CREATION OF A CONCEPTUAL MODEL OF AUTOMATED SELF-LEARNING SYSTEM OF FUNCTIONAL CONTROL AND DETECTION OF RAILWAY TRANSPORT

Abstract

Ensuring reliable and trouble-free operation of all railway and high-tech systems remains a priority task in the field of scientific developments related to operation and modernization of such complexes. The aim of the study is to develop and refine a machine learning method for the automated detection system (ADS) of functional states of railway transport components and assemblies. The objectives include: creating a glossary of feature realizations for each class of anomalies or faults; determining the minimum size of the training matrix and permissible deviations for feature implementation; developing a binary training matrix (OUFT) and optimizing its structure to improve detection accuracy. Analysis of existing NDC methods and machine learning algorithms (K-means, DBSCAN, FDBSCAN) was performed to build binary matrices and cluster features. The proposed approach optimizes the ADS training process, reduces computational complexity, and improves anomaly and fault detection in railway components and assemblies. Algorithms for parallel optimization of fault detection features and measures to enhance DSS accuracy in automated diagnostics were proposed. The developed machine learning method and ADS structure allow the creation of error-free decision rules for the diagnosis of railway transport components. This approach improves the accuracy and reliability of decision support systems and automated detection systems for functional anomalies.