Branch of the Federal State Budgetary Educational Institution of Higher Education National Research University “MEI” in Smolensk, Smolensk, Russia.

М. I. Dli, Head of the Department of Information Technology in Economics and Management, Doctor of Technical Sciences, Professor, e-mail: midli@mail.ru
А. Yu. Putchkov, Associate Professor of the Department of Information Technology in Economics and Management, Candidate of Technical Sciences, Associate Professor, e-mail: putchkov63@mail.ru
А. М. Sokolov, Researcher, Candidate of Technical Sciences, e-mail: andreisokol98@gmail.com
М. Yu. Vorotilova, Postgraduate Student of the Department of Information Technology in Economics and Management, e-mail: rita.vorotilova@mail.ru

An original software model is proposed for detecting anomalies in the technological data of a conveyor-type roasting machine used in the preparation of raw materials for the production of non-ferrous metals, as well as estimating the useful life of the equipment. The solution of these tasks is performed within the framework of predictive analytics, the purpose of which in this case is to forecast emergency situations and negative trends in the dynamics of pellets heat treatment and to use these forecasts to implement preventive measures and scheduled maintenance. The relevance of the research task is the need to improve the accuracy of predictive analytics models for modern non-ferrous metal production through the introduction of advanced digital and intelligent data analysis technologies. This creates additional competitive advantages for production facilities, since the traditional statistical models provide acceptable solutions only within a narrow range of possible dynamic variations of the controlled technological process. The novelty of the research results lies in the developed multichannel structure of a neural network predictive analytics model based on an ensemble of autoencoders. The model detects and localizes anomalies in the technological zones of the roasting machine. A distinctive feature of the model is that the dynamics of the error in restoring input data by an autoencoder is used to predict the time of anomaly occurrence, which is interpreted as an estimate of the useful life of the equipment and is implemented based on the recursive least squares method. Simulation experiments have been conducted to demonstrate the ability of the proposed predictive analytics model to detect anomalies and forecast the useful life of the roasting machine. The economic effect of the implementation of the proposed predictive analytics model will be reflected in reduced costs associated with unplanned repairs and the elimination the consequences of emergency situations involving the roasting machine.
The work was carried out with the financial support of the Ministry of Science and Higher Education of the Russian Federation within the framework of the state assignment project No. FSWF-2026-0010.

1. Apata O., Munda J. L., Migabo E. M. Artificial Intelligence for Predictive Maintenance and Performance Optimization in Renewable Energy Systems: A Com prehensive Review. Energies. 2026. Vol. 19, Iss. 536. DOI: 10.3390/en19020536
2. Dyakonov N. A., Logunova O. S. Process control systems based on predictive analytics: design. Electrotekhnicheskiye sistemy i kompleksy. 2021. No. 1. pp. 58–64.
3. Coil C., Faber K., Sniezynski B. et al. Distance-based change point detection for novelty detection in concept-agnostic continual anomaly detection. J. Intell. Inf. Syst. 2025. DOI: 10.1007/s10844-025-00949-1
4. Rahmati F., Gharaei R. H., Nezamabadi-pour H. ARDOD: adaptive radius density-based outlier detection. Evol. Intel. 2024. Vol. 17. DOI: 10.1007/s12065-024-00953-4
5. Zahoor A., Abbasi W., Babar M. Z. et al. Robust IoT security using isolation forest and one class SVM algorithms. Sci Rep. 2025. Vol. 15. DOI: 10.1038/s41598-025-20445-4
6. Bhatia S., Jain A., Li P., Kumar R., Hooi B. MStream: fast anomaly detection in multi-aspect streams. Proceedings of the Web Conference 2021 (WWW ‘21). Association for Computing Machinery. New York, NY, USA,
pp. 3371–3382. DOI: 10.1145/3442381.3450023
7. Al Amin Sarker M., Jayaraj I. A., Shanmugam B. et al. A review of artificial intelligence techniques for anomaly detection in smart grid. Artif Intell Rev. 2026. Vol. 59, Iss. 69. DOI: 10.1007/s10462-025-11429-x
8. Lee J. H., Kim M. Y. Manufacturing quality management based on TimeGAN and Seq2Seq models with magnetic press machine data. Int. J. Control Autom. Syst. 2025. Vol. 23. pp. 1199–1209. DOI: 10.1007/s12555-024- 0448-5
9. Rusanovsky M., Beeri O., Oren G. An end-to-end computer vision methodology for quantitative metallography. Sci Rep. 2022. Vol. 12. 4776. DOI: 10.1038/s41598-022-08651-w
10. Simakov А. S., Masko О. N., Nikolaev М. Yu. Technical vision system for quartz raw material quality assessment. Tsvetnye Metally. 2025. No. 3. pp. 81–88.
11. The CMS ECAL Collaboration. Autoencoder-Based Anomaly Detection System for Online Data Quality Monitoring of the CMS Electromagnetic Calorimeter. Comput Softw Big Sci. 2024. Vol. 8, Iss. 11. DOI: 10.1007/s41781-024-00118-z
12. Zhan K., Wang C., Zheng X. et al. Seq2Seq-based GRU autoencoder for anomaly detection and failure identification in coal mining hydraulic support systems. Sci Rep. 2025. Vol. 15. 542. DOI: 10.1038/s41598-024-84130-8
13. Puchkov A. Yu., Dli M. I., Lobaneva E. I. A method for estimating the useful life of equipment based on neural networks. Izvestiia SPbGTI (TU). 2021. No. 59. pp. 107–112.
14. Apata O., Munda J. L., Migabo E. M. Artificial intelligence for predictive maintenance and performance optimization in renewable energy systems: a comprehensive. Review. Energies. 2026. Vol. 19, Iss. 2. 536. DOI:
10.3390/en19020536
15. Sholle F. Deep learning in Python. 2nd Ed. Saint Petersburg : Piter, 2023. 576 p.
16. Bobkov V. I. The heat-technological process optimization of a moving dense multilayer weight of phosphorite pellets drying process by the energyresource efficiency criterion. Systems of Control, Communication and Security. 2018. No. 2. pp. 56–68.