Parametric Dynamic Mode Decomposition with multi-linear interpolation for prediction of thermal fields of Al 2 O 3 -water nanofluid flows at unseen parameters

Abhijith M S, Sandra S · 2025

Read Parametric Dynamic Mode Decomposition with multi-linear interpolation for prediction of thermal fields of Al 2 O 3 -water nanofluid flows at unseen parameters by Abhijith M S, Sandra S (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-18571.txt.

Introduction

Abhijith M S and Sandra S built a fast stand-in for expensive fluid simulations, predicting how a heated nanofluid's temperature field changes with flow speed and particle loading. The fluid, water carrying aluminum oxide nanoparticles, runs through a flat channel whose bottom wall is heated evenly. Their model learns the field's main patterns by dynamic mode decomposition, then interpolates between trained conditions for settings never simulated.

The simpler version, interpolating over flow speed alone, missed the true solver's temperatures by at most three hundredths of a percent at a test condition. A two-parameter version, varying particle share too, stayed within about two tenths of a percent on temperature, with heat transfer rates off by about six percent at worst. Both run in a fraction of full solver time. The authors suggest such surrogates can scan cooling-system design options before committing to heavy computation.

Catalog record

CREATORAbhijith M S, Sandra S
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE39 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-18571

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Parametric Dynamic Mode Decomposition with multi-linear interpolation for prediction of thermal fields of Al 2 O 3 -water nanofluid flows at unseen parameters Abhijith M S, Sandra S Abstract The study proposes a data-driven model which combines the Dynamic Mode Decomposition with multi-linear interpolation to predict the thermal fields of nanofluid flows at unseen Reynolds numbers (Re) and particle volume concentrations ( ϵ \epsilon ). The flow, considered for the study, is laminar and incompressible. The study employs an in-house Fortran-based solver to predict the thermal fields of Al 2 O 3 -water nanofluid flow through a two-dimensional rectangular channel, with the bottom wall subjected to a uniform heat flux. The performance of two models operating in one- and two-dimensional parametric spaces are investigated. Initially, a DMD with linear interpolation (DMD-LI) based solver is used for prediction of temperature of the nanofluid at any Re > > 100. The DMD-LI based model, predicts temperature fields with a maximum percentage difference of just 0.0273%, in comparison with the CFD-based solver at Re =960, and ϵ \epsilon = 1.0%. The corresponding difference in the average Nusselt numbers is only 0.39%. Following that a DMD with bi-linear interpolation (DMD-BLI) based solver is used for prediction of temperature of the nanofluid at any Re > > 100 and ϵ \epsilon > > 0.5%. The performance of two different ways of stacking the data are also examined. When compared to the CFD-based model, the DMD-BLI-based model predicts the temperature fields with a maximum percentage difference of 0.21 %, at Re = 800 and ϵ \epsilon = 1.35%. And the corresponding percentage difference in the average Nusselt number prediction is only 6.08%. All the results are reported in detail. Along side the important conclusions, the future scope of the study is also listed. A Preprint Keywords nanofluid \cdot machine learning \cdot DMD \cdot linear interpolation \cdot parametric dynamical mode decomposition \cdot partial differential equations 1 Introduction Humanity’s enduring curiosity about comprehending and exploring dynamical systems, stems from the desire to uncover the fundamental laws that govern the evolution of the universe. The research in the field of computational fluid dynamics often not only requires predicting the systems behaviour for future time instants, but also understanding their parametric dependencies to forecast behaviour for unseen parameters. The introduction of parameters expands the dimension of the problem space, necessitating high computational efficiency, to solve these equations across varying parametric values Cao et al. (2024) . We can mitigate this curse of dimensionality by using the technique of reduced order modelling (ROM) Huhn et al. (2023) , which employs mathematical approaches to derive smaller models without compromising solution accuracy. Several techniques fall under ROM, including reduced basis methods (RBM) Quarteroni et al. (2015) ; Chakir et al. (2019) , proper orthogonal decomposition (POD) Zokagoa and Soulaïmani (2012) ; Xiao et al. (2012) ; Xiao et al. (2015) ; Berkooz et al. (1993) , and proper generalised decomposition Chinesta et al. (2011) . Dynamic mode decomposition (DMD) Schmid (2010) , on the other hand, is a non intrusive method Huhn et al. (2023) that focuses on system dynamics. It processes high dimensional sequential data, captures and simplifies complex behaviours Schmid (2022) , including those in non-linear systems Rozza et al. (2018) ,and identifies and represents the system as a linear combination of dominant structures within the data. Originally developed in the fluid dynamics community Schmid (2010) , DMD has expanded to offer a wide range of applicability across multiple domains, including robotics Berger et al. (2015) , financial modelling Mann and Kutz (2016) ; Unnithan et al. (2019) , blood flow physics Habibi et al. (2020) , image segmentation Sikha et al. (2018) , image classification Rahul-Vigneswaran et al. (2019) , DNA sequence classification Aadhithya et al. (2023) and more. As already mentioned, here the intention is to predict the flow/ thermal behaviour at unseen parameters using DMD. There have been several attempts to mitigate the challenges associated with parameterized non linear dynamical systems, particularly through non-intrusive analysis Yu et al. (2019) that offer potential solutions. The literature on non-intrusive analysis of parameterized non-linear dynamical systems within fluid dynamics, focuses on three core aspects: surrogate modelling Hoang et al. (2022) ; Li and Yang (2023) ; Nikolopoulos et al. (2021) , sampling strategy and reduced order modelling (ROM) Hess et al. (2023) ; Duan and Hesthaven (2024) ; Sun et al. (2023) . An approach involving DMD and regression for predicting the snapshots of data at future time and unseen parameters is reported in Andreuzzi et al. (2023) . They considered the data to involve high fidelity snapshots taken at uniform time intervals and across various parametric configurations. These snapshots are then undergone dimensionality reduction using POD and subsequently reorganized to serve as input for the DMD algorithm. DMD is applied to predict the reduced snapshots for future time steps. Finally, the parametric manifold is reconstructed using regression technique that approximates the relationship between the parameters and the reduced snapshots. They reported that for problems involving flow prediction employing Navier-Stokes equation, the dimensionality reduction leads to inadequate representation of the true state. This study introduces a novel approach that combines Dynamic Mode Decomposition (DMD) with linear interpolation to predict the thermal fields of nanofluid flows. The flow under consideration is laminar and incompressible. To simplify the complexity of temporal data, only quasi-steady-state data for different parameters is used. Inorder, to avoid the limitations of reduced-order representations, such as those encountered in Proper Orthogonal Decomposition (POD) Andreuzzi et al. (2023) , this study does not consider any dimensionality reduction. The study employs an in-house Fortran-based solver to predict the thermal fields of Al 2 O 3 -water nanofluid flow through a two-dimensional rectangular channel (referred to as the ’CFD-based’ model), with the bottom wall subjected to a uniform heat flux. The simulation data is then utilized to develop a DMD with linear interpolation model capable of predicting thermal fields for nanofluids at previously unseen parameters. The intention here is to propose a DMD with multi-linear interpolation based models capable of predicting the thermal fields (and also the heat transfer performance) in a two dimensional channel with Al 2 O 3 -water nanofluid flow at different values of reynolds numbers and nanoparticle volume concentration. Initially the governing equations used in the in-house solver are discussed in Section - 2 . Following that the core idea behind the DMD - multi-linear interpolation based model is discussed in Section - 3 . Prior to the results and discussions in the Section - 6 , the grid independence and validation of the in-house solver are also presented (see sections 4 and 5 ). 2 Governing Equations The in-house Fortran based solver considers a laminar, incompressible, unsteady, two-dimensional homogeneous model for the nanofluid flows. The governing equations for mass, momentum and energy can be given as shown below: 2.1 Continuity Equation \nabla.{\bf v}=0 (1) 2.2 Momentum Equation \rho_{nf}\left(\frac{\partial{\bf v}}{\partial t}+\left({\bf v}.\nabla\right){\bf v}\right)=-\nabla p+\mu_{nf}\nabla.\left[\left(\nabla{\bf v}+\nabla{\bf v}^{T}\right)\right] (2) 2.3 Energy Equation \rho_{nf}c_{p,nf}\left(\frac{\partial\theta}{\partial t}+\left({\bf v}.\nabla\right){\theta}\right)=k_{nf}\nabla.\left(\nabla\theta\right) (3) where, v and \theta represent the velocity (in m/s) vector and temperature (in K) respectively. The thermo-physical properties of the nanofluid used in the above governing equations; such as, the density ( \rho_{nf} ,

Read the complete text