A.Yu. Chesalov1
1 CEO Atlansis Software (Tver, Russia)

1 achesalov@mail.ru

Modern industrial enterprises require fundamentally new solutions aimed at predicting equipment failures and timely elimination of abnormal emergency situations, managing repair costs, and optimizing and improving maintenance strategies. Existing automated predictive maintenance systems have various functional limitations. This requires the development of new architectural solutions to create intelligent systems capable of processing large amounts of heterogeneous data in real time, predicting failures with high accuracy, and optimizing maintenance processes, in close interconnection with Industrial Internet of Things peripherals and in the context of using new peripheral artificial intelligence technologies. This paper investigates the feasibility of using the large language model OpenThinker2-32B as an auxiliary subsystem of an automated predictive maintenance system for multistage technological processes. This model, which has 32 billion parameters, combines the implementation of a set of machine learning methods and deep neural network algorithms to perform the analysis of historical and current data obtained from sensors of the Industrial Internet of Things. In turn, methods must be defined to reduce uncertainties in data collection and analysis to provide more accurate predictions, expert recommendations, and maintenance opinions for industrial equipment.

The objective of the article is to study the possibility of applying and adapting the OpenThinker2-32B model for solving predictive maintenance problems for small and medium industrial enterprises, which include analysis of historical data, prediction of failures, reduction of uncertainty and preparation of expert recommendations using the Dempster-Shafer theory of evidence, optimization of schedules and maintenance processes. Another one objective is to determine the feasibility of applying the large language model algorithm in the concept of convergent architecture of the automated predictive maintenance system to improve the accuracy of predictions and its integration with expert, analytical, forecasting and decision support systems.

The results of this analysis show that the OpenThinker2-32B large language model contains a wide range of algorithms that can be used in a variety of predictive or prescriptive maintenance tasks. These algorithms can efficiently perform data analysis, identify potential faults and predict equipment failures, and optimize maintenance schedules and processes at the lowest cost and lowest TCO for the computing infrastructure. Using the Dempster-Shafer theory of evidence, new algorithms can be implemented to reduce uncertainty for more accurate predictions and to provide expert advice and insight. With its flexibility, high performance and low cost of ownership, OpenThinker2-32B is a powerful tool for automating industrial processes and improving their efficiency for small and medium-sized industrial enterprises.

The results of the study can be used to design industrial automated predictive or prescriptive maintenance systems.

Chesalov A.Yu. Analysis of feasibility of using the OpenThinker2-32B model in automated predictive maintenance systems for small and medium-sized industrial enterprises. Neurocomputers. 2025. V. 27. № 5. P. 56–70. DOI: https://doi.org/10.18127/j19998554-202505-07 (in Russian)

1. Large language model market [Elektronnyj resurs]. URL: https://www.marketsandmarkets.com/Market-Reports/large-language-model-llm-market-102137956.html (data obrashcheniya: 13.03.2025).
2. Large language model (LLM) market – A global and regional analysis [Elektronnyj resurs]. URL: https://bisresearch.com/industry-report/ large-language-model-market.html (data obrashcheniya: 13.03.2025).
3. Obzor vidov luchshikh II-modelej 2024: sravnenie i rejting populyarnykh LLM [Elektronnyj resurs]. URL: https://iaassaaspaas.ru/rating/ai/obzor-vidov-luchshih-ii-modeley-2024-sravnenie-i-reyting-populyarnyh-llm (data obrashcheniya: 12.03.2025). (in Russian)
4. Chto takoe kontekstnoe okno v II i kak ono vliyaet na rabotu yazykovykh modelej [Elektronnyj resurs]. URL: https://blog.colobridge.net/ 2025/04/context-window-in-ai/ (data obrashcheniya: 12.05.2025). (in Russian)
5. Malyshko V. Bol'shie yazykovye modeli: chto eto takoe i kak oni rabotayut [Elektronnyj resurs]. URL: https://just-ai.com/blog/bolshie-yazykovye-modeli-chto-eto-takoe-i-kak-oni-rabotayut (data obrashcheniya: 02.05.2025). (in Russian)
6. Agarwal R., Singh A., Zhang L.M. et al. Many-shot in-context learning. 2024 [Elektronnyj resurs]. URL: https://arxiv.org/abs/2404.11018 (data obrashcheniya: 12.05.2025).
7. Ukaz Prezidenta Rossijskoj Federatsii ot 30.03.2022 № 166 «O merakh po obespecheniyu tekhnologicheskoj nezavisimosti i bezopasnosti kriticheskoj informatsionnoj infrastruktury Rossijskoj Federatsii» [Elektronnyj resurs]. URL: http://publication.pravo.gov.ru/Docu­ment/View/0001202203300001 (data obrashcheniya: 12.05.2025). (in Russian)
8. Qwen2.5-32B-Instruct [Elektronnyj resurs]. URL: https://huggingface.co/Qwen/Qwen2.5-32B-Instruct (data obrashcheniya: 12.05.2025).
9. OpenThoughts2-1M [Elektronnyj resurs]. URL: https://huggingface.co/datasets/open-thoughts/OpenThoughts2-1M (data obrashche­niya: 12.05.2025).
10. Open-thoughts [Elektronnyj resurs]. URL: https://github.com/open-thoughts/open-thoughts (data obrashcheniya: 09.05.2025).
11. Chujko D. Bol'shie yazykovye modeli kak instrument dlya analiza tekhnicheskoj dokumentatsii i resheniya IT-intsidentov [Elektronnyj resurs]. URL: https://habr.com/ru/articles/895114/ (data obrashcheniya: 02.05.2025). (in Russian)
12. Yumaguzin U.F., Bashirov M.G. Prognozirovanie ostatochnogo resursa oborudovaniya predpriyatij neftegazovoj otrasli. Fundamental'nye issledovaniya. 2014. № 3 (chast' 2). S. 277–280. URL: https://fundamental-research.ru/ru/article/view?id=33623 (data obrashcheniya: 15.03.2025). (in Russian)
13. Kazarinov L.S., Shnajder D.A., Khasanov A.R. Metod otsenki tekushchego sostoyaniya kontroliruemogo oborudovaniya v zadache operativnogo planirovaniya remontno-profilakticheskikh rabot. Vestnik Yuzhno-Ural'skogo gosudarstvennogo universiteta. Ser.: Komp'yuternye tekhnologii, upravlenie, radioelektronika. 2006. № 14 (69). S. 84–87. (in Russian)
14. Kazarinov L.S., Shnajder D.A. Metod uprezhdayushchego upravleniya slozhnymi tekhnologicheskimi kompleksami po kriteriyam energeticheskoj effektivnosti. Sb. trudov «Upravlenie bol'shimi sistemami». 2011. № 32. S. 221–240. (in Russian)
15. Shnajder D.A. Avtomatizirovannye sistemy uprezhdayushchego upravleniya po kriteriyam energeticheskoj effektivnosti (v teploenergeticheskikh kompleksakh metallurgicheskikh predpriyatij). Diss. … dokt. tekhn. nauk. Chelyabinsk. 2011. (in Russian)
16. Gantseva E.A., Kaladze V.A., Polyakov A.M. Formirovanie ekspertnogo vyvoda na osnove teorii svidetel'stv. Vestnik VGTU. 2015. T. 11. № 1. S. 42–46. (in Russian)
17. Chto takoe bol'shaya yazykovaya model' (LLM)? [Elektronnyj resurs]. URL: https://aws.amazon.com/ru/what-is/large-language-model/ (data obrashcheniya: 02.05.2025). (in Russian)
18. Uchim bol'shie yazykovye modeli opisyvat' produkty dannykh [Elektronnyj resurs]. URL: https://habr.com/ru/companies/sberbank/ articles/795617/ (data obrashcheniya: 02.05.2025). (in Russian)
19. Palyukh B.V., Chesalov A.Yu. Sovremennye podkhody k sozdaniyu avtomatizirovannykh sistem prognoziruemogo obsluzhivaniya mnogostadijnykh tekhnologicheskikh protsessov v promyshlennosti. Sb. nauch. trudov XXVII Rossijskoj nauch. konf. «Inzhiniring predpriyatij i upravlenie znaniyami» (IP&UZ – 2024). 28–29 noyabrya 2024 g. M.: FGBOU VO «REU im. G.V. Plekhanova». 2024. T. 1. S. 351–357. (in Russian)
20. Ivanov V.K., Palyukh B.V. Demonstrator programmnoj platformy dlya sovmestnogo ispol'zovaniya algoritmov teorii svidetel'stv i nejronnykh setej v nechetkikh sistemakh. Programmnye produkty i sistemy. 2021. T. 34. № 4. S. 511–523. (in Russian)
21. Vinogradova N.V., Ivanov V.K., Palyukh B.V., Sotnikov A.N. Sovremennye napravleniya razvitiya i oblasti prilozheniya teorii Dempstera-Shafera (obzor). Iskusstvennyj intellekt i prinyatie reshenij. 2018. № 4. S. 32–42. (in Russian)
22. Dyrmovskij D. Bol'shie yazykovye modeli – bol'shie vozmozhnosti dlya rosta biznesa [Elektronnyj resurs]. URL: https://www.speechpro.ru/ media/news/bolshie-yazykovye-modeli-bolshie-vozmozhnosti-dlya-rosta-biznesa (data obrashcheniya: 02.05.2025). (in Russian)
23. Palyukh B.V., Chesalov A.Yu. Metodologicheskij podkhod k tsifrovoj transformatsii predpriyatij otrasli proizvoditelej oborudovaniya nerazrushayushchego kontrolya. Sb. nauch. trudov XXVI Rossijskoj nauch. konf. «Inzhiniring predpriyatij i upravlenie znaniyami» (IP&UZ – 2023). 29–30 noyabrya 2023 g. M.: FGBOU VO «REU im. G.V. Plekhanova». 2023. S. 256–260. (in Russian)
24. OpenThinker2-32B [Elektronnyj resurs]. URL: https://huggingface.co/open-thoughts/OpenThinker2-32B (data obrashcheniya: 13.05.2025).
25. Transformers [Elektronnyj resurs]. URL: https://huggingface.co/docs/transformers/index (data obrashcheniya: 12.05.2025).