Reducing scrap products and unnecessary rework has always been a goal of the manufacturing industry. With the increasing data availability and the developments in the
field of artificial intelligence (AI) for industrial applications, machine learning (ML) has been applied to radial-axial ring rolling (RARR) to predict product quality [1]. However,
the accuracy of these predictions is currently still limited by the quantity and quality of the data [2]. In order to apply supervised learning to predict part quality and possible
scrap parts, there must be plenty of datasets logged for both good and scrap parts. One suitable way to increase the number of datasets is to utilize simulation strategies
to generate synthetic datasets. However, in the hot ring rolling field, there is no fast simulation method that can be used to generate a sufficiently large synthetic database
of rolled parts with form or process errors. The research on transfer learning between different mills and datasets has offered a new idea of taking a cold ring rolling process
as the object of study [2]. Next it will investigate the extent to which the cold ring rolling can be used as a similar process for future transfer of models and results to radial-axial
ring rolling. Compared to RARR, the cold ring rolling is a process under room temperature and contains complete radial forming instead of simultaneous forming in the radial
and axial directions. The simpler forming mechanism makes it possible to build a semi-analytical model, which takes much less time compared to conventional FEMapproaches
under acceptable accuracies. Furthermore, the smaller ring geometry, simplified rolling process and reduced energy consumption mean that in-house experiments can be conducted to verify the quality of the synthetic data based on confidence intervals.
| Identifer | oai:union.ndltd.org:DRESDEN/oai:qucosa:de:qucosa:88293 |
| Date | 28 November 2023 |
| Creators | Wang, Qinwen, Seitz, Johannes, Lafarge, Rémi, Kuhlenkötter, Bernd, Brosius, Alexander |
| Publisher | Technische Universität Dresden |
| Source Sets | Hochschulschriftenserver (HSSS) der SLUB Dresden |
| Language | English |
| Detected Language | English |
| Type | doc-type:conferenceObject, info:eu-repo/semantics/conferenceObject, doc-type:Text |
| Rights | info:eu-repo/semantics/openAccess |
| Relation | 06, urn:nbn:de:bsz:14-qucosa2-882872, qucosa:88287, urn:nbn:de:bsz:14-qucosa2-882924, qucosa:88292 |
Page generated in 0.0018 seconds