Page 70 - MSAM-4-3

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

A B

C D

E

Figure 13. Process parameter dependence plots in the width prediction task: (A) Wire feed speed (v ), (B) Welding speed (v ), (C) Arc length correction
w
t
(l), (D) Pulse correction (f), and (E) Laser power (p)
the trained U-net model, which segments and crops the ∆H = H−H ,H’=H+∆H (XI)
real
bead and online extracts the actual bead width W and
real
height H . After comparing the measured and desired The trajectory of next-generation inspection and
real
dimensions, Equations X and XI compute the required monitoring architectures has already been examined
correction to obtain new targets W’ and H’; these targets in previous works. 49,50 Both studies underscore the
feed the accurate bead predictor, which derives revised considerable potential of AI-driven technologies to deliver
process parameters and refreshes the G-code. The machine more robust defect detection and real-time intervention,
resumes deposition under the updated parameters, the thereby substantially enhancing the reliability of additive
camera repeats image acquisition, and the cycle iterates manufacturing processes. Although these strategies have not
in a closed loop until the bead size remains inside the yet been implemented in the present study, subsequent work
specified error band. will focus on designing and validating the corresponding
control mechanisms with the aim of establishing a fully
∆W = W−W W’=W+∆W (X) automated closed-loop quality control system.
real,

Volume 4 Issue 3 (2025) 13 doi: 10.36922/MSAM025220036