Examination of efficient CFD analysis mesh creation based on prediction of vortex radius by machine learning

Matsushita, Kentaro ; Ezure, Toshiki ; Fujisaki, Tatsuya*; Imai, Yasutomo*; Tanaka, Masaaki  

Development of evaluation method for cover gas entrainment (GE) by vortices generated at free surface in upper plenum of sodium-cooled fast reactor (SFR) is required. An evaluation method by predicting vortices from flow velocity distribution obtained by 3D CFD analysis is developed, and partial refinement of analysis mesh from the viewpoint of improving efficiency of 3D CFD analysis is examined. In this study, a method of predicting vortex radius that can occur in a system by machine learning and using the predicted value as a reference value for analysis mesh before applying refinement (initial mesh). CFD analysis was performed for cylinder column system where wake vortex occurs by changing parameters such as cylinder diameter and inlet flow velocity, and supervised learning was performed using the obtained variables and vortex radius was predicted. As a result, it was obtained that initial mesh size can be efficiently determined by using the predicted vortex radius as a reference.