Robot for the Classification of Damaged Concrete Supports in Non-Ballasted Railway Tracks in Tunnels: Modal Analysis and Deep Learning Methods

Low Vibration Track (LVT) is a type of non-ballasted railway technology that has been installed for more than 1,800 km around the world. This system consists of a concrete block supported by a micro-cellular pad, all integrated within a rubber boot embedded in an unreinforced concrete slab. Due to heavy railway traffic, the concrete blocks can be damaged and may develop different types of cracks, such as vertical or horizontal fissures. Identifying such damage is particularly challenging because the block is surrounded by a rubber boot and embedded in the concrete slab. Even when fissures are visible on the upper surface of the block, they may not accurately reflect the internal condition of the block. In this work, we first present a modal analysis method to identify damaged blocks. Numerical and experimental results show that variations in the eigenfrequencies reveal the presence of damage, particularly in the first two modes (corresponding to bending and torsion). Based on these results, we developed a classification method that assigns each block to one of five damage levels, ranging from healthy to fully ruptured. A robot, called COBRA, was designed to deploy this method in the Channel Tunnel. This robot collects measurements and automatically classifies the vibration signals of each block. However, the measurement data may contain noisy signals that affect classification accuracy. To improve the overall performance, a deep-learning approach was employed to detect and filter out noisy data. This study demonstrates that more than 95% of noisy signals can be correctly identified, significantly improving the robot’s performance.