Sensitivity analysis on the drift ratio of bridge piers using a neural network approach

This paper aims to perform a sensitivity analysis (SA) to identify the parameters that are most relevant to the drift ratio of bridge piers as a first step toward improving seismic resilience. The investigated parameters represent (1) seismic source and attenuation effects, (2) structural geometric effects, (3) site conditions, and (4) column-scale effects. The SA is conducted through dataset-based exploration and by using an artificial neural network (ANN), a supervised machine learning algorithm. Parameter relevance is evaluated using statistical metrics—including the F-test score, residual standard deviation, and coefficient of determination—as well as synaptic weight contributions and the interpretation of drift-versus-parameters trend curves. Because near-fault ground-motion recordings are scarce in low-seismicity regions, point-source stochastic simulations are used to generate synthetic data. Six artificial datasets (ADS) are created to train the ANN. The input variables include moment magnitude (Mw), hypocentral distance ( R _hyp ), average shear-wave velocity in the upper 30 m of soil ( V _s30 ), column height ( H ), column diameter ( D ), and column spacing ( R ). The output variable is the peak pier drift. The datasets differ in their H/D ratios and in whether local site effects are included. Six ANN models are trained to predict drift and conduct the SA. The results show that R _hyp is the most influential parameter (maximum F-test score = 56%), followed by M _w (maximum F-test score = 32%), the H/D ratio, and V _s30 when site effects are considered. In contrast, column spacing ( R ) has only a marginal influence for this configuration. The resulting ANN model, based on the dominant parameters, provides rapid estimates of pier drift and supports preliminary assessment of bridge seismic performance.