Application of minimum spanning tree for data-driven modeling and analysis of turbulent heat transfer
최소신장트리를 활용한 난류 유동 분석 및 데이터 기반 모델링
- 주제어 (키워드) ANN , Minimum spanning tree , Parameter selection , Physical modeling , Turbulent duct flow , Turbulent Prandtl number; ANN , 인자선별 , 물리모델링 , 난류 덕트 , 난류 프란들 계수
- 발행기관 서강대학교 일반대학원
- 지도교수 강성원
- 발행년도 2024
- 학위수여년월 2024. 2
- 학위명 석사
- 학과 및 전공 일반대학원 기계공학과
- 실제URI http://www.dcollection.net/handler/sogang/000000077084
- UCI I804:11029-000000077084
- 본문언어 영어
- 저작권 서강대학교 논문은 저작권 보호를 받습니다.
초록
Dealing with complex physical phenomena such as turbulent heat transfer requires consider- ation of numerous physical factors such as geometry, flow conditions and locally varying flow properties. Despite the analytical difficulties, turbulent heat transfer can be observed in many in- dustrial settings such as heat exchanges, gas turbines, etc. For such high-dimensionality problem, data-driven modeling has proven to be useful due to the flexibility of artificial neural networks (ANN); however, to train such ANN model with high generalization and low overfitting the di- mensionality of the training space must be reduced. In this paper, minimum spanning tree (MST) algorithm is utilized to identify important parameters for modeling of turbulent Prandtl number and is compared with a few existing parameter selection methods. The ANN results of MST show the highest accuracy, indicating the algorithm is more effective in identifying crucial parameters and tackling the issue of parameter redundancy.
more목차
1 Introduction 1
2 Minimum Spanning Tree algorithm 4
2.1 Introduction to MST 4
2.2 Modeling of manufactured problem and parameter set 6
3 Analysis of turbulence phenomena using MST 8
3.1 Description of physical problem and database of channel flow . 8
3.2 MST results of turbulent budget terms at different regions of the channel 10
3.3 Description of physical problem and database of duct flow 11
3.4 MST results of turbulent budget terms in duct flow . 11
4 Data-driven modeling of turbulent flow applications using MST 15
4.1 Data-driven modeling of Prt in turbulent duct flow using MST . 15
4.1.1 Description of the physical problem and database . 15
4.1.2 Optimization of ANN hyper-parameters and preprocessing of the database 16
4.1.3 Past empirical models of Prt 18
4.1.4 Selection of input parameters and reduced modeling of Prt using MST . 19
4.1.5 Comparison of MST results with existing parameter selection methods . 24
4.2 Data-driven modeling of directional Prt in turbulent duct flow using MST . 25
4.2.1 Motivation of modeling directional Prt 25
4.2.2 Reduced modeling results of directional Prt at duct quadrant 28
References 38
