Study of informative value of features in rail condition monitoring

Introduction. The multidimensionality and the proximity of class boundaries due to external disturbances significantly complicate rail condition monitoring. Many informative signs are taken into account during the monitoring. Expansion of the a priori character alphabet in complex recognition system leads to a significant increase in economic losses. The emergence of measurement errors of many features and an increase in the processing time of information worsen the real time monitoring of railways. The article deals with the diminution of the dimensionality of the feature space of the recognizing system in rail condition monitoring and facilitates the formation of a working set of characteristics with the indication of the most informative combinations. Materials and Methods. The informativeness evaluation of the signs was carried out with the help of correlation coefficients and a trained classifier with a decisive function. The arguments of the classifier were the input and output electrical parameters of the four-pole of the rail line. The Kolmogorov-Gabor polynomial of the second degree of complexity, trained by solving an incompatible system of equations, was used as a decisive function. Mathematical and technical calculations were carried out in Mathcad programming. Results. In summary, these results show that the most informative indicators for a reliable classification of rail condition are input and output parameters of the four-port network of the rail line. All classes are reliably recognized. The recognition quality functions exceed 1.2. Conclusions. The results obtained in the study confirm the feasibility of approaches in forming a work set of features. The findings of this article can be used in the development of the principles of pattern recognition.