Abstract
A structural analysis model to represent the dynamic behavior of building structure is required to develop a semi-active seismic response control system. Although the finite element method (FEM) is the most widely used method for seismic response analysis, when the FEM is applied to the dynamic analysis of building structures with nonlinear semi-active control devices, the computational effort required for the simulation for optimal design of the semi-active control system can be considerable. To solve this problem, this paper used recurrent neural network (RNN) to make a time history response simulation model for building structures with a semi-active control system. Example structures were selected of an 11-story building structure with a semi-active tuned mass damper (TMD), and a 27-story building having a semi-active mid-story isolation system. A magnetorheological damper was used as the semi-active control device. Five historical earthquakes and five artificial ground motions were used as ground excitations to train the RNN model. Two artificial ground motions and one historical earthquake, which were not used for training, were used to verify the developed the RNN model. Compared to the FEM model, the developed RNN model could effectively provide very accurate seismic responses, with significantly reduced computational cost.
Highlights
Much research on the development of seismic response reduction technologies has been conducted to date
The simulation results of Example 1 show that the errors of the Long Short-Term Memory networks (LSTM) and gated recurrent units (GRU) recurrent neural network (RNN) models were almost similar, presenting an accuracy that was improved by about 6%
finite element method (FEM) model model and GRU RNN models were almost similar, presenting an10accuracy that was improved by model about 6%
Summary
Much research on the development of seismic response reduction technologies has been conducted to date. If the number of degrees of freedom (DOF) of the EF model is large, or the time-step of the state–space analysis is small for stable solving of the differential equations, the computational time for simulation tests of the semi-active control system could be considerable In this case, the solution search area of soft computing techniques including evolutionary algorithm may be narrowed to significantly reduce the optimization time, resulting in failure to find the global optimal solution. The solution search area of soft computing techniques including evolutionary algorithm may be narrowed to significantly reduce the optimization time, resulting in failure to find the global optimal solution To solve this problem, the effective dynamic response simulation model of the building structure with a semi-active control system is developed using deep learning technique. RNN model was effective in providing very accurate seismic responses, with significantly reduced computational cost
Published Version (Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have