The primary objective of this thesis is to develop a Wall-Modeled Large-Eddy Simulation (WMLES) model capable of efficiently simulating the behavior of particles immersed in a turbulent flow, balancing computational cost and accuracy. To achieve this, we utilized the open-source library CaNS (Canonical Navier-Stokes), designed for parallel computing and solving the Navier-Stokes equations with FFT-based solvers. This model divides the flow into an outer layer, where the flow is solved directly using LES, and an inner layer, where the flow near the wall is modeled to reduce computational load. The research begins with a comprehensive literature review of existing models and methodologies for simulating turbulent flows and particle behavior. Following this, the implementation of the CaNS library is detailed, highlighting its features such as hybrid MPI/OpenMP parallelization, FFTW and CuFFT for computing Fourier transforms, and GPU acceleration. The numerical methods used, including a second-order finite difference pressure correction scheme and a three-step Runge-Kutta scheme for time advancement, are discussed. A significant portion of the thesis is dedicated to defining the system and flow characteristics. The study analyzes a fully developed turbulent air flow with a friction Reynolds numbers of 1000 and grid resolutions. The particles, introduced randomly into the flow, are characterized by different Stokes numbers, which influence their behavior within the flow. The inner layer modeling is approached using a Langevin model, which incorporates stochastic elements to simulate the fluctuating velocities near the wall. However, challenges arise in blending this model with the LES results, particularly in the transition zone. To address these, various transition zones and blending techniques are explored, including the use of a custom sigmoid function to smoothly transition between the two models. The results indicate that while the LES model accurately captures the flow behavior in the outer layer, the Langevin model requires careful tuning to accurately represent particle behavior near the wall. A novel function is introduced to better model the wall-normal direction, incorporating average velocity trends and local velocity variations. The final model demonstrates improved accuracy in simulating particle behavior across different Stokes numbers, validated through comparisons with DNS data. This thesis contributes to the field by providing a robust WMLES model that effectively simulates turbulent flow and particle dynamics, offering a balance between computational efficiency and accuracy. The findings have implications for various engineering applications, where understanding particle behavior in turbulent flows is crucial.
L'obiettivo principale di questa tesi è sviluppare un modello di Wall-Modeled Large-Eddy Simulation (WMLES) capace di simulare efficacemente il comportamento delle particelle immerse in un flusso turbolento, bilanciando costi computazionali e accuratezza. Per raggiungere questo obiettivo, abbiamo utilizzato la libreria open-source CaNS (Canonical Navier-Stokes), progettata per il calcolo parallelo e per risolvere le equazioni di Navier-Stokes con solutori basati su FFT. Questo modello divide il flusso in uno strato esterno, dove il flusso è risolto direttamente utilizzando LES, e uno strato interno, dove il flusso vicino alla parete è modellato per ridurre il carico computazionale. La ricerca inizia con una revisione completa della letteratura sui modelli e le metodologie esistenti per la simulazione dei flussi turbolenti e del comportamento delle particelle. Successivamente, viene descritta l'implementazione della libreria CaNS, evidenziando le sue caratteristiche come la parallelizzazione ibrida MPI/OpenMP, FFTW e CuFFT per il calcolo delle trasformate di Fourier e l'accelerazione GPU. I metodi numerici utilizzati, inclusi uno schema di correzione della pressione a differenze finite di secondo ordine e uno schema di avanzamento temporale Runge-Kutta a tre passaggi, sono discussi. Una parte significativa della tesi è dedicata alla definizione del sistema e delle caratteristiche del flusso. Lo studio analizza un flusso d'aria turbolento completamente sviluppato con specifici numeri di Reynolds e risoluzioni della griglia. Le particelle, introdotte casualmente nel flusso, sono caratterizzate da diversi numeri di Stokes, che influenzano il loro comportamento all'interno del flusso. La modellazione dello strato interno è affrontata utilizzando un modello di Langevin, che incorpora elementi stocastici per simulare le velocità fluttuanti vicino alla parete. Tuttavia, sorgono sfide nell'integrare questo modello con i risultati LES, in particolare nella zona di transizione. Per affrontarle, vengono esplorate varie zone di transizione e tecniche di blending, incluso l'uso di una funzione sigmoide personalizzata per una transizione fluida tra i due modelli. I risultati indicano che, mentre il modello LES cattura accuratamente il comportamento del flusso nello strato esterno, il modello di Langevin richiede una messa a punto accurata per rappresentare correttamente il comportamento delle particelle vicino alla parete. Viene introdotta una funzione innovativa per modellare meglio la direzione normale alla parete, incorporando tendenze di velocità media e variazioni di velocità locale. Il modello finale dimostra un'accuratezza migliorata nella simulazione del comportamento delle particelle su diversi numeri di Stokes, validata attraverso confronti con i dati DNS. Questa tesi contribuisce al campo fornendo un modello WMLES robusto che simula efficacemente il flusso turbolento e la dinamica delle particelle, offrendo un equilibrio tra efficienza computazionale e accuratezza. I risultati hanno implicazioni per varie applicazioni ingegneristiche, dove comprendere il comportamento delle particelle nei flussi turbolenti è cruciale.
Development of a Lagrangian solver for simulating particle dispersion in turbulent flows
MAZZER, ERICA
2023/2024
Abstract
The primary objective of this thesis is to develop a Wall-Modeled Large-Eddy Simulation (WMLES) model capable of efficiently simulating the behavior of particles immersed in a turbulent flow, balancing computational cost and accuracy. To achieve this, we utilized the open-source library CaNS (Canonical Navier-Stokes), designed for parallel computing and solving the Navier-Stokes equations with FFT-based solvers. This model divides the flow into an outer layer, where the flow is solved directly using LES, and an inner layer, where the flow near the wall is modeled to reduce computational load. The research begins with a comprehensive literature review of existing models and methodologies for simulating turbulent flows and particle behavior. Following this, the implementation of the CaNS library is detailed, highlighting its features such as hybrid MPI/OpenMP parallelization, FFTW and CuFFT for computing Fourier transforms, and GPU acceleration. The numerical methods used, including a second-order finite difference pressure correction scheme and a three-step Runge-Kutta scheme for time advancement, are discussed. A significant portion of the thesis is dedicated to defining the system and flow characteristics. The study analyzes a fully developed turbulent air flow with a friction Reynolds numbers of 1000 and grid resolutions. The particles, introduced randomly into the flow, are characterized by different Stokes numbers, which influence their behavior within the flow. The inner layer modeling is approached using a Langevin model, which incorporates stochastic elements to simulate the fluctuating velocities near the wall. However, challenges arise in blending this model with the LES results, particularly in the transition zone. To address these, various transition zones and blending techniques are explored, including the use of a custom sigmoid function to smoothly transition between the two models. The results indicate that while the LES model accurately captures the flow behavior in the outer layer, the Langevin model requires careful tuning to accurately represent particle behavior near the wall. A novel function is introduced to better model the wall-normal direction, incorporating average velocity trends and local velocity variations. The final model demonstrates improved accuracy in simulating particle behavior across different Stokes numbers, validated through comparisons with DNS data. This thesis contributes to the field by providing a robust WMLES model that effectively simulates turbulent flow and particle dynamics, offering a balance between computational efficiency and accuracy. The findings have implications for various engineering applications, where understanding particle behavior in turbulent flows is crucial.File | Dimensione | Formato | |
---|---|---|---|
TESI_MAZZER_ERICA_2055010.pdf
embargo fino al 05/09/2025
Dimensione
12.95 MB
Formato
Adobe PDF
|
12.95 MB | Adobe PDF |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/69388