FUEL MODELS FOR GAS TURBINE DEVELOPMENT

Published on: 

BY USING ACCURATE FUEL MODELS, COMBUSTOR DESIGN EFFECTIVENESS CAN BE IMPROVED

Gas turbine engine designers face design challenges as the fuels landscape continues towiden.Aviation gas turbine regulationswill force dramatic reductions in pollutant emissions at both high and low power conditions in the air and on the ground. Power producing gas turbines are continually being asked to operate cleanly at low power conditions due to the addition of other variable power sources such as wind and solar. Achieving these goals requires a thorough understanding of how to design combustors to operate with both conventional and modern fuel options.

Computer simulation using Computational Fluid Dynamics (CFD) is a cost effective way to test new design techniques, but it must incorporate accurate fuel models; otherwise, simulation has limited usefulness. Those using improved fuel models are seeing improvements in the accuracy of their combustion simulations, allowing them to test a wider range of fuels and designs before building a prototype.

Jet fuel, for example, contains hundreds of different compounds that all contribute in specific ways towards ignition, flame propagation and pollutant formation. Similarly, Liquefied Natural Gas (LNG) has higher amounts of hydrocarbon gases that are heavier than methane and this can negatively affect combustion performance. The development of better fuel models and simulation tools, therefore, can facilitate better computer-aided engine and fuel design for petroleum and alternative fuels.

The promise of CFD

The early promise of CFD-based design approaches was that the predictable nature of the calculations would enable studies of multiple combustor designs prior to prototyping. As part of the development flow, simulation could predict important system attributes such as stability and emissions performance and reduce the need for costly experimental testing.

Results vary widely, however, among organizations. Further, engine manufacturers are spending more on computer hardware and software licenses to feed larger, more complex combustion CFD simulations with little tangible evidence of useful, predictive results for key design criteria such as CO and unburned hydrocarbons (UHC) at low load with non-traditional fuels and blends. This has given rise to greater use of experimental testing and higher demand for overbooked test facilities.

These software tools worked wellenough when the objective was to simply drive the combustor’s flame temperatures down in an effort to reduce NOx. Now that ultra-low NOx combustion equipment dominates the market, the industry is facing a completely new set of challenges such as Lean Blow Off (LBO), ignition, flashback, fuel flexibility and emissions. That necessitates a better understanding of how fuel models are used in combustion simulation.

Combustion is typically envisioned as a single step where fuel and oxygen burn to produce carbon dioxide and water with some pollutants. The reality is that when fuel burns, it undergoes a transformation consisting of thousands of chemical reactions with shortly lived radical species. These dictate combustion performance such as ignition, extinction and pollutant formation.

In addition, advanced combustion devices have different conditions than legacy systems due to increased flue gas recirculation (FGR), lean premixing and other combustion staging techniques. As an example, the GRI-Mech model was developed and validated using 1990’s era boiler technology and works quite well for that application. However, it has proved deficient in pre-dicting NOx emission from newer gas turbines and burner systems.

Some have chosen to use combustion models in conjunction with look-up tables or progress variables that simplify the complex chemistry. This allows the use of slightly more detailed fuel models, but the simplifications that occur result in poor prediction of emissions and combustion stability. These problems are typically addressed by using a simple kinetic post-processor that predicts well when the majority of the NOx is thermally generated. However, most ultra-low NOx combustors have already had thermal NOx substantially reduced leaving other pathways such as prompt and third-body as substantial contributors to the total NOx emissions.

Ignition delay is a critical parameter that is required in order to determine the location of the flame and combustion stability. Figure 1 shows some ignition delay results obtained using the GRI-Mech model for natural gas. It is compared against a more accurate model created from the Model Fuels Consortium (MFC) database. The X-axis on the figure is the inverse of temperature, so low flame temperatures present in combustion devices are represented on the right side of the curve. The model fails to predict ignition delay for pressures that are typical of gas turbine combustors. More importantly, the ignition delay predicted shows its greatest error in conditions that are present in ultra low NOx combustion on the right side of the curve. GRI-Mech proves successful only at 8 atm for higher temperatures that are typical only of older combustion systems.

NOx results comparing GRI-Mech to improved chemistry models are shown in Figure 2, comparing NOx predictions over a pressure range comparable to that of most gas turbine combustors. As with the ignition delay results, GRI-Mech is successful in predicting NOx at low pressures that correspond to the conditions present in older boiler and burner designs (i.e., high flame temperature and lowpressure). It under predicts NOx in designs because it lacks the details required to reflect all of the ways that NOx can be produced. The MFC chemistry model predicts NOx over this pressure range showing how recent advances in combustion kinetics can substantially improve emissions predictions compared to older models.

Advertisement

Flame speed is another key measure of the quality of a fuel model. Figure 3 shows an MFC chemistry model compared to regular models for liquid Jet-A fuel.

Creating CFD fuel models

Accurate fuel models can contain thousands of species and tens of thousands of reactions. Unfortunately, CFD is only able to handle fuel models with less than 50 species for direct solution of chemistry or up to 100 species if a less direct chemistry solution such as a flamelet table or progress variable approach is used. Many techniques have been developed to reduce the size of fuel. Software tools have automated these various techniques to allow reduction to be focused on a specific target (e.g. ignition delay, flame speed, NOx, and so on.) within certain error tolerances. This can introduce errors as species and reactions are removed from the model.

This can be illustrated by looking at model reductions for a liquid fuel model for the popular alternative jet fuel produced by converting coal to liquid with the Fischer-Tropsch process. The full fuel model contains 949 species and provides excellent agreement with the fundamental kinetic measurements of flame speed, ignition delay and emissions formation. Using a software tool and defining the acceptable error margin of 5%, various emissions reductions of different sizes can be generated. Reductions in size from the full fuel model of nearly 90% can be obtained while managing error to less than 5%. It is interesting to note, however, that the error for emissions predictions increases when the reduction is continued further to reach sizes that CFD can accommodate.

The bottom line is that fuel models can be constructed and reduced while maintaining a reasonable level of accuracy for flame shape and location, but the inaccuracies in emissions predictions introduced make CFD impractical. Accurate fuel chemistry can be applied with Equivalent Reactor Networks (ERN) to combustion problems.

Emissions predictions have been achieved for over 30 years through the use of idealized chemical reactor modeling using chemistry simulation software packages such as Chemkin. But these packages do not take into account effects of the complex 3-D flow field and geometry. Building ERNs to represent the local chemical reactions in appropriate regions of the geometry is a proven method of incorporating the effects of both the flow and the kinetics in a single simulation.

It is critical that the ERN be a true representation of the actual combustor flow field in order for the simulation to be useful. Once the ERN is created through a careful devolution of the combustor flow field, a detailed chemistry model can be used to provide an understanding of chemical behavior and performance. Using a CFD case as the basis for ERN creation dramatically improves the quality of the results. Traditionally, expert personnel would be required to create the ERN manually; a process so time consuming that it is not practical in the design process.

Automating the creation of the ERN, therefore, can eliminate some of the challenges in using this approach. Software such as the Reaction Design Energico simulation package can minimize the time it takes to build ERNs from a reacting-flow CFD solution. A series of filters are applied to the CFD or user-defined variables to generate the ERNfor prediction of combustion performance, including exit emissions. Once the ERN is created, the resulting fuel model can be used to predict the emissions of trace species such as NOx, CO and unburned hydrocarbons (UHC). The ERNcan also be employed in a parametric variation of operating conditions and fuel composition to determine how such variations would affect performance.

Replacing costly tests

Combustion simulation has the potential to eliminate costly and time consuming experimental tests that look at parameter variations such as fuel-air ratio, fuel splits, fuel composition, and so on. A typical reacting flow CFD case takes at least three days to converge on a solution with some cases taking up to a week just for one data point. It is easy to see how CFD alone rarely eliminates experimental tests when duplicating a parameter variation from an experiment. It could take a month or more. Solving an ERN in a software tool like Energico typically takes less than a few hours.

Predicting emissions from combustion equipment is a critical capability in order to guarantee performance. Figure 4 shows a comparison of ERN emissions results with experimental measurements for a single fuel injector from a low-NOx, industrial, gas-turbine engine. The ERN is able to predict the NOx emissions at the conditions in the CFD case (represented by the circle). A real value of a well constructed ERN is demonstrated through parametric variations of the ERN inputs to increase the fuel/air ratio (i.e., increase combustor exit temperature) yielding NOx predictions that are in agreement with the experimental results. Similarly results are obtained when looking at the impact of pilot fuel split on NOx formation. The impacts of fuel/air ratio and pilot fuel split on NOx formation are common combustor experiments. The ERN simulation in these cases was conducted in a couple of hours and its results can be used to replace experimental tests that cost upwards of $100,000 to perform.

ERNs have also proven useful in simulating the effects of fuel composition variation on emissions for an industrial gas turbine as is shown in Figure 5. Gaseous fuel compositions with widely varying amounts of CH4, CO, CO2 and H2 are input into the ERN and their impact on emissions of NOx are shown. The ERN predicted the correct trends for NOx emissions with the fuel composition variation and came close to predicting the actual values as well.

A CFD simulation is evaluated by running a large number of CFD cases with different mesh sizes, combustion models, and turbulence models, and comparing the results to experimental data. ERNs provide a second opinion. An example of this can be seen in the industrial gas turbine combustor shown in Figure 6.The top image is theCFDresult showing a small diffusion pilot— temperatures are inaccurate and predict very little NOx and far too much CO. The lower image shows a larger pilot flame, amore precise fuel chemistry, and the NOx and CO emissions are reliable to within 10%. Looking at the temperature distribution between the two images can help understand why there is such a difference. The flow field nearest the diffusion pilot clearly gets hotter and larger when using the fuel model owing to increased NOx and better CO oxidation. In this case, the CFD engineer can see where the result is deficient and can focus on refining the mesh or improving boundary conditions.

CFD alone cannot incorporate the fuel chemistry accuracy that is required to meet combustion challenges. Advances in fuel chemistry allow the prediction of real fuel effects, but CFD cannot handle the complexity. Automatic ERN creation from a CFD solution is an efficient way to incorporate the required fuel chemistry complexity for emissions predictions. Parameter variations of fuel/air ratio, fuel splits and fuel composition can be performed on the ERN for NOx, CO and UHC exit emissions, without requiring the development of a new CFD case. The increased accuracy of the combined ERN and CFD simulations reduces the number of tests required to perfect a design.

Author

Scott Drennan is Director ofApplications Engineering at Reaction Design of San Diego, CA. He can be reached at sdrennan@reactiondesign.com. www.reactiondesign.com