Abstract Artificial neural network models are developed to predict strong ground motion duration of subduction events for soft and firm soils. To train the artificial neural network a database with a total of 3153 seismic records with two horizontal components for interplate and inslab earthquakes is employed. The principal component method is used to carry out a dimensionality reduction of the input parameters to develop the artificial neural network models. The predicted values of the strong ground motion duration trained by the artificial neural network models are compared with those estimated with empirical expressions. In general, the strong ground motion duration predicted with the artificial neural networks follows the same tendency of that calculated with the empirical equations, although in some cases, the strong ground motion duration predicted by using the artificial neural network models presents sudden changes in its behavior. For this reason, it is recommended to carry out several verifications of the trained artificial neural network models before using them for further engineering applications, for example the simulation of synthetic records or the evaluation of seismic damage indices.

Resumen Se desarrollan modelos de red neuronal artificial para predecir la duración del movimiento fuerte del terreno de eventos de subducción en suelos firme y blando. Para entrenar la red neuronal artificial se emplea una base de datos con un total de 3153 registros sísmicos con dos componentes horizontales para eventos de interplaca e intraslab. El método de componente principal es usado para realizar una reducción dimensional de los parámetros de entrada para desarrollar los modelos de red neuronal artificial. Los valores predichos de la duración del movimiento fuerte del terreno por la red neuronal entrenada son comparados con aquellos estimados con expresiones empíricas. En general, la duración del movimiento fuerte del terreno predicha con la red neuronal artificial sigue la misma tendencia que la calculada con las ecuaciones empíricas, aunque en algunos casos, ésta presenta cambios repentinos en su comportamiento. Por esta razón, es recomendado llevar a cabo varias verificaciones de los modelos entrenados de la red neuronal artificial antes de usarlos para más aplicaciones ingenieriles, por ejemplo, la simulación de registros sintéticos o la evaluación de índices sísmicos de daño.

Abstract The main objective of this work is to carry out a parametric study of stay cables of a bridge under simulated spatially correlated buffeting forces. This vibration mechanism is simulated with an Autoregressive and Moving Average (ARMA) model, and applied to mathematical models of the stay cables of the tallest cable-stayed bridge in Mexico. The use of auxiliary damping devices to mitigate vibration is evaluated. The analysis results showed that the use of a more realistic model to represent and characterize the variation in space and time of the fluctuating wind, which is not capture by the white noise or harmonic functions, is advantageous and offers a new alternative to evaluate the structural response of stay cables without and with dampers. The implications of the analyses results in the structural behavior of the stay cables of the bridge are discussed.

Abstract Use of Tuned Mass Dampers (TMDs) to reduce the wind-induced torsional response of structures has been investigated in the literature by assuming that the wind excitations can be approximated by harmonic forces or white noise random processes; however, such an assumption is not realistic. Further, wind load effects are correlated. This study is focused on the effectiveness of different linear/nonlinear TMDs configurations to reduce the wind induced response. For the analyses, the structure is modeled as a multi-degree of freedom system under correlated wind load effects. The results show that the selection of optimum dampers is affected by the consideration of the correlated wind load effects.