Identification of Creep Camber State of 32-m Box Girders on High-Speed Railway and Prediction of Geometry of Track over Girder

The creep camber of box girders significantly lowers track regularity, making it very important to perform track geometry precise adjustment or ballast tamping operations by maintenance machinery on box girder sections at the right timing. The appropriate maintenance operation should be conducted after the creep camber reaches a stable state. To address this challenge, in this study, power and frequency spectra were used to identify the typical creep camber section of 32-m box girders on a railway line based on the inspection data of high-speed track geometry cars. Also, wavelet decomposition, moving average filtering, and the length error minimization principle were adopted to position both ends of each girder. The Mann–Kendall trend test was conducted to determine whether the creep camber reached a stable state, and ARIMA (1,1,1) was employed to predict track surface values at the ends of box girders with still-developing creep camber. The validity and reliability of the proposed framework were verified using the inspection data on the Beijing–Shanghai high-speed railway track $\mathrm{K}313+84\sim \mathrm{K}691+895$ during the period of 2012–2016. The results show that the proposed framework can accurately identify the typical creep camber section based on inspection data and precisely position both ends of a 32-m box girder. The determined creep camber states were consistent with the manually obtained results, and the proposed method can predict track surface values at the girder ends for the next 5 months more accurately than the long short-term memory networks (LSTM) and supported vector machine (SVR).