Implementing HZDR Interactive Baseline Closure Concept Using Fuzzy Logic and Snakemake Workflows


Implementing HZDR Interactive Baseline Closure Concept Using Fuzzy Logic and Snakemake Workflows

Evdokimov, I.; Hänsch, S.

Euler-Euler multiphase simulations imply numerical challenges and require taking into account many nuances. As pointed out by the authors of the Baseline concept [1], the successful framework for multiphase CFD should reflect a contemporary view on underlying physical phenomena and predict well arbitrary flow configurations. The Baseline methodology [1] specifies a particular meta-algorithm for proposing new sub-models and validating them on a large number of cases (re-interpretation of the original Baseline strategy [1] is shown on the Figure). Although original work presents a clear specification of methodology in general aspects, it lacks the most important definition of “overall improvement”.
The definition of overall improvements is not a single challenge. In addition, simultaneous evaluation of the whole case set becomes important when the number of cases approaches several dozen setups with at least 3 different plots per setup. Meanwhile, human evaluation of all generated plots may be viewed as a serious burden for the research of new better EE models at scale. To address both challenges we propose implementation of the Fuzzy Logic Controller (FLC) and the workflow architecture for scalable simulations and results evaluation of hundreds of cases (potentially).
FLC is applied universally for each case and it ensures that all requirements are fixed. It relies on two metrics calculated for each plot (i.e. void fraction or gas velocity). These are average relative error and Pearson coefficient serving as input crisp values. As an output value, FLC provides the Goodness value which is a crisp output and indicates badly fitting simulation data (with G=0) or perfectly fitting results (with G=1). It uses fuzzy sets “low”, “medium”, and “high” (similar to example in [2]) for each error and Pearson coefficient intervals and helps to achieve better flexibility and clarity than some artificially constructed function.
FLC is an important component of the post-processing Snakemake workflow(more details available in [3]). The combination allows improvements on many levels. Those are 1) explicit specification of the “improvement”, 2) productivity gain in a sense of aggregation of results in a single report, 3) extensible post-processing system which may help to analyze hundreds of cases simultaneously. The demonstrated approach may be also applied to other fields of study where multiparametric models must be tested against multiple simulation setups.

Keywords: fuzzy logic; workflow; CFD; multiphase

  • Open Access Logo Poster (Online presentation)
    Multiphase Flow Workshop 2021, 08.-10.11.2021, Online, Germany

Downloads

Permalink: https://www.hzdr.de/publications/Publ-33119
Publ.-Id: 33119