ALGORITHM FOR POSITIONING A MOVING RAILWAY TRAINS USING REMOTE VIBROACOUSTIC SENSING TECHNOLOGY

Many train positioning technologies utilizing remote vibroacoustic sensing rely on frequency-based and threshold-based signal analysis methods. The vibroacoustic signal propagating through optical fibers laid along railway tracks suffers from significant contamination by external noise, leading to substantial positioning inaccuracies. Consequently, developing a method to accurately localize moving railway vehicles with desired precision remains a critical challenge. The aim of the study is to develop an effective method for processing and analyzing vibroacoustic signals to enable positioning of a moving railway trains with the required accuracy. Materials and methods. The study was conducted using vibroacoustic signal recordings generated by freight trains exceeding 800 meters in length traveling along a 42 km railway section. This section is equipped with a vibroacoustic sensing system. For processing the vibroacoustic signals, the difference method, fast Fourier transform method, and grayscale image segmentation techniques were employed. The original signal was pre-filtered using the difference method. Subsequently, a fast Fourier transform method was applied along the time axis, followed by the calculation of the normalized sum of the discrete fast Fourier transform amplitude moduli. The resulting two-dimensional array of spectrograms was represented as an image. The task of identifying train boundaries was reduced to selecting a threshold that would separate the image into two classes: “train” and “background”. As the threshold value was chosen the one at which the inter-class variance between “background” and “train” reached its maximum. Results. Based on the unique frequency characteristics of each railway rolling stock, an algorithm has been developed to determine the real-time location of moving trains. The applied non-parametric threshold selection method enables the evaluation of threshold optimality using current data without requiring additional input parameters. The process begins with filtering the original reflectograms by calculating the difference between adjacent traces. This step is essential for removing static noise. Next, a time-domain fast Fourier transform is performed followed by the computation of the normalized sum of fast Fourier transform amplitudes. By treating the fast Fourier transform result as an image (where pixels represent normalized sums of absolute amplitudes from the discrete fast Fourier transform for each fiber optic sample), an image segmentation method can be applied to separate the “background” and “train” classes. The method relies on calculating an optimal threshold that maximizes the inter-class variance between “background” and “train”. During the comparison of each pixel’s brightness with the derived threshold, pixels are assigned values of 0 or 1, producing a binary array. Post-processing is then applied to the binary array, where all isolated 1s are set to 0. Subsequently, train boundaries are determined: the left boundary is identified as the first 1 from the start of the binary array, and the right boundary is the first 1 from the end of the array. To smooth the train boundaries, a least squares approximation method is applied. Using the proposed method on processed reflectogram recordings, the boundaries of moving trains along a railway section were identified with a deviation of within 20 meters. Conclusions. The proposed algorithm enables the positioning of both passenger and freight railway trains, including those composed of mixed carriages. Freight trains exhibit a higher susceptibility to wheel pair defects, which introduce distortions into vibroacoustic signals. Processing vibroacoustic signals in the frequency domain and analyzing a specific frequency range characteristic of moving trains mitigates the described effect, as well as dynamic environmental factors. The use of a non-parametric method for threshold calculation allows evaluating the optimality of the selected threshold value based on the criterion of maximum variance between the “background” and “train” classes. Computing frequency characteristics at equal short intervals during train movement enables adaptation to the current conditions of the railway track section.