One of the most interesting problems concerning combustion phenomena for the latest generation of engines is the introduction of high pressure gas into a much lower pressure environment. This situation, intentionally designed inside civilian ground combustors, does not differ too much from physical phenomena equally known as for example the outflow of high pressure gas from exhaust nozzles of supersonic aircraft in surrounding air, or the leakage of content present in pressurized tanks both in the space as well as in the domestic environment due to possible damage or puncture. In the most modern combustion chambers this phenomenon is mainly sought because of the need to make an important reduction to the amount of pollutants engendered by chemical reactions, which can be strongly limited by modifying, for example, the chemical reagents, requiring in this way often a high injection pressure to ensure sufficient flow penetration and correct mixing of the substances. The important experimental studies produced starting from Ashkenas and Sherman, and whose analysis was subsequently reproduced in an even more detailed and in-depth manner by Velikorodny and Kudriakov and others, allowed to develop experimental equations able to characterize the undergoing physical process. These researches have highlighted a number of considerable problems from the fluid dynamics and thermodynamic point of view, since physics indicates how by changing the NPR (Nozzle Pressure Rateo), ie the relationship between nozzle discharge pressure and the one of surrounding environment, the phenomenology changes considerably. Indeed, it will be seen how, since the flow is injected in a strongly underexpanded form, it spontaneously undergoes an expansion outside the nozzle, which subsequently entails the generation of a series of normal and oblique interacting shock waves, which lead to complex formations that complicate a framework in which the strong presence of turbulent phenomena, mixing and chemical reactions make extremely difficult to obtain an accurate simulation. Ashkenas and Sherman, as well as some other authors, show how the change of NPR value identifies the progressive transition from a physical phenomenon characterized by the appearance of a series of expansions and progressive compressions, to a very different one characterized by a single expansion followed by a brutal compression due to the presence of a normal shock wave commonly defined as mach disc. Nowadays, though, the use of computational softwares guarantees considerable advantages from a cost, time and safety point of view compared to experimental counterparts, while managing to produce solutions in many cases completely superimposable to the results obtained in the laboratory. However, within the OpenFOAM software, problems of strongly underexpanded flows are commonly treated through the use of an explicit solver (rhoCentralFoam), which is considerably slow also due to the own subset of discretized equations. Another computational complication is evident, namely the strong grading of the discretized mesh commonly used in the geometries of interest. Very often the motors have small intake nozzles compared to the much larger mixing and combustion chambers, normally involving the development of computationally inefficient meshes within the geometry or the application of different solvers to be used on different portions of the mesh. The objective is therefore to develop the necessary modifications to allow a study of the main parameters of the considered problem through simulation by means of an implicit solver (rhoPimpleFoam in its transonic version), thus allowing the increase in the fluid dynamic timestep size, and obtaining therefore a considerable decrease in computational cost and time, bypassing the fluid dynamics problems related to the discontinuities introduced by the shock wave systems. This is possible thanks to the creation of new adaptive boundary conditions able to compensate for these issues, thus ensuring stability and convergence of the correct solution. Moreover, whenever chemical reactions are present, the number of equations that must be solved within every discretize cell increases. Chemical ODEs (Ordinary Differential Equations) must be accounted and solved for a massive number of cells, due to the small dimension of the latter. Since within OpenFOAM such process takes place cyclically, the latter becomes progressively inefficient as the number of cells increases. In order to increment the computational efficiency this calculation should be performed in a parallelized manner. Thus, following the example of other authors who have gained advantage by producing a development of operations by parallel computing on NVIDIA graphics cards, although for very different problems, the speed up of the active chemistry required for the introduction of a different newly created method for ODEs treatment, that exploited GPU (Graphic Processing Unit) computational acceleration by means of a GPGPU (General-purpose computing on Graphics Processing Unit) approach. All this is possible through the use of NVIDIA hardware architecture CUDA (Compute Unified Device Architecture), that through its development environment allows to write functions able to perform parallel calculation on the NVIDIA GPUs. Please note that to develop such a resolution not only a total porting but a revision and transformation of the main steps for the treatment of ODEs has been performed, and a link between OpenFOAM and extern "C" CUDA functions were produced as the language of the latter is not by default compatible with the first.
Uno dei problemi più interessanti riguardanti i fenomeni di combustione all'interno dei motori di ultima generazione è da riferirsi all'introduzione di gas ad altra pressione all'interno di un ambiente a pressione molto inferiore. Questa situazione, volutamente ricercata all'interno dei comuni combustori, non differisce molto da fenomeni fisici altrettanto noti come ad esempio lo sfiato in atmosfera di gas ad alta pressione da ugelli di scarico di aeromobili supersonici, o la perdita di contenuto presente in serbatoi pressurizzati sia in ambiente spaziale che quello terrestre a causa di possibile danneggiamento o foratura. All'interno dei comuni combustori moderni questo fenomeno è voluto principalmente per necessità di ottenere importante riduzione della quantità di inquinanti generati dalle reazioni chimiche, che possono essere limitati fortemente modificando, ad esempio, i reagenti chimici, richiedendo al contempo un'alta pressione di immissione per assicurare sufficiente penetrazione del flusso e corretto rimescolamento delle sostanze. Gli importanti studi sperimentali prodotti a partire da Ashkenas e Sherman, e le cui analisi furono successivamente riprodotte in modo ancora più dettagliato e approfondito da Velikorodny, Kudriakov e altri, hanno permesso di sviluppare equazioni sperimentali in grado di caratterizzare il processo fisico considerato. Queste ricerche hanno anche evidenziato una serie di problemi considerevoli dal punto di vista della fluidodinamica e della termodinamica, poiché la fisica indica come cambiando l'NPR (Nozzle Pressure Rateo), cioè la relazione tra la pressione di scarico dell'ugello e quella dell'ambiente circostante, la fenomenologia cambi considerevolmente. Infatti si vedrà come, dato che il flusso viene iniettato in forma fortemente sottoespansa, quest'ultimo debba subire spontaneamente un'espansione all'esterno dell'ugello. Ciò comporta la generazione di una serie di onde d'urto interagenti, normali e oblique, che portano a complesse formazioni, le quali contribuiscono a complicare una struttura in cui la forte presenza di fenomeni turbolenti, di miscelazione e di continuo sviluppo di reazioni chimiche rendono estremamente difficile ottenere una simulazione accurata. Ashkenas e Sherman, così come alcuni altri autori, mostrano come il cambiamento del valore di NPR identifichi la progressiva transizione da un fenomeno fisico caratterizzato dall'apparizione di una serie di espansioni e compressioni progressive, ad uno molto diverso caratterizzato da un'unica espansione seguita da una brutale compressione dovuta alla presenza di un'onda d'urto normale comunemente definita come "mach disc". Oggigiorno, comunque, l'uso di software computazionali garantisce notevoli vantaggi dal punto di vista del costo, del tempo e della sicurezza rispetto alle controparti sperimentali, riuscendo a produrre soluzioni in molti casi completamente sovrapponibili ai risultati ottenuti in laboratorio. Tuttavia, all'interno del software OpenFOAM, i problemi riguardanti flussi fortemente sottoespansi vengono comunemente trattati mediante l'uso di un risolutore esplicito (rhoCentralFoam), che è considerevolmente lento anche a causa del proprio sottoinsieme di equazioni discretizzate. Un'altra complicazione computazionale è evidente, vale a dire il repentino cambiamento di dimensione delle celle nelle mesh discretizzate che tipicamente vengono usate nelle geometrie di interesse. Molto spesso i motori hanno ugelli di immissione ben più piccoli rispetto alle camere di miscelazione e di combustione. Situazione, quest'ultima, che normalmente comporta lo sviluppo di mesh computazionalmente inefficienti all'interno della geometria o l'applicazione di diversi solutori da utilizzare su diverse porzioni della stessa. L'obiettivo è quindi quello di sviluppare le modifiche necessarie per consentire uno studio dei principali parametri del problema attraverso una simulazione mediante risolutore implicito (rhoPimpleFoam nella sua versione transonica), consentendo così l'aumento della dimensione della discretizzazione temporale fluidodinamica, ottenendo quindi una considerevole diminuzione dei costi e dei tempi computazionali, aggirando i problemi fluidodinamici relativi alle discontinuità introdotte dai sistemi di onde d'urto. Ciò è possibile grazie alla creazione di nuove condizioni al contorno adattative in grado di compensare questi problemi, garantendo così stabilità e convergenza della soluzione corretta. Inoltre, ogni volta che sono presenti reazioni chimiche, aumenta il numero di equazioni che devono essere risolte all'interno di ogni cella discretizzata. Le ODE (equazioni differenziali ordinarie) chimiche devono essere contabilizzate e risolte per un numero elevato di celle, a causa della piccola dimensione di quest'ultime. Poiché all'interno di OpenFOAM tale processo avviene ciclicamente, quest'ultimo diventa progressivamente inefficiente all'aumentare del numero di celle. Per aumentare l'efficienza computazionale tale calcolo dovrebbe essere eseguito in modo parallelo. Quindi, seguendo l'esempio di altri autori che, sebbene per problemi molto diversi, hanno tratto vantaggio dallo sviluppo di operazioni mediante calcolo parallelo su schede grafiche NVIDIA, l'accelerazione della risoluzione della chimica attiva ha richiesto lo sviluppo di un nuovo metodo sofisticato per il trattamento delle ODE, il quale sfruttando l'accelerazione computazionale offerta dalla GPU (Graphic Processing Unit) permetta di costruire un approccio GPGPU (General-purpose computing on Graphics Processing Unit). Tutto ciò è possibile grazie all'architettura hardware NVIDIA CUDA (Compute Unified Device Architecture), che attraverso il suo ambiente di sviluppo consente di scrivere funzioni in grado di eseguire calcoli paralleli sulle GPU NVIDIA. Si noti che per sviluppare tale risoluzione non solo si è reso necessario eseguire una conversione totale delle funzioni, ma si è dovuto effettuare una revisione e in alcuni casi persino una completa riscrtittura dei passaggi principali per il trattamento delle ODE. Inoltre, un collegamento tra OpenFOAM e le funzioni CUDA è stato prodotto mediante creazione di una funzione "extern C" poichè i due linguaggi non sono compatibili di default.
Fast algorithms for highly underexpanded reactive spray simulations
GHIOLDI, FEDERICO
2017/2018
Abstract
One of the most interesting problems concerning combustion phenomena for the latest generation of engines is the introduction of high pressure gas into a much lower pressure environment. This situation, intentionally designed inside civilian ground combustors, does not differ too much from physical phenomena equally known as for example the outflow of high pressure gas from exhaust nozzles of supersonic aircraft in surrounding air, or the leakage of content present in pressurized tanks both in the space as well as in the domestic environment due to possible damage or puncture. In the most modern combustion chambers this phenomenon is mainly sought because of the need to make an important reduction to the amount of pollutants engendered by chemical reactions, which can be strongly limited by modifying, for example, the chemical reagents, requiring in this way often a high injection pressure to ensure sufficient flow penetration and correct mixing of the substances. The important experimental studies produced starting from Ashkenas and Sherman, and whose analysis was subsequently reproduced in an even more detailed and in-depth manner by Velikorodny and Kudriakov and others, allowed to develop experimental equations able to characterize the undergoing physical process. These researches have highlighted a number of considerable problems from the fluid dynamics and thermodynamic point of view, since physics indicates how by changing the NPR (Nozzle Pressure Rateo), ie the relationship between nozzle discharge pressure and the one of surrounding environment, the phenomenology changes considerably. Indeed, it will be seen how, since the flow is injected in a strongly underexpanded form, it spontaneously undergoes an expansion outside the nozzle, which subsequently entails the generation of a series of normal and oblique interacting shock waves, which lead to complex formations that complicate a framework in which the strong presence of turbulent phenomena, mixing and chemical reactions make extremely difficult to obtain an accurate simulation. Ashkenas and Sherman, as well as some other authors, show how the change of NPR value identifies the progressive transition from a physical phenomenon characterized by the appearance of a series of expansions and progressive compressions, to a very different one characterized by a single expansion followed by a brutal compression due to the presence of a normal shock wave commonly defined as mach disc. Nowadays, though, the use of computational softwares guarantees considerable advantages from a cost, time and safety point of view compared to experimental counterparts, while managing to produce solutions in many cases completely superimposable to the results obtained in the laboratory. However, within the OpenFOAM software, problems of strongly underexpanded flows are commonly treated through the use of an explicit solver (rhoCentralFoam), which is considerably slow also due to the own subset of discretized equations. Another computational complication is evident, namely the strong grading of the discretized mesh commonly used in the geometries of interest. Very often the motors have small intake nozzles compared to the much larger mixing and combustion chambers, normally involving the development of computationally inefficient meshes within the geometry or the application of different solvers to be used on different portions of the mesh. The objective is therefore to develop the necessary modifications to allow a study of the main parameters of the considered problem through simulation by means of an implicit solver (rhoPimpleFoam in its transonic version), thus allowing the increase in the fluid dynamic timestep size, and obtaining therefore a considerable decrease in computational cost and time, bypassing the fluid dynamics problems related to the discontinuities introduced by the shock wave systems. This is possible thanks to the creation of new adaptive boundary conditions able to compensate for these issues, thus ensuring stability and convergence of the correct solution. Moreover, whenever chemical reactions are present, the number of equations that must be solved within every discretize cell increases. Chemical ODEs (Ordinary Differential Equations) must be accounted and solved for a massive number of cells, due to the small dimension of the latter. Since within OpenFOAM such process takes place cyclically, the latter becomes progressively inefficient as the number of cells increases. In order to increment the computational efficiency this calculation should be performed in a parallelized manner. Thus, following the example of other authors who have gained advantage by producing a development of operations by parallel computing on NVIDIA graphics cards, although for very different problems, the speed up of the active chemistry required for the introduction of a different newly created method for ODEs treatment, that exploited GPU (Graphic Processing Unit) computational acceleration by means of a GPGPU (General-purpose computing on Graphics Processing Unit) approach. All this is possible through the use of NVIDIA hardware architecture CUDA (Compute Unified Device Architecture), that through its development environment allows to write functions able to perform parallel calculation on the NVIDIA GPUs. Please note that to develop such a resolution not only a total porting but a revision and transformation of the main steps for the treatment of ODEs has been performed, and a link between OpenFOAM and extern "C" CUDA functions were produced as the language of the latter is not by default compatible with the first.File | Dimensione | Formato | |
---|---|---|---|
2019_4_Ghioldi.pdf
accessibile in internet solo dagli utenti autorizzati
Descrizione: Testo della tesi in formato PDF
Dimensione
4.24 MB
Formato
Adobe PDF
|
4.24 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/10589/146075