This thesis work investigates and optimizes the implementation of two speed – up strategies based on the In Situ Adaptive Tabulation (ISAT) and the Cell Agglomeration (CA) algorithms, to reduce the computational cost related to the detailed description of the chemistry in the simulation of fluidized bed reactors via CFD – DEM and Euler – Euler models. In particular, these two modeling approaches are crucial steps in the context of the hierarchical modeling of fluidized bed reactors, which has been reported in literature as a key modeling strategy to manage the enormous amount of solid phase to be tracked in these units (i.e. up to 500 ton, corresponding to 1013 particles). Indeed, the aim of the hierarchical strategy is to derive engineering correlations describing the particle scale phenomena by means of the low hierarchy CFD – DEM simulations of lab scale reactors to enable the industrial reactor simulation by means of the high hierarchy Euler – Euler approach without a significant loss of information. Despite the potentiality of this strategy, the effectiveness of CFD – DEM and Euler – Euler modeling approaches in the context of reactive fluidized beds is limited by the huge computational cost caused by the number of computational cells and catalytic particles for which Euler – Euler and CFD – DEM must respectively provide a detailed description of the chemistry (i.e. in the order of 105-107 particles or cells for lab scale CFD – DEM and industrial scale Euler - Euler simulations). In order to reduce the computational effort of the chemistry, this work implements both ISAT – based and the CA – based speed – up strategies into both modeling approaches. On the one hand, the ISAT – based strategy is applied to a solution algorithm that decouples the contribution of the heterogeneous chemical reaction and of the interphase transport in each cell or particle. Thus, the computational gain provided by this strategy is strongly affected by the maximum time step allowed to accurately split the phenomena, since it limits the time step usable to solve the transport and chemistry in the solid phase of the bed. On the other hand, the CA – based strategy is applied to the solution of the two phenomena coupled, thus avoiding all the accuracy and computational gain issues related to the splitting of the phenomena. Therefore, a detailed investigation of the performances of the two implemented strategies has been carried out. In particular, the speed – up obtained applying the strategies to CFD – DEM and Euler – Euler solvers have been assessed in computationally affordable case studies (with 10,000 reactive particles for the CFD – DEM framework and 40,000 reactive cells for the Euler – Euler framework). In the CFD – DEM case, the ISAT time step imposed by the splitting is 1/3 of the CA time step, anyhow the ISAT strategy won providing a 12.5-fold reduction of the chemistry computational cost with respect to the 4-fold provided by the CA. In the Euler – Euler case, the splitting imposes an ISAT time step equal to 1/20 of the CA one, due to the longer adopted simulation time step with respect to the CFD – DEM. Thus, the CA won providing a 25-fold reduction of the chemistry computational cost with respect to the 5.5- fold provided by the ISAT. Therefore, the results have shown that the number of time steps required by ISAT strategy for each time step adopted in case of the application of CA one leads to an ISAT or CA winning situation depending on the characteristic time of the transport and reaction phenomena happening in the reactor. However, the testing of the two speed – up strategies have proven that a relevant reduction of the computational cost is achievable my means of either ISAT or CA strategy, from both the CFD – DEM and Euler – Euler modeling approaches, paving the way for their effective application to lab and industrial cases of technical relevance, as shown for the CFD – DEM simulation of a 200,000 particles fluidized bed for the Oxidative Coupling of Methane process which has been speed – up 37 times by means of the algorithms implemented in this work.
Questo lavoro di tesi investiga ed ottimizza l’implementazione di due strategie di speed – up, basate sugli algoritmi In Situ Adaptive Tabulation (ISAT) e Cell Agglomeration (CA), per ridurre i tempi computazionali dovuti ad una descrizione dettagliata della chimica nelle simulazioni di reattori a letto fluido via modelli CFD – DEM ed Euleriano – Euleriano. In particolare, questi due approcci modellistici sono elementi cruciali per una modellazione gerarchica dei reattori a letto fluido, che, come riportato in letteratura, è la strategia chiave per gestire l’elevata quantità di fase solida che deve essere tracciata in queste unità (i.e. fino a 500 ton, corrispondenti a 1013 particelle). Infatti, l’obiettivo della strategia gerarchica è la derivazione di correlazioni ingegneristiche che descrivano i fenomeni che avvengono alla scala delle particelle catalitiche attraverso simulazioni CFD – DEM di reattori di laboratorio per rendere possibili le simulazioni di reattori industriali attraverso un approccio Euleriano – Euleriano senza una significativa perdita di informazioni. Nonostante la potenzialità di questa strategia, l’efficacia dei modelli CFD – DEM ed Euleriano – Euleriano per simulare letti fluidi reattivi è limitata dall’enorme tempo computazionale causato dalle numerose celle computazionali e particelle catalitiche per i quali i modelli Euleriano – Euleriano e CFD – DEM devono rispettivamente fornire una descrizione accurata della chimica (i.e. nell’ordine di 105-107 particelle o celle per simulazioni CFD – DEM di reattori in scala di laboratorio e Euleriane – Euleriane di reattori industriali). Per ridurre il costo computazionale della chimica, questo lavoro implementa le strategie basate sull’ISAT e sul CA in entrambi i modelli. Da una parte, la strategia basata sull’ISAT è applicata su un algoritmo che disaccoppia i contributi della chimica eterogenea e del trasporto interfase in ogni cella o particella. Perciò, il guadagno computazionale fornito da questa strategia è fortemente dipendente dal massimo time step permesso per disaccoppiare i fenomeni, visto che limita il time step utilizzato per risolvere il trasporto e la chimica della fase solida. Dall’altra parte, la strategia basata sul CA è applicata su un algoritmo che risolve i due fenomeni accoppiati, quindi evita tutte le problematiche di accuratezza e di guadagno computazionale correlate al disaccoppiamento dei fenomeni. Per questi motivi, le performance delle due strategie implementate sono state dettagliatamente investigate. In particolare, lo speed – up ottenuto applicando le strategie sui solver CFD – DEM ed Euleriano – Euleriano è stato validato in casi studio computazionalmente sostenibili (con 10,000 particelle reattive per il framework CFD – DEM e 40,000 celle reattive per il framework Euleriano – Euleriano). Per il CFD – DEM, il time step dell’ISAT, imposto dal disaccoppiamento, è 1/3 del time step del CA, ad ogni modo l’ISAT vince riducendo i tempi computazioni della chimica di 12.5 volte contro le 4 volte fornite dal CA. Invece per l’Euleriano – Euleriano, il disaccoppiamento impone un time step dell’ISAT uguale ad 1/20 di quello del CA, a causa dei time steps di simulazione più lunghi rispetto al CFD – DEM. Quindi, il CA vince riducendo il costo computazionale della chimica di 25 volte contro le 5.5 volte fornite dell’ISAT. Perciò, i risultati hanno mostrato che il numero di time step richiesti dall’ISAT per ogni time step utilizzato nell’applicazione del CA genera la vittoria dell’ISAT o del CA, che quindi dipende dai tempi caratteristici del trasporto e delle reazioni che avvengono nel reattore. Comunque, i test delle due strategie di speed – up hanno dimostrato che una rilevante riduzione del tempo computazionale è possibile tramite l’ISAT o il CA, per entrambi i modelli CFD – DEM ed Euleriano – Euleriano, spianando la strada per la loro applicazione su casi di laboratorio ed industriali rilevanti, come mostrato con la simulazione CFD – DEM di un letto fluido per il processo di Oxidative Coupling of Methane, composto da 200,000 particelle, che è stato velocizzato di 37 volte mediante gli algoritmi implementati in questo lavoro.
Development and optimization of speed-up strategies for the reactive CFD simulation of catalytic fluidized beds
MICALE, DANIELE
2018/2019
Abstract
This thesis work investigates and optimizes the implementation of two speed – up strategies based on the In Situ Adaptive Tabulation (ISAT) and the Cell Agglomeration (CA) algorithms, to reduce the computational cost related to the detailed description of the chemistry in the simulation of fluidized bed reactors via CFD – DEM and Euler – Euler models. In particular, these two modeling approaches are crucial steps in the context of the hierarchical modeling of fluidized bed reactors, which has been reported in literature as a key modeling strategy to manage the enormous amount of solid phase to be tracked in these units (i.e. up to 500 ton, corresponding to 1013 particles). Indeed, the aim of the hierarchical strategy is to derive engineering correlations describing the particle scale phenomena by means of the low hierarchy CFD – DEM simulations of lab scale reactors to enable the industrial reactor simulation by means of the high hierarchy Euler – Euler approach without a significant loss of information. Despite the potentiality of this strategy, the effectiveness of CFD – DEM and Euler – Euler modeling approaches in the context of reactive fluidized beds is limited by the huge computational cost caused by the number of computational cells and catalytic particles for which Euler – Euler and CFD – DEM must respectively provide a detailed description of the chemistry (i.e. in the order of 105-107 particles or cells for lab scale CFD – DEM and industrial scale Euler - Euler simulations). In order to reduce the computational effort of the chemistry, this work implements both ISAT – based and the CA – based speed – up strategies into both modeling approaches. On the one hand, the ISAT – based strategy is applied to a solution algorithm that decouples the contribution of the heterogeneous chemical reaction and of the interphase transport in each cell or particle. Thus, the computational gain provided by this strategy is strongly affected by the maximum time step allowed to accurately split the phenomena, since it limits the time step usable to solve the transport and chemistry in the solid phase of the bed. On the other hand, the CA – based strategy is applied to the solution of the two phenomena coupled, thus avoiding all the accuracy and computational gain issues related to the splitting of the phenomena. Therefore, a detailed investigation of the performances of the two implemented strategies has been carried out. In particular, the speed – up obtained applying the strategies to CFD – DEM and Euler – Euler solvers have been assessed in computationally affordable case studies (with 10,000 reactive particles for the CFD – DEM framework and 40,000 reactive cells for the Euler – Euler framework). In the CFD – DEM case, the ISAT time step imposed by the splitting is 1/3 of the CA time step, anyhow the ISAT strategy won providing a 12.5-fold reduction of the chemistry computational cost with respect to the 4-fold provided by the CA. In the Euler – Euler case, the splitting imposes an ISAT time step equal to 1/20 of the CA one, due to the longer adopted simulation time step with respect to the CFD – DEM. Thus, the CA won providing a 25-fold reduction of the chemistry computational cost with respect to the 5.5- fold provided by the ISAT. Therefore, the results have shown that the number of time steps required by ISAT strategy for each time step adopted in case of the application of CA one leads to an ISAT or CA winning situation depending on the characteristic time of the transport and reaction phenomena happening in the reactor. However, the testing of the two speed – up strategies have proven that a relevant reduction of the computational cost is achievable my means of either ISAT or CA strategy, from both the CFD – DEM and Euler – Euler modeling approaches, paving the way for their effective application to lab and industrial cases of technical relevance, as shown for the CFD – DEM simulation of a 200,000 particles fluidized bed for the Oxidative Coupling of Methane process which has been speed – up 37 times by means of the algorithms implemented in this work.| File | Dimensione | Formato | |
|---|---|---|---|
|
Tesi_Micale.pdf
non accessibile
Descrizione: Testo della tesi
Dimensione
8.28 MB
Formato
Adobe PDF
|
8.28 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/10589/150163