Page 31 - ESAM-1-4

Engineering Science in
Additive Manufacturing Machine learning for biomedical metal AM

A

B

C

Figure 4. Models and results for macrostructure quality prediction. (A) Artificial neural network model for density prediction; (B) support vector
68
machine model for density prediction; and (C) deep neural network model for surface roughness prediction. 75
68
effectively predict and control macro-topography even effective pathway for the precise control and optimization
in the challenging WAAM process. Similarly, So et al. of surface roughness, thereby forming a crucial foundation
75
used a deep neural network (DNN) to predict the surface for enhancing overall product quality and reliability.
roughness between consecutively stacked layers in the
WAAM process (Figure 4C). Wire feed speed, travel speed, 2.3. Microstructure forward prediction
and the geometric features of the previously deposited layer Microstructure serves as the critical bridge linking AM
were utilized as inputs. The DNN model exhibited excellent process parameters to macrostructure properties.
76
performance on the test set, with an RMSE of 0.03 and a Factors such as grain size, morphology, and phase
high correlation coefficient (r = 0.97) between predicted composition directly influence a material’s mechanical
and actual values, demonstrating strong predictive and bio-functional properties. However, compared with
77
capability. Surface roughness prediction through ML conventional manufacturing methodologies, the inherently
models, ranging from ensemble methods such as XGBoost intricate kinetics of AM processes can significantly alter
to hybrid systems like GA-ANFIS and deep learning the solidification behavior and grain structure of alloys
approaches like DNN, has shown considerable promise during fabrication. Therefore, accurately predicting
across different AM technologies, including LB-PBF and microstructure is central to achieving customized
WAAM. These data-driven prediction methods provide an performance in biomedical metals.

Volume 1 Issue 4 (2025) 9 doi: 10.36922/ESAM025440031