Skip to main content

Large-Eddy and System Simulation to predict Cyclic Combustion Variability in gasoline engines

Final Report Summary - LESSCCV (Large-eddy and system simulation to predict cyclic combustion variability in gasoline engines)

Executive summary:

The objective of LESSCCV was to exploit the recent possibilities of engine computational fluid dynamics (CFD) tools based on Large-Eddy Simulation (LES) to fundamentally improve the understanding of cyclic combustion variability (CCV) in gasoline engines under real operating conditions, and to provide adequate modelling.

As a starting point for the project work, multi-scale CFD tools able to simulate a whole engine, i.e. including the combustion chamber (cylinder) as well as the full intake and exhaust ducts, were developed. They were based on coupling 1D-CFD codes, which provide an adequate prediction of the flow in the intake and exhaust lines, with 3D-CFD codes using the innovative LES technique, which is necessary to accurately reproduce the complex interactions inside the cylinder between the turbulent flow and combustion chemistry which is key factor for CCV. The results achieved allow simulating the whole engine, which is necessary to accurately account for the numerous sources of CCV, which are as much related to the flow in the combustion chamber, than to the flow in the rest of the engine.

Project Context and Objectives:
Context

The operation of gasoline engines, which we call for the sake of simplicity spark-ignition engines (SIE) in what follows, is characterised by a non-repeatability of instantaneous combustion rate between different cycles at nominally identical engine operating parameters, commonly referred to as cyclic combustion variability (CCV). CCV inevitably appear over the whole engine operation range, due in particular to the unsteady, cyclic and turbulent nature of flow and combustion in SIE. As long as their amplitude is sufficiently low, engine simulation software or model based engine control can predict combustion in any given cycle using deterministic models that neglect cyclic variations.

It is common practice to consider that CCV amplitudes (measured in terms of standard deviation of IMEP, normalised by the mean value) of less than around 5% are acceptable. For higher CCV amplitudes, individual cycles will behave very differently from the statistically most probable cycle predicted by control algorithms or engine simulation tools. As a result, predictions of fuel consumption or emissions over a number of cycles may differ quite substantially from the one based on the mean engine cycle. These differences increase with the CCV amplitude, and may reach extreme levels in the case of misfires or extreme knocking cycles.

The starting point for the present LESSCCV project was the fact that:
1) the understanding of how the complex combination of different sources leads to the occurrence of CCV for a specific engine design or mode of operation is still limited;
2) there is a need to make such knowledge available in system simulation tools, which are increasingly used in the engine development and optimisation process.

LESSCCV rationale and Scientific and technical objectives

In this context, LESSCCV proposed an innovative approach based on advanced Computational Fluid Dynamics (CFD) engine simulation tools to gain a better understanding of CCV due to flow phenomena. These tools have the potential to address the numerous coupled phenomena influencing engine combustion, and give potentially access to any quantity needed to understand CCV.

The objective of LESSCCV was to make use of the recent possibilities of advanced engine CFD tools to fundamentally improve the understanding of CCV related to the flow in SIE and provide adequate modelling.

Its starting point was the fact that CCV are the result of a complex combination of flow phenomena at different scales, and which can vary from cycle to cycle:
- global phenomena, which are related to the global engine behaviour: trapped mass, intake mass flow rate, tumble ratio, overall equivalence ratio, mean cylinder pressure, mean intake charge temperature, EGR rate and mean temperature, BMEP, etc. Their spatial scales are of the order of some characteristic dimensions of the engine, and their time scales of the order of the crank angle and up to the engine cycle;
- local phenomena, which are related to local flow variables: temperature, pressure, composition, turbulence, velocity, dissipation rate, etc. Their spatial scales range from some micrometers up to some millimetres, their time scales from some microseconds up to the crank angle.

Both scales strongly interact, rendering the understanding of CCV complex. This is especially true in the combustion chamber, where the non-linear response of combustion is a key contributor to important CCV levels.

1D-CFD simulation approaches, whereby the flow inside the engine is supposed to be mainly dependent on the axial flow direction and time, allow addressing global flow phenomena, and their related impact on CCV. Yet they cannot reproduce local flow effects. On the other hand, 3D-CFD simulation tools, and in particular the emerging Large-Eddy Simulation (LES) technique, resolve the three-dimensional flow structures, and allow addressing local instantaneous flow phenomena and their role in the occurrence of CCV. However they can hardly include global phenomena for reasons of high computational overhead. The rationale of work within LESSCCV is to combine both in an integrated approach aimed at understanding the origins of CCV and modelling their impact on the global engine behaviour.

1D-CFD codes are fast (down to real time performances) and can thus be applied to the simulation of complete engines under realistic operation. But they can only render the effects of global phenomena on CCV, mainly those related to the unsteady overall flow in the intake and exhaust systems, and inside the fuel injection system (FIS). Local phenomena inside the combustion chamber are oversimplified, not allowing to account for their impact on CCV. Attempts to introduce local effects in 1D-CFD combustion chamber models exist. Their principle is to statistically perturb the mean in-cylinder quantities in each cycle in order to reproduce the experimentally observed CCV levels. The type and amplitude of the perturbations are the result of a fitting procedure to mimic experimental findings, and not based on physical knowledge. The achievable reproduction of CCV characteristics is thus low.

On the other end of the spectrum of simulation tools, 3D-CFD has the potential to accurately resolve the local phenomena inside the combustion chamber. The innovative Large-Eddy Simulation (LES) 3D-CFD technique has recently shown its potential to predict effects of CCV resulting from the local in-cylinder flow in SIE. The standard RANS approach to 3D-CFD may be used to study the response of the combustion chamber, but this method based on simulating the most probable engine cycle is not very accurate to predict cyclic variability. The shortcoming of LES is that the necessary computational times are orders of magnitude higher than for 1D-CFD tools, so that they cannot be applied to the simulation of complete engines.

LESSCCV proposed developing innovative models to be combined with existing 1D-CFD SIE combustion chamber models to achieve a better reproduction of CCV and their impact on global engine behaviour. The aim was to base these new models on physical knowledge on the interaction between local and global flow effects, gained from the project's LES work.

The 1D-CFD models for reproducing CCV were to take the form of stochastic or deterministic cyclic perturbations of the input parameters for 1D-CFD combustion chamber models. The characteristics of these perturbations were to result from the analysis of the LES findings, and depend on the type of SIE and the operating points studied. The aim was that by applying these models in mutli-cycle 1D-CFD simulations of an engine, they allow reproducing experimental findings on CCV with only minor fitting of model constants, as a result from the fact that their formulation relies on a deeper physical understanding than possible at the start of LESSCCV.

In this sense, LESSCCV aimed at achieving 1D-CFD software that predicts the occurrence of CCV without any a priori knowledge from experiments. The predictivity of 1D-CFD is naturally limited, as it is based on many simplifying assumptions. LESSCCV rather proposed research on models which can be combined with existing or slightly adapted combustion chamber models, in order to allow a more accurate reproduction of the characteristics and levels of CCV in regions of the engine operation where they are known to be important. After the project, this could be instrumental in assisting in the development of control strategies using the achieved knowledge and aiming at limiting the negative impact of CCV.

For LES studies aimed at gaining a deeper understanding of the causes of CCV in SIE, it was essential addressing the coupling between global and local phenomena, as CCV potentially result from a combination of both. In this sense, a coupling between 1D-CFD - addressing global phenomena in the intake and exhaust systems as well as in the FIS - and LES - addressing local phenomena in the combustion chamber – was to be achieved in LESSCCV.

The objective of LESSCCV was to exploit the recent possibilities of engine computational fluid dynamics (CFD) tools to fundamentally improve the understanding of cyclic combustion variability (CCV) in gasoline engines under real operating conditions, and to provide adequate modelling.

The detailed scientific and technical objectives of work in LESSCCV were as follows:
- Develop multi-scale CFD tools able to predict CCV in SIE, based on coupling innovative 3D-CFD tools based on Large-Eddy Simulation (LES) and comparable techniques like PANS to address local phenomena, with 1D-CFD tools used to simulate full engines under realistic operation, to address global phenomena;
- Apply the achieved multi-scale CFD tools to the study of CCV in indirect injection (II) SI, direct injection (DI) SI and CAI engines to acquire a basic understanding of the origins of CCV and of their impact, as a result of both global and local phenomena;
- Complement the multi-scale CFD findings by detailed studies of local effects of CCV related to the early flame kernel growth at the spark plug using Direct Numerical Simulation (DNS), and of the interactions between flow in the FIS and a fuel spray using a multi-scale CFD study of injection into a constant volume bomb;
- Based on the findings achieved during the multi-scale CFD studies and on outcomes from detailed studies of local effects of importance to CCV, propose models which combine with combustion chamber models to reproduce effects of CCV in multi-cycle 1D-CFD simulations of full engines, in regions in which CCV were found to be important. This work will mainly rely on existing combustion models, or on necessary evolutions in terms of spark- or auto-ignition or turbulence modelling. The chosen approach aims at addressing cyclic variations of heat release rate (and thus cylinder pressure), the impact on pollutants being the result of applying existing approaches for their kinetics;
- Apply the developed CCV models in industrial 1D-CFD software, and perform case studies to assess the improved reproduction of the characteristics and effects of CCV they allow. Explore how this can contribute to further improve the design and operation of advanced SIE.

LESSCCV exclusively addressed CCV related to flow phenomena. Other sources, like variations of thermodynamic conditions outside the engine, vibrations, electrical and magnetic sources, or variations of the nominal geometry of different engines due to manufacturing uncertainties or wear effects were not covered.

The LESSCCV partnership brought together 2 research centres (IFPEN as coordinator and FEV), 3 major European engine simulation software vendors and engineering companies (AVL, LMS-Imagine and RICARDO), and 3 Universities (Czech Technical University, University of Western Macedonia and Politecnico di Milano) from 7 European countries, all internationally recognised for their expertise in engines and simulation.

The LESSCCV project originated from work within the Task Force "Modelling and Simulation" of EARPA (see http://www.earpa.eu online) to which most project partners participated.

Project Results:

The results generated within the LESSCCV project can be summarised as follows, distinguishing 5 main types of results:
1 - Coupled 1D CFD and 3D LES codes
2 – Detailed studies of local flow phenomena of importance to CCV
3 – LES studies of CCV in 3 types of SIE
4 – Capitalising LES knowledge in the form of CCV models for 1D-CFD
5 – Explore potential benefits from applying the developed CCV in engine and vehicle simulations

1 - Coupled 1D CFD and 3D LES codes

In order to accurately capture all possible sources of CCV, and in particular the ones related to the interactions between the unsteady flow in the intake and exhaust ducts and the combustion chamber, it appeared essential for the LESSCCV work to develop strategies allowing to run 1D CFD codes, that predict the flow in the ducts that are too long and complex to be modelled by LES, coupled with the LES code, that describes flow and combustion inside the cylinder and in its immediate neighbourhood, where the accuracy of LES is required and leads to important computational time requirements. Such coupling strategies are essential to enable study of the detailed causes and effects of CCV in SIE in a CPU-time efficient way.

Two groups of partners therefore developed coupling strategies between 1D and 3D LES codes.

The first group brought together AVL with Ricardo and worked on commercially available CFD codes. They delivered coupling methods based on the 1D-CFD tools BOOST (edited and commercialised by AVL) and WAVE (edited and commercialised by Ricardo) on a crank-angle resolved basis with the industrial 3D-CFD code FIRE (edited and commercialised by AVL and used in LESSCCV to perform LES of DI-SIE). The work was based on adopting an advanced code coupling technology and employing a numerically robust and stable implementation practice of a non-reflecting boundary condition treatment between the 1D and 3D CFD domains. The developed coupling methodology enables an easy modelling of the individual parts of the pipe-work / combustion chamber assembly as either 1D or 3D domain. This ensures efficient and user friendly handling of coupled 1D- and 3D-CFD calculations for analysis and assessment of the CCV cause-and-effect mechanisms of single-cylinder and multi-cylinder arrangements.

The activities and achievements of the work can be summarized as follows:
1) definition of the coupling strategy between the 3D-CFD code FIRE and the 1D-CFD tools BOOST and WAVE considering physical and numerical aspects, as well as definition of targets in terms of results accuracy and parallel performance to be achieved;
2) modification/extension of the CFD-code FIRE in order to be seamlessly coupled with WAVE and BOOST;
3) set up of a coupled version of FIRE and the involved 1D-CFD tool BOOST and execution of basic functionality tests on serial and parallel platforms in order to ensure numerical consistency of the results;
4) validation of the developed 1D/3D-CFD coupling approach for BOOST, FIRE and WAVE in terms of proper propagation of acoustic waves, species transport, etc. across the domain interfaces on the basis of selected test-cases in order to ensure its applicability in the porject's LES studies.

The second group brought together IFPEN and PoliMi and worked on research codes. They delivered two strategies to combine the 1D CFD code GASDYN developed by PoliMi, with the AVBP LES code developed by IFPEN. This has been done by implementing an effective coupling between both codes. After the initial analysis of different possible numerical approaches, a simple procedure has been chosen as a good compromise between robustness and complexity. Different evolutions of GASDYN had to be achieved, in order to yield a coherent thermodynamic description between both codes, and in order to allow frequent restarts, rendered necessary by AVBP's moving mesh strategy.

With regard to the development of interfaces to allow coupled 1D/3D simulations between GASDYN and AVBP, a few approaches have been explored. Finally, the approach chosen for the code coupling is based on the switching of boundary conditions of inlet and outlet, due to its simplicity and easy implementation in AVBP. In this approach the solution of the Riemann problem is involved, since it is necessary to determine the direction of the flow at the interface. The HLLC approximate Riemann solver has been implemented in order to treat the presence of different chemical composition (contact surfaces). The solution of the Riemann problem returns the value of the volumetric flux at the interface and its direction. On the basis of the flow direction, AVBP uses a proper set of boundary conditions based on the NSCBC theory. When the flow is directed outward from the 1D model, a set of boundary conditions for inlet flows is used in AVBP exploiting the velocity and the gas temperature passed by the 1D model. In the case of flow directed outward from the 3D model, the gas temperature is imposed as well as the total pressure.

The two-way coupling between AVBP and GASDYN was finally achieved by means of the open-source external coupler PALM. Much work was spent to find a common configuration to make AVBP, PALM and GASDYN work together in the same environment.

Although the developed coupling method could be demonstrated on a simple application, it turned out that much more work would have been required to apply it to the project's LES studies. Therefore the partners applied an alternative approach consisting of chaining both codes, the 1D CFD code thus being run first and providing boundary conditions for the LES. Although this did not allow accounting for feedbacks from the cylinder to the ducts, it proved to yield sufficiently accurate results in the project's LES studies.

2 – Detailed studies of local flow phenomena of importance to CCV


The impact of two phenomena was studied in terms of their potential impact on CCV:
a/ early flame kernel growth and its interaction with the flow at the spark;
b/ direct fuel injection in the cylinder, its potential cyclic to cycle variability, and its interactions with the in-cyljnder aerodynamics.

a/ UOWM undertook DNS (direct numerical simulation) studies of early flame kernel development.

The objective was to acquire a more detailed understanding of certain aspects of initial flame kernel growth after spark-ignition. This plays a role in the occurrence of CCV, and supported the LES and system simulation modeling of the other project partners. Three different configurations were studied with DNS.

The 1st setup was used to characterize the relationship between flame displacement speed and stretch rate (curvature and strain) in the absence of turbulence. The functional dependence of flame displacement speed and total flame stretch has been characterized with the use of the Markstein number at the low stretch rate limit. Markstein numbers were calculated for a variety of unburned mixture temperatures (298K and 750K), pressures (1 to 10 bar) and equivalence ratios (0.8 to 1.4) using both reduced step and detailed chemical mechanisms. The major conclusion from the simulations performed for this setup is that the Markstein number strongly depends on the temperature isosurface upon which it is measured/calculated. Defining a local 'Markstein length' at various positions within the flame zone, this length tends asymptotically to the burnt Markstein length value inside the reaction zone; however, when moving away from the reaction zone and towards the fresh mixture its behaviour is more abrupt. This result agrees well with the asymptotic theory which predicts flame accumulation and transverse fluxes along the flame front. In collaboration with Prof. M. Matalon from University of Illinois at Urbana-Champaign, we performed a detailed comparison of our numerical results from the 1st setup with recently derived composite asymptotic expansions for the mass flux through the flame. One of the major outcomes of this collaboration, which is currently in preparation for publication, is that the Markstein number in the unburned side of the flame diverges as we move towards the unburned side of the flame front, due to the exponential nature of the preheat zone. From this we can conclude that it cannot be clearly defined and does not have a physical meaning as it does not converge to a finite value. Therefore, only the Markstein number of the burned side should be taken into account when modelling the effects of stretch on early flame kernel development.

The 2nd setup was used to investigate how the flame surface area is affected by convection near the spark plug. It was found that in the absence of turbulence, the increase of convection velocity (from 1, 2, 3, 5, up to 7 times the laminar flame speed) at the spark location does not affect the qualitative way the flame area evolves in time, which is roughly linear for the time period considered. It is important to note that an ignition delay is observed when the convection velocity exceeds 2 times the laminar flame speed. For large convection velocity, the observed ignition delay was 0.8 crank angle degrees for an engine running at 3000rpm. Given the fact that the convection velocity measured in actual engine operation is 7 to 10 times the lmainar flame speed, the ignition delay due to high convection velocity at the spark could be playing a role in the occurrence of CCV. In addition, it was found that the calculated ignition delay depends strongly on the chemical scheme used. Thus, in order to properly assess ignition delay times, the chemical scheme used in the numerical simulations should be ignition delay-calibrated with experimental measurements. In another direction, the 2nd setup was extended to account for both convection and turbulence. The ignition delay was affected by the turbulent field in the vicinity of the flame ignition kernel; this effect could not be easily quantified and will require further investigation. It is important to note that at the first stages of ignition, the spark energy deposition causes local preheating of the mixture and the evolution of the kernel does not behave as if propagating in a fresh mixture with constant temperature. This observation is an element that should be accounted for in model for the spark ignition, in order to account for preheated regions and homogenous ignition phenomena.

The objective of the 3rd setup was to investigate how the ignition kernel is affected by initial turbulence with varying intensity and length scales. It was found that for higher turbulence intensity levels (of the order of 1 to 3 times laminar flame speed), flame displacement speed is slightly decreased, but the total flame area is significantly increased, resulting to an overall increase in Heat Release Rate (HRR). Moreover, ignition time is significantly affected by turbulence time scale, i.e. delay in ignition is observed as turbulence intensity increases or length scale decreases. Using hotspot initial conditions (no preheating of the reacting mixture), the dependence of the displacement speed with stretch rate correlates well with the results of the 1st setup, in both 2D and 3D simulations. Using a spark source term in the energy equation (with preheating of the reacting mixture and more relevant to engine conditions), variation in ignition time was obtained. In addition, for fuel rich conditions (F/A eq. ratio of 1.2 and 1.4) and for the same turbulent characteristics as in the stoichiometric and lean conditions, no ignition was observed. This indicated that the increase of turbulent intensity could result in ignition failure for fuel rich mixtures. On the other hand, the turbulent integral length scale does not seem to have a significant effect on this behaviour.

Furthermore, in collaboration with the group of Dr. P. Fischer of Argonne National Labs (ANL), we also performed 3D simulations of spherically expanding flame kernels in the presence of initial turbulence. It is important to mention that to our knowledge, these are the first spherically expanding flame DNS simulations in a turbulent environment using a spherical computational domain with outflow boundary conditions -and not periodic-, allowing for proper flame development at constant pressure and without influence from neighbouring flames due to periodicity. One of the main outcomes of these numerical simulations was that the mean flame displacement speed can be predicted from the total mean stretch rate of the flame kernel, at least for small stretch rates. It was found that although the flame kernel is propagating inside a turbulent environment of the order of flame velocity and thickness, the mean quantities of the flame (flame speed and flame stretch) follow approximately the same linear relationship as the ones calculated from the simplified laminar case (1st setup), at least at low stretch rates.

b/ LMS and IFPEN studied the possible contributions of the FIS on CCV.

LMS used actual FIS models for both Diesel and Gasoline direct injection applications in order to support its investigations. The strategy was to conduct the first observations on a simple demo model and to refine, in a second stage, the analyses using complete FIS models including comprehensive models for the HP pump with its control valve, the rail (and a possible extra pressure control device) and the injectors. This allowed analysing the dynamic couplings and the propagation of the hydraulic pressure waves in the FIS and their impact on the injection rates and fuel delivery. For this type of study, steady-state boundary conditions are generally applied in terms of actuator controls, mainly because it is difficult to get access to the details of the control strategies and information about the impact of the electrical board net of the actuator characteristics.

The analysis conducted in the framework of the project showed that in these conditions, no CCV can be observed. In other words, the physics associated to the FIS itself does not lead to any cyclic variation. This is due mainly to the fact that the internal dynamics of the FIS are faster at the scale of the engine cycle and we generally obtain a constant cycle-to-cycle pattern for a given injector. On the contrary, the geometry of the system and its dynamics can lead to cylinder to cylinder variability but this type of variability was not in the scope of LESSCCV.

IFPEN aimed at studying with LES the interaction of a fuel spray with the engine aerodynamics in order to evaluate its impact on CCV in GDI engines. This initial objective could not be fully achieved as planned, as more important spray model development work than initially foreseen had to be undertaken, and as reliable detailed experimental data for validating and supporting LES were not available.

A first study concerned the exploration, using an Eulerian LES spray model, of the impact of variations of the flow at the injector nozzle exit on the fuel mass fraction fields generated by a single hole round injector. This work was based on performing LES of fuel injections into a quiescent environment of a constant volume, high pressure and temperature vessel. Although the results from LMS had indicated that the variations of FIS flow were small, the complex physical phenomena taking place close to the nozzle exit introduce a level of variability of the outflow (cavitation, turbulence). The performed LES indeed showed that the impact of small injection rate variations during the opening of the injector had a relatively limited impact on the spray and on the resulting fuel mass fraction distribution. However it appeared that for the short injecting durations studied, the impact of small variations of the shutdown rate affected more significantly the spray and fuel concentration fields. Furthermore, the turbulent nature of the nozzle outflow was found to lead to important variations from shot to shot of local instantaneous spray characteristics that could affect CCV in late injection strategies. However this could not be verified due to the absence of dedicated experimental data.

A second part of the work consisted of developing a LES spray model able to simulate a 6-hole solenoid injector experimentally studied in the FUI project MAGIE. First, the Eulerian spray model used for the single hole studies was extended, and first simulations of a 6-holes injector indicated encouraging results. Nevertheless, the Eulerian spray model proved to be highly prone to numerical instablities. These numerical issues did not allow performing the planned validation with the experimental data. An alternative approach was then decided, and consisted of adapting the Lagrangian spray model of AVBP to LES of the MAGIE multi-hole injector. This required an important amount of development work. First LES of the MAGIE injector indicated that this approach indeed allowed a stable LES, and shall be a good basis for future LES work dedicated to study GDI combustion. Here also, the absence of related data and the missing time in the frame of LESSCCV did not allow actually applying the developed model to the study of the impact of spray variations on CCV.

The achieved work contributed essential elements for the ability of the AVBP code to study engine sprays, which will find its direct exploitation in the frame of two French national research projects aimed at studying abnormal combustion in downsized GDI (ANR/TDM ICAMDAC) and at the spray/wall interaction during fast transients in such engines (ANR/TDM ASTRIDE), as well as in a bilateral LES research project with a major European OEM under construction.

3 – LES studies of CCV in 3 types of SIE

A central achievement of LESSCCV is the successful study with LES of the sources and effects of CCV in three different types of SI engines. The amount and ambition of the performed research work is unique at the world level, and opens exciting future perspectives for the exploitation of LES as advanced engine design tool.

IFPEN, assisted by PoliMi, applied the chaining strategy developed based on the AVBP LES code and the 1D CFD code GASDYN to the LES of a single cylinder indirect injection engine (IIE) studied experimentally by IFPEN in the French ANR/PREDIT project SGEmac. The interest of this engine was that the acquired database was specifically designed to include so-called stable and unstable operating points, i.e. exhibiting respectively low and high levels of CCV. The performed work was to validate the prediction by LES of the experimentally observed CCV levels, and then exploit the detailed information available from the LES to identify the sources of CCV, having in view the exploitation of this understanding in the CCV model development for system simulation that was another key objective of LESSCCV.

Work started by setting up the computations for both the AVBP LES code and the 1D CFD code GASDYN, and to apply them, in a chained approach whereby the 1D CFD code allowed setting the boundary conditons for the 3D code, to the simulation of the motored reference operating point of the database. This allowed identifying different methodological elements for ensuring a stable and accurate prediction. In particular, it proved important not to limit too much the 3D CFD computational domain in order to capture accurately the flow reversals appearing especially during exhaust valve opening. Extending the 3D domain to include a longer part of the intake and exhaust ducts also allowed imposing the boundary conditions coming from the 1D CFD code in a proper way, as the local flow conditions far from the cylinder essentially were one dimensional in nature. It proved also important to refine the mesh around the valves and especially in the valve seats. In the latter region, an adapted artificial viscosity scheme had to be used to be able to render the shocks that form during exhaust valve opening. Other aspects concerned adaptations of the wall laws, as well as using the newly developed sub-grid scale LES model SIGMA. An explicit centred Lax-Wendroff scheme of second order accuracy was used for solving convection. The LES predictions for the simulated 25 consecutive cycles in terms of cylinder pressure, trapped mass and in-cylinder mean and RMS flow fields were found to accurately reproduce experimental findings.

Once the methodology for the motored reference case had been validated, work concerned its extension to be able to predict the reference fired operating point of the database, exhibiting low levels of CCV, as well as two unstable points with high CCV levels. The latter were obtained from the reference point by adding an important intake air dilution, or by strongly reducing the fuel/air equivalence ratio of the fresh charge. In order to yield accurate results, minor adaptations of the ECFM-LES premixed combustion model had to be performed, and especially the resolved stretched source terms in the FSD equation was modified. Spark-ignition was simulated using the ISSIM spark-ignition model. More than 20 consecutive cycles were simulated for the stable reference, as well as for each of the two unstable points. No artificial perturbations were introduced in these simulations: The boundary conditions at the interfaces between the 3D and 1D domains were imposed as predicted by the 1D CFD approach, and were repeated unchanged for all simulated cycles; Combustion was fully simulated, owing to the unique capacity of the ISSIM model, so that the performed LES can be considered as a numerical experiment.

The LES predictions for the three fired operating points were compared with available experimental data in terms of trapped mass, time evolution of cylinder pressure (mean and RMS), as well as IMEP (mean and RMS). These comparisons fully validated the LES in terms of their reproduction of the experiments. In particular, the experimentally observed levels of CCV for the three points were reproduced qualitatively and quantitatively. The LES naturally yielded the lowest CCV levels for the reference point, and predicted the impact of dilution and leaning out the fresh mixture without any additional tuning of model coefficients. The standard deviation of IMEP was quite accurately captured by the LES for all cases, confirming that the employed approach is capable of retaining the physics leading to these CCV. Plotting Matekunas diagrams for the crank angle of maximum pressure for the 3 cases showed that the LES predicted the change in nature of the dependency from preceding cycles between the stable and unstable points, as found experimentally.

Finally, the LES results were analysed in order to identify the sources for the observed CCV and support system simulation model development (see below). The overall scheme for the appearance of instabilities was found to be related to the cycle to cycle variations of the flow at spark ignition and during the combustion. These variations are mostly related to the turbulent nature of the engine flow, which naturally introduces a statistical variation of local flow quantities from cycle to cycle. Although the exact path for these variations are hardly identifiable, it appeared that a factor greatly contributing to these effects is the tumbling nature of the intake flow generated in a SI engine. This tumble was found to exhibit a processing movement around the tumble rotation centre, and shows important variations from cycle to cycle. As a result, the compression by the piston leads to a turbulent flow near top dead centre combustion that shows significant variations in terms of local turbulence intensity, flow scales and local resolved velocity from cycle to cycle. Owing to the highly non-linear nature of combustion, in which small variations in local thermodynamic and flow conditions can lead to important variations of heat release rate, these small variations are amplified, ultimately leading to the observed CCV. The sensitivity of the combustion to variations in flow and turbulence conditions is higher, the lower the laminar flame speed is, i.e. the higher the dilution rate or the lower the fuel/air equivalence ratio. This explains the important increase of CCV of both unstable points as compared to the stable reference point, and appears to be accurately captured by the used LES models. Further analysis showed that in the diluted case, which was characterised by the lowest laminar flame speed of the three cases, CCV was the highest and combustion was very sensitive to all flow characteristics, with not one of them standing out as having a major impact compared to the others. In the lean operating point, the flame proved to oppose more resistance to wrinkling by turbulence, and at the same time turned out to be more sensitive to local flow directions at the spark plug. Indeed, it appeared that in this case, the slowest engine cycles corresponded to those where the flow direction at the spark at ignition was such that it pushed the initial flame kernel towards the spark plug cavity. As a result of the geometrical (and in reality also thermal) limitation resulting from this, the flame propagation exhibited some preferential directions, which appeared as being one additional factor of instability for this operating point.

AVL and JBRC applied the coupled 1D/3D-CFD tool developed at the project start to LES of multiple engine cycles of a DI SI-engine in order to study the impact of the instantaneous cycle-resolved flow and mixture conditions on the space and time resolved flame propagation process, and hence on the overall rate of heat release that finally is responsible for the cyclic dispersion of the in-cylinder pressure traces.

The engine under consideration was a passenger-car type single-cylinder research engine with direct injection and variable valve timing for both the intake and exhaust ports. For the present study two typical operating conditions, representing cases with considerably different CCV levels were selected to investigate their causes.

In the simulations the engine configuration was represented by a coupled 1D/3D-CFD model (optionally with unidirectional or bidirectional coupling), with the combustion chamber and the intake/exhaust ports (including parts of the runners) represented by a 3D-CFD computational model and the intake/exhaust pipework set-up adopting a 1D-CFD approach, enabling proper capturing of the pressure waves in the intake and exhaust ducts.

The LES flow model in the 3D-CFD FIRE simulation was based on the Smagorinsky approach, validated by comparing LES/PANS results with measurement data for idealized duct flow configurations.

Simulation of the flame propagation process was based on the LES extension of the coherent flame approach published by IFPEN, adopting proper modelling of the unresolved sub-grid scale strain and curvature effects. In the absence of a dedicated LES model in FIRE, spark ignition was not computed, and early flame kernel growth skipped. This was replaced by an ad-hoc approach consisting of prescribing the initial flame surface density in a sphere around the spark location, with its magnitude stochastically varied for the individual engine cycles in order to mimic the impact of sub-grid scale wrinkling effects on the early flame kernel growth rate. The allowable range of the initial flame kernel radii and initial flame surface densities was specified on the basis of detailed LES simulations of the flame initiation and flame kernel growth process, and exploiting the DNS study performed by UOWM. The former were performed as separate FIRE calculations prior to the 3D-CFD engine studies for a variety of flow and mixture composition parameter variations under idealized but engine like conditions, and then provided for their subsequent use during the engine simulation runs as correlation and in tabulated form, respectively.

The simulation of the liquid fuel spray propagation was based on a discrete droplet model, using suitable sub-models for atomization, secondary droplet break-up, evaporation and wall interaction. The spray characterization in terms of shape and tip penetration evolution was achieved via proper parameterization of the droplet initial conditions in terms of droplet size distribution and injection velocity based upon measured data available from one of the members of the project's Industrial Stakeholder Board.

Validation of the coupled 1D/3D-CFD BOOST/FIRE computational model and its parameterization was obtained via comparison of the calculated in-cylinder pressure traces for several consecutive engine cycles based on RANS simulations and comparison with the corresponding measured mean in-cylinder pressure traces.

Further validation of the adopted methodology was achieved by analysing the impact of different boundary condition treatments on the resulting flow field structure and by numerous sensitivity studies on the influence of numerical parameter variations on the calculation results accuracy. For the LES based study of the causes of cycle-to-cycle variations, the 1D/3D-CFD coupled simulations were run for more than 20 cycles for each of the selected operating points in order to obtain a statistically relevant set of information on the in-cylinder flow, mixture formation and combustion quantities. The validation studies and multi-cycle engine calculations were performed in close cooperation between AVL and JBRC using the coupled BOOST/FIRE tool.

Calculation results of the cycle-resolved flame propagation process clearly revealed that individual combustion cycles differed considerably from each other in terms of both flame shape and preferential propagation direction. It was clearly visible from the results that the differences in preferential burning direction and flame size and shape characteristics were already initiated at the very early beginning of the combustion process.

By comparing the calculated cylinder pressure traces with the corresponding measured data it could be clearly seen that the adopted LES approach was well capable of reflecting the degree of cyclic dispersion between the different cycles. When comparing the cycle-to-cycle variations of peak in-cylinder pressure and the crank-angle degree of its appearance with the corresponding ROHR analysis based on the measured pressure traces, it could be seen that the range of scatter was matched well, even on a quantitative basis. A similarly good agreement could be observed for the timing of 10%, 50% and 90% mass fraction burned.

To assess the sensitivity of the calculated cycle-specific rate-of-heat-release (ROHR) and cylinder pressure data on different engine and operating specific parameters, such as e.g. in cylinder flow field, residual gas content and mixture composition, the achieved computational results were analysed in detail. Since both the fuel and residual gas concentration fields at the spark plug location and also in major parts of the cylinder were found to be spatially well homogenized for all engine cycle, it was concluded that in the cases under investigation the cycle-to-cycle combustion variations were not caused by cycle-specific mixture composition variations.

In order to identify other potential sources of cycle-to-cycle variation, virtual experiments were conducted with respect to the ignition and early flame kernel formation process. These results clearly showed that when applying identical initial flame kernel radii and initial flame surface density values for all combustion cycle, the cyclic variability of ROHR and cylinder pressure got considerably reduced. Hence, it was concluded from these results that the flow field structure and intensity at the spark plug at the time of ignition strongly governs the cycle-specific initial flame kernel size and flame front wrinkling intensity, and hence exerts a strong influence on the cycle-specific early flame kernel growth rate. It was this mechanism that finally could be identified as a major source of the CCV in the DI SI-engine analysed in this study. Another important source of CCV is the interaction of local turbulence structures with large scale one(s) (e.g. tumble). This mechanism induces that even the motored regime (no combustion) shows relatively significant CCV of velocity vectors. In the studied engine, the interaction of the fuel spray with the engine aerodynamics was not found to have a significant impact on CCV, essentially as an early injection, low injection pressure strategy was used. This will probably not hold for other injection strategies and will have to be the subject of future work.

In addition, it could be demonstrated that the developed 1D/3D-CFD approach based on AVL BOOST and AVL FIRE is well capable of reflecting the underlying physical processes governing CCV in SI-engines and hence turns out to be a suitable tool to support future activities to analyse and optimize SI-engine performance with respect to CCV.

FEV performed LES studies of Gasoline Controlled Auto Ignition (CAI) combustion. CAI systems have a high potential of fuel consumption and emissions reduction for gasoline engines at part load operation. The controlled auto-ignition is initiated by reaching the thermal ignition conditions at the end of compression. The combustion propagation of CAI is essentially controlled by chemical kinetics, and thus differs significantly from conventional premixed combustion, since it enables efficient combustion even under very lean dethrotteled operation conditions.

Consequently, the CAI combustion process is determined by the thermodynamic state, and can be controlled by a high amount of residual gas, the gas temperature, and a certain stratification of air, residual gas, and fuel.

Due to the high amount of residual gas in the combustion chamber, the preceding cycle and its combustion has a high effect on the next cycle by directly affecting the thermodynamic state of the in-cylinder charge. In a DI-CAI engine this is not only due to the strong temperature dependence of the auto-ignition process, but also due to the effect of hot residual gas on the fuel evaporation, which generates cyclic fluctuations of the mixture formation and stratification. Stratification effects of fuel and residual gas directly affect the rate of heat release in CAI operation.

FEV's objective was to investigate the effects leading to CCV in a DI-CAI engine by using LES with the industrial CFD tool StarCD. The work scope therefore focused on extending the CFD code by specific sub-models. Among those was the development and implementation of a LES ignition model for CAI. The model is based on reduced chemical kinetics and accounts for the effect of air, fuel, residual gas and temperature stratification on CAI auto-ignition.

This enabled the analysis of the subsequent simulation results to improve the understanding of the sources of CCV in such engines, especially the role of stochastic perturbation of the in-cylinder charge and mixture distribution, and the role of preceding cycles, which directly affect the state of the residual gas content.

Furthermore, a LES injection sub-model for the DI injector, based on an experimental database obtained in an injection chamber, was implemented and verified. Finally a LES auto-ignition and combustion model for CAI was developed and implemented. This model is based on tabulated reduced chemical kinetics and accounts for the effect of air, fuel, residual gas and temperature stratification on CAI auto-ignition and kinetically controlled combustion process.

First validations with experimental results were conducted by simulating a CAI engine operation at 2000 rpm and 3 bar IMEP in CCR mode for which experimental data was available. It was shown that consecutive LES calculations can resolve cyclic fluctuations of the in-cylinder flow, residual gas mass fraction, mixture fraction and enthalpy. Thereby LES predicted a better mixture homogenization than RANS, which is assumedly effected by a significant higher level of charge motion predicted by LES in comparison to RANS. Fluctuations of the in cylinder mass after gas exchange were noticeable and may probably be caused by a high sensitivity of the calculated gas exchange to processes during intermediate compression. This will have to be further analyzed in detail in further investigations.

A first validation of the LES combustion model showed an acceptable agreement with- the test bench data, when the in-cylinder values are discretized into thermodynamic phase space dimension. A comparison between discretization in mass fraction vs. a discretization in enthalpy has shown that enthalpy classification leads to better results in terms of lower pressure gradients during CAI combustion, which agrees to the experimental observation.

Due to the high amount of residual gas in the combustion chamber, the preceding cycle and its combustion has a high effect on the next cycle by directly affecting the thermodynamic state of the in-cylinder charge. It was shown that this is not only due to the strong temperature dependence of the auto-ignition process, but also to the effect of hot residual gas on the fuel evaporation, which generated cycle-to-cycle fluctuations of the mixture formation and stratification. Stratification effects of fuel and residual gas directly affect the rate of heat release in CAI operation.

The pressure curves predicted by LES showed cycle-to-cycle fluctuations in all cases. Furthermore, the pressure gradients of the individual cycles were similar to the averaged measured pressure traces. When averaging the LES cycles, the agreement of the pressure gradient and peak pressure to the measured average values wre very good, thus validating the overall modelling approach.

4 – Capitalising LES knowledge in the form of CCV models for 1D-CFD

Another key element of the research strategy performed by the partners of LESSCCV is the proposal, based on the project's LES work, of CCV models for system and 1D CFD simulations able to study full engines, drivetrains and vehicles. The originality of this work was the usage of the detailed understanding gained in the LES to propose reduced models retaining the essential physics of CCV and to render its main effects, at a fraction of the costs of a LES. Such models can then serve as basis for supporting the development of designs and control strategies aimed at reducing the negative impact of CCV.

All the models developed within LESSCCV were based on the same principle: Their starting point were reduced combustion models for SIE as available in system simulation and 1D CFD codes.

One can distinguish two main types of such reduced combustion models:
1/ Mathematical models: They are simply based on using mathematical relations with numerical parameters that have to be fitted in order to reproduce experimental observations. Such models (as the well known Wiebe model) are not actually based on physical reasoning, and a priori they cannot predict engine combustion without any knowledge from experiments. Such models, once fitted, are nevertheless useful to provide predictions of combustion characteristics over a whole engine map based on some discrete experimentally studied points;
2/ Physical models: They are attempting to render as much as possible in a reduced way the complex physics of engine combustion. They are either based on phenomenological observations, or on the reduction of 3D models, and have the advantage that they a priori provide predictive capabilities, the parameter fitting effort being reduced as compared to purely mathematical approaches, even when applied to different engines. Nevertheless, these models are parameterised with modelling constants, but also with same numerical parameters that need to be fitted to reproduce accurately specific engines.

In practice, reduced combustion models can also be combinations of both approaches.

5 – Explore potential benefits from applying the developed CCV in engine and vehicle simulations

LMS implemented the CCV model developed by IFPEN in the AMESim platform and coupled it with the CFD1D library it had developed. LMS participated to the validation of the retained approaches through the completion of dedicated test cases for this new model, and defined a methodology to help for the model calibration and parameter setting.

Potential Impact:
To start with, all partners of the LESSCCV project prepared a statement concerning the impact of the LESSCCV project work and an outline of their exploitation, as follows:

AVL and JBRC

Better understanding of CCV phenomenon is the expected outcome of the project. After processing huge amount of data, it seems that global thermodynamic properties are not so important – local in-cylinder values are critical in terms of CCV (these are difficult to be modelled by 0D/1D tools). The CCV may seem to be a random process, however it is not the case – strong cross-correlations may be present. Turbulence properties and initial flame kernel development were identified as the most critical parameters in terms of CCV.

FEV

As an outcome of this research work, simulation tools have been developed that can be further used to predict CCV in CAI engines. The tools serve as a means for advanced research, allowing to study in detail specific topics of the microscopic interactions between turbulence, liquid injection, mixing and combustion in the combustion chamber of piston engines. Here, the full three dimensional resolution of the in-cylinder flow field, fuel mixing, and chemical reactions is available as an in-depth research and analysis tool for future engine development work.

IFPEN

In terms of LES applied to piston engines, the LESSCCV project has allowed IFPEN to further develop its research 3D CFD code AVBP in terms of coupling to a 1D CFD code, validating a simulation methodology for predicting CCV in a PFI-SIE, and developing a direct fuel injection model. A first exploitation of these achievements is to use it as a basis for further research in the domain of piston engine LES, and to widely publish the achieved results in reference international journals and conferences. This will be achieved in collaborative funded French research projects, and during a newly created task in the frame of the Groupement Scientifique Moteur (GSM) that brings together IFPEN, PSA Peugeot-Citroën and Renault. The aim of this research will be to build on the achievements of LESSCCV and to extend the application domain to cover other non-cyclic aspects of SI combustion, and in particular abnormal combustion and their link to CCV, wall film formation in fast transients of GDI engines, or fuel spray/aerodynamics interactions in GDI engines.

LMS

The involvement of LMS in the LESSCCV project permitted to strengthen its scientific network. LMS and IFPEN have been working on the development of internal combustion engine modelling features in LMS Imagine.Lab AMESim for almost 10 years and the LESSCCV project allowed a progress in the modelling capabilities of the software. First of all, LMS was able to improve its 1D CFD solver for gas dynamics simulation. The upgraded numerical core and the new orifice/throttle physical models were pushed in a new version of the CFD1D library in the standard product for the release 11 delivered to the market in 2012. This increased the tool capacity and hence the potential business related to the application of AMESim for engine design activities at car manufacturers and suppliers. On the other hand, the AMESim platform gets benefit of the development of the upgraded CFM1D-CCV model by IFPEN which would be implemented in standard as soon as in 2013.

PoliMi

The results achieved during the LESSCCV project are going to be exploited in different ways during the next years. First of all, a few journal publications will be prepared in collaboration with the partners to disseminate the main scientific advances. Moreover, the involvement of PoliMi's group in this EU project on LES and CCV has promoted a new growing interest for this field in Milan, so that currently two researchers are actively working in this area, which will represent a main topic in the future. Stable collaborations have been developed between PoliMi and most of the partners, both industrial and academic, which will be fundamental to increase the research work on LES (and DNS) applied to I.C. engines carried out in Milan. Moreover, 2-3 new PhD students from Italy and EU/extra-EU countries are going to begin their research activity on LES for I.C. engines in the Department of Energy during the next year.

RICARDO

The exploitation of the Ricardo work concentrates on the use of the developed models in commercial software tools which are licensed to OEM vehicle manufacturers across the world. For example the 1D-3D co-simulation between the Ricardo WAVE 1D CFD tool and the AVL FIRE 3D CFD tool is available now in the latest commercial releases of these tools.

The new CCV combustion model work is planned to be included in a future commercial release of WAVE tool with further test examples, documentation and marketing material. The theoretical detail behind the new CCV combustion model has already been published in public as a presentation at the LES4ICE Conference Nov 2012. There is also an intended peer-reviewed paper (combined with IFPEN and PoliMi) for a paper comparing the CCV combustion model approaches applied to the SGMmac engine measurements.

UOWM

The results obtained during the DNS work contribute to provide an improved understanding and insight on the flame displacement speed for the case of a propagating premixed stretched flame front. This will assist the scientific community to bridge the gap between experimental and simulation data in premixed flame propagation, by a consistent definition of the flame displacement speed and the clarification of the proper location inside the flame that the Markstein length should be calculated. “Ambiguity” on this subject has been reported by several authors in the literature and many of the published papers furnish this matter. Furthermore the results can be used towards the development and validation of improved spark ignition and early flame kernel development models for LES codes used in engine simulation. Moreover, the developed DNS tool based on the nek5000 platform, which has been developed and extended during the project, can be used in the future by researchers to study flame-turbulence interactions of propagating flame fronts in a turbulent environment.

In summary, the exploitation of the research work performed in the LESSCCV project and of the achieved results, as well as their potential impact, can be outlined in terms of three main categories:

- Innovative CFD tools:
The LES tools that were developed within the project are at the forefront of CFD research in the domain of ICE combustion worldwide, as exemplified in the LESSCCV presentations held in the LES4ICE'12 workshop at the project end, and with more publications to come. They indeed represent a breakthrough in terms of their demonstrated ability to address non-cyclic phenomena not accessible to today's standard industrial CFD codes.

- Contributions to greening engines and vehicles:
The developed CFD tools and methodologies for accounting for CCV in SIE, and their exploitation in commercially available software tools largely used in the European automotive industry, could have a positive impact on the environmental performances of the future engines that they could contribute designing and optimising.

- Advancement of combustion research:
In terms of fundamental research, the results generated by LESSCCV did not yet lead to publications in journals referenced in the WOS, but different SAE papers have been published, which albeit not referenced as a refereed journal, is submitted to a review prior to publication, and receives a wide audience in automotive research and industry worldwide. Furthermore a number of communications were performed by the partners in different scientific conferences, allowing to widely disseminating the research work performed in LESSCCV. At least 2 publications in refereed journals are also planned after the project end.

The exploitation of the results from LESSCCV shall therefore bring benefits for the European Union concerning different societal and economic aspects:

- Competitiveness of European Industry: In the long-term, the performed research and achieved results will support the European automotive industry in proposing highly innovative products with outstanding environmental performances that are essential for supporting and further improving European competitiveness in a highly competitive domain. Another domain where LESSCCV can be expected to have a positive long-term impact is simulation software. Europe is amongst the leaders worldwide, and the innovative and advanced software tools that shall be commercially exploited after the project end will definitively contribute reinforcing that position.

Overall, LESSCCV thus shall have a positive impact in terms of employment in Europe, both in the research and industrial sectors, although at this point they are impossible to estimate.

- Contributing to a durable and greener road transport: By supporting the development of greener, more efficient and cleaner, engines and powertrains, LESSCCV will have a long-term positive impact on the development of elements for a sustainable and greener European road transport. Projections from ERTRAC in 2012 indeed indicate that by 2030, as much as 85% of powertrains for European road transport will be based on ICE. This underlines the importance of further improving their concepts, which in particular will be achievable by addressing phenomena so far not directly accounted for. Amongst those are the CCV subject of LESSCCV, but also many other non-cyclic phenomena it will be essential to understand and master in order to reach the goal of a cycle to cycle optimal control of ICE combustion. The work on LES and its capitalisation in system simulation in LESSCCV can be expected to be one key tool to reach this goal.

- Contribute to improve the health of citizens; indirectly, by contributing to further reduce pollutant emissions and increasing fuel efficiency of the still widely used ICE, LESSCCV shall also contribute to reduce overall the negative impact of road transport on the health of European citizens.

List of Websites:

http://projet.ifpen.fr/Projet/jcms/c_8580/lessccv