Application Of BP Neural Network To Predict The Extrusion Load And Exit Temperature Evolution During Extrusion Process

Fast and accurate prediction of the extrusion load and exit temperature is of importance in choosing extrusion equipment, designing extrusion die and selecting extrusion process parameters, and it's useful for producing high-quality extrusion efficiently. There are a number of conventional methods that have been established to determine the extrusion load and exit temperature, for example through theoretical analysis, empirical equations and nomography. These methods however offer limited efficiency and accuracy in dealing with the extrusion process involving multiple, often interacting parameters. Due to approximating ability to arbitary non-linear mapping, the Back-Propagation artificial neural networks (BP ANN) has broad application in extrusion process of metal. In present paper, the influence of the shape factor on the extrusion load and exit temperatue in extrusion process was studied. The BP artificial neural networks, ANN1 was built to predict the peak value of extrusion load and rise of exit temperatue, and ANN2 was built to predict the extrusion load and exit temperature evolution in the extrusion process. Fast and accurate prediction of the extrusion load and exit temperature was realized by the built BP artificial neural networks.36 groups of extrusion processes of AZ31 magnesium alloy profiles were simulated under the extrusion conditions of shape factor (1~1.45),extrusion ratio (8.1, 17.4 and 26.4) and ram speed (1, 2 and 4 mm/s) by using DEFORM-3D software package, and the simulation results were analyzed. The shape factor, extrusion ratio, ram speed and extrusion stroke were set as the input data, and the peak value of extrusion load, rise of exit temperature and exit temperature evolution were taken as the output data. Those data were chosen as the training samples for the built BP artificial neural networks ANN1 and ANN2 with MATLAB, which were determined by comparing the performance of the network with adjusting the network parameters. After training, the networks were verified by experiments. The results of study demonstrate that the extrusion load and exit temperatue increase with the increasing of shape factor under the same extrusion condition; the predicted results agree well with the simulated ones, the differences of prediction results exhibit low value, which can satisfy the request of industry, the verified ANN1 and ANN2 possess the performance of promotion; based on the verified ANN1, the peak value of extrusion load and rise of exit temperature were predicted under the extrusion conditions of extrusion ratio 26.4, ram speed 2 mm/s and 21 individual shape factors, there is a good agreement between predictions and simulations with the maximal error of 3.5 %; the curves of extrusion load and exit temperature were predicted by the verified ANN2 under extrusion conditions of 3 individual shape factors, extrusion ratio 17.4 and ram speed 4 mm/s, and the ram speed curve for isothermal extrusion for AZ31 magnesium alloy profile was obtained at 420℃under extrusion conditions of extrusion ratio 8.1 and shape factor 1.12; the curves of predicted extrusion load under the conditions of shape factor 1.03, extrusion ratio 8.84, ram speed of 3 and 4 mm/s by the verified ANN2 agree with the experimental measurements with the average error of 6.3 %, so it has a practical application value.