Page 61 - MSAM-4-3

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

Figure 1. Schematic illustration of the laser-arc hybrid additive manufacturing system

A yield their own prediction outputs. Thereafter, particle
swarm optimization (PSO) iteratively updates the weights
of the models, driving the weighted prediction error
downward until convergence and ultimately producing an
B ensemble prediction that fuses the four models. Finally, in
accordance with Equation II, the mean absolute percentage
error is calculated for each base model as well as for the
ensemble.

y
1 n y  i
i
MAPE 100% (II)
C n i1 y i
Particle swarm optimization is a global optimization
technique inspired by swarm intelligence, which locates
optima by modeling the coordinated, iterative movement of
many particles within a multidimensional search domain.
Every particle acts as a potential solution; its velocity and
Figure 2. Schematic diagram shows the quantification of weld-bead position are continually updated with reference to its
width and height. (A) Actual morphology of a weld bead in the personal best and the swarm’s global best, which hastens
training set. (B) 3D point cloud obtained through line-laser scanning. convergence and elevates search efficiency.
(C) Extracted weld-bead contour
To simplify the weight optimization for multi-model
subjected to hyperparameter optimization and training ensembles, we introduce, under the PSO paradigm, a one-
on the training and validation datasets. The tuned models dimensional discrete index encoding (ODIE) that assigns
are subsequently deployed on the test set, where they each weights to the base regression models with high efficiency.

Volume 4 Issue 3 (2025) 4 doi: 10.36922/MSAM025220036