Seismic response of bridges employing knowledge-enhanced neural networks for the lumped plasticity modelling of RC piers

Abstract

To facilitate seismic analysis of bridges, especially on a regional scale, this study established a parametric finite element model of bridges incorporating simplified component elements. It employs a knowledge-enhanced neural network (KENN) to calibrate the parameters of the lumped plasticity model of pier columns. Along with a database of historical experimental results, the influence of the key characteristics of reinforced concrete columns on model parameters are investigated and formulated as physical laws to supervise KENN training. The developed KENN model was then developed, yielding root mean square errors within the range of [0.027, 0.209]. These errors are slightly larger than those of the purely data-driven neural network, yet the KENN model aligns more consistently with the physical principles. Further, to demonstrate its accuracy and efficiency, the proposed methodology was applied for the rapid seismic response analysis of typical bridges.