Methods to increase computational efficiency of CALPHAD-based thermodynamic and kinetic models employed in describing high temperature material degradation. (June 2016)

• Record Type: Journal Article
• Authors: Pillai, R.; Galiullin, T.; Chyrkin, A.; Quadakkers, W.J.
• Is Part Of: Calphad. Volume 53(2016)
• Page Start: 62
• Page End: 71
• Publication Date: 2016-06
• Subjects: CALPHAD -- Thermodynamic and kinetic modelling -- Interdiffusion -- MPI -- Interpolation; Phase diagrams -- Data processing -- Periodicals; Thermochemistry -- Data processing -- Periodicals; Diagrammes de phases -- Informatique -- Périodiques; Thermochimie -- Informatique -- Périodiques; Thermodynamica; Electronic journals; 541.363
• DOI: 10.1016/j.calphad.2016.03.004
• Languages: English
• ISSNs: 0364-5916

Abstract

Abstract: Coupled thermodynamic and kinetic models rely on the thermodynamic and mobility databases which are compiled using critically assessed thermodynamic and diffusivity data acquired from various sources of experimental data. A continuous influx of experimental thermodynamic and kinetic data means that the respective databases have not only increased in complexity but also in size. The time and computational effort for the equilibrium calculations increases with increasing number of components and phases to be considered. In the present work, the applicability of a few methods was investigated to increase the computational efficiency of coupled thermodynamic and kinetic models. Three cases of varying complexities in terms of the number of phases, alloying elements and phenomena to be modelled were considered for demonstration. The distribution of the intensive thermodynamic calculations on multiple computing cores using MPI (Message Passing Interface) was undertaken. The interpolation scheme for dynamic storage of thermodynamic data available in the commercial software DICTRA was employed on a single computing core and the resulting performance was compared with the MPI computations. Additionally, the interpolation scheme was also parallelised to test its scaling capability in comparison to the computations performed solely with MPI. A linear scaling of computation speeds was observed with parallelisation of the thermodynamic calculations with MPI. However, the degree of scaling was dependent on the complexity of the calculation. The interpolation scheme on a single core in comparison with MPI on 48 cores was 20 times faster in one case but about 20–50 times slower in the other two cases. A parallelisation of the interpolation scheme improved its performance in the other two cases. However, the computational scaling was still poor compared to the MPI computations. Abstract : Highlights: Methods to increase computational efficiency of thermodynamic–kinetic models. MPI protocol was employed for accelerating the equilibrium calculations. Almost linear scaling in simulation times was obtained with parallelisation methods. Performance of the interpolation scheme was dependent on the complexity of the case.