Opin vísindi

Lagrangian Particle Tracking Data of a Straining Turbulent Flow Assessed Using Machine Learning and Parallel Computing

Lagrangian Particle Tracking Data of a Straining Turbulent Flow Assessed Using Machine Learning and Parallel Computing


Title: Lagrangian Particle Tracking Data of a Straining Turbulent Flow Assessed Using Machine Learning and Parallel Computing
Author: Hassanian, Reza   orcid.org/0000-0001-5706-314X
Riedel, Morris   orcid.org/0000-0003-1810-9330
Helgadottir, Asdis   orcid.org/0000-0002-6653-1600
Costa, Pedro   orcid.org/0000-0001-7010-1040
Bouhlali, Lahcen
Date: 2022-05-25
Language: English
Scope: 104
University/Institute: Háskóli Íslands
University of Iceland
School: Verkfræði- og náttúruvísindasvið (HÍ)
School of Engineering and Natural Sciences (UI)
Department: Iðnaðarverkfræði-, vélaverkfræði- og tölvunarfræðideild (HÍ)
Faculty of Industrial Eng., Mechanical Eng. and Computer Science (UI)
Series: Proceedings of 33rd Parallel Computational Fluid Dynamics (ParCFD);2022
Subject: Vélrænt nám; Samhliðavinnsla; Gervigreind; HPC; AI; Turbulent flow; Parallel computing; Lagrangian particle tracking
URI: https://hdl.handle.net/20.500.11815/3515

Show full item record

Citation:

R. Hassanian, M. Riedel, P. Costa, A. Helgadottir, L. Bouhlali, "Lagrangian Particle Tracking Data of a Straining Turbulent Flow Assessed Using Machine Learning and Parallel Computing," in Proceedings of 33rd Parallel Computational Fluid Dynamics (ParCFD) 2022. Alba, Italy, 2022.

Abstract:

This study aimed to employ artificial intelligence capability and computing scalability to predict the velocity field of the straining turbulence flow. Rotating impellers in a box have generated the turbulence, subsequently subjected to an axisymmetric straining motion, with mean nominal strain rates of 4s^-1. Tracer particles are seeded in the flow, and their dynamics are investigated using high-speed Lagrangian Particle Tracking at 10,000 frames per second. The particle displacement, time, and velocities can be extracted using this technique. Particle displacement and time are used as input observables, and the velocity is employed as a response output. The experiment extracted data have been divided into training and test data to validate the models. Support vector polynomial regression (SVR) and Linear regression were employed to see how extrapolation for the velocity field can be extracted. These models can be done with low computing time. On the other hand, to create a dynamic prediction, Gated Recurrent Unit (GRU) is applied with a high-performance computing application. The results show that GRU presents satisfactory forecasting for the turbulence velocity field and the computing scale performed on the JUWELS and DEEP-EST and reported. GPUs have a significant effect on computing time. This work presents the capability of the GRU model for time series data related to turbulence flow prediction.

Files in this item

This item appears in the following Collection(s)