Long-Period Dynamic Characteristics of Embedded Track Using Machine Learning

Different from traditional fastener systems, embedded track is a rail placed in the groove and wrapped by a variety of polymer materials, thus realizing the longitudinal continuous support and vibration noise reduction. Due to its superior dynamic characteristics, it has been initially used in trams, subways and high-speed railways. With the promulgation of the Noise Law, its demand is also increasing. However, its structure and mechanism are relatively complex, and its dynamic characteristics changes with the service life. In addition, its performance is difficult to measure directly and service life is as long as 30 years or more. In order to analyze the dynamic characteristic changes of the embedded track throughout its life cycle, fatigue tests are performed by subjecting the embedded track to sinusoidal excitation of different amplitudes and periods. This allows to simulate its service process during the whole life cycle. Meanwhile, the vibration response of embedded track at different stages is collected. Unfortunately, it is difficult to judge the performance and state of embedded track according to the vibration response directly. In order to solve this problem, this paper proposes an embedded track long-period dynamic response analysis method based on machine learning. This method can evaluate the performance change of the embedded track without any label only based on the dynamic response. Among them, self-supervised deep learning networks are used to autonomously extract deep features of the vibration response. These features are then classified by clustering algorithms into different phases of the service life. Finally, the change law of vibration and noise performance of embedded track in different stages is explored. The proposed method can determine the performance status of the pre-embedded track based on the field vibration response test results. It also estimates the decay process of track performance with service life and determines the maintenance cycle according to the performance requirements.