An artificial neural network model for predicting the behaviour of semi-rigid joints in fire

This paper presents an artificial neural networking (ANN) model developed to predict the behaviour of semi-rigid bare-steel joints at elevated temperature. Data for three flush end-plate and one flexible end-plate joints were considered. Sixteen parameters which included geometry of the joint's components, material properties of the joint, joint's temperature and the applied moment were used as the input variables for the model whilst the joint's rotation was the main output parameter. Data from experimental fire tests were used for training and testing the model. In total, fifteen different test results were evaluated with 331 and 61 cases were used for training and testing the developed model, respectively. The model predicted values were compared with actual test results. The results obtained indicated that the model can predict the moment-rotation behaviour in fire with very high accuracy. The coefficients of determination (R²) for training and validation of the model were 0.964 and 0.956, respectively.