Study On Prediction Of Shear Mechanics Characteristics Of Rock Joints Based On BP Neural Network

In the current era of great development of engineering construction,the construction and safety maintenance of water conservancy and hydropower,public transportation,and energy mining are more complexity.It is scientific and reasonable to accurately assess the stability of large rock slopes and underground rock mass engineering structures.The supporting system is an important prerequisite for preventing geological disasters and engineering accidents to ensure the safety of the public and properties.Joint fissures are developed in natural rock masses,and their deformation and instability are primarily controlled by slippage and failure of rock joints.Therefore,it is of great significance to accurately predict the deformation and failure of rock joints.The current theoretical models are developing rapidly,but they are generally faced with shortcomings,such as being limited to their experimental work and ineffective for predicting other's work.Therefore,it is necessary to obtain a more accurate and robust model for predicting and evaluating mechanical and deformation characteristics of rock joint shear.The article analyzes many factors that affect the shear characteristics of joints,and uses artificial intelligence technology to predict and evaluate on this basis.In this paper,a larger sample size is obtained by collecting shear experiments of previous scholars.A stable topological structure is obtained by a cross-validation algorithm.The BP neural network was optimized by MBGD,Adam and RMSprop,and then predict the shear behavior.In order to further broaden the scope of application of the model and overcome the influence of differences in rough quantification methods.This paper conducts direct shear test on dry,indoor humidity,surface immersion,and fully saturated samples to simulate rock slopes in short-term and long-term rain immersion.According to the experimental results,the effects of initial asperity angle,normal stress,and water content on the strength,deformation characteristics,and failure modes of red sandstone joints are explored.The PCA algorithm and BP neural network algorithm are combined to predict the resistance under different water content.The main findings of this article are as follows:(1)Through the test results of the two-dimensional and the three-dimensional shear strength model,it is found that compared with the reconstructed input parameters,the original parameters are easier to extract the model's features.The prediction is more accurate,and it is significantly better than the theoretical model.The single factor sensitivity analysis results of the two types of models show that normal stress is the most important factor affecting the shear strength.(2)Through the analysis of the prediction results of the full curve of the shearing process,the neural network model reflects the characteristics of the experiment: with normal stress and roughness.As the normal stress increases,the peak shear stress and residual shear stress increase.Moreover,the dilatancy decreases with the increase of the normal stress.(3)The red sandstone significantly softens when exposed to water;the uniaxial compressive strength,tensile strength,and elastic modulus have a negative exponential function relationship with the water content.When water content()increase,the peak shear stress and the contribution of asperity to shear behavior decrease.When 4),? and increase,then joint surface bulge transitions from surface wear to gnawing failure,the crack initiation position transfers from the upper part to the lower part,the damaged area,and damage volume increases.Finally,comparing PCA?BP neural network and BP neural network performance in predicting shear strength,PCA?BP neural network has achieved good advantages in terms of test accuracy and model complexity.