The transition to sustainable energy systems requires alternative fuels that reduce carbon emissions while remaining compatible with existing combustion technologies. This thesis develops a comprehensive framework for the development, validation, and optimization of chemical kinetic mechanisms, with applications to two promising e-fuels: oxymethylene ethers (OME2-5) and ammonia (NH3). The framework combines structured data management, quantitative model evaluation, and systematic parameter optimization within a unified computational environment. At the foundation of this work is the SciExpeM (Scientific Experiments and Models) infrastructure, a centralized platform designed for combustion research data and kinetic models. The system implements FAIR principles (Findable, Accessible, Interoperable, Reusable) to ensure data traceability, reproducibility, and controlled access. It integrates OpenSMOKE++ for reactor simulations, the Curve Matching methodology for model validation, and OptiSMOKE++ for parameter optimization. This integration creates an automated workflow that replaces manual comparison and ad hoc tuning with reproducible and systematic procedures. A central contribution of the thesis is the development of an objective framework for kinetic model validation. Combustion mechanisms are traditionally assessed through visual comparison between experimental and simulated curves, a practice that is inherently subjective and limited in scope. The Curve Matching methodology addresses this limitation by first reconstructing continuous representations of discrete experimental data using smoothing splines with generalized cross-validation. Agreement between simulations and experiments is then quantified through five complementary dissimilarity indices that evaluate both function-level and derivative-level behavior. Statistical and clustering analyses identify experimental conditions where discrepancies are largest and locate deficiencies within the operating space. This structured diagnostic process directly guides mechanism improvement. For oxymethylene ethers, the thesis demonstrates that chemical lumping combined with reaction-class-based optimization enables compact and accurate kinetic models suitable for computational fluid dynamics applications. The proposed lumped formulation is organized into 14 reaction classes and scales linearly with molecular chain length, avoiding the rapid growth characteristic of detailed mechanisms. Kinetic parameters are optimized within physically motivated uncertainty bounds using OptiSMOKE++, with adjustments applied coherently across the entire OME2-5 family to prevent species-specific compensating errors. An integrated reduction--optimization strategy treats skeletal reduction as the first stage of a two-step process: an initial 63% reduction in mechanism size is followed by systematic recalibration to recover predictive accuracy. The ammonia study focuses on pressure-dependent kinetics and the role of bath gas composition in third-body recombination reactions. Four key reactions (NH2+H, H+NO, NH2+NH2, H+O2) are parameterized to account for collider-specific efficiencies based on high-level theoretical calculations. The reactions are reformulated from PLOG to falloff format with temperature-dependent enhancement factors. Three mixture rules of increasing complexity --- Independent Colliders (IC), Linear Mixing Rule with Pressure dependence (LMR-P), and Linear Mixing Rule with Reduced Pressure (LMR-R) --- are evaluated. Although LMR-R is theoretically the most rigorous, predicted macroscopic quantities differ by less than approximately 10% under typical combustion conditions, indicating that simpler formulations are sufficient within current experimental uncertainties. In addition to deterministic optimization, the thesis introduces the Augmented Ensemble Kalman Filter (AEnKF) for kinetic parameter estimation. Unlike classical approaches that minimize a static objective function, AEnKF updates both state variables (species concentrations and temperature) and kinetic parameters as time-resolved measurements become available, while enforcing consistency with the governing kinetic model. Applied to ammonia oxidation in shock tubes, this approach demonstrates how sequential data assimilation can support mechanism refinement in a dynamically consistent manner. Overall, this work establishes a reproducible and systematic approach to combustion mechanism development. For OMEs, it shows that a single compact mechanism can represent the entire OME2-5 family while remaining computationally tractable. For ammonia, it clarifies the impact of bath gas composition on recombination kinetics under practical conditions. More broadly, the framework replaces subjective evaluation and manual parameter tuning with structured, quantitative, and reproducible procedures for combustion research.
La transizione verso sistemi energetici sostenibili richiede carburanti alternativi che riducano le emissioni di carbonio rimanendo compatibili con le tecnologie di combustione esistenti. Questa tesi sviluppa un framework completo per lo sviluppo, la validazione e l'ottimizzazione di meccanismi cinetici chimici, con applicazioni a due e-fuel promettenti: gli ossimetilene eteri (OME2-5) e l'ammoniaca (NH3). Il framework combina una gestione strutturata dei dati, una valutazione quantitativa dei modelli e un'ottimizzazione sistematica dei parametri all'interno di un ambiente computazionale unificato. Alla base di questo lavoro si trova l'infrastruttura SciExpeM (Scientific Experiments and Models), una piattaforma centralizzata progettata per i dati della ricerca sulla combustione e i modelli cinetici. Il sistema implementa i principi FAIR (Findable, Accessible, Interoperable, Reusable) per garantire la tracciabilità dei dati, la riproducibilità e il controllo degli accessi. Integra OpenSMOKE++ per le simulazioni dei reattori, la metodologia Curve Matching per la validazione dei modelli e OptiSMOKE++ per l'ottimizzazione dei parametri. Questa integrazione crea un flusso di lavoro automatizzato che sostituisce il confronto manuale e la taratura ad hoc con procedure riproducibili e sistematiche. Un contributo centrale della tesi è lo sviluppo di un framework oggettivo per la validazione dei modelli cinetici. I meccanismi di combustione sono tradizionalmente valutati tramite confronto visivo tra curve sperimentali e simulate, una pratica intrinsecamente soggettiva e di portata limitata. La metodologia Curve Matching affronta questa limitazione ricostruendo dapprima rappresentazioni continue dei dati sperimentali discreti mediante spline di smoothing con validazione incrociata generalizzata. La concordanza tra simulazioni ed esperimenti viene poi quantificata attraverso cinque indici di dissimilarità complementari che valutano il comportamento sia a livello di funzione che di derivata. Analisi statistiche e di clustering identificano le condizioni sperimentali in cui le discrepanze sono maggiori e localizzano le carenze all'interno dello spazio operativo. Questo processo diagnostico strutturato guida direttamente il miglioramento del meccanismo. Per gli ossimetilene eteri, la tesi dimostra che il lumping chimico combinato con l'ottimizzazione basata su classi di reazione consente di ottenere modelli cinetici compatti e accurati, adatti alle applicazioni di fluidodinamica computazionale. La formulazione lumped proposta è organizzata in 14 classi di reazione e scala linearmente con la lunghezza della catena molecolare, evitando la crescita rapida caratteristica dei meccanismi dettagliati. I parametri cinetici sono ottimizzati entro limiti di incertezza fisicamente motivati mediante OptiSMOKE++, con aggiustamenti applicati in modo coerente all'intera famiglia OME2-5 per evitare errori compensativi specifici per specie. Una strategia integrata di riduzione–ottimizzazione tratta la riduzione scheletrica come prima fase di un processo in due fasi: una riduzione iniziale del 63% delle dimensioni del meccanismo è seguita da una ricalibrazione sistematica per recuperare l'accuratezza predittiva. Lo studio sull'ammoniaca si concentra sulla cinetica dipendente dalla pressione e sul ruolo della composizione del gas di trasporto nelle reazioni di ricombinazione a terzo corpo. Quattro reazioni chiave (NH2+H, H+NO, NH2+NH2, H+O2) sono parametrizzate per tenere conto delle efficienze specifiche dei collisori sulla base di calcoli teorici ad alto livello. Le reazioni sono riformulate dal formato PLOG al formato falloff con fattori di enhancement dipendenti dalla temperatura. Tre regole di miscelazione di complessità crescente --- Independent Colliders (IC), Linear Mixing Rule with Pressure dependence (LMR-P) e Linear Mixing Rule with Reduced Pressure (LMR-R) --- sono valutate. Sebbene LMR-R sia teoricamente la più rigorosa, le quantità macroscopiche previste differiscono di meno di circa il 10% nelle tipiche condizioni di combustione, indicando che formulazioni più semplici sono sufficienti entro le incertezze sperimentali attuali. Oltre all'ottimizzazione deterministica, la tesi introduce il Filtro di Kalman d'Insieme Aumentato (AEnKF) per la stima dei parametri cinetici. A differenza degli approcci classici che minimizzano una funzione obiettivo statica, AEnKF aggiorna sia le variabili di stato (concentrazioni delle specie e temperatura) sia i parametri cinetici man mano che le misurazioni temporalmente risolte diventano disponibili, garantendo al contempo la coerenza con il modello cinetico governante. Applicato all'ossidazione dell'ammoniaca in tubi a shock, questo approccio dimostra come l'assimilazione sequenziale dei dati possa supportare il raffinamento del meccanismo in modo dinamicamente coerente. Nel complesso, questo lavoro stabilisce un approccio riproducibile e sistematico allo sviluppo di meccanismi di combustione. Per gli OME, dimostra che un singolo meccanismo compatto può rappresentare l'intera famiglia OME2-5 rimanendo computazionalmente trattabile. Per l'ammoniaca, chiarisce l'impatto della composizione del gas di trasporto sulla cinetica di ricombinazione in condizioni pratiche. Più in generale, il framework sostituisce la valutazione soggettiva e la taratura manuale dei parametri con procedure strutturate, quantitative e riproducibili per la ricerca sulla combustione.
A data-driven framework for continuous chemical kinetic model development
Dinelli, Timoteo
2025/2026
Abstract
The transition to sustainable energy systems requires alternative fuels that reduce carbon emissions while remaining compatible with existing combustion technologies. This thesis develops a comprehensive framework for the development, validation, and optimization of chemical kinetic mechanisms, with applications to two promising e-fuels: oxymethylene ethers (OME2-5) and ammonia (NH3). The framework combines structured data management, quantitative model evaluation, and systematic parameter optimization within a unified computational environment. At the foundation of this work is the SciExpeM (Scientific Experiments and Models) infrastructure, a centralized platform designed for combustion research data and kinetic models. The system implements FAIR principles (Findable, Accessible, Interoperable, Reusable) to ensure data traceability, reproducibility, and controlled access. It integrates OpenSMOKE++ for reactor simulations, the Curve Matching methodology for model validation, and OptiSMOKE++ for parameter optimization. This integration creates an automated workflow that replaces manual comparison and ad hoc tuning with reproducible and systematic procedures. A central contribution of the thesis is the development of an objective framework for kinetic model validation. Combustion mechanisms are traditionally assessed through visual comparison between experimental and simulated curves, a practice that is inherently subjective and limited in scope. The Curve Matching methodology addresses this limitation by first reconstructing continuous representations of discrete experimental data using smoothing splines with generalized cross-validation. Agreement between simulations and experiments is then quantified through five complementary dissimilarity indices that evaluate both function-level and derivative-level behavior. Statistical and clustering analyses identify experimental conditions where discrepancies are largest and locate deficiencies within the operating space. This structured diagnostic process directly guides mechanism improvement. For oxymethylene ethers, the thesis demonstrates that chemical lumping combined with reaction-class-based optimization enables compact and accurate kinetic models suitable for computational fluid dynamics applications. The proposed lumped formulation is organized into 14 reaction classes and scales linearly with molecular chain length, avoiding the rapid growth characteristic of detailed mechanisms. Kinetic parameters are optimized within physically motivated uncertainty bounds using OptiSMOKE++, with adjustments applied coherently across the entire OME2-5 family to prevent species-specific compensating errors. An integrated reduction--optimization strategy treats skeletal reduction as the first stage of a two-step process: an initial 63% reduction in mechanism size is followed by systematic recalibration to recover predictive accuracy. The ammonia study focuses on pressure-dependent kinetics and the role of bath gas composition in third-body recombination reactions. Four key reactions (NH2+H, H+NO, NH2+NH2, H+O2) are parameterized to account for collider-specific efficiencies based on high-level theoretical calculations. The reactions are reformulated from PLOG to falloff format with temperature-dependent enhancement factors. Three mixture rules of increasing complexity --- Independent Colliders (IC), Linear Mixing Rule with Pressure dependence (LMR-P), and Linear Mixing Rule with Reduced Pressure (LMR-R) --- are evaluated. Although LMR-R is theoretically the most rigorous, predicted macroscopic quantities differ by less than approximately 10% under typical combustion conditions, indicating that simpler formulations are sufficient within current experimental uncertainties. In addition to deterministic optimization, the thesis introduces the Augmented Ensemble Kalman Filter (AEnKF) for kinetic parameter estimation. Unlike classical approaches that minimize a static objective function, AEnKF updates both state variables (species concentrations and temperature) and kinetic parameters as time-resolved measurements become available, while enforcing consistency with the governing kinetic model. Applied to ammonia oxidation in shock tubes, this approach demonstrates how sequential data assimilation can support mechanism refinement in a dynamically consistent manner. Overall, this work establishes a reproducible and systematic approach to combustion mechanism development. For OMEs, it shows that a single compact mechanism can represent the entire OME2-5 family while remaining computationally tractable. For ammonia, it clarifies the impact of bath gas composition on recombination kinetics under practical conditions. More broadly, the framework replaces subjective evaluation and manual parameter tuning with structured, quantitative, and reproducible procedures for combustion research.| File | Dimensione | Formato | |
|---|---|---|---|
|
Dinelli_PhD_thesis.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
11.11 MB
Formato
Adobe PDF
|
11.11 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/10589/256518