Генерация синтетических диагностических признаков вибрации подшипниковых узлов на основе вариационного автокодировщика

P G Krinitsin; S V Chentsov

doi:10.22213/2410-9304-2026-3-65-71

Authors

P. G. Krinitsin Siberian Federal University
S. V. Chentsov Siberian Federal University

DOI:

https://doi.org/10.22213/2410-9304-2026-3-65-71

Keywords:

vibration diagnostics, bearing, augmentation, autoencoder, generation, imbalance, classification, feature

Abstract

Ensuring the reliability of industrial equipment requires diagnostic systems capable of detecting faults at early stages. Under real operating conditions, the vast majority of recorded signals correspond to the healthy state of machinery, while data regarding faulty modes are extremely scarce. This leads to a pronounced imbalance in training datasets, significantly reducing the efficiency of machine learning models in vibration-based diagnostics of bearing units. The objective of this work is to develop and programmatically implement a method for generating synthetic diagnostic vibration features of bearing units to ensure training data balancing in technical diagnostic tasks. The proposed method is based on a Variational Autoencoder (VAE) - a probabilistic generative model that approximates the posterior distribution of a latent variable. The loss function incorporates reconstruction error and Kullback-Leibler divergence, with a weighting coefficient selected to account for the limited training sample size. A reparameterization trick is employed to ensure the differentiability of sampling from the latent space. For experimental verification, vibration signals from the bearing units of an ELG-400 bucket elevator, operated at a petroleum coke transportation site, were utilized. The software system is implemented in a specialized technical computing environment and includes modules for filtering, spectral analysis, feature extraction, and synthetic data generation. The synthetic data reproduce the mean values of five key diagnostic features: root mean square (RMS) vibration, standard deviation, peak value, crest factor, and kurtosis. The proposed approach effectively addresses training data imbalance in industrial equipment diagnostics, enhancing the robustness and generalization capability of diagnostic models without requiring additional real-world measurements.

Author Biographies

P. G. Krinitsin, Siberian Federal University

Post-graduate

S. V. Chentsov, Siberian Federal University

DSc in Engineering, Professor

References

Recent advances in the application of deep learning for fault diagnosis of rotating machinery using vibration signals / B. A. Tama, M. Vania, S. Lee, S. Lim // Artificial Intelligence Review. 2023. Vol. 56, No. 5. P. 4667-4709. DOI 10.1007/s10462-022-10293-3.

A systematic literature review of deep learning for vibration-based fault diagnosis of critical rotating machinery: Limitations and challenges / O. Matania, I. Dattner, Ja. Bortman [et al.] // Journal of Sound and Vibration. 2024. Vol. 590. P. 118562. DOI 10.1016/j.jsv.2024.118562.

Diagnosis of unbalanced bearing fault samples based on cycle generative adversarial networks / R. Huang, L. Ma, J. Hu [et al.] // Neural Computing & Applications. 2025. Vol. 37, No. 19. P. 13447-13467. DOI 10.1007/s00521-025-11265-8.

Mathematical Description of Dynamic Processes Occurring in Rolling Bearings Used in Oil-And-Gas Sector Rotor Machines as the Basis for Their Vibration-Based Diagnostics / E. I. Velichko, D. A. Inozemtsev, A. V. Muzykantova, M. S. Stepanov // IOP Conference Series: Earth and Environmental Science, Virtual, Online, 10-12 января 2022 года. Virtual, Online, 2022. P. 042050. DOI 10.1088/1755-1315/988/4/042050.

A Review on Vibration Monitoring Techniques for Predictive Maintenance of Rotating Machinery / M. Romanssini, P. C. C. De Aguirre, L.Compassi-Severo, A. G. Girardi // Eng. 2023. Vol. 4, No. 3. P. 1797-1817. DOI 10.3390/eng4030102.

Оценка состояния подшипников качения по спектрам вибрации / К. Б. Пальчик, И. М. Кругова, Д. В. Огурцов, Н. А. Халилов // Эксплуатация морского транспорта. 2022. № 2 (103). С. 117-122. DOI 10.34046/aumsuomt103/23.

Nikitin, Yu. Spectral Density-Based Pseudo-Random Sequence Generator for Bearing Vibration Simulation / Yu. Nikitin, S. Trefilov, P. Lekomtsev // AIP Conference Proceedings, Izhevsk, 24-26 ноября 2021 года. Izhevsk: Published by AIP Publishing, 2023. P. 020023. DOI 10.1063/5.0111186.

Wang, L. Research on bearing fault diagnosis based on machine learning and SHAP interpretability analysis / L. Wang, M. Wu // Scientific Reports. 2025. Vol. 15, No. 1. P. 41242. DOI 10.1038/s41598-025-25083-4.

Novelty Detection and Fault Diagnosis Method for Bearing Faults Based on the Hybrid Deep Autoencoder Network / Yu. Zhao, H. Hao, Yu. Chen, Yu. Zhang // Electronics. 2023. Vol. 12, No. 13. P. 2826. DOI 10.3390/electronics12132826.

Conditional variational autoencoder and GAN-based approach for long-tailed fault diagnosis for the motor system / Huang M., Sheng C., Rao X. // Measurement. 2024. Vol. 240.

Колосова З. А., Колосова О. П., Шнайдер Д. А. Получение набора вибрационных сигналов подшипника качения с разной степенью развития локального дефекта наружного кольца // Advanced Engineering Research (Rostov-on-Don). 2025. Т. 25, № 3. С. 242-255. DOI 10.23947/2687-1653-2025-25-3-242-255.

Komorska, I. Condition Monitoring Using a Latent Space of Variational Autoencoder Trained Only on a Healthy Machine / I. Komorska, A. Puchalski // Sensors. 2024. Vol. 24, No. 21. P. 6825. DOI 10.3390/s24216825.

Fault diagnosis using variational autoencoder GAN and focal loss CNN under unbalanced data / W. Li, D. Liu, Ya. Li [et al.] // Structural Health Monitoring. 2024. DOI 10.1177/14759217241254121.

Huang, M. Conditional Variational Autoencoder and generative adversarial network-based approach for long-tailed fault diagnosis for the motor system / M. Huang, Ch. Sheng, X. Rao // Measurement. 2025. Vol. 242. P. 116116. DOI 10.1016/j.measurement.2024.116116.

Learning Dynamics in Linear VAE: Posterior Collapse Threshold, Superfluous Latent Space Pitfalls, and Speedup with KL Annealing / Ichikawa Y., Hukushima K // arXiv. 2023.

A Low Complexity Rolling Bearing Diagnosis Technique Based on Machine Learning and Smart Preprocessing / A. Fort, E. Landi, M. Mugnaini, V. Vignoli // Sensors. 2023. Vol. 23, No. 17. P. 7546. DOI 10.3390/s23177546.

Pei, Di. A Novel Method for Bearing Fault Diagnosis under Variable Speed Based on Envelope Spectrum Fault Characteristic Frequency Band Identification / Di. Pei, J. Yue, J. Jiao // Sensors. 2023. Vol. 23, No. 9. P. 4338. DOI 10.3390/s23094338.

Mueller, Ph. N. Attention-enhanced conditional-diffusion-based data synthesis for data augmentation in machine fault diagnosis / Ph. N. Mueller // Engineering Applications of Artificial Intelligence. 2024. Vol. 131. P. 107696. DOI 10.1016/j.engappai.2023.107696.

Review of imbalanced fault diagnosis technology based on generative adversarial networks / H. Chen, J. Wei, H. Huang [et al.] // Journal of Computational Design and Engineering. 2024. Vol. 11, No. 5. P. 99-124. DOI 10.1093/jcde/qwae075.

A novel deep learning framework for rolling bearing fault diagnosis enhancement using VAE-augmented CNN model / Yu. Wang, D. Li, L. Li [et al.] // Heliyon. 2024. Vol. 10, No. 15. P. e35407. DOI 10.1016/j.heliyon.2024.e35407.