Feasibility of Deep Neural Network Surrogate Models in Fluid Dynamics

Niels Christian Bender, Henrik C. Pedersen, Torben Ole Andersen

Research output: Contribution to journalJournal articleResearchpeer-review

Abstract

This paper studies reduced-order-models for the fluid flow problem of a digital valve, and whether it may efficiently be formulated by a deep Artificial Neural Network (ANN) to model e.g. the valve flow, flow-induced force, stiction phenomena and steep local pressure gradients that arise before plunger impact, which may otherwise require CFD to be accurately modeled. Several methodologies are investigated to evaluate both the required computation time and the accuracy. The accuracy is benchmarked against CFD solutions of flows and forces. As basis for comparison an analytical model is proposed where some fitting parameters are allowed, and the equation is tested outside its fitting range. A similar model is built as a deep ANN which is trained with data from the analytical model to investigate the amount of data required for an ANN and its fitting capabilities. The results show that in higher dimensions the required training data can be maintained low if data is structured by a Latin Hypercube, otherwise the amount becomes infeasible. This makes an ANN surrogate feasible when compared to a look-up table, and may be expanded to higher dimension where dynamical effects are included. However, the required data and computational cost for this is too extensive for the valve design considered as basis for the analysis. Instead, for this specific problem, the derived analytical model is sufficient to describe the valve dynamics and reduces the computation time significantly.
Original languageEnglish
JournalModeling, Identification and Control
Volume40
Issue number2
Pages (from-to)71-87
Number of pages16
ISSN0332-7353
DOIs
Publication statusPublished - Apr 2019

Fingerprint

Surrogate Model
Fluid Dynamics
Fluid dynamics
Network Model
Artificial Neural Network
Neural Networks
Neural networks
Analytical models
Analytical Model
Computational fluid dynamics
Higher Dimensions
Stiction
Latin Hypercube
Pressure gradient
Reduced Order Model
Look-up Table
Flow of fluids
Pressure Gradient
Fluid Flow
Computational Cost

Keywords

  • Artificial Neural Networks
  • CFD
  • Digital Valves
  • Flow-induced Forces
  • Reduced Order Models
  • Lumped Parameter Models

Cite this

@article{b01731812ea44839a00ab16be85a006c,
title = "Feasibility of Deep Neural Network Surrogate Models in Fluid Dynamics",
abstract = "This paper studies reduced-order-models for the fluid flow problem of a digital valve, and whether it may efficiently be formulated by a deep Artificial Neural Network (ANN) to model e.g. the valve flow, flow-induced force, stiction phenomena and steep local pressure gradients that arise before plunger impact, which may otherwise require CFD to be accurately modeled. Several methodologies are investigated to evaluate both the required computation time and the accuracy. The accuracy is benchmarked against CFD solutions of flows and forces. As basis for comparison an analytical model is proposed where some fitting parameters are allowed, and the equation is tested outside its fitting range. A similar model is built as a deep ANN which is trained with data from the analytical model to investigate the amount of data required for an ANN and its fitting capabilities. The results show that in higher dimensions the required training data can be maintained low if data is structured by a Latin Hypercube, otherwise the amount becomes infeasible. This makes an ANN surrogate feasible when compared to a look-up table, and may be expanded to higher dimension where dynamical effects are included. However, the required data and computational cost for this is too extensive for the valve design considered as basis for the analysis. Instead, for this specific problem, the derived analytical model is sufficient to describe the valve dynamics and reduces the computation time significantly.",
keywords = "Artificial Neural Networks, CFD, Digital Valves, Flow-induced Forces, Reduced Order Models, Lumped Parameter Models",
author = "Bender, {Niels Christian} and Pedersen, {Henrik C.} and Andersen, {Torben Ole}",
year = "2019",
month = "4",
doi = "10.4173/mic.2019.2.1",
language = "English",
volume = "40",
pages = "71--87",
journal = "Modeling, Identification and Control (Online)",
issn = "0332-7353",
publisher = "Norwegian Society of Automatic Control",
number = "2",

}

Feasibility of Deep Neural Network Surrogate Models in Fluid Dynamics. / Bender, Niels Christian; Pedersen, Henrik C.; Andersen, Torben Ole.

In: Modeling, Identification and Control, Vol. 40, No. 2, 04.2019, p. 71-87.

Research output: Contribution to journalJournal articleResearchpeer-review

TY - JOUR

T1 - Feasibility of Deep Neural Network Surrogate Models in Fluid Dynamics

AU - Bender, Niels Christian

AU - Pedersen, Henrik C.

AU - Andersen, Torben Ole

PY - 2019/4

Y1 - 2019/4

N2 - This paper studies reduced-order-models for the fluid flow problem of a digital valve, and whether it may efficiently be formulated by a deep Artificial Neural Network (ANN) to model e.g. the valve flow, flow-induced force, stiction phenomena and steep local pressure gradients that arise before plunger impact, which may otherwise require CFD to be accurately modeled. Several methodologies are investigated to evaluate both the required computation time and the accuracy. The accuracy is benchmarked against CFD solutions of flows and forces. As basis for comparison an analytical model is proposed where some fitting parameters are allowed, and the equation is tested outside its fitting range. A similar model is built as a deep ANN which is trained with data from the analytical model to investigate the amount of data required for an ANN and its fitting capabilities. The results show that in higher dimensions the required training data can be maintained low if data is structured by a Latin Hypercube, otherwise the amount becomes infeasible. This makes an ANN surrogate feasible when compared to a look-up table, and may be expanded to higher dimension where dynamical effects are included. However, the required data and computational cost for this is too extensive for the valve design considered as basis for the analysis. Instead, for this specific problem, the derived analytical model is sufficient to describe the valve dynamics and reduces the computation time significantly.

AB - This paper studies reduced-order-models for the fluid flow problem of a digital valve, and whether it may efficiently be formulated by a deep Artificial Neural Network (ANN) to model e.g. the valve flow, flow-induced force, stiction phenomena and steep local pressure gradients that arise before plunger impact, which may otherwise require CFD to be accurately modeled. Several methodologies are investigated to evaluate both the required computation time and the accuracy. The accuracy is benchmarked against CFD solutions of flows and forces. As basis for comparison an analytical model is proposed where some fitting parameters are allowed, and the equation is tested outside its fitting range. A similar model is built as a deep ANN which is trained with data from the analytical model to investigate the amount of data required for an ANN and its fitting capabilities. The results show that in higher dimensions the required training data can be maintained low if data is structured by a Latin Hypercube, otherwise the amount becomes infeasible. This makes an ANN surrogate feasible when compared to a look-up table, and may be expanded to higher dimension where dynamical effects are included. However, the required data and computational cost for this is too extensive for the valve design considered as basis for the analysis. Instead, for this specific problem, the derived analytical model is sufficient to describe the valve dynamics and reduces the computation time significantly.

KW - Artificial Neural Networks

KW - CFD

KW - Digital Valves

KW - Flow-induced Forces

KW - Reduced Order Models

KW - Lumped Parameter Models

U2 - 10.4173/mic.2019.2.1

DO - 10.4173/mic.2019.2.1

M3 - Journal article

VL - 40

SP - 71

EP - 87

JO - Modeling, Identification and Control (Online)

JF - Modeling, Identification and Control (Online)

SN - 0332-7353

IS - 2

ER -