The Characterization And Prediction Of Particles Falling Behaviors In The Hopper

Particles can be seen everywhere in daily life.Moreover,a hopper has very wide and vital applications in handing the granular materials in daily life and the industrial production.So,the full understanding of the particle flow inside a hopper and segregation behaviors is of great importance to control and optimize the discharge process.Therefore,this paper was analyzed the particle flowing pattern from the basic research and the industrial exploration.Firstly,by employing experimental and numerical methods,the influence of particle packed pattern on the transient particle flow is investigated in terms of the particle-scale kinetics and structure.For the mono-sized particles packed pattern,despite the similar particle-scale structure,smaller particles achieve greater kinetic energy conversion efficiency,which help shorten the discharge time.For the binary-sized particles uniform mixing pattern,the interaction between particles increases the individual kinetic energy and transient average coordination number of large particles,while decreases that of small ones.Then the in-between kinetic energy and the disperse structure are reached.For the layer by layer mixing pattern,the strong percolation effect caused by the upper small particles hinders the increase of the individual kinetic energy at the beginning of the discharge process,and the transient average coordination number at the layer interface abruptly reaches 8.By contrast,when the small particles are placed at the bottom,more particles are active in the larger space,and subsequently,a looser structure is achieved in a shorter period.At the same time,most of the current research on particles uses DEM method,but when using DEM to simulate particles'behaviors,it will consume a lot of computing resources and the calculation time is longer.So,this paper proposes a combination of DEM and neural network to shorten the calculation time of DEM and save the computing resources.Based on the picture data,the convolutional neural network(CNN)is used to predict the discharge time of the particles in the wedge-shaped hopper and Alexnet-FC is the suitable convolutional neural network structure for this paper.This network structure can accurately predict the discharge time of the particles in the wedge-shaped hopper and the R~2 value can be up to 0.998.Moreover,by comparing the running efficiency of the DEM simulation time and the training time on the GPU and CPU,it can be seen that training the neural network on the GPU can significantly save DEM simulation time and saves computing resources,but the DEM simulation speed is faster than the neural network training speed on the same CPU machine.Secondly,for the bell-less top blast furnace,significant mass and size segregation will occur at the throat.This paper combines DEM simulation and artificial neural network(ANN)to predict particle segregation.Using those methods for the combination of basic research and industrial application.Firstly,DEM was used to explore the segregation of particles in the charging process and quantitatively describe its segregation characteristics.Then,based on DEM simulation data,an ANN network model is trained to predict the particle segregation behaviors in the radial direction of the furnace throat.The results show that the average standard deviation in each ring is 0.249.So,when the initial particle mass is determined,the mass distribution in the radial direction is also fixed at different initial mass ratios.The mass from the center to the edge increased from18.888kg to 27.560kg,and finally decreased to 21.968kg when the mass ratio is 0.3.For the mass ratio of small particles in the same ring,as the initial mass ratio of small particles increases,the smaller particles in the same ring is also increasing.Therefore,the particle mass first increases and then decreases in the radial direction.For the distribution of the particle mass ratio,small particles are more likely to gather in the center area,while large particles are concentrated in the edge area.Moreover,by comparing with the simulation results,it is proved that the established artificial neural network model can effectively predict the segregation behaviors of particles.