Research on Binary Feature Encoding Approach of ANN and It's Application to Cold Extrusion Process Design (ID: 8-129)

摘要

In this paper, binary feature encoding approach, ANN and their applications to cold extrusion process design were studied. Feature encoding approach, which means how to transform the geometric and forming information of the cold extruded parts into data that ANN can identify and deal with, is one of the key techniques of applying ANN to cold extrusion process design system. Here, the binary encoding approach, which was utilized to express features of the cold extruded parts in ANN-based cold extrusion process design system, was put forward and improved accroding to some principles. The encoding formats of geometric shape features, geometric relation features, precision features, and basic forming process features were encoded under the consideration of encoding precision and network running efficiency. This encoding approach has advantages as follows: the number of features and feature codes is reduced; the difference between feature codes with same types is extended, and features can be identified easily in the process of ANN mapping; feature codes become more even and reasonable; the running precision of ANN is improved in a certain extent.The cold extruded parts can be expressed almost completely by feature codes together with feature parameters. The training and testing sample sets of ANN were selected using TAGUCHI method. In the end, the approach and ANN-based cold extrusion process design system were validated through a typical cold extruded sample.