• Lyubov S. Lomakina, Nizhny Novgorod state technical university named after R.E. Alekseev
• Alisa N. Dvitovskaya, Nizhny Novgorod state technical university named after R.E. Alekseev
• Kirill A. Korelin, Nizhny Novgorod state technical university named after R.E. Alekseev

The article investigates the application of ensemble machine learning methods for diagnosing and forecasting the states of technical and technological objects under conditions of noisy data, nonlinear dependencies, and high-dimensional feature spaces. The relevance of the work is driven by the need to enhance the reliability of industrial systems through minimizing accident risks and optimizing operational processes. Traditional approaches demonstrate insufficient accuracy in complex scenarios, motivating the use of ensemble technologies that combine predictions from multiple models to achieve robust results. The primary focus is on Bagging, Boosting, and Stacking methods, their mathematical foundations, and practical implementation. An experiment was conducted using an ensemble of convolutional neural networks (CNNs) for classifying defects in metal microstructure. The results showed an increase in prediction accuracy with an increasing number of classes compared to a single model, confirming the effectiveness of ensembles in reducing error variance and correcting model bias. The proposed approach demonstrates potential for integration into industrial systems, enhancing diagnostic reliability and operational safety of complex technical systems.

technical object diagnostics, state forecasting, ensemble methods, bagging, boosting, stacking, convolutional neural networks

2025-12-01