Prediction analysis of deflection in the construction of composite box-girder bridge with corrugated steel webs based on MEC-BP neural networks

Abstract

This paper presents an efficient computational framework addressing the deflection prediction problem for prestressed concrete (PC) rigid frame bridge with corrugated steel webs (CSWs) erected by the balanced cantilever method. The framework consists of three stages of analysis: (1) obtain the datasets for establishing the prediction model, in which the input samples and output response are generated by the Latin hypercube sampling (LHS) technique and high-fidelity three-dimensional (3D) finite element (FE) model, respectively; (2) based on the datasets, an optimized neural network model that integrates with mind evolutionary computation (MEC) algorithm and back-propagation (BP) algorithm is established, then the model performance is evaluated by the statistical criteria; and (3) the well-trained MEC-BP model is used to predict the deflection for CSWs-PC bridge construction. The applicability and efficiency of the framework are demonstrated with a real CSWs-PC bridge. Results show that the proposed method achieves much better prediction performance than the conventional BP model, and can deliver an efficient prediction tendency for the deflection control of a CSWs-PC bridge during the complex construction process.