Particle Resolved Direct Numerical Simulation And Submodel Study Of Pulverized Coal Combustion

The combustion process of gas-solid two-phase systems is crucial in various industrial ap-plications.However,the experimental method in this field faces technical and scale limitations.To overcome these limitations,the numerical simulation method is widely utilized.Numerical simulations,known for providing comprehensive data with lower resource consumption,are essential in the study of two-phase combustion.Typically,large-scale numerical simulations of gas-solid two-phase combustion employ the point source particle model.This model describes the interaction between particles and the fluid using point-source models,but these models don’t solve these interactions directly.Because of some assumptions of the models,some small-scale effect are ignored,which cause some error.Therefore,we need to develop more accurate mod-els to achieve a balance between computational resource consumption and model accuracy.In this paper,a full-resolved direct numerical simulation platform for reacting particles is constructed,which takes into account the particle volume effect and directly analyzes the heat and mass transfer on the particle surface and in the boundary layer.In this way,the reaction rate of pulverized coal particles on the particle scale is investigated,as well as the factors affecting the force and the interaction mechanism between the particles.First,this paper investigates the effect of coke combustion on particle traction.The particle drag force is affected by both gas-phase and heterogeneous-phase reaction rates.The heterogeneous reaction leads to the cre-ation of Stefan flow,which supplements the shear layer and reduces the viscous forces while suppressing the formation of the vortex.The homogeneous reaction,on the other hand,leads to a high-temperature zone at the flame surface around the particles,where the high temperature significantly increases the viscosity coefficient,and the change in the species further leads to an increase in the viscosity coefficient.The products of the homogeneous and heterogeneous reactions accumulate in the rear of the particle,increasing the pressure drop around the parti-cle.The reactions lead to two competing processes of increasing pressure drop and decreasing viscous forces,with the effect of increasing pressure drop dominating at larger particle sizes,increasing the trailing coefficient of the particles.In addition to this,the particle and fluid tem-perature difference leads to a significant difference in particle surface viscosity and incoming flow,and high-temperature particles are accordingly subjected to larger viscous forces.Based on the simulation results,three dimensionless numbers characterizing the effect of the reac-tion are summarized to incorporate the effect of char combustion into the drag force coefficient empirical correlation.Based on the DNS database of fully-resolved simulations,we quantitatively analyze the error levels of Baum&Street model which introduced by assumptions such as”no gas-phase reactions”and”diffusion coefficients are uniformly distributed in the flow field”.The analysis of the species transport balance shows that the gas-phase reaction within the particle boundary layer has a large influence on the species transport,which makes the particle reaction rate cal-culated based on the empirical Sherwood number correlation significantly larger.In order to improve the reaction rate model,we introduce two parameters,the ratio of gas-phase reaction and species diffusion,and the Reynolds number.We use the Random Forest method to obtain more accurate Sherwood number models and use it to improve the accuracy of the Baum&Street model.In addition,the gas-phase reaction within the particle boundary layer and the vortex with reaction products can lead to the discrepancy between the velocities and compo-nents obtained by grid interpolation in the point source simulation and the actual situation.To address the above problems,the machine learning method is utilized to obtain a model of drag force and reaction rates in relation to the spatial scale of the particle and the grid,as well as relative positions,which effectively corrects the errors brought by the point source assump-tion.Furthermore,due to the relatively small size difference between the grids and particles in point source simulations and the influence of particle boundary layer composition on grid interpolation,it leads to discrepancies between interpolated components and inflow velocities.Addressing these challenges,machine learning methods are used to represent spatial scales of particles in relation to particle offset within the grid and the angle of deviation from the flow.By training machine learning models,the study effectively corrects errors arising from assumptions about material properties and point source modeling.Based on the single-particle model,we explore the mechanism of particle pairwise interac-tions on reaction rates and forces.We first disassembled the pairwise interaction into the nozzle effects,direct inter-particle flame interaction,and the wake of products and high temperature.When exploring the effect of two-particle interactions on the reaction rate,we found that the particle reaction rate was less affected by the nozzle effect,and the species transport due to the convection had less effect on the gas-phase reaction.Meanwhile,the direct flame surface inter-action did not occur at distances greater than 2D_pbecause the flame surface action was confined to the boundary layer.So,the wake of products and high temperature is the most important fac-tor in the reaction rate.The effect of the pairwise interaction on the forces is mainly caused by the nozzle effect and wake flow.Meanwhile,by comparing the trailing force coefficients of burning and cold particles,we conclude that the effect of pairwise interactions on the drag force is much more obvious than the effect on reactions on the drag force.Based on the conclusion of the quantitative analysis,we summarize the dimensionless numbers and further develop a regression model for the effect of two-particle interaction on force and reaction rate by random forest methods.