Dissertation for obtaining the scientific degree of Doctor of Philosophy in specialty 151 – automation and computer-integrated technologies (field of knowledge – 15 Automation and instrumentation). – National Technical University "Dnipro Polytechnic" of the Ministry of Education and Science of Ukraine, Dnipro, 2023.
The dissertation includes an analysis of the weighing processes in rail transport, types of moving objects, metrological standards, and scientific works. The results have demonstrated that existing systems have shortcomings such as insufficient accuracy (even when adhering to the recommendations provided by the system developers, including object velocity within 3-12 km/h and absence of wheel defects). The accuracy of wagon identification is around 95%, but in the case of deviating from recommendations (increasing train speed to 15 km/h), accuracy significantly decreases. Real-time data processing is lacking, and reliance on specific system-adapted templates or additional systems beyond the weighing platform is necessary.
Based on the conclusions drawn from the analysis of existing solutions and considering the task of developing a system to address the challenges of object identification and weighing during motion, a research plan has been formulated.
Data collection was conducted from real single-platform weighing systems, which served as the basis for further research. The train speed and railway vehicle speed were calculated at minimum and maximum values, in accordance with metrological standards, to facilitate subsequent testing of research outcomes.
It was found that the passage of a railway moving object is described by step-like signals, with duty cycle, quantity, and amplitude depending on axle count, number of bogies, and object weight.
Approximation methods were explored for the real data obtained from weighing systems, along with artificial intelligence techniques for future use in object identification tasks. The timeliness and relevance of this research stem from the systematic increase in cargo turnover by enterprises. Used polynomial approximation for processing data from strain gauge railway weighing systems, a system of linear equations was obtained. These equations accurately reconstruct experimental data with minimal error, derived from the existing enterprise system.
In this way, a comprehensive approach to data analysis and utilization of mathematical models has been established, contributing to the enhancement of railway transportation processes and identification of the wagons.
Using numerical methods, an algorithm for approximating the passage of railcars has been developed for various combinations of bogies and individual wagon bogies, employing the Heaviside function. It has been determined that the average approximation error for the passage data of a two-axle wagon using 6th-degree polynomials is approximately 10.66% (for the entire passage) and 1.3% for partial passage approximation. Meanwhile, approximating with the Heaviside function for the same passage yields an error of 0.5%. Additionally, polynomial approximation hinders solution standardization and increases data processing time (critical for real-time systems) due to the necessity of segmenting the data.
Normalizing sensor readings from conditional units, obtained from the cumulative box, to a value range of [0; 1] allows for a proportional description of the moving rail object. This avoids the dependence of final results on wagon or locomotive speed, enhancing the accuracy of wagon identification within the moving train through the utilization of the percentage of axle presence on the weighing platform (onset/exit). The ability to determine wagon type with the same number of axles but differing characteristics in the inter-axle space and wagon base has emerged.
