PSI - Issue 77

A. Polanský et al. / Procedia Structural Integrity 77 (2026) 529–536 Adam Polanský / Structural Integrity Procedia 00 (2026) 000 – 000

535

7

Table 3. Evaluation of the models testing accuracy

Input size [ px ]

MiniVGGNet [ % ]

Resnet50 [ % ]

ConvNeXT-Tiny [ % ]

96,4 ± 0,6

400x400

90,8 ± 0,8

92,6 ± 1,5

96,8 ± 0,8

800x800

94,3 ± 1

95,8 ± 0,9

Table 4. Performance metrics of the tested models Input size [px] Model True Negative

False Positive

False Negative

True positive

Precision [%]

Recall [%]

F1 [%] 91,4 93,4 93,0 95,9 96,4 96,9

400X400 800X800 400X400 800X800 400X400 800X800

MiniVGGNet MiniVGGNet

84,4 88,0 87,4 93,6 95,2 95,0

15,6 12,0 12,6

2,8 1,8 2,2 2,0 2,4 1,4

97,2 98,2 97,8 98,0 97,6 98,6

86,2 89,1 88,6 93,9 95,3 95,2

97,2 98,2 97,8 98,0 97,6 98,6

Resnet 50 Resnet 50

6,4 4,8 5,0

ConvNeXT Tiny ConvNeXT Tiny

6. Conclusion In this paper, we present a novel protocol for automatic assessment of the SLM quality layer. The protocol is based on CNN, which classifies each layer separately. In the preliminary experiment, three different CNN models were trained to perform binary classification of powder bed pictures. The best accuracy was achieved by the ConvNeXt-Tiny model, reaching 96,8 % accuracy. In the following steps, we will focus on the segmentations of common defects occurring within layers and account for the position of printed parts. To achieve a more universal model, it’s necessary to collect extra images, ideally from different SLM printers. The segmentation is related to basic research focusing on how the mentioned defects affect the real material. Specifically, maraging steel 18Ni300 will be used. Other sensors, such as an accelerometer mounted on the recoater, will be used to obtain additional information about the printing process itself. 7. Acknowledgment This research “Using machine learning for quality control in additive manufacturing” was supported by the project CZ.02.01.01/00/23_021/0009165 “Development of digital twins of structural components with support of on -line monitoring of their operational l oading and simulations in laboratory conditions” funded by the Johannes Amos Comenius Programme. The project is co-funded by the European Union. 8. References

[1] Fukushima, K., 1980. Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position . Biological Cybernetics, 36(4), pp.193 – 202.

[2] Krizhevsky, A., Sutskever, I. & Hinton, G.E., 2012. ImageNet Classification with Deep Convolutional Neural Networks . In: F. Pereira, C.J.C. Burges, L. Bottou & K.Q. Weinberger, eds., Advances in Neural Information Processing Systems , 25. Lake Tahoe, Nevada, 3 – 6 December 2012, pp.1097 – 1105.