关键词: 干湿循环, 力学特性, 砂岩, 本构模型, 神经网络

Abstract: Based on uniaxial compression test of sandstone under the different drying-wetting cycles, the laws of drying-wetting effect on mechanical behaviors of sandstone, such as deformation behavior, intensity behavior and failure characteristic, were analyzed. The results show that the elastic modulus and the peak strength of sandstone trend to decrease with the increase of drying-wetting cycles, and its reduction range is changed from large to small; moreover, the decrease of above data can not develop limitlessly but there is a constant value as critical value; in this test, the constant values of elastic modulus and the peak strength are experiment data of 20th drying-wetting cycles. The failure characteristic of sandstone is also affected by drying-wetting cycles, the less drying-wetting cycles are, the more obvious brittle damage is; that are, with the strengthen of drying-wetting effect, the damage law of sandstone present a transformation from brittleness to ductility. On the basis of these tests, taking the strain and drying-wetting cycles as input layer and the stress as output layer, the three layers neural network constitutive model, whose structure is 2-12-1, was proposed. Through the massive samples learned and inspected, the model can well describe the mechanical properties of sandstone under the cyclic drying-wetting effect, which also confirms that it is feasible and reliable for the neural network method to establish the rock constitutive model. At last, the intact function relations between the mechanical behavior and the cyclic drying-wetting effect were built.

Key words: drying-wetting cycle, mechanical behavior, sandstone, constitutive model, neural network