprising new inputs. The outputs generated from both systems were distinct and approximately equal, indicative of a successful training network and conclusive of effectively predicting seismic responses of multi-storey steel structures that can be used for performance-based seismic engineering. The proposed data-driven methodology can be used to support time consuming nonlinear dynamic simulations. It can further be a significant part in developing structural digital twins representing the dynamic response of structures.