.png){kind=icon}
.png){kind=icon}
Return
Multi-objective optimization of stamping process based on physics-guided extra trees for large thin-walled parts
.png){kind=icon}
PRE
AI
X
T
J
X
Y
S
DOI：10.1007/s10999-026-09950-0.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Stamping of large thin-walled metal covers involves complex multi-objective conflicts among cracking, wrinkling, and springback, where purely data-driven models often suffer from poor generalization under limited sample conditions. To address this, a Physics-Guided Extra Trees (PG-ET) collaborative optimization framework is proposed. The framework reformulates the optimization problem by introducing defect-oriented physical descriptors derived from plastic deformation and frictional instability mechanisms. First, a physics-guided feature pool containing high-order coupling descriptors is constructed, followed by target-specific feature decoupling using permutation importance analysis. Subsequently, Yeo-Johnson target transformation and the fully randomized partition mechanism of Extra Trees are combined to improve robustness against feature collinearity. The developed PG-ET model achieves high prediction accuracy: the absolute error for thinning is as low as 0.0024, the coefficient of determination (R2) for thickening reaches 0.9662, and the R2 for springback prediction reaches 0.8637. Finally, the PG-ET surrogate is integrated with Non-dominated Sorting Genetic Algorithm II (NSGA-II) and Technique for Order Preference by Similarity to an Ideal Solution (TOPSIS) decision-making to identify the optimal process parameters. Industrial validation on a 55-inch television (TV) backplane demonstrates that, while preventing cracking, the proposed framework reduces macro-warpage by 48.5% and maximum thickening by 26.6%. The results indicate that embedding physically interpretable deformation descriptors into surrogate modeling effectively enhances optimization reliability for complex thin-walled forming.
Keywords:
Sheet metal forming
Surrogate model
Physics-guided machine learning
Multi-objective optimization
Feature engineering
Journal
IF:
3.6
Papers:
215
Citations:
1.5K
Organization
Cited Papers