INTRODUCTION
1.3 Previous Work
Notable progress has been made in understanding the fundamental contributors to the combustion dynamic response. A large portion of the work has focused on the effects of unmixedness as driven by the acoustic field. Shih et al. (1996) performed experiments to evaluate the contribution of unmixedness to unstable combustion. Indeed, it was found that fluctuations in fuel distribution resulting in localized variations in equivalence ratio were a strong driver for combustion instability. Later, simulations performed by Lieuwen et al. (1998) also showed a strong coupling between combustor oscillations and resulting variations in mixture fractions. Around the same time, Anderson et al. (1998) and Torger et al. (1998) performed experiments assessing the coupling between an applied acoustic field and fuel injectors under cold-flow conditions.
Work by Cohen et al. (2001, 2003) then investigated modulating the fuel flow in a combustor to actively control the unmixedness. It was shown that substantial control authority was achievable using this method and that the coupling mechanism between the fluctuating heat release and the modulated fuel was strongly related to the mixing process. Similar (although not so dramatic) results had been observed years earlier by Zsak (1993) and Kendrick (1995) in an experimental dump combustor. Further work by Hardalupas et al. (2002) and Demare et al. (2004) focused on acoustic actuation of reactant inlet streams in order to enhance mixing and combustion.
The method was also used to approach issues related to flame attachment and combustor flame- out. Other work in acoustically enhanced combustion had been performed by Cadou et al. (1991) and Cadou (1996) on a double-ledge dump combustor. Cadou also found enhanced combustion in some regimes when the system was driven near resonance.
Somewhat related work was performed by Richards et. al. (1998, 2005). This research focused on passive techniques for control of combustion instabilities. Among other things, Richards investigated transport delays in premixed or jet-mixed systems. The result was the development of the τ-f product method for relating frequencies with positive Rayleigh index to the convection transport time delay from the fuel injector to the flame stabilization point. This has proved to be remarkably powerful in real industrial applications and has been used successfully by various large gas turbine engine manufacturers.
Overall, however, most work has focused directly on combustion dynamics. Many techniques have certainly been employed. From the standpoint of system identification, one novel method in particular is worth highlighting. Work by Seywert and Culick (1999), and Seywert (2001) involved the passive identification of the net combustion response in a system by the measurement of noise intrinsic to the combustor operation. The technique is based on Burg’s method (also known as the maximum entropy method) and can provide a level of system identification in situations where only pressure data are available.
Nonetheless, a great deal of experimental work has focused on direct examination of the combustion dynamics of various systems by observing, in some fashion, the relation between the unsteady heat release and the fluctuating portion of the pressure or velocity field. Most of these methods fall into two distinct categories: those using chemiluminescence to infer the heat release rate and those using laser induced fluorescence (LIF). These, in turn, can be divided, into two sub- categories: Those experiments in which the combustion system was naturally unstable and those in which the system was acoustically forced. Table 1-1 shows the resulting matrix of this work.
Frequencies at which observations were made, whether naturally occurring or forced, are shown.
Much of the early chemiluminescence work used photomultiplier tubes (PMT) to observe the luminescence signal. In experiments by Sterling (1991), Chen et al. (1993) and Kappei et al.
(2000), a slit was incorporated to provide some level of spatial resolution. Typically, the slit was
Chemiluminescence PLIF
Naturally Unsteady
Poinsot et al. (1987) (440-590 Hz)
Sterling and Zukoski (1991) (188 Hz)
Kendrick (1995) (188 Hz)
Broda et al. (1998) (1750 Hz)
Kendrick et al. (1999) (235 Hz, 355 Hz)
Venkataraman et al. (1999) (490 Hz)
Kappei et al. (2000) (370-460 Hz)
Lee et al. (2000) (1750 Hz)
Cadou et al. (1991) (43 Hz)
Shih et al. (1996) (400 Hz)
Cadou et al. (1998) (328 Hz)
Kopp-Vaughan et al. (2009) (158 Hz)
Boxx et al. (2010) (308 Hz)
Steinberg et al. (2010) (308 Hz)
Caux-Brisebois et al. (2014) (289-418 Hz)
Acoustic Forcing
Chen et al. (1993) (300 Hz, 400 Hz)
Durox et al. (2002) (Up to 400 Hz)
Pun et al. (2002) (< 60 Hz)
Kulsheimer et al. (2002)(< 350 Hz)
Schuller et al. (2003) (Up to 400 Hz)
Kim et al. (2010) (100-400 Hz)
Shin et al. (2011) (250 Hz)
Cadou et al. (1998) (360 Hz, 420 Hz)
Pun et al. (2000, 2001, 2003) (22-55 Hz)
Santhanam et al. (2002) (200 Hz)
Balachandran et al. (2005) (40-310 Hz)
Kang (2006) (22-400 Hz)
Bellows et al. (2007) (130 Hz, 410 Hz)
Thumuluru et al. (2009) (130 Hz, 270 Hz)
Kim et al. (2009) (514 Hz)
Yilmaz et al. (2010) (85 Hz, 125 Hz, 222 Hz)
Coats et al. (2010) (200-1200 Hz)
Table 1-1: Previous work in oscillating flames related to investigation of combustion dynamics.
oriented to provide resolution in the streamwise direction. Experiments by others, however, used 2-D imaging systems (historically CCD based) to obtain spatially-resolved luminescence signals.
This includes work by Kendrick (1995), Broda et al. (1998), Kendrick et al. (1999), Venkataraman et al. (1999), Pun (2001, 2002), Schuller (2003) and Kang (2006), among others. In recent years, high speed CMOS imagers with and without coupled intensifiers have grown in availability and popularity. In many cases, this allows direct observation of the unsteady behavior without resorting to phase-locking.
The work using chemiluminescence listed in Table 1-1 spans a range of combustor types.
Experiments by Sterling et al. (1991) and Kendrick (1995) involved a single-ledge dump combustor that exhibited a natural instability at 188 Hz. Experimental research by Chen et al. (1993) incorporated premixed flames and utilized the same equipment used by Sankar et al. (1990). These experiments were specifically designed to simulate combustion of solid rocket propellants and utilized only two forcing frequencies. Work by Durox et al. (2002) and Schuller et al. (2003) involved acoustic forcing of a flame at frequencies up to 400 Hz. However, the flame was operated in an open field (versus a drive chamber) and the CH* luminescence response data collected contained only magnitude information. Research by Pun et. al. (2002) examined a single element, jet-mixed burner operating on landfill surrogate gas. The burner was forced at frequencies from 22 to 55 Hz and phase resolved chemiluminescence image data were collected. Kim et al. (2010) applied inlet forcing to a premixed gas turbine combustor at frequencies from 100 Hz to 400 Hz.
Both bulk and image CH* chemiluminescence data were collected and flame transfer functions were computed.
In contrast to CH* chemiluminescnce, others have chosen to use OH* chemiluminescence as a measure of heat release rate. Lee et al. (2000) examined a naturally unstable, premixed combustor with inlet swirl. OH PLIF was used for flame visualization; however, OH* luminescence was used as the unsteady heat release metric. Steinberg et al. (2010) imaged OH* chemiluminescence in an unstable, dual-swirl combustor oscillating at 308 Hz. Kulsheimer et al.
(2002) performed experiments on a premixed, swirl-stabilized burner. The reactant mixture was modulated at frequencies up to 350 Hz and with amplitudes up to 70% of the mean flow. Imaging of OH* luminescence was used as a measure of the dynamic response.
One fundamental disadvantage with the chemiluminescence measurement technique is the lack of depth of field resolution. Generally speaking, measured fluorescence is integrated over the depth of the flame. This difficulty is essentially eliminated by the use of planar laser induced
fluorescence (PLIF). Here, fluorescence is restricted to the plane of the laser sheet, which is coincident with the object plane of the imaging lens. Additionally, this spectroscopic method allows selected chemical species to be stimulated into fluorescence by optical pumping at specific wavelengths, typically in the ultraviolet. The first documented demonstration of PLIF of OH radicals in a flame was performed by Dyer and Crosley in 1982. Since then, the technique has been improved and refined with technical advances in lasers, optical coatings and intensified imaging systems being major contributors.
PLIF measurements have been used as a locator of the reaction zone and as a metric for heat release in numerous combustion experiments. Bergthorson (2005) used CH PLIF in experiments on a stagnation burner under steady conditions to measure the reaction zone location, thickness and CH concentration profile. CH PLIF was used by Coats et al. (2010) to estimate flame front area as a measure of heat release rate in an acoustically-forced, lean premixed, multi-port laminar burner. Cadou et al. (1991), Shih et al. (1996), Pun et al. (2000, 2001 and 2003), Ratner et al. (2002), Kang (2006) and Yilmaz et al. (2010) used OH PLIF as a direct measure of heat release in unsteady (naturally unstable and forced) experiments. OH PLIF was also used in naturally unstable combustion experiments by Kopp-Vaughan et al. (2009), Boxx et al. (2010), Steinberg et al. (2010) and Caux-Brisebois et al. (2014), as well as inlet-driven, unsteady experiments by Santhanam et al. (2002), Bellows et al. (2007), Thumuluru et al. (2009) and Kim et al. (2009); in these cases, PLIF data were used for flame visualization and/or flame surface area estimation. Heat release rates (where determined) were either measured by collection of CH* or OH* chemiluminescence or estimated from the computed flame sheet area. PLIF signal intensity itself was not used as a heat release rate metric.
Paul et al. (1998) demonstrated the use of OH PLIF in combination with CH2O PLIF to produce an improved measure of the heat release rate. This combination was later used by Balachandran et al. (2005) to evaluate unsteady heat release in an inlet-driven, turbulent, premixed
burner. Additionally, Cadou et al. (1998) used nitric oxide seeding and NO PLIF to measure the temperature field within a double-ledge dump combustor. Ratner et al. (2003) used NO PLIF in experiments on an acoustically forced combustor to measure the unsteady production of nitric oxide.
Much of the experimental work performed on unsteady combustion in gas phase systems has revolved around specific combustors and usually ones that are naturally unstable. This may, in part, be the result of desperation. It is not surprising that attention (and consequently, money) is focused on commercial combustors that exhibit an instability problem and are pacing a delivery schedule. (Managers can be quite sensitive to this.) Such work (generally speaking) concentrates either on a specific problem inherent to a particular combustor, or on the generalized, large-scale behavior of the examined systems in order to generate some global insight into their dynamics.
Regardless, these experiments have helped to improve general understanding of the combustion instability phenomena, and have also served to validate experimental techniques such as those used by numerous researchers including Pun et al, Ratner et al. and Kang. Nonetheless, little experimental work has focused specifically on fundamentally simple burners and examining the combustion dynamics therein. Most of the burners or combustors involved in previous research have been complex in nature, involving various levels of unmixedness (in premixed systems) and any combination of turbulent flows, recirculation, swirl, and so on. With such systems and with incomplete information about the fields involved (pressure, temperature, velocity, reactant concentrations, etc.) it is difficult to arrive at definite conclusions about some fundamental mechanisms.
Great strides have been made in recent years in the development of high speed imaging systems (intensified CMOS units in particular) and pulse-burst Nd:YAG lasers (Jiang et al. 2010;
Den Hartog et al. 2010). These systems allow short cinematographic sequences to be captured in various laser-based imaging techniques. This allows detailed investigation of flame evolution in
systems that do not exhibit strict periodic behavior. Nonetheless, to this point the investigative power of these systems has yet to be fully leveraged.
Finally, one would be remiss in not mentioning work in this research area accomplished through numerical means. A great deal of simulation work has focused on response characterization of complex, gas-turbine derived combustors. Examples include work by Roux et al. (2005), Selle et al. (2006), and Huber and Polifke (2009). The combustion systems simulated in these cases were turbulent and employed some level of inlet swirl. Roux and Selle used large eddy simulation (LES) in their numerical methods while Huber and Polifke used a commercial CFD package to execute unsteady Reynolds-averaged, Navier-Stokes (URANS) simulations.
Nonetheless, other numerical work has focused on acoustically driven behavior in simpler systems. Single-element, turbulent, bluff-body stabilized burners have been explored by Armitage et al. (2006), Han and Morgens (2015), and Han, Yang and Mao (2016). Various numerical codes were used, employing both URANS and LES. In all unsteady cases, inlet velocity forcing was applied.
Finally, there are numerical simulations of acoustically forced laminar flames. These have the greatest relevance to the work at hand. Most of this work involves the study of one of two types of flames: the inverted ‘V’ flame with a central rod, and the Bunsen style flame. Inverted ‘V’
flames have been studied by Kaufmann, Nicoud and Poinsot (2002), and Birbaud et. al. (2008).
Acoustically driven Bunsen style flames have been simulated by Truffin and Poinsot (2005), Duchaine (2012), and Silva et al. (2015). Both styles of flames were simulated by Schuller, Durox and Candel (2003), where the results were then compared against previously analytically generated transfer functions. The simulation codes employed were typically compressible DNS Navier- Stokes solvers, with the most notable one being AVBP developed jointly by CERFACS and IFP.