Abstract

In this paper, the potential of building more accurate and robust models for the prediction of the ultimate pure bending capacity of steel circular tubes using artificial intelligence techniques is investigated. Therefore, a database consisting of 104 tests for fabricated and cold-formed steel circular tubes are collected from the open literature and used to train and validate the proposed data-driven approaches which include the Random Forest methodology in two variants: the original version in which the control parameters are manually updated, and an enhanced RF-PSO variant, where Particle Swarm Optimization is used for optimizing these parameters. The data set has four input parameters, namely the tube thickness, tube diameter, yield strength of steel and steel elasticity modulus, while the ultimate pure bending capacity is considered as the target output variable. The obtained results are compared to the real test values through various statistical indicators such as the root mean square error and the coefficient of determination. The results indicate that the proposed enhanced model can provide an accurate solution for modelling the complex behavior of steel circular tubes under pure bending conditions.