About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Forming and Joining of Advanced Sheet Metal Materials
|
| Presentation Title |
Machine learning-based toolpath compensation for springback control in DSIF
|
| Author(s) |
Shayan Darzi, Matthew D. Adams, Punnathat Bordeenithikasem, Babak Raeisinia, Brad L. Kinsey, Jinjin Ha |
| On-Site Speaker (Planned) |
Shayan Darzi |
| Abstract Scope |
Double-sided incremental forming (DSIF) offers die-less production of customized sheet-metal parts, but the elastic springback that follows trimming often distorts the final geometry. This study proposes a machine-learning framework to predict post-trimming springback so that it can be pre-emptively compensated at the toolpath level. 3D scans acquired after trimming are converted into point-series descriptors capturing local geometry features, i.e., surface curvature. These descriptors are paired with anisotropic yield parameters obtained from tensile tests. Three surrogate architectures, i.e., feed-forward artificial neural network (ANN), long short-term memory network (RNN), and convolutional neural network (CNN), are trained on this combined feature set to forecast the normal displacement field expected after trimming. By mapping predicted deviations back onto the CAD surface, the method generates a compensated toolpath that targets shape fidelity in subsequent DSIF runs. This framework significantly reduces material loss due to trial-and-error iterations in producing accurate parts using the DSIF process. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, Shaping and Forming, |