Simulation data generating algorithm for railway turnout fault diagnosis in big data maintenance management system

Abstract

Currently, the identification of turnout failures mainly depends on the use of intelligent models. However, speed-up turnouts cannot provide enough fault samples for model training. A fault diagnosis model with insufficient fault samples can cause serious safety problems due to underfitting. In this paper, we are aiming at proposing a speed-up turnout fault generation algorithm (SFGA) to address the fault-insufficient problem. The algorithm analyzes data collected from the big data management system (BDMS), and then generates 11 common faults for speed-up turnout based on Bayes Regression, Harmonic Superposition, and Fixed Constraint models. Furthermore, this paper employs derivative dynamic time warping (DDTW) to calculate similarities between simulation faults and real faults for evaluation. An experiment based on real data collected from the Guangzhou Railway Bureau in China demonstrates that all simulation faults generated by SFGA are efficient, and can be used as a training set for fault diagnosis methods of speed-up turnouts.