Forming state recognition in deep drawing process with machine learning

Abstract

In press processing, quality inspection of a product is often carried out for each lot in the post process stage. When a failure occurs, it may result in a large number of defective products due to the fast processing speed. In order to prevent this, it is ideal to immediately stop the processing just after the defect occurs. Therefore, confirming the processing state in-process is required. This study proposes a new quality inspection method for deep drawing processes by using the count rate of acoustic emission (AE) signals. To analyze the AE count, deep learning, which is a multilayered neural network, is employed to recognize defects during a deep drawing process. The material used was a ductile material of cold rolled steel plate and is relatively difficult to find cracks during the deep drawing process. Characteristics were clarified by analysis of AE counts at plastic deformation and brake of material, and performing the forming state recognition experiment by a multilayer neural network (deep learning) showed a maximum recognition rate of 97.3%. High recognition rate was obtained despite the small number of data used.

Keywords

• sheet metal forming
• acoustic emission
• deep learning
• deep drawing
• data acquisition
• sensor