Community Research and Development Information Service - CORDIS


DYNAMO Report Summary

Project ID: 620180
Funded under: FP7-JTI
Country: United Kingdom

Final Report Summary - DYNAMO (DYNAMO Design methods for durabilitY aNd operAbility of low eMissions cOmbustors)

Executive Summary:
Lean burn combustor systems are a key technology to reduce NOx emissions for future aero engine gas turbines. The ability to maintain the desired combustor metal temperature is critical to achieving acceptable durability. The levels of fuel-air premixing inherent in lean burn designs makes them susceptible to thermo-acoustics instabilities which will have a drastic impact on the durability of the combustor.

The overall aim of this project is to develop validated methodologies for the prediction of combustor temperature and thermo-acoustics instabilities to allow confident design of the combustion system of a demonstrator engine at Technology Readiness Level 6.

The first work package focussed on cooling and radiative heat transfer. It used Computational Fluid Dynamics to highly resolve the flow in two types of complex combustor liner. These were then used to create and calibrate two sub-grid scale models for the CFD that allowed the effect of pedestals and impingement-effusion to be represented without explicitly resolving the geometrical details. This allowed the simulation of the cooling tiles with significantly less computational resource requirements making it suitable for rapid design calculations. Radiative heat transfer to the combustor is strongly influenced by the soot concentrations close to the fuel injector, and simulations were carried out to test the sensitivity of this radiative load to the choice of models used to represent combustion and soot production. This was also used with a conjugate heat transfer analysis to compute the metal temperature of part of the fuel injector. The overall results were found to be relatively insensitive to the modelling approach chosen.

The second work package further developed a design tool for the automated analysis of a combustor design. This used links to CAD software to automatically create the fluid volume, generate the CFD mesh, apply network determined boundary conditions and finally calculate the fluid flow and post-process the data. The main focus here was the development of coupling to a FEA system for mechanical stress and thermals in order to also automate these tasks. The accurate knowledge of metal temperature in the combustor is the key to determining durability. Also incorporated was an automatic link to an external network model representing the combustor cooling tiles. The overall process was demonstrated on several combustor geometries.

The final work package focussed on the thermoacoustics behaviour of the combustor. This had two strands: the first looked at an isolated, isothermal single fuel injector and its aerodynamic response to a plane wave. This was validated against experimental data and compared to previous simulations using a different CFD code. The work was then extended by considering how the spray droplet size and velocity varies due to the varying flow speed in the fuel injector. This took an existing droplet model and modified and calibrated with experimental data. A real combustor contains of multiple fuel injectors and includes reacting flow. This second strand looked at different modelling approaches for this problem, in particular comparing Unsteady RANS and Large Eddy Simulation for forced and unforced annular simulations. LES was found to be preferable as it fully captured the unsteadiness, particularly of the flame.

Project Context and Objectives:
Lean burn combustor systems are a key technology to reduce NOx emissions for future aero engine gas turbines. The ability to maintain the desired combustor metal temperature is critical to achieving acceptable durability of the engine. However, the high overall pressure ratios of the compressor and the large air mass flow through the combustor fuel injector results in limited amounts of cooling air, which itself can be relatively hot. The levels of fuel-air premixing inherent in lean burn designs also makes them potentially susceptible to thermo-acoustics instabilities, and this will also have a drastic impact on the durability of the combustor. The overall aim of this project is to develop validated methodologies for the prediction of combustor temperature and thermo-acoustics instabilities to allow confident design of the combustion system of a demonstrator engine at Technology Readiness Level 6.
To achieve this, the project is composed of three technical work packages: Cooling, Smart System for Thermal Analysis of Combustors and Thermoacoustics.

The overall objectives are:

Cooling Workpackage
Development of a sub-grid scale model for RANS CFD prediction of pedestal combustor tiles based upon experimental data and reference CFD solutions.

Development of a sub-grid scale model for RANS CFD prediction of impingement-effusion combustor tiles based upon high resolution Large Eddy Simulation CFD solutions.

Assessment of sensitivity of radiation heat transfer to combustion model, soot model and radiation model.

Smart System for Thermal Analysis Workpackage
Object oriented implementation in a CAD environment of tools to encapsulate current best design and analysis practice so as to support an automated analysis workflow for gas turbine combustor design. Linking of 1D flow solvers, two and three-dimensional CFD meshing, URANS flow solvers, Paraview post-processing and FEA based thermal and stress analysis.

Thermoacoustics Workpackage
Simulation of thermoacoustic instabilities in a gas turbine combustor through the prediction of the response of fuel injectors to acoustic waves and the unsteady behaviour of a full annular combustor. Evaluation of compressible unsteady RANS simulation method compared to Large Eddy Simulation techniques so as to ascertain the trade-off of potentially improved accuracy versus computational cost. This will require:

Determination of the response of the flow from a fuel injector due to acoustic plane waves in an isothermal, atmospheric environment as measured experimentally.
Incorporation of experimental spray response as time dependent boundary conditions to compute an improved Flame Transfer Function.
Demonstration of the ability to capture fully-coupled thermoacoustic interaction in the annular chamber, involving circumferential modes including both propagating and standing waves.

Project Results:

Pedestal Tiles
Both RANS and LES CFD have been used to simulate pedestal tiles. The results are very similar in the impingement region and again to the right as the flow progresses through the pins, however, in the intermediate zone the flow is quite different. Experimental data is only available at two stations where the solutions are quite similar, although in general, the LES is better than all the RANS solutions. The pedestal tile model can be used to replace all the pins, but in doing so the complex flow in the feed and impingement is lost. A hybrid approach is proposed where the impingement region pins are resolved and the remainder modelled. The hybrid approach matches the fully resolved solution between the first and second pin and then the fully modelled for the remainder of the domain.

Impingement Effusion Tiles
Impingement-effusion cooling is an alternative to the pedestal tiles discussed in the previous section. Calculations have been carried out using RANS, SST-SAS and LES modelling approaches. The finest meshes for LES contain 40 million cells.
LES captures the highly unsteady flow within the tile. The flow in the feed (top part) contains no vorticity, and no turbulence, however, the impingement of the feed hole jets onto the hot side wall creates large amounts of turbulence and recirculation, which is then discharged through the lower angled effusion holes.
There was relatively little experimental data for this case, but comparisons of mass flow through the effusion holes for two LES grid and experiments shows good consistency. In addition, comparison of mean and fluctuating velocity within the impingement gap shows only a small difference between the SST-SAS and the LES results.
The LES data has then been used to help calibrate the sub-grid scale model. The sub grid scale model avoids the requirement to mesh each hole of the feed and effusion in the tile. As an example this grid using the sub-grid scale model has 20,000 cells in comparison with 20 million cells for the fully resolved LES. When comparing static pressure across the tile, this modelled approach gives excellent agreement with the LES.

Radiation Modelling Recommendations
The parameters that were varied in order to investigate sensitivity were:
• Sauter Mean Diameter (SMD)
• Sensitivity to chemistry
• Conjugate heat transfer
Grids were generated using ICEMCFD with a dominant hex core approach. The mesh sizing parameters were based upon best practice approaches supplied by Rolls-Royce. In total the mesh contains 25 million cells.
For this study both steady and unsteady incompressible equations for the reacting flow are solved using PRECISE-UNS. These include the basic mass, momentum, enthalpy transport equations and the mass conservation equation for species. Combustion is simulated with the Flamelet Generated Manifold Model (FGM). This model assumes that the flame structure is unique. Therefore, it can be computed using a laminar flame code using detailed chemistry. This is done using the laminar flamelet code CHEM1D using the boundary conditions of temperature and pressure provided by Rolls-Royce Derby. Thus, two operating conditions are considered in this study, namely, Maxi Take-off (MTO) and Top of Pilot (TOP), respectively. We made use of the Imperial College (IC) model for the modelling of soot. Finally, turbulence is computed using the RNG k−ε model. The velocity components are evaluated through the second-order linear upwinding scheme. The second-order Gamma scheme is used for progress variable (pgr), variance of the progress variable (vpgr), mixture fraction (fmi), variance of mixture fraction (gfmi), enthalpy (h) and the turbulence variables respectively. For URANS, a time step of t= 1 μs was used. To study the sensitivity to chemistry two different mechanisms were utilized.
A considerable number of calculations have been carried out exploring the design parameters defined previously. These have been used as input to the radiation model and a conjugate heat transfer calculation for a portion of the fuel spray injector (Figure 16) in order to predict metal temperatures. The radiative heat transfer was found to be relatively insensitive to sensible variations in the chosen parameters.

Smart System for Thermal Analysis:
Lean Burn Aerothermal Network Capability
The was primarily concerned with the development of an aerothermal network generation capability for lean burn combustors. Prometheus automatically generates such networks by post-processing an automatically generated CFD fluid volume. This volume is itself generated by Prometheus from a provided combustor CAD assembly (or sketchbook) consisting of the relevant combustor casing and injector geometries.
Prior to the DYNAMO program the Prometheus design system was capable of dealing only with rich burn combustors. In order to cope with lean burn combustors, which are topologically different in a number of respects to rich burn combustors, a new lean burn combustor class was created within the Prometheus software. This class inherited all of the generic feature recognition capabilities developed for rich burn combustors but at the same time offered considerable scope to include new geometry recognition and manipulation capabilities specifically for lean burn combustors.
The initial task was to therefore extend the hint based feature recognition routines embedded within Prometheus to cope with a number of new features found within a lean burn sketchbook to successfully generate a CFD fluid volume. This included coping with, for example, staggered combustor tiles, a parametric combustor cowling and most significantly a completely new fuel spray nozzle topology. With routines defined to correctly identify surfaces from the CAD parts to include within the fluid volume an additional set of geometry manipulation routines were developed to cope with the presence of additional outlets on the combustor rear inner casing (CRIC), generate additional post-processing planes within the annuli and combustor exit and a parameterised inlet plane.

The automatic creation of the lean burn aerothermal network follows an almost identical process to that for a rich burn combustor. The feature identification routines run to generate the CFD fluid volume classifies the surfaces of the combustor into groups which can be post-processed to create the various features defining the aerothermal network. A rolling ball search is performed along the inner and outer combustor annuli, prediffuser and flametube from which duct heights are calculated. The topology of the network representing the injector is selected depending on the nature of the combustor cowling, for example, if the cowling is present or not, and the network parameters for the injector are populated with quantities extracted during the fluid volume creation. Finally network “bits” are included to represent the effusion cooling on the tiles with the number, size and angle of the holes being extracted from the combustor preliminary design spreadsheet.

Lean Burn CFD
Having developed an automated CFD fluid volume generation capability this was extended to automatically generate the scripts to mesh this volume, run and then post-process the CFD simulation.
The hint based feature recognition process employed during the fluid volume generation not only identified surfaces present within the volume but used this to extract important information driving mesh generation. The location of the fuel spray nozzle, the height of the passages within the fuel spray nozzle and the bounds of the flametube, for example, are all employed in generating an appropriate ICEM meshing script according to the inbuilt meshing best practice.
In this case an ICEM script is generated to first convert the fluid volume NX part file into an ICEM .tin file. Density boxes are placed around the injector with the number of boxes and their size automatically defined to enable a smooth transition from the cells within the fuel spray nozzle to the global mesh size. Cells within the passages are sized to ensure the requested minimum number of elements is produced across the smallest passage gap. In a similar manner, a series of density boxes are positioned downstream of the fuel spray nozzle spanning the flametube between the two periodic planes. The number of these boxes is automatically defined to ensure that the mesh refinement is only within the flametube and not the annuli and is therefore dependent on the sector angle or number of burners and the thickness of the combustor skin and tiles on the inner and outer walls.

With mesh refinement zones automatically defined the script generates a tetrahedral mesh and then performs a hexahedral conversion and a series of smoothing operations. An automated script creation process for lean burn CFD simulation was also developed. These scripts are automatically created using the mass flow splits parsed from the aerothermal network analysis and information on the effusion holes, inlet pressures and temperatures and fuel mass flows etc. from the combustor preliminary design spreadsheet. This results in three separate CFD simulation scripts for the proprietary CFD solver. The first script runs a RANS simulation of the lean burn combustor for 6000 iterations. This is followed by a second script which runs a URANS simulation for 4000 time steps. Finally a third script is called to run the simulation further and post-process radiation for use in any future thermal simulation.

In addition to the creation of the mesh and running the CFD simulation,a number of post-processing utilities were developed to automatically extract important results from, not just a lean burn simulation, but any Prometheus generated CFD simulation. Based around Paraview these scripts calculate pressure losses, extract parameter values from the mesh on a set of named surfaces and extract near wall properties at a specified near wall distance. Both of the scripts extract surface properties to a point item file, a file of unstructured nodal coordinates and corresponding parameter value, for example, temperature, pressure, heat transfer coefficient or incident radiation. This file format enables these parameters to be mapped onto thermal models.

Lean Burn FEA
This aimed to develop an entirely new capability and automate FEA within the system. Given the goal of a generic combustor design system, i.e. one for both lean and rich burn combustors the developments within WP3.3 attempted to ensure as much generality as possible. While lean burn systems are the focus of this DYNAMO WP the FEA automation framework was developed to ensure it could be applied to not only rich burn system but individual combustor components and even other engine components.

The FEA analysis process commences from either an individual CAD component part or a database of parts which could include the combustor sketchbook used in the above CFD fluid volume creation. These CAD parts each have a set of FEA boundary conditions stored as either face or solid body attributes describing convecting zones, voids, applied forces etc. The Prometheus system then automatically combines these parts into a simulation assembly. During the creation of this assembly, boundary condition parameters can be updated if necessary, joint faces are automatically determined and any necessary splits are created and the geometry is sectored if required. The simulation assembly is then automatically meshed and a corresponding FEA journal file is created.
The automatic creation of joint faces (either mechanical or thermal joints) also defines a set of meshing constraints to ensure that the nodes on both sets of faces match. Additional local control of the mesh size can be specified by the user through edge, face or body attributes with global mesh scaling factors provided to permit mesh convergence studies to be performed with ease. The sectoring operation not only manipulates the geometry but automatically ensures matching nodes between corresponding periodic faces.

With the FEA mesh generated the automatically generated journal file can be used to load in additional information extracted from other simulations. Unstructured maps of temperature, pressure or incident radiation produced using the scripts developed previously can be imported into the model and used. Likewise, lower fidelity, constant pressures or axial variations in pressure, derived from the aerothermal network model, can be used to define pressures on some of the casings for example. Ultimately the FEA workflow concludes with a successful thermal simulation.
Although the above framework is illustrated using a combustor simulation the processes developed are completely generic and can be used to link any CAD part to not only a thermal analysis but a mechanical or even a thermo-mechanical analysis.
The CFD fluid volume generated for a tiled lean burn combustor contains no explicit representation of the effusion holes or the cavity between the tile and the combustor skin. A prediction of the flow in this region is important in any thermal analysis of the combustor to ensure that the heat transfer coefficients (HTCs) are correct. A proprietary Digital Impingement Effusion Tile (DIET) tool creates and solves a “2D” aerothermal network for this cavity flow and produces a graphical item file of, for example, HTC which can be used in subsequent thermal analyses. However, prior to the DYNAMO project the creation of the inputs to DIET, which required information on the location, angle and size of each impingement and effusion hole as well as the location and size of other geometry features such as the pins, spacers side rails and overall dimensions of the tile, had to be created entirely by hand. Any automated combustor thermal analysis capability therefore required a means of automatically generating DIET inputs from a CAD file of the tile and the cold skin. Typically the combustor sketchbook geometry is not of sufficient fidelity to include either an explicit representation of either the effusion or impingement holes. To create this geometry the proprietary tile standard feature toolkit was used. With an appropriately detailed representation of the tile and skin additional Prometheus functionality was included to employ the hint based feature recognition routines to automatically generate the inputs to DIET.

With a generic framework in place for automating FEA simulations the ability to automatically create, run and post-process quantities like HTC and incident radiation from a CFD simulation and the ability to automatically generate inputs to DIET in order to calculate tile cavity properties, a thermal simulation of either a single tile or the complete combustor could be performed.


Single Fuel Injector Response
A simplified orific problem was initially studied as it contains many flow features of the full fuel injector, but removes the complication of meshing the complex geometric detail of a fuel injector. This allowed the testing and development of numerical procedures within the CFD code and comparison with experimental data and previous simulations in a different CFD code. Steady acoustic fluctuations are then imposed at the downstream boundary and the response of the observed velocity fluctuation through the orifice is characterised as an impedance. These were then compared to previous calculations as a verification test.

A realistic fuel injector was also been simulated using a high resolution LES with the WALE sub grid scale turbulence model. Comparison of the predicted impedance in was made with an older OpenFOAM simulation and experimental data. For resistance, the OpenFOAM simulations are in very good agreement with experiment, whilst the CFD simulations underpredict at the medium Strouhal number and over-predict at the higher Strouhal number. For reactance, both simulations show the correct trend of increasing with Strouhal number, but the current CFD shows an overprediction of the level and whilst the older OpenFOAM results show an under-prediction.

Experimental Spray Response
A dynamic spray injection model based on local flow velocity was devised to incorporate the changes in droplet distribution with variation in flow due to acoustic fluctuations. Two different approaches were investigated. The first is based on a single probe where a change in the distribution of the droplet is based on fluctuation at one particular point. The second approach is based on multiple circumferential points in an equally spaced ring-shaped injection region. Although multi-point injection in a ring-shaped region provides a realistic estimation of local flow variation across the circumference, a better understanding of the effect of changes in swirl ratio, velocity magnitude and liquid properties on spray droplet distribution can be obtained from the single probe approach. In addition, experimental data has been collected in a previous project for a particular location using Phase Doppler Anemometry (PDA) for acoustic forcing. Once a clearer understanding of changes in droplet distribution due to various parameters is obtained, these estimations can easily be replicated in the multi-point ring-shaped injection approach. The coefficients of the modified Gepperth model is scaled according to various experiments conducted at number of circumferential points at PDA facility in Loughborough.
The simulation relates how a variation in local velocity will lead to a variation in droplet sizes. The model is then calibrated so that the experimentally observed distributions are obtained.

Fully Annular Rig
The rig contains 18 burners in total, with options also for 12 and 15-burner configurations. Each of the burners contains a bluff body with a conical end. The exit blockage ratio is set to 50%. A six-vane swirler is fitted around each bluff body at a distance of 10mm from the exit of the burner. The fuel is ethylene which is supplied premixed with air at atmospheric pressure and temperature. A CAD representation of the entire 3D annular geometry was constructed in a modular fashion to allow individual components to be extracted as required. Meshing was completed using the ANSYS-ICEM mesher. The unstructured mesh included refined areas around salient features such as the injectors, bluff bodies and swirlers.

Single Burner Investigation
About four times more cells were required for a geometry with a swirler than for one without it.
The aim was then to calibrate a swirl boundary condition to simplify the model and save on mesh and computing resources. The geometry was cut above the plenum to focus on the parts most affected by swirl. Using the k-ω SST turbulence model, the new geometry was run both with and without a swirler included. The inlet velocity was chosen to match the experimental outlet velocity of 18 m/s. A cold flow case was run for each model using steady RANS, and results were obtained for the swirling and non-swirling configurations.

A swirl inlet velocity boundary condition was applied at the exit plane of the injector, and cold flow simulations were run without a swirler included in the geometry. The swirl inlet boundary condition does not account for the swirler vane wakes observed in the case with an actual swirler included, and the importance of the wakes in cold and ignited cases is left to future work. Similar simulation settings to the previous swirling flow cases were used. The flow was found to behave as expected with accelerations at the area contractions out of the injector and decelerations through the chamber. The observed exit velocity of around 18 m/s is comparable to the previous cases as well as to the available experimental results.

Full Burner Investigation
The original CAD for this case was cut at the same plane as the single burner case above. Meshing produced a total cell count of 1,035,499. This mesh was used to run isothermal cases for both swirling and non-swirling inlet velocity conditions. Each of the inlets was treated separately with the swirl originating from the centre point of each burner at the inlet plane. Cold flow simulations were run with similar inlet conditions to the single burner cases. The inlet temperature was set to a constant 300K, pressure was atmospheric and an inlet axial velocity of 18 m/s was defined for the non-swirling case at the inlet plane. For the swirling case, the same inlet velocity components were used as for the single burner boundary condition. For both cases the flow features were well represented, and the swirl physics was accurately captured.

A premixed ethylene-air mixture was used for a reacting case, as in the previous experiments, with an inlet temperature of 300K. No acoustic forcing was prescribed. A swirl velocity inlet was defined. The instantaneous LES simulation captures the asymmetry present in the experimental data, while the RANS results remain relatively symmetric.

The results from the RANS full burner reacting case presented in the previous section were used here to initialise the flow. A time-dependent oscillating velocity boundary condition was imposed at each of the inlets. An amplitude of 0.65 and frequency of 160Hz were defined in order to directly compare the results with LES data provided by Dr Chin Yik Lee.
Flame front rollup is visible during the oscillating cycle. This is due to the high oscillation amplitude. As in the cold flow and unforced reacting simulations, the LES seems to capture the asymmetry present in the flow whereas the URANS results remain fairly symmetric.

Potential Impact:
The work undertaken in the DYNAMO project is focussed on increasing the durability of future low emissions gas turbine combustion systems. These low emission systems have a significant reduction in NOx which is a major contributor to local air quality at low altitudes and global warming at high altitudes. The work undertaken here reduces the risk of these low emission systems having unforeseen durability problems. For an aircraft engine, lack of durability can lead to increased maintenance costs and departure delays, and in the extreme case, if leading to combustor failure a loss of life.

In particular, three aspects of the work allow better design of low emissions combustors and hence accelerates their in service date:
Improved analysis of combustor liner cooling due to the CFD modelling improvements and a better understanding of the sensitivity of radiative heat transfer.
An automated analysis tool for the design of a combustor that includes fluid and thermal analysis.
Improved capability for thermoacoustic analysis of combustors.

The main dissemination activities so far have been presentation at conferences and workshops. These have included the Greener Aviation 2016 conference in Brussels. More specialist dissemination events have been held including presentations at annual reviews of research to industry. Several journal articles are under preparation and one has been accepted. None are currently recorded in Template A1 as these have not been published yet and do not have a DOI.

Exploitation is via the industrial partner who has access to the results and source code of the design tools. These are expected to be rapidly implemented into their design process.

List of Websites:
Professor Gary Page, Chair in Computation Aerodynamics +44 (0)1509 227205


Christopher Malins, (Grants and Contracts Officer)
Tel.: +44 1509 222421
Record Number: 197231 / Last updated on: 2017-04-10
Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top