Application of ANN to the hot extrusion of magnesium alloy sheets

Abstract

During the hot extrusion of magnesium alloy, its structure is hexagonal close-packed, which has fewer slide surfaces than aluminium alloy. Experiments show that if the hot extrusion of magnesium alloy sheets is performed under a die that has a high extrusion ratio of 35.9, and a fixed speed is adopted, all of the final products carry defects. Restated, sound sheets with zero-defects cannot be obtained. Further experiments were performed on the extrusion of magnesium alloy sheets by the multi-speed method (MSM), and sound sheets were obtained. However, when the multi-speed method of extruding was used, adjusting the initial speed at the appropriate time directly affected the results of experiments. This paper addresses an experiment designed by the method of orthogonal array (OA), and uses magnesium alloy AZ31 and AZ61 as outer OAs, whereas the factors selected as inner OAs are the temperature of the container, the temperature of the material, the initial speed of extrusion, the final speed and the lubricant. Moreover, the OA of L18 is selected to conduct experiments repeatedly and then apply artificial neural networks (ANN) to learn the results of the experiment. To forecast the temperature of the material for an experiment between 340 and 390 °C, curves of the timing of an adjustment of the initial speed of extrusion by the multi-speed method with increments of 5 °C should be plotted. Experiments are performed to confirm the accuracy of ANN analysis.