Details
| Original language | English |
|---|---|
| Title of host publication | Material Forming ESAFORM 2025 |
| Subtitle of host publication | The 28th International ESAFORM Conference on Material Forming – ESAFORM 2025 |
| Editors | Pierpaolo Carlone, Luigino Filice, Domenico Umbrello |
| Pages | 889-898 |
| Number of pages | 10 |
| ISBN (electronic) | 978-1-64490-359-9 |
| Publication status | Published - 2025 |
| Event | 28th International ESAFORM Conference on Material Forming, ESAFORM 2025 - Paestum, Italy Duration: 7 May 2025 → 9 May 2025 |
Publication series
| Name | Materials Research Proceedings |
|---|---|
| Volume | 54 |
| ISSN (Print) | 2474-3941 |
| ISSN (electronic) | 2474-395X |
Abstract
Fluctuations in process parameters are a significant cost driver in hot forging. Particularly in multi-stage processes, fluctuations in the early process steps can lead to components having to be declared as scrap. Especially in the processing of hybrid components, these fluctuations pose a major problem, as the manufacturing costs are higher and the joining zone properties, which are very susceptible to process fluctuations, have a strong influence on the properties of the resulting component. The aim of this work is to develop an AI-supported solution for inline parameter optimisation. This approach allows for the compensation of process fluctuations by adjusting subsequent process steps, in order to still achieve end products that meet the requirements. FE-Simulations were carried out, whereby the boundary conditions for the induction current and the press speed were each varied by multiplying the original time curves with a normal distributed factor in order to simulate process noise. The resulting data was used to train a machine learning model that predicts the maximum first principal stress as indicator for the condition of the joining zone and the contact from the semi-finished product to the workpiece as indicator for mould filling. An evolutionary algorithm was used to optimise the press speed and the stroke in order to maximise contact and minimise the maximum first principal stress. Finally, the prediction time was minimised while maintaining the prediction accuracy. The approach presented promises a significant reduction in waste by enabling dynamic and predictive adjustment of process parameters in real time. This not only leads to an increase in efficiency, but also to a reduction in costs in the manufacturing process.
Keywords
- Cup Backward Extrusion, Machine Learning, Process Optimisation
ASJC Scopus subject areas
- Materials Science(all)
- General Materials Science
Cite this
- Standard
- Harvard
- Apa
- Vancouver
- BibTeX
- RIS
Material Forming ESAFORM 2025: The 28th International ESAFORM Conference on Material Forming – ESAFORM 2025. ed. / Pierpaolo Carlone; Luigino Filice; Domenico Umbrello. 2025. p. 889-898 (Materials Research Proceedings; Vol. 54).
Research output: Chapter in book/report/conference proceeding › Conference contribution › Research › peer review
}
TY - GEN
T1 - AI supported process optimisation in a multi-material cup backward extrusion process chain
AU - Ortlieb, Eduard
AU - Wester, Hendrik
AU - Uhe, Johanna
AU - Behrens, Bernd-Arno
N1 - Publisher Copyright: © 2025, Association of American Publishers. All rights reserved.
PY - 2025
Y1 - 2025
N2 - Fluctuations in process parameters are a significant cost driver in hot forging. Particularly in multi-stage processes, fluctuations in the early process steps can lead to components having to be declared as scrap. Especially in the processing of hybrid components, these fluctuations pose a major problem, as the manufacturing costs are higher and the joining zone properties, which are very susceptible to process fluctuations, have a strong influence on the properties of the resulting component. The aim of this work is to develop an AI-supported solution for inline parameter optimisation. This approach allows for the compensation of process fluctuations by adjusting subsequent process steps, in order to still achieve end products that meet the requirements. FE-Simulations were carried out, whereby the boundary conditions for the induction current and the press speed were each varied by multiplying the original time curves with a normal distributed factor in order to simulate process noise. The resulting data was used to train a machine learning model that predicts the maximum first principal stress as indicator for the condition of the joining zone and the contact from the semi-finished product to the workpiece as indicator for mould filling. An evolutionary algorithm was used to optimise the press speed and the stroke in order to maximise contact and minimise the maximum first principal stress. Finally, the prediction time was minimised while maintaining the prediction accuracy. The approach presented promises a significant reduction in waste by enabling dynamic and predictive adjustment of process parameters in real time. This not only leads to an increase in efficiency, but also to a reduction in costs in the manufacturing process.
AB - Fluctuations in process parameters are a significant cost driver in hot forging. Particularly in multi-stage processes, fluctuations in the early process steps can lead to components having to be declared as scrap. Especially in the processing of hybrid components, these fluctuations pose a major problem, as the manufacturing costs are higher and the joining zone properties, which are very susceptible to process fluctuations, have a strong influence on the properties of the resulting component. The aim of this work is to develop an AI-supported solution for inline parameter optimisation. This approach allows for the compensation of process fluctuations by adjusting subsequent process steps, in order to still achieve end products that meet the requirements. FE-Simulations were carried out, whereby the boundary conditions for the induction current and the press speed were each varied by multiplying the original time curves with a normal distributed factor in order to simulate process noise. The resulting data was used to train a machine learning model that predicts the maximum first principal stress as indicator for the condition of the joining zone and the contact from the semi-finished product to the workpiece as indicator for mould filling. An evolutionary algorithm was used to optimise the press speed and the stroke in order to maximise contact and minimise the maximum first principal stress. Finally, the prediction time was minimised while maintaining the prediction accuracy. The approach presented promises a significant reduction in waste by enabling dynamic and predictive adjustment of process parameters in real time. This not only leads to an increase in efficiency, but also to a reduction in costs in the manufacturing process.
KW - Cup Backward Extrusion
KW - Machine Learning
KW - Process Optimisation
UR - http://www.scopus.com/inward/record.url?scp=105008061484&partnerID=8YFLogxK
U2 - 10.21741/9781644903599-95
DO - 10.21741/9781644903599-95
M3 - Conference contribution
SN - 9781644903599
T3 - Materials Research Proceedings
SP - 889
EP - 898
BT - Material Forming ESAFORM 2025
A2 - Carlone, Pierpaolo
A2 - Filice, Luigino
A2 - Umbrello, Domenico
T2 - 28th International ESAFORM Conference on Material Forming, ESAFORM 2025
Y2 - 7 May 2025 through 9 May 2025
ER -