关键词: 颗粒流模型, 正交试验, 宏细观参数关系, 神经网络算法, 互演计算

Abstract: Mesoscopic discrete element method on the basis of particle flow model is one of the latest numerical methods in recent years, and plays an important role in the area of discontinuous medium of geotechnical engineering. Nevertheless, its application in actual engineering is limited because of complex interrelationship between macroscopic and mesoscopic parameters. Based on the theory of orthogonal test, we take an unbonded material as the test object and design 9  samples which are tested in biaxial compression under 3 different confining pressures. In this test, 4 factors are considered and each varies in 3 levels, which can reflect the change of macroscopic property of medium caused by different mesoscopic parameter groups, and can avoid the flaw of controlling variables method which makes some variables constant. According to test results, the sensitivity of mesoscopic parameters are discussed and principles about parameter adjusting are obtained. In the end, mutual-deduction of macroscopic and mesoscopic parameters is achieved through neural network algorithm. This research could offer reference for the use of software PFC2D/3D (Particle Flow Code in 2 Dimensions/3 Dimensions) in actual engineering.

Key words: particle flow model, orthogonal test, relationship between macroscopic and mesoscopic parameters, neural network algorithm, mutual deduction method