A comparative study between the ANN-based and the current methods for seismic response estimation in the case of single degree of freedom systems

Abstract

The target displacement is one of the most critical performance indicators in the seismic vulnerability assessment. The Nonlinear Time History Analysis (NL-THA) is the most reliable method for calculating the seismic response of any building by solving the differential equation of motion. However, this procedure is considered time-consuming, and it needs expertise to perform. For that reason, many codes and standards have proposed and adopted various methods and procedures to estimate and predict the seismic response and target displacement. These alternative methods represent some prediction uncertainties. Machine Learning (ML) algorithms became an exciting tool in earthquake engineering due to their performance and prediction simplicity. This paper compares the target displacement prediction of a novel Artificial Neural Networks (ANNs) method to the Displacement coefficient Method (DCM) adopted by FEMA-356, the Modified Coefficient Method (MCM) adopted by FEMA-440, and the NL-THA. The comparison is performed to 10 Single degrees of Freedom (SDOF) with different vibration periods and yielding forces (fy). The ANN model uses the SDOF characteristics and the ground motion (GM) parameters to estimate the maximum inelastic response. The results show a high performance of the ANNbased method in terms of Mean squared Errors (MSE), Mean Relative Error (MRE), and Mean Absolute Error (MAE).