The Additive Manufacturing (AM) paradigm and its particular realization - Laser Powder Bed Fusion (LPBF) - rapidly incorporates modern prototyping and fabrication routines. However, the printed specimens' structural defects affect the method. Cracks are the most crucial of them. The acoustic emission (AE) monitoring system is a promising monitoring method for L-PBF, capable of detecting cracks and differentiating them from noise (e.g., inert gas circulation pump). We constructed the dataset of AE events employing a structure-born high-frequency sensor for high-strength Al-based alloy. The AE events in the dataset were labeled manually in the binary setup (crack - noise)[1].

The experimental dataset of the AE events was employed for training and validation of the ML models. The events were represented as finite feature vectors of standardized moments and principal components of waveforms and spectra. We performed a 6-fold cross-validation (CV) procedure for model selection among Logistic Regression, Support Vector Machine, RandomForest, and Gaussian Process Classifier. The CV showed that the Gaussian Process Classifier shows the best accuracy (~99 %) utilizing the least number of features. The developed ML-based approach allows for in-situ detection of the cracks [2].
We plan to continue our study by extending the dataset with data for new alloys and exploring latent space descriptions of the AE events within autoencoders.

References
[1] M. Seleznev, T. Gustmann, J. M. Friebel, U. A. Peuker, U. Kühn, J. K. Hufenbach, H. Biermann, A. Weidner, Additive Manufacturing Letters, 2022, 3, 100099
[2] D. Y. Kononenko, V. Nikonova, M. Seleznev, J. van den Brink, D. Chernyavsky, Additive Manufacturing Letters, 2023, 5, 100130