Résumé:
The focus of this study is to analyze the effects of earthquake parameters on the ground motion characteristics and structural behavior. The intensity measures of the ground motion investigated are the peak ground acceleration, significant duration, and the mean period. These measures were strategically selected to take into account the main features of the ground motion, such as the amplitude of the motion, cumulative effect, and frequency content. As for the engineering demand measures investigated, the inelastic response spectra and hysteretic energy demand spectra are
selected to evaluate the nonlinear behavior of structures. For each of the parameters considered, a predictive model is developed, tested, and finally used to perform a sensitivity analysis.
The strong motion database developed in this study includes 1104 records, collected from the Kiban Kyoshin Network (KiK-Net) from 10 events. The selected events have a depth less than 13 km, a magnitude between 4.8 and 7.3 and an epicentral distance ranging between 15 to 200 km.
The Artificial Neural Network ANN technique is used as an alternative to regression methods. Compared to the existing attenuation models, in addition to the earthquake independent parameters used for attenuation relationships, a new aspect is considered in this dissertation called
directionality. An analysis of the effect of directionality on the Peak Ground Acceleration (PGA) was performed, and it was found that their effect could cause an increase in the PGA that may reach up to 35%. Therefore, a radial angle parameter has been included in the input of the predictive model. The performance criteria used indicate that the predicted values of the intensity measures by the neural network are in good accordance with the observed ones. Finally, a sensitivity analysis for the earthquake parameters was performed in order to quantify the influence of each parameter on the intensity measures and structural behavior using the synaptic weights of the validated ANN.
The ANN model with only one hidden layer and a limited number of neurons, makes it easy to implement it in a spreadsheet or a simple computer program using the synaptic matrices and the bias vector, so that it can be routinely integrated in engineering applications and for Seismic Hazard Analysis studies.
