Page 43 - IJAMD-1-1

International Journal of AI
for Material and Design ML for quality improvement in L-PBF

Figure 12. Flowchart of the work proposed by Ansari et al. 59

Figure 13. Schematic of a convolutional neural network model of the work proposed by Ansari et al. 59

The original labeled data assists in the evaluation of the were trained to detect flaws in layer-wise images, with the
clustering model during the K-means algorithm stage. objective of predicting the presence of defects in printed
67
Some research works have sought to predict the products. An additional approach was presented by
occurrence of faults in the final product by analyzing in situ Pandiyan et al., involving two generative CNN architectures
collected data. For example, Terry et al. utilized the height based on variational auto-encoder (VAE) and general
data of high-resolution powder bed surfaces obtained from adversarial networks (GAN). These architectures were
a laser profilometer to predict the quality of products, applied and designed to predict the existence of defects
distinguishing between faulty or nominal products using a by analyzing in situ acoustic signatures. The models were
68
shallow ANN. In the work by Snow et al., NN and CNN semi-supervised and trained with acoustic signals from
66
Volume 1 Issue 1 (2024) 37 https://doi.org/10.36922/ijamd.2301