CPU+GPU Heterogeneous Parallel Computing Of Dendritic Growth Of Steel

Steel is currently the most widely used metal material by mankind,and it is also the metal material with the largest production scale and the most mature production technology.Although the current industrial methods such as continuous casting technology for steel production have been very mature,the performance of steel materials is still affected by defects such as segregation.Dendritic growth,fluid flow,dendritic motion and solute diffusion have a great influence on the formation of macrosegregation in the cast slab.In the current research,although the dendrite growth and solid phase transformation during the solidification of smaller samples can be observed through X-ray in-situ observation technology,the solidification of steel is a continuous solidification process of high-temperature molten steel,and its products are large in scale and opaque.Therefore,the in-situ observation of the dendrite growth behavior during the solidification of steel has not yet been achieved.Due to the rise of computer technology in recent years,especially the emergence of high-performance devices suitable for parallel computing such as GPU,numerical simulation methods have attracted more and more attention.The rise of simulation methods such as the Phase Field(PF)method,Cellular Automaton(CA)method,and the Lattice Boltzmann method(LBM)has allowed people to have more methods and higher efficiency to explore the effect of the external field on the branch during the solidification of steel.The influence of crystal growth behavior and motion.By comparison,the phase field method is adopted to couple the phase field,solute field and temperature field,and a phase field model for dendritic growth of binary,multicomponent alloy and a multiphase field model for inclusion reaction are established.The dendritic growth of binary alloy of peritectic steel under isothermal and non-isothermal conditions,dendritic growth during solidification of ternary alloy and peritectic phase transformation during solidification are simulated respectively,and the effects of different parameters on the growth reaction process are discussed.The main research contents and results are as follows:(1)Two-dimensional and three-dimensional PF-LBM models are established,and the accuracy of the model is tested using the classical analytical model and the force motion behavior of the particles.The process of dendrite growth during steel solidification is simulated.When the undercooling degree is 15K and the simulation of dendrite growth behavior is extended from two-dimensional to three-dimensional,the steady growth rate of the dendrite tip increases from 4.3×10-3m/s to 5.3×10-3m/s,an increase of 1.2 times,the phenomenon of solute enrichment becomes weaker,and the solute layer becomes thinner.(2)The use of CPU+GPU heterogeneous parallel computing provides efficient computing efficiency for the solution of the PF-LBM model in three-dimensional space.For the current model,when the total number of grids is set to an integer multiple of 32 and the thread block size is configured to 128,the best acceleration effect will be obtained.The larger the number of grids,the better the advantages of heterogeneous parallel computing.When the number of grids is 3843,the speedup ratio reaches 1567 times.After using the segmented memory loading method to remove the physical limitation of GPU memory,the speedup ratio reached 1700 times when the number of grids was 5123,and the powerful computing power of CPU+GPU heterogeneous parallel computing was more fully utilized.(3)In the simulation of steel solidification dendrite growth,the existence of forced convection will affect the dendrite morphology and solute distribution,resulting in asymmetry of dendrite growth.When the melt convection exists,the melt will flow around the dendrite and transport the solute on the upstream region of the dendrite to the downstream region,so that the solute concentration gradient of the upstream region is greater than the downstream region,and the growth rate is higher.the secondary dendrites are more developed,and the growth of the dendritic arms on the downstream region is inhibited,the secondary dendrites will not grow,and the asymmetry index increases sharply.(4)The motion of the dendrites in the flowing melt will have a greater impact on the dendrite morphology.Melt convection promotes the acceleration of the dendrite,and the relative velocity between the dendrite and the melt gradually decreases to 0.The influence of the melt convection on the dendrite morphology basically disappears,and the dendrite maintains an equiaxed crystal shape.In the static melt,the dendrites will sink downward under the action of gravity,agitate the melt and form a vortex near the upward dendrite arms.The relative movement between the dendrite and the solution will transport the solute to the vicinity of the upward dendrite arm,inhibiting the growth of the upward dendrite arm,and the downward dendrite arm grows fastest.The swirling flow will not only drive the dendrites to make centrifugal movement around the swirling center,but also spin the dendrite,which causes the melt impact on each dendrite arm to be more uniform,and the dendrite morphology still maintains various symmetry.(5)The growth behavior of multiple dendrites will affect each other.When multiple dendrites grow together,the dendrite gap becomes the most serious area of solute enrichment and the final solidification position,and the probability of segregation is higher.In the presence of melt convection,since the melt can pass through the gaps of dendrites,the protective effect of upstream dendrites on downstream dendrites is relatively weak.Melt convection has a greater impact on the growth behavior of all dendrites in the computational domain and the dendrite morphology is asymmetric.In the densely distributed areas of dendrites,the transport of solute by the melt is weakened,solute enrichment is serious,and the growth of dendrites is inhibited.