A two-level machine learning approach for predicting thermal striping in T-junctions with upstream elbow
Name
A two-level machine learning approach for predicting thermal striping in T-junctions with upstream elbow.pdf
Description
Published version
Size
3.99 MB
Format
Adobe PDF
Checksum (MD5)
5943db5bbb69a8b3d4e33ae73eb90744
Author(s)
Wang, Yu-Jou
Baglietto, Emilio
Shirvan, Koroush
Date Issued
June 2, 2024
Journal
Numerical Heat Transfer, Part B: Fundamentals
Publisher
Taylor & Francis
Citation
Wang, Y. J., Baglietto, E., & Shirvan, K. (2024). A two-level machine learning approach for predicting thermal striping in T-junctions with upstream elbow. Numerical Heat Transfer, Part B: Fundamentals, 85(6), 662–682.
Version
Final published version
Abstract
Thermal striping is a phenomenon characterized by oscillatory mixing of non-isothermal streams, which is commonly seen in industrial processes such as nuclear coolant piping, petrochemical plants and liquefied natural gas transportation. The oscillatory mixing of hot and cold fluid can produce thermal field fluctuations and pose a potential risk of high-cycle thermal fatigue failures. Predicting and evaluating spatiotemporal fluctuations in thermal striping often requires high resolution and massive computational power. Although there have been extensive studies using machine learning algorithms on surrogate modeling, research focused on spatiotemporal fluctuation predictions is very limited. Due to the high dimensionality, it often requires complex algorithms with a large amount of high-fidelity training data, which limits the adoption of such methods for industrial applications. In this research, a two-level machine learning framework based on turbulence coherent structures is proposed and its application to a practical problem is demonstrated. The two-level design leverages vortex identification and local bias correction techniques, efficiently reducing the number of full-order simulations required for training. In the first level, well-organized coherent structures are extracted by performing Proper Orthogonal Decomposition on local parameters and then a tree-based machine-learning model is used to down-select the reference structures for the field reconstruction. In the second level, a parameterized convolution neural network is trained to predict the bias introduced by reference structures approximation. The demonstration of the methodology shows that the method can accurately capture the fluctuation frequencies and amplitudes of the spatiotemporal fields in a highly variational setting. Based on the vortex identification method, the methodology is expected to be applicable to general phenomenon driven by large coherent structures.
MIT Department
Massachusetts Institute of Technology. Department of Nuclear Science and Engineering
Terms of Use
Creative Commons Attribution-NonCommercial-NoDerivatives
Persistent DSpace Link
DOI of Published Version
https://doi.org/10.1080/10407790.2023.2253362
