A novel reconstruction method of temperature field for thermomechanical stress analysis of arch dams

Abstract

Temperature distribution on the dam surface plays an important role in reliable thermomechanical stress analysis of arch dams. In this paper, a novel reconstruction method based on a convolutional neural network is proposed for the temperature field on the downstream surface of arch dams. The environmental variables, including solar radiation, mountain shadows, ambient temperature, and weather conditions, are considered in the reconstruction method. The temperature field of the Xiaowan dam is predicted using the proposed method and it agrees well with the monitoring data in the spatial and temporal scale. The results show that the environmental variables significantly affect the temperature on the downstream surface, and result in temporal-spatial variability of nonuniform temperature distribution. In addition, the proposed method solves the data sparseness and low-frequency acquisition problems that may miss the peak temperature values in the current monitoring system. The thermomechanical stress analysis based on the reconstructed method and finite element method is conducted, and its results illustrate that the proposed reconstructed method can give more realistic performance of the dam and predict possible thermal cracking on the dam surface.