Institutional Repository of Vilnius University
Lauko eilučių duomenų sintetinimas naudojant generatyvinius konkuruojančius tinklus /
Abstract
dc:descriptionThis work examines hypoid gear vibration time series and their synthesis using generative neural networks. The aim of the study is to synthesize these vibration signals using generative models, thereby increasing the volume of available experimental data. Four models were tested: two generative adversarial networks (TimeGAN and SeriesGAN) and two autoencoderbased models (a simple autoencoder and a variational autoencoder). The experimental results showed that training the generative adversarial network (GAN) models on the given vibration data was unstable – they failed to produce highquality synthetic signals. Meanwhile, the simple autoencoder only learned to reconstruct the training dataset signals (with almost no error) and could not generate any new sequences. In contrast, the variational autoencoder successfully generated statistically realistic hypoid gear vibration signals, although the diversity of the generated examples remained limited. In summary, the VAE approach proved effective in partially achieving the stated goal, whereas applying GAN models for this purpose was unsuccessful due to training difficulties. The study concludes that the application of generative synthesis methods to mechanical vibration data requires further research – in order to increase the diversity of generated signals and to stably train more complex models (e.g., GANs), specialized techniques or improved model architectures are necessary.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Pučinskas, Donatas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD239032076&prefLang=en_US