Abstract
An attempt to bridge the gap between capabilities offered by advanced full-field microstructure evolution models based on the cellular automata method and their practical applications to daily industrial technology design was the goal of the work. High-performance parallelization techniques applied to the cellular automata static recrystallization (CA-SRX) model were selected as a case study. Basic assumptions of the CA-SRX model and developed modifications allowing high-performance computing are presented within the paper. Particular attention is placed on the development of the parallel computation scheme allowing numerical simulations even for a large volume of material. The development of new approaches to handle communication within the distributed environment is also addressed in the paper as a means to obtain higher computational efficiency. Evaluation of model limits was based on the scalability analysis. The investigation was carried out for the 3D and 2D case studies. Therefore, the complex static recrystallization cellular automata simulation taking into account the influence of recovery, nucleation based on accumulated energy, and the progress of recrystallization as a function of stored energy and grain boundary mobility with high-performance computing capabilities is now possible. The research highlighted that parallelization is more effective with an increasing number of cellular automata cells processed during the entire simulation. It was also proven that the developed parallelization scheme and communication mechanism provides a possibility of obtaining scaled speedup over 700 times for 2D and over 800 times for 3D computational domains, which is crucial for future applications in industrial practice. Therefore, the presented approach’s main advantage is based on the possibility of running the calculation based on input data obtained directly from high-resolution 3D imaging of the microstructure. With that, the full immersion of the experimental results into the numerical model is possible. The second novelty aspect of this work is related to the identification of the quality of model predictions as a function of model size reductions.
Highlights
Modern approaches to the production of complex metallic components are increasingly supported by advanced laboratory research [1,2] and the constant development of more and more accurate numerical models [3,4,5]
Some of the solutions are focused on parallelization for processors within a single computing unit [20], more advanced are those designed for massively parallel computing centers [21] or modern graphics cards [22]
Approaches based on a distributed memory and the message passing interface (MPI) standard are intensively applied to CA algorithms [27,28]
Summary
Modern approaches to the production of complex metallic components are increasingly supported by advanced laboratory research [1,2] and the constant development of more and more accurate numerical models [3,4,5]. The solution provides an efficient way to use a single computational unit with a high number of cores offered by new processors, but it can only use the amount of memory available within this particular unit This is a significant limitation when simulations of large 3D computational domains generated directly on experimental data are considered. Full field modeling of microstructure evolution based on the direct input from experimental 3D imaging in an acceptable time should be possible Development of such a parallel 3D full-field CA static recrystallization (SRX) model for scientific, and industrial applications, was the goal of this research
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
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
