2019 Fiscal Year Final Research Report
High-precision turbulent modeling by machine learning
Project/Area Number |
17K14588
|
Research Category |
Grant-in-Aid for Young Scientists (B)
|
Allocation Type | Multi-year Fund |
Research Field |
Fluid engineering
|
Research Institution | Osaka University |
Principal Investigator |
Shimizu Masaki 大阪大学, 基礎工学研究科, 助教 (20550304)
|
Project Period (FY) |
2017-04-01 – 2020-03-31
|
Keywords | Turbulent model / Machine learning |
Outline of Final Research Achievements |
We have reduced the computational cost of fluid dynamics by using various data-driven methods in machine learning. Numerical calculation of turbulence requires an enormous calculation cost, but it became significantly small by solving the model of the minimum necessary variables constructed by deep learning. Using dynamic mode decomposition, which is one of data-driven algorism, we devised a method that enables efficient control of object motion in a fluid by using modes decomposed from flow data. The reinforcement learning algorithm was modified to be suitable for flow control, enabling low-cost reinforcement learning.
|
Free Research Field |
流体力学
|
Academic Significance and Societal Importance of the Research Achievements |
乱流の数値計算は様々な工学的製品の設計開発において用いられ,効率的で高精度な性能評価が求められる.物理的な面から乱流計算の効率化が進められてきたが,近年は機械学習による効率化が急速に進みつつある.本研究では乱流が少数自由度で記述できるという性質を用いて,機械学習の様々なアルゴリズムを流体工学の問題に応用した.これにより,物理から演繹的には導出不可能な方法によって,【研究成果の概要】で述べた最適化手法が可能になり,更なる工学製品の省エネの実現に結び付くことが期待できる.
|