The analysis of the time-dependent flow field in the complex anatomy of the human nasal cavities is becoming more and more important owing to its diagnostic value, making it a crucial tool for patient-specific surgery planning of the Ear, Nose and Throat medical doctors. Fast as well as accurate numerical tools are required for use in a clinical setting. In this work, we present for the first time a finite-differences computer code that solves the incompressible Navier--Stokes equations by taking as input the STL file generated by a patient-specific CT scan. The code is based on an innovative (existing) immersed-boundary method, is second-order accurate in space, and is written in the CPL language. Quality and performance are assessed on a laminar steady flow and a turbulent flow, computed for the anatomy of an healthy patient reconstructed from a CT scan. A comparison with the identical case computed with OpenFOAM is carried out. Results of the two codes are coincident, in terms of global quantities (flow rate, pressure drop, nasal resistance), numerical accuracy (which turns out to be second-order in space as measured by the convergence of the flow rate at increasing grid size) and statistics of the flow fields. The fields are compared in terms of both mean quantities and second-order statistics of velocity and pressure. A detailed analysis of the code performance has been carried out. Presently, the code is still available in a serial version only, and there is significant further optimization to be done before progressing to the parallel version, as revealed by a detailed profiling. However, already at this stage the serial performance provides a 50 times speed-up in comparison to the same problem run with OpenFOAM, with an absolute speed of 1.6 E-7 seconds per time step and per point on an Intel Core i5 desktop CPU. The memory footprint is also reduced by almost 5 times, allowing a clinically significant DNS of a case to be run to create well-converged statistics overnight on a single core of a desktop computer.
L'analisi di un flusso instazionario all'interno della complessa anatomia della cavità nasale umana ha un valore diagnostico sempre più importante ed è uno strumento cruciale per l'otorinolaringoiatra per una pianificazione chirurgica specifica per il paziente. Per l'uso in ambito clinico sono necessari strumenti numerici rapidi e precisi. In questo lavoro, presentiamo per la prima volta un codice numerico alle differenze finite che risolve le equazioni di Navier-Stokes incomprimibili usando come input un file STL generato a partire da una TAC di un paziente. Il codice si basa su un innovativo (ed esistente) metodo a contorno immerso, accurato al secondo ordine nello spazio ed è scritto in linguaggio CPL. La qualità e le prestazioni del codice sono valutate su un flusso laminare stazionario e un flusso turbolento, entrambi calcolati sull'anatomia di un paziente sano ricostruito da una scansione TC. Viene effettuato un confronto sullo stesso caso calcolato con OpenFOAM. I risultati dei due codici sono coincidenti in termini di grandezze globali (portata, salto di pressione, resistenza nasale), accuratezza numerica (che risulta essere di secondo ordine nello spazio, come misurato dalla convergenza della portata al variare della dimensione della griglia) e statistiche dei campi di flusso. I campi di velocità e pressione vengono confrontati sia in termini di quantità medie che di statistiche di secondo ordine. È stata effettuata un'analisi dettagliata delle prestazioni del codice. Al momento, il codice è ancora disponibile solo in versione seriale e, come rivelato da una dettagliata profilazione, è possibile compiere un'ulteriore e significativa ottimizzazione prima di passare alla versione parallela. Tuttavia, già in questa fase, le prestazioni seriali forniscono incremento di velocità di 50 volte rispetto allo stesso problema eseguito con OpenFOAM, con una velocità assoluta di 1.6 E-7 secondi per passo temporale e per punto di griglia su una CPU desktop Intel Core i5. Anche la memoria occupata è ridotta di quasi 5 volte, consentendo di effettuare una DNS clinicamente significativa e ottenere in una notte statistiche a convergenza utilizzando un singolo core di un computer desktop.
A direct numerical simulation code for the flow in the human nose
Vecchietti, Lorenzo
2020/2021
Abstract
The analysis of the time-dependent flow field in the complex anatomy of the human nasal cavities is becoming more and more important owing to its diagnostic value, making it a crucial tool for patient-specific surgery planning of the Ear, Nose and Throat medical doctors. Fast as well as accurate numerical tools are required for use in a clinical setting. In this work, we present for the first time a finite-differences computer code that solves the incompressible Navier--Stokes equations by taking as input the STL file generated by a patient-specific CT scan. The code is based on an innovative (existing) immersed-boundary method, is second-order accurate in space, and is written in the CPL language. Quality and performance are assessed on a laminar steady flow and a turbulent flow, computed for the anatomy of an healthy patient reconstructed from a CT scan. A comparison with the identical case computed with OpenFOAM is carried out. Results of the two codes are coincident, in terms of global quantities (flow rate, pressure drop, nasal resistance), numerical accuracy (which turns out to be second-order in space as measured by the convergence of the flow rate at increasing grid size) and statistics of the flow fields. The fields are compared in terms of both mean quantities and second-order statistics of velocity and pressure. A detailed analysis of the code performance has been carried out. Presently, the code is still available in a serial version only, and there is significant further optimization to be done before progressing to the parallel version, as revealed by a detailed profiling. However, already at this stage the serial performance provides a 50 times speed-up in comparison to the same problem run with OpenFOAM, with an absolute speed of 1.6 E-7 seconds per time step and per point on an Intel Core i5 desktop CPU. The memory footprint is also reduced by almost 5 times, allowing a clinically significant DNS of a case to be run to create well-converged statistics overnight on a single core of a desktop computer.| File | Dimensione | Formato | |
|---|---|---|---|
|
TesiVecchietti.pdf
non accessibile
Dimensione
24.97 MB
Formato
Adobe PDF
|
24.97 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/183423