Reduction of Defect Rate in Impact Extrusion by Slide Motion Control Using Machine Learning

Abstract

Intelligent slide motion control was achieved by using a hydraulic press capable of controlling slide motions in combination with artificial intelligence. Computer-aided engineering (CAE) was used to clarify the factors that cause extrusion defects and slide motions with a high ratio of good products for impact extrusion. Next, experiments were conducted to verify the effect of slide motions with a high ratio of good products, and the results obtained by CAE were confirmed to be correct. Then, a convolutional neural network was constructed with a capability of using the elastic strain of the punch in the extrusion initial stage obtained in the experiments as input data for machine learning to predict the extrusion quality, and this network was implemented in a press machine. We confirmed that when a defect was predicted, the machine automatically switched to a slide motion with a high ratio of good products. By using this intelligent slide motion control, we can change the extrusion of a predicted defective product to that of a good product by adjusting the slide motion to reduce the incidence of defective products.