Construction and analysis of model to predict the technical state of railway car axle boxes using intelligent predictive analytics methods

O. V. Druzhinina^a, E. R. Korepanov^a, I. V. Makarenkova^a, V. V. Maksimova^a, A. A. Petrov<sup>b

a</sup> Federal Research Center "Computer Science and Control" of the Russian Academy of Sciences, 44-2 Vavilov Str., Moscow 119333, Russian Federation
^b Yelets State University named after I. A. Bunin, 28 Kommunarov Str., Yelets 399770, Lipetsk Region, Russian Federation

Abstract: The paper is devoted to the study of the problem of constructing and analyzing a model for predicting the technical state of axle boxes of railway cars based on the use of artificial intelligence methods. The relevance of this problem is related to the need to create and improve high-tech and energy-efficient data analysis tools for diagnosing the technical condition of elements and systems of transport infrastructure. It is proposed to use the LSTM (Long Short-Term Memory) neural network architecture to predict the state when processing sequential data (time series). Synthetic datasets for neural network training are generated using the developed simulation stochastic model of thermal control of axle boxes. The performed computer modeling in the PyTorch environment allowed to conduct a comparative analysis of the results of computational experiments and to evaluate the effectiveness of LSTM training in the framework of the problem under consideration. The constructed predictive analytics model can serve as the basis for the ABITech Thermal Forecast Module, a software package for diagnosing the technical state of axle boxes.

Keywords: data analysis, predictive analytics, computer modeling, neural networks LSTM, time series, machine learning algorithms, technical state assessment, intelligent transport systems.

Received: 20.06.2025
Accepted: 15.10.2025

DOI: 10.14357/08696527250407