LAUSR

Abstract The damage detection for the steel-spring vibration isolator (SVI) of floating-slab track (FST) is a challenging task in urban rail transit since failure signals are difficult to perceive through existing methods. This work proposes a damage detection method for the SVI based on convolution neural network (CNN). A deep residual network which reduces the degradation risk is established as the damage conditions classifier, and the economical sensor deployment is investigated for the first time to monitor all SVIs. Using vibration responses generated via vehicle-FST coupled dynamic simulations, the network extracts damage-sensitive features from raw data to identify the damaged SVIs. For network training and testing, the multiple data sets are constructed under various scenarios. The influence of complex sensor deployment positions on detection performance is revealed, and a general rule about the number of sensors is inferred. A full-scale experiment is implemented to demonstrate the feasibility of the proposed method.