Deep Convolutional Neural Network to Assist Die Design for Flow Balance of Aluminum Hollow Extrusion

The demand for aluminum alloy extrusion is increasing and the requirements for the accuracy and geometric complexity of extrusion products are becoming more stringent. However, the traditional extrusion die design often relies on the designer's experience, therefore, a lot of die refinement works are still needed. To avoid this situation, deep learning technology, namely deep convolution neural network (DCNN), is applied to obtain suitable design parameters for extrusion die. In this study, DCNN in extrusion die design is mainly divided into two parts: suggesting to porthole structure at the upper die for hollow extrusion and determining the bearing length for balancing the metal flow in hollow extrusion process. First, the cross-sectional shapes of the products from the over 100 extrusion literatures are used as training data set to establish four neural network training models which belong to image classification. Second, the training label for the object detection model is built by assigning a complexity factor for the die bearing length of each feature on the product profile. Next, the design parameters provided by DCNN models are used to design the extrusion die as a validation test. Finally, a finite element analysis based on the obtained die design is conducted to verify the flow balance of extrusion process.