Prediction of freezing damage in high-speed railway tunnels under airflow influence in cold regions

Tunnels in cold regions will encounter serious frost damage during operation. Based on the Tongsheng Tunnel, an unsteady heat transfer model of surrounding rock, tunnel lining, and wind flow was established. The nu-merical method was verified by a field test. The effects and sensitivity of airflow factors on the tunnel tem-perature field were studied, especially the freezing depth (XF) and heat-adjusting depth (XH). Then a BP neural network was developed and trained to predict tunnel freezing conditions. The results show that: (1). XF and XH are negatively correlated with T (average air temperature in winter) at the same position and positively corre-lated with S (average wind speed in winter). However, the relationship between XF, XH and D (dominant wind direction in winter) is related to L (distance from the entrance). XF and XH are positively correlated with D when L is less than 150 m, but negatively correlated with D when L exceeds 150 m. (2). The order of sensitivity to the effect of the tunnel temperature field is T > S > D. (3) The tunnel direction should be avoided parallel to the wind direction, and should be intersected perpendicular to the wind direction or at a large angle as much as possible. (4). Finally, by neural network, using T, S, D, and L to predict XF and XH of the tunnel arch bottom is feasible and efficient. Consequently, the BP neural network is capable of predicting the freezing damage location in tunnels in a convenient and efficient way.