Crack Detection using Empirical Mode Decomposition and Intrinsic Time-scale Decomposition Methods in Turnout Rails

This study investigates crack detection in the turnout tip rails by analyzing the acoustic emission (AE) signals generated by crack propagation, which can help monitor turnout crack. As such crack predominantly occurs when train moves along the railway track, the damage signals contain a large number of wheel-rail noises. Therefore, a critical aspect of this research is the extraction of the subtle damage signals from the predominant wheel-rail noises. In our approach, wheel-rail noises at varying speeds were recorded on the turnout tip rail to establish a baseline of train movement noises. Meanwhile, three-point fatigue tests were conducted on U71Mn steel rail specimens to capture the characteristic AE signals of crack propagation. These typical AE signals were then artificially mixed with the noises of moving trains, simulating real-world conditions. Subsequently, the mixed signals were decomposed using two methods: Empirical Mode Decomposition (EMD) and Intrinsic Time-scale Decomposition (ITD). Accordingly, we extracted the frequency feature of the center of gravity from the reconstructed signals, employing it as a discriminant between noises and crack-related AE signals. A comparative analysis of the two decomposition methods reveals that EMD outperforms ITD method in computational efficiency and accuracy, specifically, more effective in distinguishing AE signals at a train speed of 70 km/h. This finding suggests that EMD method is better suited for real-world turnout crack monitoring applications.