10.14489/td.2026.10.pp.028-039

2026, 10 October

DOI: 10.14489/td.2026.10.pp.028-039

Orlov A. A., Shcherbakov V. M., Budadin O. N., Fedotov M. Yu., Krylova E. V., Ksenda E. A., Kozelskaya S. O.
DEVELOPMENT AND RESEARCH OF A HYBRID NEURAL NETWORK MODEL WITH FORECASTING OF THE RESIDUAL RESOURCE OF COMPLEX-SHAPED METAL OBJECTS AND THEIR INTERNAL STRUCTURE (ON THE EXAMPLE OF LOADED ELEMENTS OF A POWERFUL GAS-TURBINE UNIT)
(pp. 28-39)

Abstract. A gas turbine unit (GTA) is a power plant that includes a gas turbine engine, an electric generator and auxiliary systems designed to generate electricity. The main elements of the GTA are: a gas turbine engine (compressor, combustion chamber, turbine), an electric generator (or centrifugal supercharger), a fuel supply system, an oil system, a cooling system, an automatic control system, as well as auxiliary equipment (starting device, gearbox, frame). This paper addresses the issue of improving the accuracy of predicting the residual resource of complex-shaped and internally structured metal objects (using the example of the rotor bearings of the GTD-110M gas turbine unit). For this purpose, a hybrid neural network model has been developed that combines time series analysis and statistical features. At the first stage, sliding windows with a length of 100 minutes of temperature, pressure and vibration measurements are formed, to which the first and second differences are added to take into account the dynamics of degradation. To extract time dependencies, a bidirectional LSTM layer is used with an attention mechanism that highlights critical wear phases. In parallel, classical statistical indicators (mean, standard deviation, minimum, maximum) are calculated from each window, which are combined with the contextual attention vector and fed into the regression network. Training and testing were conducted on synthetic data simulating three scenarios of bearing degradation: uniform wear, accelerated wear, and abrupt deterioration. A comparison with the linear temperature trend extrapolation model showed that the hybrid model reduces the mean absolute error (MAE) by 65.8 %, the RMS error (RMSE) by 67.3 %, and the coefficient of determination R2 increases from 0.42 to 0.89. The most significant gain (74.1 % according to MAE) was achieved in the accelerated wear scenario, which poses the greatest danger in real operation. The error distribution of the hybrid model is significantly narrower and centered near zero, which confirms its stability. The results obtained allow us to recommend the developed architecture for integration into the operator's automated workplace system for parametric diagnostics of the GTD-110M in order to switch from routine preventive repairs to maintenance based on the actual condition.

Keywords: gas turbine unit (GTA), residual resource prediction, complex-shaped metal object and internal structure, rotor bearing, hybrid neural network model, bidirectional LSTM, attention mechanism, parametric diagnostics.

+ - About the Authors Click to collapse

A. A. Orlov (Federal State Budgetary Educational Institution of Higher Education "National Research University "MEI", Moscow, Russia, Russian Academy of Engineering, Moscow, Russia, Joint-stock company «Scientific, Educational, Industrial and Environmental Enterprise «Karat», Staroe Shatkino village, Russia) E-mail: Данный адрес e-mail защищен от спам-ботов, Вам необходимо включить Javascript для его просмотра.
V. M. Shcherbakov (Federal State Budgetary Educational Institution of Higher Education "National Research University "MEI", Moscow, Russia) E-mail: Данный адрес e-mail защищен от спам-ботов, Вам необходимо включить Javascript для его просмотра.
O. N. Budadin (JSC Central Research Institute of Special Machine Building, Khotkovo, Russia) E-mail: Данный адрес e-mail защищен от спам-ботов, Вам необходимо включить Javascript для его просмотра.
M. Yu. Fedotov (Russian Engineering Academy, Moscow, Russia) E-mail: Данный адрес e-mail защищен от спам-ботов, Вам необходимо включить Javascript для его просмотра.
E. V. Krylova (Federal State Budgetary Educational Institution of Higher Education "National Research University "MEI", Moscow, Russia, Russian Academy of Engineering, Moscow, Russia, Joint-stock company «Scientific, Educational, Industrial and Environmental Enterprise «Karat», Staroe Shatkino village, Russia) E-mail: Данный адрес e-mail защищен от спам-ботов, Вам необходимо включить Javascript для его просмотра.
E. A. Ksenda (Federal State Budgetary Educational Institution of Higher Education "National Research University "MEI", Moscow, Russia) E-mail: Данный адрес e-mail защищен от спам-ботов, Вам необходимо включить Javascript для его просмотра.
S. O. Kozelskaya (JSC Central Research Institute of Special Machine Building, Khotkovo, Russia) E-mail: Данный адрес e-mail защищен от спам-ботов, Вам необходимо включить Javascript для его просмотра.

+ - References Click to collapse

1. Korablev, Yu. A. (2025). Prediction of the remaining useful life of technological equipment using the deep learning LSTM method. Izvestiya YuFU. Tekhnicheskie nauki, 5(247), 277–288. [in Russian language].
2. Orlov, A. A., Krylova, E. V., Shcherbakov, V. M., & Avdeev, A. D. (2025a). Intelligent monitoring and diagnostics system for gas turbine elements (ISMEGTU). Izvestiya vysshikh uchebnykh zavedeniy. Problemy energetiki, 27(2), 211–221. [in Russian language]. https://doi.org/10.30724/1998-9903-2025-27-2-211-221
3. Shcherbakov, V. M., Krylova, E. V., & Sirichenko, A. A. (2025). Intellectualization of automatic parametric diagnostics of powerful energy equipment. In Radioelectronics, Electrical Engineering and Energy: Abstracts of the 31st International Scientific and Technical Conference of Students and Postgraduates (p. 988). OOO "Tsentr poligraficheskikh uslug "Raduga". [in Russian language].
4. Orlov, A. A., Krylova, E. V., Shcherbakov, V. M., & Avdeev, A. D. (2025b). Intelligent monitoring and diagnostics system for gas turbine elements (ISMEGTU). Izvestiya vysshikh uchebnykh zavedeniy. Problemy energetiki, 27(2), 211–221. [in Russian language]. https://doi.org/10.30724/1998-9903-2025-27-2-211-221
5. Shcherbakov, V. M., & Pytkina, E. A. (2024). Intellectualization of automated parametric diagnostics of high-power gas turbine engine elements. In Radioelectronics, Electrical Engineering and Energy: Abstracts of the 30th International Scientific and Technical Conference of Students and Postgraduates (p. 1040). OOO "Tsentr poligraficheskikh uslug "RADUGA". [in Russian language].
6. Bekarevich, A. A. (2014). Development of the theory of thermal non-destructive testing of complex shaped metal products and the development of automated diagnostics and operational reliability assessment tools. ID "Spektr". [in Russian language].
7. Bekarevich, A. A., Valiakhmetov, S. A., Budadin, O. N., et al. (2014). Automated thermal quality control and diagnostics of the technical condition of gas turbine blades with operational reliability assessment. ID "Spektr". [in Russian language].
8. Aksenov, E. V. (2025). Development of an identifier for detecting faults in gas turbine engine rotor bearings using a computational-experimental method. In Prospects for the development of engine engineering: Proceedings of the International Scientific and Technical Conference named after N. D. Kuznetsov (pp. 205–208). Samara University Press. [in Russian language].
9. Galyshev, D. V., Yakovenko, A. D., Ibraeva, O. L., & Shestakov, A. L. (2025). Verification of the reliability of technical monitoring measurement readings using a cascade of autoencoders. Vestnik YuUrGU. Ser. Vychislitel'naya matematika i informatika, 14(4), 5–24. [in Russian language].
10. Khramtsov, I. V., Shobey, M. M., Sudakov, I. A., et al. (2025). Diagnostics of rolling bearing defects by the envelope spectrum analysis of vibration and acoustic signals. Zavodskaya laboratoriya. Diagnostika materialov, 91(10), 50–58. [in Russian language].
11. Cao, H., Xiao, W., Sun, J., et al. (2024). Fusing data- and model-driven methods for RUL prediction in smart manufacturing systems. In 43rd Chinese Control Conference (CCC) 2024 (pp. 6945–6949). IEEE Computer Society.
12. Brahimi, L., Hadroug, N., Iratni, A., & Hafaifa, A. (2025). Advancing predictive maintenance for gas turbines: An intelligent monitoring approach with ANFIS, LSTM, and reliability analysis. Computers & Industrial Engineering, 191(2), Article 110094.
13. Razzaq, W., & Zhao, Y.-B. (2025). CARLE: a hybrid deep shallow learning framework for robust and explainable RUL estimation of rolling element bearings. Soft Computing, 29(23 24), 6269–6292.
14. Chen, S., Li, Z., Wang, Y., & Liu, X. (2025). Stage-based remaining useful life prediction for bearings using GNN and correlation-driven feature extraction. Machines, 13(1), Article 43.
15. Wang, H., Liu, J., Zhang, W., & Chen, Y. (2024). A hybrid deep learning network for diagnosing multipoint faults in rolling bearings under variable operating conditions. Journal of Mechanical Science and Technology, 38(11), 5989–6003.
16. Zhou, F., Li, X., & Wang, T. (2025). A remaining useful life prediction method for rolling bearings based on hierarchical clustering and Transformer-GRU. Applied Sciences, 15(10), Article 5369.
17. Fu, S., Avdelidis, N. P., & Plastropoulos, A. (2025). Novel hybrid prognostics of aircraft systems. Electronics, 14(11), Article 2193.
18. Huang, Z., Li, Y.-F., Xu, Z., et al. (2025). A hybrid remaining useful life prediction method for bearings with Bootstrap framework. Journal of Mechanical Science and Technology, 39, 3609–3619.
19. Sun, Y., Wang, H., Li, J., et al. (2025). Remaining life prediction modeling method for rotating components of complex intelligent equipment. Electronics, 14(1), Article 136.
20. Zhao, R., Li, Y., Zhang, W., & Liu, H. (2025). Remaining useful life prediction of cutting tools based on a depthwise separable TCN-BiLSTM model with temporal attention. Lubricants, 13(11), Article 507.
21. Sharma, J., Mittal, M. L., & Soni, G. (2025). Attention based convolutional gated recurrent unit model for turbofan engine remaining useful life prediction. In Risk, Reliability and Resilience in Operations Management (Chapter 10, pp. 201–214). Elsevier.

+ - Purchase digital version of a single article Click to collapse

This article is available in electronic format (PDF).

DOI: 10.14489/td.2026.10.pp.028-039

Copy the article DOI and follow the link https://id-spektr.ru/product/pokupka-elektronnoy-stati-iz-zhurnala-kontrol-diagnostika

Please specify the article DOI in the order comments.