Font Size: a A A

The Bubble Dynamics Study Based On CUDA And Lattice Boltzmann Method

Posted on:2016-01-29Degree:MasterType:Thesis
Country:ChinaCandidate:Y H QinFull Text:PDF
GTID:2370330464953626Subject:Computer software and theory
Abstract/Summary:PDF Full Text Request
The movement of bubbles is common in nature and engineering applications of the classic two-phase flow phenomenon.Bubble dynamics play an important role in real life and engineering applications.In recent years,many scholars use the theoretical analysis and experimental methods to study the dynamics of the bubble.however,since these methods have certain limitations.coupled with the complexity of bubble movement,research in this area is greatly restrictions.With the development of computer hardware and software technology,with the rise of low cost,high accuracy,high computational efficiency of numerical simulation methods such bubble dynamics based research carried out rapidly.Lattice Boltzamnn method(Lattice Boltzmann Method,LBM)is a numerical simulation method has mesoscopic characteristics,its algorithm is simple.easy to handle border with natural parallelism,since birth,it has been widely used in multiphase flow,microscale research stream,chemical reaction flow,flow in porous media and other aspects.Based on CUDA parallel technology,is a kind of high performance computing technology,which is developing rapidly in recent years.CUDA technology,make full use of the advantage of the GPU multi-core multiple-pass,simple programming,high efficiency,greatly promoted the GPU performance boost,and it is quickly used in the field of parallel computing and image processing.At present.CUDA technology has been in medical image,computational fluid mechanics,environmental engineering,and other fields are widely used.This paper proposes a LBM model which combines with the pseudopotential model and the free energy model.Model adopted standard momentum evolution equation and calculation method of the equilibrium speed of the pseudopotential model,the evolution process is simple,do not need to solve complex partial differential equation,using the interface capturing equation of the free energy model,require less discrete speed.Model incorporates the advantages of the pseudopotential model and the free energy model,can adapt to simulate two-phase flow of big density,but also has small spurious speed.In this paper,the model is used to study dynamic characteristics of air bubbles,and use the parallel technology based on CUDA to optimize the LBM program.First,model validation.Verify Laplace's law,it took a total of 12 different simulation solution for the bubble radius external pressure difference,the theoretical analysis and simulation curve curve exactly.Possessive flow field parameters point to the distance from the center of the bubble sequence there is a certain relationship between the radius of the bubble model taken 30 lattice units simulation,simulation results consistent with the theoretical analysis of the curve,the curve showing anti "Z"shape.Radius of the bubble 20 grid units in the initialization state,and the abscissa parallel sequence parameter value after grid point on a straight line as the center of the bubble theory transactions;at steady state,and the abscissa parallel straight line through the center of the bubble the value of the order parameter on the grid as analog data,the data is plotted in two different states,the theoretical curve is consistent with the simulation.The above experiments show that the simulation results with the theoretical model of the basic agreement.Next,Relationship single bubble with model parameters.Surface tension is a gas-liquid contact surface discontinuity resulting from a force.Bubbles in different interface thickness,the size of the surface tension are also different variations.In this paper,bubble radius defined as 20 grid units,taking 10 different interface thickness calculating the surface tension of bubbles.Simulation results show that,in a certain range,as the interface thickness increases,the surface tension of the bubble closer to theoretical values.Virtual speed exists but do not exist in actual when LBM simulates..the larger the virtual speed,the more unstable the simulation.Papers take 10 different surface tension calculated radius of 20 lattice units maximum virtual speed when the bubble reaches stable.The results show that with the increase in surface tension.the greater the speed of virtual simulation produced.Therefore,When LBM simulation,try to get small and suitable surface tension.Then study the consolidation of air bubbles.First study the factors that affect the merger of two bubbles.Migration factor,interface thickness.surface tension and other parameters are important factors that influence the bubbles merge.In this paper,one of the two parameters is fixed,change the third parameter,study the influence of various.parameters on the combined of the bubbles.The simulation results show that the greater the migration coefficient,two bubbles merging faster,simulation results agree with the literature[34]well.This paper also studied the interface thickness,the effect of surface tension of bubbles merge,the greater the thickness of the interface,the two bubbles merging faster;the greater the surface tension,the two bubbles merge faster.The researchers then three symmetrically placed bubbles,four symmetrically placed the consolidation of bubbles,the bubbles can be merged together and merge mto the middle of four symmetrical bubble liquid containing large bubbles.Finally,Use CUDA technology to optimize LBM program.Dimension division in the optimization process,Grid values as large as possible,the value of block is try to make the block is an integer multiple of the number of threads 32.Through the Grid and reasonable division of block size,the program received the highest 24.6876 times speedup.In the memory access optimization process,the paper focuses on global memory and constant memory optimization.Global memory optimization is to satisfy the merger access conditions.constant memory optimization is the procedure is always the same amount of memory to define constants.Optimization program memory' access by 25.41 times speedup.In instruction optimization process,paper use arithmetic instruction optimization method,the program for all low clock cycle operand arithmetic instructions into high clock cycle operand arithmetic instructions,procedures to obtain the highest 26.872 times speedup.Comprehensive optimization process,the program will more than several single optimization program together.because the effects of various optimization program optimization is not linear superposition,the program received the highest 29.931 times speedup.Good optimization results greatly enhance the efficiency of LBM simulation program.
Keywords/Search Tags:LBM, CUDA, parallel computing, optimization
PDF Full Text Request
Related items