Comparative evaluation of 3-d numerical and artificial neural network models for predicting extrusion pressure and product deflection

摘要

This paper presents the comparative study of the developed 3-D numerical and artificial neural network (ANN) models for predicting extrusion pressure and product deflection of aluminium extrudes. An extensive experimental program was undertaken to extrude a aluminuim (Al) alloy on ELE Compact-1500 compression machine. The developed 3-D numerical model for the prediction of extrusion pressure and product deflection was based on two technological parameters (die bearing length, die diameter and other parameters as coefficient of friction, flow stress and two, constants. The Artificial neural network model for extrusion pressure and product deflection was developed based on groups of experiments carried out as samples, Eight (8) parameters (die bearing length, radius of curvature, slip angle, die angle, die ratio ram displacement, pocket depth and die diameter) were used as inputs into the network architecture of 8 [4-3]_2 2 in predicting the extrusion pressure and product deflection. After series of network architectures were trained using different training algorithms such as Levenberg-Marquardt, Bayesian Regulation, Resilient Backpropagation using MATLAB 7.9.0 (R20096), the LM8 [4-3]_2 2 was selected as the most appropriate model. The predicted extrusion pressure and product deflection by 3-D numerical and ANN models were compared with the measured values using some statistical indicators such as correlation coefficient (R), mean absolute error (MAE), root means square error (RMSE), and Nash-Scutcliffe efficiency (NSE). The results showed that the ANN model presented a better prediction, but the developed 3-D numerical model was found to be equally useful for extrusion pressure and product deflection.