Study on hysteresis performance of four-limb CFST latticed column-box girder joints based on GA-BP neural network

Abstract

This study investigates the seismic performance and energy dissipation capacity of the concrete-filled steel tube (CFST) latticed column-box girder joints. A low-cycle reciprocating load test was carried out on the CFST latticed column-box girder joints with three different connection forms. Based on the test results, a genetic algorithm (GA) is developed in combination with the back propagation neural network (BPNN) model for prediction research. Using a training set of joints with single-limb diagonal and cross-diagonal braces (training set), the development of load-displacement hysteresis curves of joints with transverse diaphragm plate (test set) under low-cycle reciprocating loads is predicted. The test results are compared with those obtained from the traditional algorithms and finite element simulations. The results on the joint hysteresis performance show that all three types of joints have a good seismic performance and energy dissipation capacity. Compared with the traditional algorithms, the GA-BP model predicts more accurate results and is therefore used to predict the load-displacement hysteresis curves of CFST latticed column-box girder joints, which is shown to be convenient and feasible.