DNN surrogate model based cable force optimization for cantilever erection construction of large span arch bridge with concrete filled steel tube

Abstract

Determining an ideal set of erection cable forces is crucial for concrete filled steel tube (CFST) arch bridges built using the cable-based cantilever erection method. As the span of these bridges increases, the intricacy and nonlinearity of their cable-based construction systems become more pronounced. For cable force optimization in such systems, the linear methods yield considerable inaccuracy, while the nonlinear methods based on the finite element model compromise on efficiency. To address this challenge, an erection cable force optimization method based on deep neural network (DNN) surrogate model is proposed. Firstly, the finite element model is utilized to generate force-alignment samples to depict the optimization problem. The mapping inherent in these force-alignment samples is learned by a DNN surrogate model to decouple the optimization process from the finite element analysis. Subsequently, random erection cable forces within plausible constraints are fed into the surrogate model to predict their corresponding arch rib alignments. The set of cable forces with an arch rib alignment closest to the target alignment is identified as the optimal result. Finally, the proposed method is applied to a 310 m span CFST arch bridge, and the maximum alignment error is limited within 11 mm.