Intelligent Modeling Method For Repose Angle Of Granular Materials Using DEM Data

Granular materials exist widely in nature and engineering,and are widely used in agriculture,construction,pharmaceutical and other industries.The repose angle is an important macroscopic parameter characterizing the flow behavior of granular materials.The value of repose angle is influenced by many factors including particle diameter,friction and density of the granular material.Due to importance,much effort has been devoted to the study of repose angle.With the continuous rapid development of high-performance computing technology,discrete element method(DEM)has become a powerful simulation method for studying the repose angle.However,DEM needs to calculate the trajectories and velocities of all particles in each time step,resulting in high computational cost,slow simulation process,and limitations in practical usages.In order to solve this problem,an intelligent prediction model for the repose angle is proposed using data-driven method,based on DEM simulation data.The established intelligent model is then used to replace the DEM model to quickly predict the value of repose angle when the input parameters are changed,so that long-time DEM simulation can be avoided and the computing efficiency dramatically improved.The main work and conclusions are as follows:(1)A parametric study is conducted to find out the main simulation parameters(particle diameter,density,friction,shear modulus,poisson's ratio etc.)that influence the repose angle.The possible range of each parameter is also determined.By use of Latin hypercube sampling method,200 sets of input parameters with random and uniform distribution in high-dimensional parameter space are obtained.(2)The obtained 200 sets of parameters are used as inputs of a DEM model for granular material filled in a holly cylinder.The simulated image of the repose angle is obtained,based on which the value of the repose angle is obtained.The obtained 200 sets of data,i.e.200 input-output pairs,are obtained for training of the intelligent models,with the first 100 as training sets,and the last 100 as testing sets.(3)Based on the training set,the support angle vector(SVM)is used to establish an intelligent model for the repose angle.In the training process of SVM model,the particle swarm optimization algorithm is adopted to optimize the model parameters,including the penalty coefficient C and kernel function g.The results show that the SVM model has a decision coefficient of0.96 on the training set.The testing dataset is then used to analyze the predictive performance of the trained SVM model.It is found that the SVM model has the advantages of high prediction accuracy and fast calculation speed: the decision coefficient between SVM predicted and DEM simulated result is as high as 0.94,while the time needed by SVM model to predict a repose angle is dramatically shorter than that of DEM simulation(0.17 seconds vs.40 hours).(4)The established SVM intelligent model is compared with other models,including BP neural network model and the Kriging model.The SVM model shows better performance in the index of determining coefficient,root mean square error and average absolute percentage error.Moreover,the proposed SVM intelligent model for predicting the repose angle is tested by physical experiments.The error between the model predicted value and the experimentally measured value is found to be within ±1°,which demonstrates the effectiveness and reliability of the proposed SVM intelligent model.The proposed SVM model is effective in rapid prediction of the repose angle of granular materials.It avoids the problem of numerous computations occurring in DEM simulation,and can be expanded in the modeling of other macroscopic features of the granular material.