At St. Petersburg Electrotechnical University "LETI", an algorithm has been developed for automatic recognition and classification of types of irregularities on railway rails. This was reported by the university's press service.
In the course of the research, methods were actively used that help analyze signals in the time and frequency domains, including Fourier analysis and wavelet transform. One of the significant results of the project was a new algorithm for classifying rail track irregularities, which makes it possible to increase the reliability of detecting defects on rail rolling surfaces.
Specialists studied the system of short and impulse irregularities to identify rail surface defects and measure their parameters. The developed classifier was tested on a sample of 150 signals. The results showed 85% accuracy in determining the type of joint and 15% of irregularities requiring further investigation.
The algorithm is based on the analysis of vibration signals, which allows distinguishing between bolted and welded joints. Studies have shown that a welded joint is characterized by low-frequency oscillations (140-200 Hz), while a bolted joint is characterized by high-frequency oscillations (1400-2400 Hz).
The new system not only performs measurements, but also classifies irregularities.
It is important to note that there were no incorrect defect definitions. Thus, the algorithm can be used in other flaw detector car systems and when installing a system for measuring short and impulse irregularities on regularly running trains.
Earlier www1.ru reported that Skoltech and MIPT developed an algorithm for quickly searching for highly efficient alloys using AI.
Read materials on the topic:
A machine learning system was used to study the properties of metallic hydrogen at MIPT
In Kazan, they learned how to melt metal using sound waves
In Moscow, neural networks have been adapted to predict defects in steel for nuclear power plants