Applied machine learning to estimate length of separation and reattachment flows as parameter active flow control in backward facing step / Ahmad Fakhri Giyats, Mohamad Yamin and Cokorda Prapti Mahandari
Recently, large amounts of data from experimental measurements and simulations with high fidelity have extensively accelerated fluid mechanics advancement. Machine learning (ML) offers a wealth of techniques to extract data that can be translated into knowledge about the underlying fluid mechanics....
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Faculty of Mechanical Engineering Universiti Teknologi MARA (UiTM)
2023
|
| Subjects: | |
| Online Access: | https://ir.uitm.edu.my/id/eprint/84115/1/84115.pdf https://doi.org/10.24191/jmeche.v20i3.23904 https://ir.uitm.edu.my/id/eprint/84115/ https://doi.org/10.24191/jmeche.v20i3.23904 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Recently, large amounts of data from experimental measurements and simulations with high fidelity have extensively accelerated fluid mechanics advancement. Machine learning (ML) offers a wealth of techniques to extract data that can be translated into knowledge about the underlying fluid mechanics. Backward-Facing Step (BFS) is well-known for its application to fluid mechanics, particularly flow turbulence. Typically, a numerical approach can be used to understand the flow phenomena on BFS. In some
instances, numerical investigations have a computational time limitation. This paper examines the application of ML to predict reattachment length on BFS flow. The procedure begins with a simulated meshing sensitivity of 1.27 cm in step height. This numerical analysis was conducted in the turbulent zone with a Reynolds number between 35587 and 40422. OpenFOAM® was used to perform numerical simulations using the turbulence model of k-omega shear stress transport. ML employed information in the form of Velocity and Pressure at every node to represent the type of turbulence. Using Recurrent Neural Networks (RNNs) as the most effective model to predict reattachment length values, the reattachment length was predicted with a Root Mean Square Error of 0.013. |
|---|