Rapid prediction of the ultimate moment of flush endplate connections at elevated temperatures through an artificial neural network

Abstract

Steel connections are vulnerable structural components, and their failure can lead to the collapse of the whole structure in a fire. Hence, it is crucial to predict the connection behaviors accurately to enable a precise prediction of structural performance in the fire. This paper proposes an efficient approach for predicting the ultimate moment of flush endplate connections at elevated temperatures using the finite element method and an Artificial Neural Network (ANN). Firstly, a finite element model considering both geometric and material nonlinearities is developed and verified with existing experimental results conducted by others. Based on the validated finite element model, an ABAQUS plug-in, namely COinFIRE, is implemented in ABAQUS to facilitate the parametric study. Consequently, 886 finite models are generated and used as training data for developing the ANN model. Accordingly, the endplate thickness, the pitch of bolts, the bolt row distance, the outer edge distance, the gage distance, the bolt diameter, the number of bolt rows, and the material properties at elevated temperatures are considered input variables. Meanwhile, the output is the ultimate moment. The obtained results show that the proposed ANN model can accurately predict the ultimate moment of flush endplate connections (R2 = 1.0, 1.0, 1.0, RMSE = 0.811 kN.m, 1.627 kN.m, 1.256 kN.m, and a20-index = 1.0, 1.0, 1.0 for the training, testing and validation datasets, respectively). Finally, closed-form formulas and a graphical user interface (GUI) tool are developed to evaluate the ultimate moment of flush endplate connections at elevated temperatures for practical use. The proposed approach shows great promise as a powerful design tool for steel connections at elevated temperatures.