Page 71 - MSAM-4-3

Materials Science in Additive Manufacturing Bead geometry prediction in laser-arc AM

which its influence on weld bead formation can
undergo significant shifts in direction or magnitude.
Although the method delivers encouraging results,
PSO-EP presently depends on optimizing base model
weights within a fixed framework, thereby constraining
its flexibility and extensibility. Hence, future studies might
explore more self-adaptive or hierarchical ensemble
schemes to augment the model’s representational capacity
and generalization.

Acknowledgments
None.

Funding
This work was financially supported by the CNPC
Figure 14. The workflow of path planning
Innovation Foundation (Grant No. 2024DQ02-0306),
Innovation and Entrepreneurship Leading Talent
4. Conclusion Project of Yantai Development Zone in 2022 (Grant
In this study, we devised a PSO-driven ensemble No. 2022RC008), Natural Science Foundation of Shandong
regression method, designated PSO-EP, for accurately Province (Grant No. ZR2023QE164), Natural Science
forecasting weld-bead size in the multiphysics-coupled Foundation of Qingdao (Grant No. 23-2-1-83-zyyd-jch),
LAHAM technique. The method leverages PSO to tune and National Natural Science Foundation of China (Grant
the weightings of several base models and, in turn, elevates No. 52405359).
the aggregate prediction accuracy. The effectiveness of
PSO-EP was assessed through extensive comparisons Conflicts of interest
against individual learners (GPR, SVR, ANN, and ELM) The authors declare no competing interests.
and representative ensembles such as averaging, stacking,
and ELGA. The findings showed that PSO-EP delivers Author’s contributions
top-ranked accuracy for predicting both weld-bead width Conceptualization: Xingwang Bai
and height. Formal analysis: Youheng Fu
(1) PSO-EP demonstrated the best performance in weld- Investigation: Boce Xue, Changze Li
bead width prediction, achieving an MAE of 0.1454, Methodology: Kui Zeng
an RMSE of 0.1890, and an R of 0.9567; compared Resource: Yonghui Liu, Yanzhen Zhang
2
with the next-best SVR (R of 0.9510) as well as Writing–original draft: Hui Ma, Runsheng Li
2
averaging (R of 0.9373), stacking (R of 0.9246), and Writing–review & editing: Runsheng Li
2
2
ELGA (R of 0.9346), it improved R by 2.07%, 3.47%,
2
2
and 2.36%, respectively Ethics approval and consent to participate
(2) PSO-EP likewise excelled in predicting weld-bead Not applicable.
height, registering an MAE of 0.0505, an RMSE of
0.0571, and an R of 0.9492, substantially surpassing Consent for publication
2
ANN (R of 0.9289), SVR (R of 0.9222), and averaging
2
2
(R of 0.9387), stacking (R of 0.9378), and ELGA (R Not applicable.
2
2
2
of 0.9206), with R gains of 1.12%, 1.22%, and 3.11 %, Availability of data
2
respectively
(3) SHAP interpretability analysis indicates that weld Data are available from the corresponding author upon
bead width is primarily influenced by the combined reasonable request.
effects of welding speed, wire feed speed, and laser
power, whereas weld bead height prediction is driven References
mainly by laser power and welding speed 1. Chen X, Fu Y, Kong F, et al. An in-process multi-feature data
(4) Subsequent SHAP threshold analysis uncovered that fusion nondestructive testing approach for wire arc additive
each process parameter exhibits a threshold above manufacturing. Rapid Prototyp J. 2022;28(3):573-584.
Volume 4 Issue 3 (2025) 14 doi: 10.36922/MSAM025220036