关键词: 胶结材料, 离散单元法, 率敏性, 神经网络

Abstract: In order to establish the relationship among microscopic properties, loading rates and macroscopic responses of cemented geo-materials, the parallel bond model was employed to simulate the physical and mechanical characteristics of the cemented materials based on the discrete element method(DEM). Moreover, a series of undrained triaxial numerical tests were performed with different microscopic parameters (number, cohesion, Young's modulus, internal friction angle of bonds, and void ratio) and loading rates (1, 0.1, 0.01, 0.002 mm/min). Based on three parameters include residual strength, peak strength and the corresponding axial strain, the rate-dependent behaviors of the cemented geo-materials under different microscopic parameters were discussed. Furthermore, based on the numerical test results, BP neural network algorithm is used to establish an intelligent model for predicting the rate-dependent behavior of the macroscopic properties in cemented materials. The results show that: (1) the cemented material has a significant loading rate sensitivity characteristics, presenting a significant increase of the peak strength with the increase of loading rate, and it has a semi-logarithmic linear correlation. However the residual strength and axial strain at peak strength are less sensitive to the loading rates; (2) the rate-dependent behavior of the cemented material is mainly caused by the fragmentation of internal bonds. During the whole shearing process, the evolution of average bond breakage percentage per unit strain shows a similar trend as that of the deviatoric stress. In addition, the average bond breakage percentage increases with the increase of loading rates; (3) the proposed BP neural network model, which considers the influence of microscopic parameters and loading rates on macroscopic responses, can reasonably describe the rate-dependent behavior of the macroscopic properties in cemented materials with the relative error around 10%.

Key words: cemented materials, discrete element method, rate-dependent behavior, neural network