Application of machine learning algorithms of bayesian networks to solve the problems of diagnosing complex technical systems using discrete diagnostic parameter

Аuthors

Dorozhko I. V.

Mlitary spaсe Aсademy named after A.F. Mozhaisky, 197198, St. Petersburg, Zhdanovskaya St., 13

e-mail: vka@mil.ru

Abstract

The article suggests the use of the Bayesian network apparatus to formalize the process of diagnosing technical systems. With the help of a priori information about the composition, reliability, relationships of diagnostic features and types of technical conditions, an initial model is built. The scientific novelty of the research lies in the study of the possibility of correcting the parameters of the diagnostic model as a result of machine learning with a limited training sample size, taking into account the degree of confidence in a priori information. It is pointed out the advantages of Bayesian networks, which consist in the ability to compensate for limited experimental data by using a priori information. The influence of taking into account the degree of trust in a priori data and the amount of statistical information on the results of machine learning is investigated. The practical significance lies in the possibility of applying the proposed approach to standard racks of an automated control system for technological equipment of the launch complex. A study of structural-parametric learning of the diagnostic process model in the form of a Bayesian network has been conducted, as a result of which not only the parameters, but also the structure of the model is adjusted. This opens up opportunities to identify dependencies between diagnostic features that may not have been taken into account at the design stage. Attention is drawn to the problem of inconsistency of the initial data, which, with certain combinations of diagnostic signs, lead to the inability to draw a conclusion about the type of technical condition, i.e. to obtain a diagnostic result. These contradictions are the result of errors in setting the initial model or the presence of unaccounted-for dependencies of diagnostic features and can be resolved as a result of structural-parametric machine learning. The results obtained contribute to the development of a methodology for monitoring and diagnosing complex technical systems. The proposed approach can be used at the design and operation stages of automated control systems for technological equipment of the launch complex.

Keywords:

diagnosis, diagnostic parameters, a priori information, Bayesian network, structural-parametric machine learning

References

1. Belozerov V.A. et al. Avtomatizatsiya tekhnologicheskikh protsessov i proizvodstv (Automation of technological processes and productions), Saint Petersburg, VKA imeni A.F. Mozhaiskogo, 2014, 284 p.
2. Arsen'ev V.N. Otsenivanie kharakteristik sistem upravleniya po ogranichennomu chislu naturnykh ispytanii (Evaluation of control system characteristics based on a limited number of verification tests), Moscow, Restart, 2013, 126 p.
3. Vorontsov V.A., Fedorov E.A. Trudy MAI, 2018, no. 82, URL: https://trudymai.ru/published.php?ID=58817 (accessed 05.05.2025).
4. Dmitriev A.K. Modeli i metody analiza tekhnicheskogo sostoyaniya bortovykh system (Models and methods of analysis of the technical condition of on-board systems: training manual), Saint Petersburg, VIKU imeni A.F. Mozhaiskogo, 1999, 171 p.
5. Dorozhko I.V., Ivanov O.A. Trudy MAI, 2021, no. 118, URL: https://trudymai.ru/published.php?ID=158259 (accessed 05.05.2025).
6. Dorozhko I.V., Zakharova E.A., Osipov N.A. Trudy Voenno-kosmicheskoi akademii im. A.F. Mozhaiskogo, 2019, no. 669, pp. 216 – 223.
7. Dorozhko I.V., Osipov N.A., Ivanov O.A. Trudy MAI, 2020, no. 113. URL: http://trudymai.ru/eng/published.php?ID=118181 (accessed 05.05.2025).
8. Zavedeev A.I. Trudy MAI, 2012, no. 54, URL: https://trudymai.ru/published.php?ID=29687 (accessed 05.05.2025).
9. Kobzarev I.M. Kopkin E.V. Trudy Voenno-kosmicheskoi akademii im. A.F. Mozhaiskogo, 2019, no. 671, pp. 326 – 337.
10. Kopeika E.A., Verbin A.V. Trudy MAI, 2023, no. 128, URL: https://trudymai.ru/published.php?ID=171411 (accessed 05.05.2025).
11. Kopkin E.V., Kravtsov A.N., Myshko V.V. Analiz tekhnicheskogo sostoyaniya kosmicheskikh sredstv (Analysis of the technical condition of space assets), Saint Petersburg, VKA imeni A.F. Mozhaiskogo, 2016, 189 p.
12. Kopkin E.V., Kravtsov A.N., Myshko V.V. Kontrol' i diagnostika kosmicheskikh sredstv (Control and diagnosis of space tools: a training manual), Saint Petersburg, VKA imeni A.F. Mozhaiskogo, 2016, 198 p.
13. Mit'kin M.A., Gavrilov K.Yu. Trudy MAI, 2025, no. 141, URL: https://trudymai.ru/published.php?ID=184505 (accessed 05.05.2025).
14. Luis Enrique Sucar. Veroyatnostnye grafovye modeli. Printsipy i prilozheniya (Probabilistic Graphical Models. Principles and Applications), Moscow, DMK Press, 2021, 338 p.
15.  Tonkikh A.N. Trudy MAI, 2025, no. 141, URL: https://trudymai.ru/published.php?ID=184504 (accessed 05.05.2025).
16. Tulup'ev A.L., Nikolenko S.I., Sirotkin A.V. Osnovy teorii baiesovskikh setei. (The basics of the theory of Bayesian networks), Saint-Petersburg, Izd-vo Sankt-Peterburgskogo universiteta, 2019, 399 p.
17. Cowell R.G., Dawid A.P., Lauritzen S.L., Spiegelhalter D.J. Probabilistic Networks and Expert Systems, New York, Springer-Verlag, 1999. DOI:10.1198/JASA.2001.S424
18. GeNIe & SMILE. Decisions systems laboratory, School of Information Sciences, University of Pittsburg. URL: https://www.innovation.pitt.edu/pitt-startups/bayesfusion/.
19. Jensen F.V. Bayesian Networks and Decision Graphs, New York, Springer-Verlag, 2001. DOI:10.1007/978-1-4757-3502-4
20. Pearl J. Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference, New York, Morgan Kaufman Publ., 1991, DOI:10.2307/2026705.

Download