|Tipo di tesi||Tesi di dottorato di ricerca|
|Titolo||Valutazione delle potenzialità di diverse metodologie CFD per la simulazione della detonazione e della sua probabilità nei motori a combustione interna.|
|Titolo in inglese||Assessment of the performance of CFD methods to predict cycle resolved knocking events and knock statistics in internal combustion engines.|
|Settore scientifico disciplinare||ING-IND/08 - MACCHINE A FLUIDO|
|Corso di studi||INGEGNERIA INDUSTRIALE E DEL TERRITORIO|
|Data inizio appello||2018-03-23|
|Disponibilità||Accessibile via web (tutti i file della tesi sono accessibili)|
Negli ultimi anni le sempre più stringenti normative anti inquinamento e la necessità di ridurre i consumi di combustibile stanno innalzando sempre più gli obiettivi da raggiungere in termini di efficienza e prestazione specifica dei motori a combustione interna.
In the latest years pollutant regulations and fuel consumption concerns are pushing engine manufacturers towards the quest for higher thermal efficiency and specific power output from internal combustion engines. The maximum allowable specific performance, in modern spark ignited units, can be obtained only operating the engine as close as possible to engine knock without incurring in the harmful knock onset. This task is even more complex given the highly stochastic nature of turbulent engine flows, leading to cycle-to-cycle variability (CCV) and to random occurrence / cycle-dependency of knock. Abnormal combustion events hinder the possibility to operate at the engine theoretical optimum combustion phasing due to sporadic yet damaging knocking events. In order to consider the stochastic nature of knock and ensure engine durability, knock mitigation strategies such as spark advance reduction and mixture fuel-enrichment are employed, but these in turn negatively affect engine efficiency and fuel consumption. From a modelling standpoint, the stochastic nature of engine knock, related to combustion instability and cycle-to-cycle variability of turbulent flows, would suggest Large-Eddy Simulation (LES) as the most appropriate approach for CFD simulations. Despite this is conceptually true and several publications show the applicability of LES to both research and production units, LES still remains a very time and CPU demanding approach which can hardly be integrated in the industrial design process and timeframe for the development of new SI units. To limit computational costs and times, RANS models are usually chosen to represent the average engine behavior. Different knock models are available in literature to predict the average knock tendency of the engine though they all suffer from the intrinsic inability to account for far-from-average realizations inherently conflicting anyway with the stochastic nature of knock. This limitation can be overcome by the addition of variance equations for fundamental physical variables in the RANS framework. The information given by this kind of models is of statistical nature and it is grounded in turbulence generated variance of physical fields, which in turn affects the end-gas reaction rate towards autoignition. Conversely from LES, these models naturally neglect a series of CCV-promoting factors, e.g. those pertaining to variability in spark-ignited flame kernel onset and turbulent flame propagation variability. However, such statistics-based RANS models are able to artificially reconstruct a presumed probability of knocking cycles, which can be a very useful indication to the engine designer. In this thesis work the performance of both RANS and LES CFD methods for knock prediction is assessed with the aim to underline pros and cons of the two different approaches in both current production and research units. In particular results from standard RANS and LES knock models are compared with the results from a recently developed turbulence-based statistical knock model and experiments in order to highlight the potential of statistical models in bridging the gap between the physical stochastic nature of knock and the “average” CPU efficient knock prediction in RANS.