Prediction model for significant duration of strong motion in India

The duration of a ground motion has a bigger influence on the response of a structure. As a result, by accurately anticipating the duration of ground motion, the seismic design of structures can be controlled. As a result, the goal of this research is to develop a new prediction model for earthquake ground motion duration. Using an Indian database recorded between 1986 and 2001, an equation for predicting the significant duration (Ds5-95%) is constructed. The database consists of 148 horizontal acceleration time histories recorded on rock and soil sites with the magnitude varying from 4.5 to 7.2 and hypocentral distance from 10 to 400 km. Artificial Neural Networks (ANNs) are employed for developing the prediction model. Moment magnitude (Mw), hypocentral distance (Rhypo), site condition (S) are chosen as input parameters and Ds5-95% is chosen as the output parameter for the ANN model. A two-layer feed-forward neural network was selected to properly predict the duration of a ground motion. Levenberg-Marquardt (LM) back propagation (BP) algorithm was selected to train the network after testing. The significant duration increases as the hypocentral distance and magnitude of the earthquake increase. In rock sites, the significant duration was predicted to be higher than in soil sites. Sensitivity analysis was conducted to determine the order of importance of input variables on the output parameter. The developed ANN model was compared with the existing duration prediction models.