A new predictive model for restrained distortional buckling strength of half-through bridge girders using artificial neural network

Half-through girders are not affected by conventional lateral-torsional buckling. I-section beams of simply supported half-through girders experience compression in their top flanges and tension in their bottom flanges. In this condition, the compression flange is restrained only by the stiffness of the web, and the buckling mode is generally restrained distortional. In this study, new and efficient model is derived to predict the Restrained Distortional Buckling (RDB) strength of half-through I-section bridge girders utilizing an Artificial Neural Network (ANN). The model is developed based on a reliable database obtained from the nonlinear finite element (FE) method. To verify the accuracy of the derived model, it is applied to estimate the RDB strength of parts of the FE analysis results that were not included in the modeling process. A sensitivity analysis has been also developed to determine the importance of each input parameters. ANN model is further compared to the some existing design codes. The results indicate that the proposed model is effectively capable of evaluating the RDB load of the half-through girders. The prediction performance of the ANN model is markedly better than prediction of the AISC/LRFD and the AS4100 specifications. The ANN-based design equation can reliably be employed for pre-design applications.