Quality Assessment of CFD Software Using Workflows and Decision Trees

The present research focuses on the new efforts for leveraging so-called workflows for the management of dozens of CFD validation cases. Proposed workflows are meta-algorithms built on top of the free open-source Snakemake library. They allow manipulation of a growing database that currently contains 66 OpenFOAM cases representing various pipe and bubble column multiphase flows with supplied experimental data. The curated case set also provides documentation files. The proposed approach relies on the utilization of human-provided "keywords" for designing a featured dataset. We also use the method developed earlier for quantification of the fitness of simulated to experimental results. It leverages Fuzzy Logic for combining fitness metrics across various physical fields and produces single performance metric called "Goodness". By transforming metrics into a single crisp output value, the algorithm can assign then and additional categorized property to the simulation result of each individual case such that "degraded", "neutral" or "improved" with regard to available experimental data. At the top level, we demonstrate how successes and failures of case sets may be highlighted and analyzed with Decision Trees. The resulting tree is useful in the quality assessment of the CFD software in the final stages of solver development. Decision Trees provide a transparent way for analyzing diverse CFD validation case sets when investigating top-level CFD model changes (for example, applying a new turbulence model).