UDC 004.896
AUGMENTATION OF AUDIO DATA FROM CYBER-PHYSICAL LEVEL OF DIGITAL TWINS OF TECHNOLOGICAL SYSTEMS
O. D. Kazakov, Ph.D. (Econ.), Associate Professor, Vice-Rector for Digitalization, Head of the Department of IT, BGITU, Bryansk, Russia; orcid.org/0000-0001-9665-8138, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
N. Yu. Azarenko, Ph.D. (Econ.), Associate Professor, Master of BGITU, Bryansk, Russia;
orcid.org/0000-0001-6644-418X, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
The problem of audio data augmentation from cyber-physical level of digital twins of technological systems is considered. The aim of the work is to evaluate quality metrics of equipment health classification models trained on the same neural network architectures with the same configured hyperparameters, but on different datasets: original and extended based on the proposed approaches: 1) random blocking algorithm for sequential frequency range, i.e. blockages of certain sections of the spectrogram; 2) adding Gaussian noise to the spectrogram. Augmentation of the data obtained during the operation of pumping station AL-KO HW 3600 Easy 113798 at the stage of designing the cyber-physical level of its digital copy made it possible to increase the data set to 392 objects. The values of F-measure of Transfer learning test set on the extended data set after augmentation based on the proposed algorithms are in all cases higher than on the original data set. This study makes a certain contribution to the increased application of deep learning neural networks for diagnosing equipment failures. The proposed methods will make it possible to achieve high diagnostic accuracy with a small initial data set, and also solve the problem of improving safety and reliability of technological system operation in real conditions.
Key words: digital twin, technological system, audio data augmentation, cyber-physical level, transfer learning