Research and prediction of early-age loads in double-block ballastless track structure

Abstract

Considering the hydration reaction of cement, a finite element model(FEM) of the coupling of temperature and humidity field and chemical field in the early age of the double-block ballastless track structure was established, and a field monitoring test on the early age of track slab was carried out with the newly constructed Fuzhou-Xiamen high-speed railway line as a case to verify the correctness of the FEM. Based on the official meteorological data from China Meteorological Administration (CMA) and the FEM, the dataset of vertical temperature and humidity gradients in the early age of the track slab was constructed, and five machine learning training models, namely, machine learning (MLR), multivariate polynomial regression (MPR), support vector regression (SVR), random forest (RF), and gradient boosted regression (GBR), were adopted to train the temperature gradient model and humidity gradient model for the early age of the track slab. The research results show that: (1) under the combined effect of solar radiation, convective heat transfer, radiative heat exchange and the cement hydration reaction, the temperature loading leads to the susceptibility of the sleepers to diagonal cracking; the humidity loading leads to transverse cracking in the center region of the surface of the track slab; (2) the finite element model with a higher accuracy provided the training dataset for the machine learning model. According to the Spearman correlation coefficient, the input features of the temperature gradient model and the humidity gradient model are determined, respectively; (3) Five ML methods were used to train the temperature gradient model and humidity gradient model, both of which demonstrated good generalization ability with a coefficient of determination(R²) greater than 0.85. Among these methods, GBR and SVR exhibited the best performance; From the perspective of MAPE value, the prediction effect of the temperature gradient training model was better than that of the humidity gradient training model; (4) The curing method and the average value of the relative humidity were the important input features influencing the training models for the temperature and humidity gradient loads, respectively. This finding provided an important guideline for the construction of ballastless track structures.