Optimization of performance of artificial neural network for predicting the tensile properties of friction stir welded al-5083

In this research, the optimization of the artificial neural network (ANN) capability for predecting the tensile strength and elongation of friction stir welded Al-5083 (FS-welded Al-5083) was carried out. The effective parameters of ANN, such as the number of layers, number of neurons in hidden layers, transfer function between layers, the learning algorithm and etc. were investigated and the efficient neural network was determined to predict the tensile properties of FS-welded Al-5083. The investigations revealed that the perceptron neural network with two hidden layers and 17 neurons numbers, Lunberg-Marquardt training algorithm and Logsig transfer function for the intermediate layers and Tansig transformation function for the output layer is the most optimized neural network for the prediction. The optimized network has an optimal structure based on the minimum value of the mean square error of 0.05, the maximum total correlation coefficient of 0.93 and the line regression with an angle of 45 degrees between the actual and estimated values. Therefore, this network has a good performance for training, generalizing and estimating of tensile strength and elongation of FS-welded Al-5083.