Page 17 - ESAM-1-1

Engineering Science in
Additive Manufacturing ML in MAM monitoring and control through images

Figure 7. Common machine learning algorithms applied for in situ monitoring in metal additive manufacturing

A B

Figure 8. Diagram of the mechanism of supervised learning algorithms: (A) k-nearest neighbors (KNN); (B) gate recurrent unit (GRU)

A

B

Figure 9. Diagram of generative adversarial networks: (A) generator; (B) discriminator

between real and synthetic images, as shown in Figure 9A. high-quality synthetic data while the discriminator becomes
The discriminator leverages both labeled and unlabeled data a reliable classifier for real-world MAM monitoring tasks.
to improve its classification accuracy, while the generator
enhances the dataset by producing realistic synthetic 3.1.3. Unsupervised learning
images, as shown in Figure 9B. This approach reduces Unsupervised learning is a valuable approach for analyzing
dependency on labeled data, improves feature learning, and large volumes of unlabeled image data, such as melt
enables robust defect detection and process monitoring. The pool images, layer-wise deposition patterns, or defect
competitive training process ensures the generator produces signatures, without the need for manual annotations.

Volume 1 Issue 1 (2025) 11 doi: 10.36922/esam.8548