Abstract
To accelerate the large-scale cellular automaton (CA) simulation for grain growth, a parallel CA model for grain growth was developed. The model was implemented based on the compute unified device architecture (CUDA) parallel computing platform. The model was verified by the grain growth of a single crystal and the columnar-to-equiaxed transition (CET) of an Al-7wt% Si specimen of uniform undercooling with a constant cooling rate. The grid independence of the model was verified. The grain growth of a plate-like casting of nickel-based superalloy during directional solidification process was simulated and the obtained results of grain density at each section with different heights were compared with the experimental data. The CET transition of directional solidified Al-7wt% Si cylindrical ingot was simulated. The grain texture and cooling curves were in good agreement with experimental results from the literature. Finally, high parallel performance of the CA model was obtained and evaluated.
Highlights
Nickel-based superalloy has been widely used to produce blades which are used in aero engines and industrial gas turbines
The decentered cellular automaton (CA) algorithm was proposed by Gandin et al [3] and it was widely accepted to predict the dendritic grain growth
Nastac et al [5] proposed a stochastic modeling of microstructure formation to predict the grain structure in castings and the influences of various neighborhood configurations on nucleation and grain growth were evaluated
Summary
Nickel-based superalloy has been widely used to produce blades which are used in aero engines and industrial gas turbines. The parallel computational method for the CA model at the mesoscopic scale, like the grain envelope, is much helpful to expedite the simulation of DS process of superalloy blades. The MPI technique was utilized by Lian et al [28] to accelerate the mesoscopic CA model for the grain growth during additive manufacturing and the scaling test indicated that the parallel efficiency can reach 80% for the simulation consisting of about half a billion cells. GPU-based parallel computing technology has been widely accepted in dendrite growth simulation using phase field model, due to massive computation capacity and high memory bandwidth, where high efficiency and scalability were demonstrated [29,30,31]. (6) Status transition of the interface cells and the corresponding captured liquid cells
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
