About this Abstract |
Meeting |
2023 TMS Annual Meeting & Exhibition
|
Symposium
|
Characterization of Minerals, Metals and Materials
|
Presentation Title |
Evaluation of Feature Engineering Methods for the Prediction of Sheet Metal Properties by an Artificial Neural Network from Punching Force Curves |
Author(s) |
Marcel Goerz, Adrian Schenek, Mathias Liewald, Kim Rouven Riedmüller |
On-Site Speaker (Planned) |
Marcel Goerz |
Abstract Scope |
The part quality that can be achieved in forming and stamping processes depends on the properties of the sheet metal material to be processed. Since these properties may fluctuate considerably, however, and thus lead to the production of scrap, it is therefore important to monitor such material fluctuations during part production. For this, the ongoing digitization of production processes provides new possibilities for part or quality monitoring. In this context, a novel AI-based method for the direct determination of material parameters from punching force curves measured in production was presented in a past study of the IFU. This paper deals with the investigation of three methods for extracting features from these measuring data. In addition to domain knowledge-based feature engineering, statistical feature extraction (PCA) as well as a gradient-based method are analyzed and compared regarding their influence on the prediction accuracy of sheet metal properties by the AI (ANN) used. |
Proceedings Inclusion? |
Planned: |
Keywords |
Mechanical Properties, Machine Learning, Iron and Steel |