Neural network model for determining the critical bending moment of web-tapered beams

Resumo

In the design of web-tapered members, determining the critical bending moment is a complex task since the non-prismatic geometry of the members leads to difficulties in their stability analysis. This paper proposes a neural network regression model to obtain the critical bending moment of web-tapered members. A database of 15,000 critical bending moment values for different profile geometries was generated through numerical analyses to train and evaluate the neural network models. The model was developed using the Adam optimizer and the influence of predictor variables was examined. The findings suggest that the use of both geometric and slenderness parameters as input variables increases the model’s performance. The proposed model achieved a mean average percentage error of approximately 3% in predicting the critical bending moment. Finally, the model predictions were compared with a formulation from the literature. A significant improvement in prediction accuracy was observed using the regression model compared to the considered equations.