A method for constitutive modeling of rock and soil granular materials based on deep learning and data-driven

Abstract

The invention provides a method for establishing a constitutive model of rock-soil granular materials based on deep learning and data-driven, including: using discrete element numerical tests instead of laboratory tests to study the macro- and micro-mechanical properties of rock-soil granular materials; Numerical samples of granular materials are subjected to discrete element numerical tests under different loading paths, and a large amount of stress-strain relationship data are obtained; the accumulated absolute strain is used as the state variable of the rock-soil granular material to describe the current state of the granular material; the improved long-term Short-term memory neural network unit, build a deep learning network, input non-sequence data such as the structural characteristics and initial state of the numerical sample into the first unit of the network, input the strain increment during the loading process into the network in turn, and output the granular material The current stress and other state quantities; the noise reduction method based on singular value decomposition is used to reduce the noise of the training data to prevent the network training from not converging or the error is too large.

Description

technical field [0001] The invention belongs to the research field of constitutive models of rock and soil granular materials, and relates to the constitutive theory of rock and soil granular materials, discrete element numerical experiments and deep learning. The improved long-short-term memory neural network is used to mine the granular materials obtained based on discrete element numerical experiments. Based on the stress-strain relationship data, the data-driven constitutive relationship of rock-soil granular materials is established, which provides a new idea for the multi-scale coupling of rock-soil granular materials at the macroscopic and mesoscopic levels. Background technique [0002] A considerable part of the existing environment or objects of geotechnical engineering, water conservancy engineering, road and bridge engineering, etc. is composed of discrete granular materials, such as clay, sand, gravel, coarse-grained soil, etc., which can be collectively referred...

AI Technical Summary

Problems solved by technology

The development of the constitutive model makes it possible to more comprehensively consider the complex mechanical properties of rock and soil granular materials, but also introduces more model parameters, some parameters can be calibrated through experiments, while some parameters are more difficult to determine

[0005] In addition to the problem of model parameter values, the macroscopic phenomenological constitutive model based on continuum mechanics still has the following problems: the macroscopic phenomenological modeling idea only considers the macroscopic stress-strain relationship, and cannot describe the mechanical behavior at the particle level; Conventional test devices and test techniques cannot explore the mechanical properties of rock-soil granular materials under complex loading paths; macroscopic phenomenological constitutive models are mostly established based on traditional constitutive theories, and these theories are mostly proposed based on metal material tests, including some physics Assumptions, such as Drucker postulate, associative or nonassociative flow criterion, etc.

Images