Combustion science must necessarily go through a deep process of innovation, as only improving the energy efficiency and the fuel flexibility it will be possible to mitigate the impact of the anthropogenic activities on the climate and the environment. Because of the strong relation that is observed in chemically reacting flows between the fluid-dynamic conditions and the chemical kinetics, the use of Computational Fluid Dynamics (CFD) simulations with detailed kinetic mechanisms represents the best tool to optimize and develop novel combustion systems. In fact, while the CFD provides for the possibility to retrieve information that cannot be extracted by using experimental means (such as the turbulence-chemistry interaction and the local straining rates) and it avoids the costs associated to the scale-up process from laboratory scale experiments, the use of detailed kinetic mechanisms offers the possibility to correctly describe process conditions which are relevant from an industrial point of view (i.e., in which the chemical and mixing time scales are comparable), as well as to predict the formation of complex chemical species, such as the pollutants. Nevertheless, the use of detailed kinetic mechanisms in numerical simulations adds a considerable number of differential equations to be solved (because of the large number of species which are taken into account), and therefore increases the computational complexity of the CFD model. Thus, Machine Learning (ML) algorithms and Reduced-Order Models (ROMs) can be effectively included in the numerical description of chemically reacting flows. In fact, they can be used either to reduce the computational cost associated to the large number of equations in CFD simulations carried out with detailed chemistry, or to leverage the detailed information which can be found in massive, high-fidelity, data obtained from Direct Numerical Simulations (DNS), for model development and validation. In this Thesis, unsupervised and supervised learning algorithms were employed to design a novel adaptive-chemistry approach: the Sample-Partitioning Adaptive Reduced Chemistry (SPARC). This framework can be used to reduce the computational effort required by detailed CFD simulations thanks to a kinetic reduction accomplished in light of the local conditions of the thermochemical field. Several machine-learning algorithms, such as the Principal Component Analysis (PCA), the Local Principal Component Analysis (LPCA), and Artificial Neural Networks (ANNs) were coupled with the Direct Relation Graph with Error Propagation (DRGEP), a graph-based tool for the automatic reduction of kinetic mechanisms. The aforementioned algorithms were compared to achieve the optimal formulation of the adaptive approach, such that the best performances, in terms of accuracy and computational speed-up with respect to the CFD simulation carried out with detailed kinetics, could be obtained. Finally, PCA-based algorithms were proposed and tested to perform feature extraction and local feature selection from high-fidelity data, which were obtained by means of a DNS of a n-heptane jet reacting in air. The PCA, as well as two formulations of LPCA, and the Procrustes analysis were employed and compared with the aim to extract the main features of the turbulent reacting jet in an unsupervised fashion (i.e., to perform data mining tasks), as well as to aid the formulation of local optimized ROMs. All the codes employed to perform the unsupervised and supervised machine learning tasks in the current work were also included in an open-source Python framework, called OpenMORe, designed to perform reduction, clustering and data analysis, and specifically conceived for reacting flows. In fact, although many open-source Python software are already available, they often cannot be adapted to the user’s specific needs, unlike OpenMORe. In addition, many features such as the PCA-based clustering algorithm, or the local feature selection via PCA, are not yet available on any commercial or open-source software, to the best of the author’s knowledge.
Il mondo della combustione deve necessariamente sottoporsi ad un profondo processo di innovazione, poiché solo migliorando l'efficienza energetica e la flessibilità dei combustibili sarà possibile mitigare l'impatto delle attività antropiche sul clima e sull'ambiente. Alla luce del forte legame che si osserva nei flussi reattivi tra le condizioni fluidodinamiche e la cinetica chimica, l'uso della fluidodinamica computazionale (CFD) con meccanismi cinetici dettagliati rappresenta lo strumento migliore per ottimizzare e sviluppare nuovi sistemi di combustione. Infatti, se da un lato mediante la CFD si ha la possibilità di recuperare informazioni che non possono essere estratte utilizzando mezzi sperimentali (come l'interazione tra turbolenza e chimica e gli strain rates locali) ed evitare i costi associati al processo di scale-up, dall’altro l'uso di meccanismi cinetici dettagliati offre la possibilità di descrivere correttamente condizioni di processo rilevanti da un punto di vista industriale (ovvero in cui le scale temporali chimiche e di miscelazione sono comparabili), nonché di prevedere la formazione di specie chimiche complesse, come gli inquinanti. Tuttavia, l'uso di meccanismi cinetici dettagliati nelle simulazioni numeriche comporta un numero considerevole di equazioni differenziali da risolvere (a causa del gran numero di specie che vengono prese in considerazione), e di conseguenza aumenta la complessità computazionale del modello CFD. Pertanto, gli algoritmi di Machine Learning (ML) ed i modelli ridotti (ROM) possono essere efficacemente inclusi nella descrizione numerica dei flussi reattivi. Infatti, possono essere utilizzati sia per ridurre il costo computazionale associato al gran numero di equazioni nelle simulazioni CFD che utilizzano una cinetica dettagliata, sia per sfruttare le informazioni che possono essere trovate in dati ottenuti da Direct Numerical Simulations (DNS) per lo sviluppo e la validazione di modelli. Nella presente tesi sono stati impiegati algoritmi di unsupervised e supervised learning per proporre un nuovo approccio di chimica adattiva: il Sample-Partitioning Adaptive Reduced Chemistry (SPARC). Questo framework può essere utilizzato per ridurre il costo computazionale richiesto da simulazioni CFD che utilizzano un meccanismo cinetico dettagliato grazie a una sua riduzione realizzata alla luce delle condizioni locali del campo termochimico. Diversi algoritmi di machine learning, come la Principal Component Analysis (PCA), la Local Principal Component Analysis (LPCA) e le Artificial Neural Networks (ANN), sono stati accoppiati con il Direct Relation Graph with Error Propagation (DRGEP), un algoritmo finalizzato alla riduzione automatica dei meccanismi cinetici. I suddetti algoritmi sono stati successivamente confrontati per ottenere la formulazione ottimale dell'approccio adattivo tale da ottenere le migliori prestazioni, in termini di accuratezza e velocità di calcolo, rispetto alla simulazione CFD effettuata con un meccanismo cinetico dettagliato. Infine, sono stati proposti e testati algoritmi basati su PCA per eseguire operazioni di feature extraction e local feature selection su dati ad alta fedeltà, ossia su una DNS di un getto di n-eptano che reagisce in aria. La PCA, le due formulazioni di LPCA, e la Procrustes analysis sono state impiegate e confrontate con l'obiettivo di estrarre le principali features del getto turbolento reattivo, nonché per facilitare la formulazione di modelli ridotti ottimizzati localmente. Tutti i codici impiegati per eseguire le attività di unsupervised e supervised learning proposte in questa tesi sono stati infine inclusi in un framework Python open source, chiamato OpenMORe, progettato per poter eseguire la riduzione, il clustering e l'analisi dei dati, e specificamente concepito per applicazioni di fluidodinamica reattiva. Infatti, sebbene molti software Python open source siano già disponibili, spesso non possono essere adattati alle esigenze specifiche dell'utente, a differenza di OpenMORe. Inoltre, molte funzionalità come l'algoritmo di clustering basato su PCA, o la selezione di variabili locali tramite PCA, non sono ancora disponibili su alcun software commerciale o open source, al meglio delle conoscenze dell’autore.
Data-driven models for reacting flows simulations: reduced-order modelling, chemistry acceleration and analysis of high-fidelity data
D'ALESSIO, GIUSEPPE
2020/2021
Abstract
Combustion science must necessarily go through a deep process of innovation, as only improving the energy efficiency and the fuel flexibility it will be possible to mitigate the impact of the anthropogenic activities on the climate and the environment. Because of the strong relation that is observed in chemically reacting flows between the fluid-dynamic conditions and the chemical kinetics, the use of Computational Fluid Dynamics (CFD) simulations with detailed kinetic mechanisms represents the best tool to optimize and develop novel combustion systems. In fact, while the CFD provides for the possibility to retrieve information that cannot be extracted by using experimental means (such as the turbulence-chemistry interaction and the local straining rates) and it avoids the costs associated to the scale-up process from laboratory scale experiments, the use of detailed kinetic mechanisms offers the possibility to correctly describe process conditions which are relevant from an industrial point of view (i.e., in which the chemical and mixing time scales are comparable), as well as to predict the formation of complex chemical species, such as the pollutants. Nevertheless, the use of detailed kinetic mechanisms in numerical simulations adds a considerable number of differential equations to be solved (because of the large number of species which are taken into account), and therefore increases the computational complexity of the CFD model. Thus, Machine Learning (ML) algorithms and Reduced-Order Models (ROMs) can be effectively included in the numerical description of chemically reacting flows. In fact, they can be used either to reduce the computational cost associated to the large number of equations in CFD simulations carried out with detailed chemistry, or to leverage the detailed information which can be found in massive, high-fidelity, data obtained from Direct Numerical Simulations (DNS), for model development and validation. In this Thesis, unsupervised and supervised learning algorithms were employed to design a novel adaptive-chemistry approach: the Sample-Partitioning Adaptive Reduced Chemistry (SPARC). This framework can be used to reduce the computational effort required by detailed CFD simulations thanks to a kinetic reduction accomplished in light of the local conditions of the thermochemical field. Several machine-learning algorithms, such as the Principal Component Analysis (PCA), the Local Principal Component Analysis (LPCA), and Artificial Neural Networks (ANNs) were coupled with the Direct Relation Graph with Error Propagation (DRGEP), a graph-based tool for the automatic reduction of kinetic mechanisms. The aforementioned algorithms were compared to achieve the optimal formulation of the adaptive approach, such that the best performances, in terms of accuracy and computational speed-up with respect to the CFD simulation carried out with detailed kinetics, could be obtained. Finally, PCA-based algorithms were proposed and tested to perform feature extraction and local feature selection from high-fidelity data, which were obtained by means of a DNS of a n-heptane jet reacting in air. The PCA, as well as two formulations of LPCA, and the Procrustes analysis were employed and compared with the aim to extract the main features of the turbulent reacting jet in an unsupervised fashion (i.e., to perform data mining tasks), as well as to aid the formulation of local optimized ROMs. All the codes employed to perform the unsupervised and supervised machine learning tasks in the current work were also included in an open-source Python framework, called OpenMORe, designed to perform reduction, clustering and data analysis, and specifically conceived for reacting flows. In fact, although many open-source Python software are already available, they often cannot be adapted to the user’s specific needs, unlike OpenMORe. In addition, many features such as the PCA-based clustering algorithm, or the local feature selection via PCA, are not yet available on any commercial or open-source software, to the best of the author’s knowledge.| File | Dimensione | Formato | |
|---|---|---|---|
|
DALESSIOthesisFinal.pdf
solo utenti autorizzati dal 10/07/2022
Dimensione
26.64 MB
Formato
Adobe PDF
|
26.64 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/10589/177428