This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.
Recent reports point at the substantial reduction in NOx emissions in lean premixed gas turbine combustors. However, combustors of this type pose the inherent risk of lean blowout (LBO). This calls for an on-line early warning system that can be integrated with the combustor controls to ensure efficient and uninterrupted functioning. High reliability and a quick response time of such systems are necessary for most applications, and are particularly significant in aircraft propulsion applications. In this study we propose and investigate novel techniques aimed at developing an efficient LBO prediction system. The work characterizes the behavior of a swirl stabilized, LPG-air fueled, dump plane combustor flame near its LBO limit in terms of CH* chemiluminiscence (optical) emission intensity, total emission intensity and flame color analysis. We investigate the influence of fuel-air equivalence ratio and the mixing distance (Lfuel) in the premixer on the efficacy of the prediction methods. The acquired data points at the occurrence of various combustion modes in accordance with the equivalence ratio and fuel supply condition. Several techniques based on Fourier spectral analysis, statistical analysis and direct raw signal thresholding are presented for different combustion modes which analyze optical signal and transform it to some suitable quantitative indicator for sensing the proximity of lean blowout. It has been observed that the different sensing technique discussed here behave differently with respect to the mixing length provided for fuel and air. A detailed priori mapping of LBO for a fixed combustor configuration is required for deciding the quantitative indicator for particular sensing technique. A new RGB analysis technique has been explored for sensing the signature of LBO based on the image processing technique carried on the direct digital photography of the flame. It is observed that for all the different ports conditions which gives different mixing distances, the Red/ Blue (R/B) color intensity ratio decreases significantly towards LBO. This ratio could be a more suitable LBO precursor for sensing the incipient lean blowout.
Keywords: Lean blow out (LBO), Combustor, Image processing, Chemiluminescence, Statistics
The use of continuous combustion processes is inevitable in the present day industries and transportation systems. The growing use of combustion put a heavy penalty to the environment in the form of more pollutant emission (NOx). To achieve the lowest possible NOx, modern combustors are designed to operate in a lean premixed or partially premixed mode  that operate at low temperature. However, both premixed and partial premixed combustor are prone to inherent risk of lean blow out (LBO) as combustors are made to operate at an equivalence ratio (Î¦) in the proximity of their LBO limit (Î¦LBO). When sufficiently lean flames are subject to power setting changes, flow disturbances or variation in fuel composition, the resulting equivalence ratio (Î¦) perturbations may cause loss of combustion. Such a blowout could cause loss of power and expensive down time in stationary turbine, which increases maintenance cost and reduce productivity and engine life. The problem is particularly severe for aircraft engines where combustion is the ultimate source of engine thrust, lean blowout (LBO) causes loss of thrust and thus posses a significant safety hazard. This is especially important when power is reduced during approach and landing. To avoid such scenarios, current systems are typically operated with a wide margin above the uncertain LBO limit. Enhanced performance will require a reduction of this margin. Therefore, the ability to sense the LBO signature can provide significant payoffs. For example, a sensing system with appropriate sensitivity, reliability and time response could be used in an active control system that would allow combustors to run at lean conditions compared to the present combustors, thereby reducing NOx emission without compromising safety.
In the past much more extensive studies were conducted for the stability of combustors. Stability of a combustor can be broadly classified into static and dynamic stabilities. Dynamic stability deals with the self excited oscillations produced due to coupling between heat release fluctuations with the pressure oscillations . A combustor is considered to be statically unstable when the flame cannot be held inside the combustor. Therefore at Flashback or blowout, the flame becomes statically unstable. Static stability boundaries are usually dependent on operating parameters like flow rate, temperature and equivalence ratio (Î¦). Dynamic instability can cause fluctuations in some of these parameters and can thus drive the combustor operating conditions towards its static stability limits. Likewise, static instability can cause changes in flame shape and thus affect the heat release distribution inside the combustor and thus drive dynamic instabilities. There is much research in the field of controlling dynamic instability [3, 4], there is less work in understanding complex combustors near their static stability limits (flashback or blowout limits). Although some previous studies shed the lights on flame dynamics near static stability limits (LBO limit) like there have been several studies in simple flame that indicate there is a local loss of flame that grows to complete loss of flame [5, 6]. De Zilwa et. al.  investigated flame dynamics close to blowout in dump stabilized combustors with and without low swirl. They noticed very low frequency oscillations as the rich and lean extinction limits were approached. Murugandam and Seitzman  observed the extinction and reignition events prior to LBO in swirl stabilized, methane fueled gas turbine combustor. There are some more studies which reported flame instability or transient behavior close to blowout [9, 10].
This transition behavior, which appears to precede blow out could be used in the past for generating the LBO precursors. The choice of non-intrusive sensor is best suited which is outside from the harsh environment of combustor and captured electromagnetic and acoustic radiation produced within the combustor. Muruganandam et. al.  used OH* chemiluminescence to identify the LBO precursors which are short duration, localized flame extinction and reignition events in premixed, swirl stabilized, methane fueled combustor. They used those precursor events to develop the sensing methods which are incorporated in active control system to mitigate the impended blowout. Yi and Gutmark  used the optical emission for sensing the LBO precursors in partial premixed, liquid fueled, multiswirl stabilized gas turbine combustor. There are few more studies which used acoustic radiation rather than optical emission like Mukhopadhyay et. al.  identified moving kurtosis and threshold based techniques for early detection of flame extinction by using time series data of combustor pressure. Shashvat prakash et. al.  incorporated acoustic based sensing technique with active control system for mitigation of impended blowout. It is observed that in all the previous studies the focus is mainly on the characterization of flame dynamics towards blowout limit for fixed configuration of combustor either premixed or partial premixed. A survey of previous studies [3, 15] reported that the equivalence ratio fluctuation and the unmixedness of fuel and air influence a heat release fluctuation which is the primary cause of dynamic instability and subsequently which affects the static limit (LBO limit) of flame. As per our observation no study has been reported yet where the effect of fuel unmixedness has been seen on the behavior of sensing techniques, which motivate us to undertake experimental study on a swirl stabilized, LPG (60% propane and 40% butane) and air fueled, dump plane combustor, in order to elucidate the effects of fuel and air unmixedness on the flame dynamics near blowout limit and subsequently asses the behavior and effectiveness of different sensing techniques in the next section.
Fig.1 illustrates the experimental setup, which consists of three major elements: a combustor, mass flow control devices for fuel and air and flame imaging system consist of PMT and camera. The non-vitiated combustion air is supplied at ambient temperature from compressor to the bottom port provided on premixing tube and metered upstream of the combustor using calibrated mass flow controller (MFC) (Aalborg range 0-500 LPM). The fuel (LPG) is supplied from a pressurized cylinder fitted with needle valve to control the flow rate and metered upstream of the combustor by calibrated Aalborg mass flow controller (RANGE 0-10 LPM.). The fuel is injected to the different side ports provided at different locations on premixing tube gives different mixing lengths (Lfuel). The injected fuel mixes with air as it passes through the mixing distance (Lfuel). Finally fuel-air mixture entered the combustor through an inlet swirler in the annulus around a centre body, located just prior to dump plane in the premixing section. The inner diameter of the premixing tube is 2.3 cm, and the diameter of the center body is 0.8 cm. The inlet swirler has six vanes positioned at 600 to the flow axis. The quartz tube is provided in a combustion zone having internal diameter 6 cm. and length 20 cm. The quartz tube facilitates the optical diagnostics of the combustion.
The line of sight heat release measurement of combustion was obtained by measuring the chemiluminescence intensity from CH* radicals of flame with photomultiplier tube (PMT) of Hamamatsu make (model 931B), PMT is fitted with a 10nm bandwidth filter centred at 432nm to allow emissions from CH* radicals (432nm) alone to reach the PMT. This PMT has a built in amplifier (bandwidth of 20 kHz) to convert the current to voltage and operates from a 12VDC source. The PMT output signal (voltage) is conditioned by using NI SCXI 1050 08-channel signal conditioner, subsequently the conditioned signal from sensor was acquired to a personal computer using a PXI 8050 National InstrumentsÂ® 16 bit data, 06-channel A/D card. A total of 32768 data (N) (i.e., 215 points) were acquired in each experiment using NI software LABVIEW 7.0Â® at a sampling rate of 2000 Hz. High-speed images of the flame under stable and near blow-off condition were obtained in order to better understand the phenomenology of the flame blow-off process, and hence improve capabilities for interpreting the optical signature. A camcorder (SONY, DCR-TRV 300) is used to capture the flame video, later the images are captured from video at 33 frames per second (FPS). The global features of flame were examined using a colour digital camera. The aperture, exposure times and focal length of camera could be independently controlled to get the correct results. The color phenomenology of flame was recorded with the camera.
Figure 1. Schematic diagram of the experimental apparatus
The combustor was approached to LBO stepwise by controlling the equivalence ratio(Î¦), which is adjusted via the fuel flow rate while keeping the airflow fixed. A larger incremental reduction in Î¦ of about 0.05 is used when the combustion is well above the LBO, whereas a smaller reduction in Î¦ of about 0.02 is used when combustion is close to LBO.
Five different ports as Port1, Port 2, Port 3, Port 4 and Port 5 has been used for injecting the fuel which subsequently gives different mixing lengths (Lfuel) as 35cm, 30cm, 25 cm, 20cm and 15cm respectively for fuel and air. The experiments were carried out by using three different air flow condition which gives three different Reynolds numbers mentioned in table1 of experimental condition.
Table 1. Experimental Condition
Number of fuel injecting ports
(05) Port1, Port 2, Port 3, Port 4, Port 5
Mixing lengths (Lfuel)
35cm, 30cm, 25cm, 20cm, 15cm. respective to
Port1, Port 2,Â_ _ _ _ _Port 5.
Air flow rate (Q)
70, 80 and 85 LPM
Flow Reynolds number (Re. no.)
6645, 7480, & 7948 respectively with air flow rate (Q)
Sampling rate from PMT
A/D digitization resolution
16 bit (+- 5V)
In this section, we describe the phenomenology of the flame blowout process. High-speed images are presented in conjunction with simultaneous optic data to aid the development of data-analysis schemes described in the next section. Prior study have observed [7, 8] towards lean blowout limit, there are random instances where the flame exhibit oscillation in the combustor. These oscillations produce near flame loss event in combustor. The similar flame loss event we observed in the condition of port1, port 2 and port 3. One such flame loss event is elucidated here in Fig. 2. which plots a sequence of images captured at 33 frames per second, here port1 is used for injecting the fuel, air flow rate used as Q= 80 LPM and in this case of experiment blow out occurs at equivalence ratio 0.75 (Î¦LBO= 0.75). From Fig. 2, it is seen that, the combustor initially has a spatially combustion zone. Then the flame detaches from the centre body, showing weak reaction and moves further downstream from the combustor inlet and stabilize there. The flame begins to disappear from field of view and there is almost complete loss of flame, suggesting extinction event. From the down stream the flame packets are convicted back to inlet, which reignites the unburned fuel which entered the combustor in previous period. The fuel is rapidly consumed which exhibit intense combustion and the flame is re-establish in the combustor (reignition event). These unique extinction and reignition events span a period of several milliseconds, and they occur randomly in time, prior to LBO. As the combustor approached towards LBO limit, the frequency of these events increases and thus the time between two such events decreases closer to LBO.
Fig.2. Sequence of high speed images of global flame with Î¦= 0.75 for port1. The images are separated by 30 msec. The time sequence of images is from left to right on the top followed by left to right on the following rows. The first image shows the viewing window for optical sensing
Fig.3. Sequence of high speed images of global flame with Î¦= 0.51 for port 4. The images are separated by 30 msec. The time sequence of images is from left to right on the top followed by left to right on the following rows.
Figure 2. Time series data of CH
Fig.5. Time series data of CH chemiluminescence signal for for port2, port3, port4, and port5 for away and nearer to LBO condition (Read Top to Bottam)
Fig.4 plots typical measured time dependence of the CH* chemiluminiscence at the equivalence ratios i.e. Î¦ = 0.81 and Î¦=0.75 representing the flame condition far from and one near to LBO respectively. As expected, the mean intensity decreased with lower equivalence ratio, due to reduced heat release as the fuel flow is reduced. Also for the lean case i.e. Î¦=0.75, the chemiluminescence intensity shows some relatively high amplitude bursts (unsteady events) with the signal going both below and well above the mean value and occasionally drops to near zero value. To illustrate this behaviour in detail, a expanded plot of these events are also shown. Often, these bursts or events are characterized by an almost complete loss of chemiluminiscence signal quickely followed by intense emission from a imaged region. Thease bursts or events coincide with the occurrence of flame loss (exctinction) and re-ignition events described in the high-speed video images. Similarly Fig. 3 shows high speed images towards blowout limit for port 4 for the same experimental condition used in describing high speed images in port 1. The similar mode we observed in port 5 case also. From Fig. 3 it is observed that the flame in port 4 and port 5 does not oscillates in the entire combustor and rather the oscillations are limited to dump plane so that the flame loss are not so significant in this case.
Figure 5 plots the CH* chemiluminiscence variations for port 2, port 3, port 4 and port 5 for rich and one close to lean blowout equivalence ratio. From Fig. 5, it is clearely seen that for port 2 and port 3 case, the intensity of chemiluminiscence bursts (unstedy events) is more intense relative to mean and occasionally drops to zero value (extinction) as compared to the CH signal for port 4 and port 5 case. These bursts matches with the flame loss and reignition events due to intense oscillations of flame in entire combustor seen in port 2 and port 3 case which is not so significant in port 4 and port 5 case.
5.LBO Sensing Strategies:
Developing LBO sensing strategies with maximum sensitivity, speed and robustness requires a thorough understanding of the flame characteristics prior to blowout. The high-speed flame images obtained and analyzed in conjunction with simultaneous optic data and the transition of colour phennomelogy prior to blow off were used to aid the development of such strategy. A simple approaches are considered like statistical, Fourier spectral and direct raw signal (Thresholding ) in various confrugation of combustor obtained by changing the mixing length available for fuel and air. Apart from this a new image processing based technique, RGB analysis has been established for sensing the signature of blowoff in the next section.
5.1 Spectral Aanalysis
Fig. 6 shows the Fourier power spectrum of the CH* chemiluminiscence signal for Î¦= 0.81, 0.77 and 0.75 after excluding the D.C. signal. These curves have been normalized at each Î¦ to have the same total power. Here air flow rate Q= 80LPM, N= 32768, Î¦LBO = 0.75. From Fig. 6, it has been observed that the fractional energy in low frequencies appears to increase as LBO limit is approached. A cumulative distribution function Î¾ (i) is defined to quantify the energy distribution with the frequency 
(K= 1,2,â€¦â€¦â€¦.1000) (1)
Where Hk refers to the amplitude of CH* chemiluminiscence at the kth pin in fast Fourier transform (FFT) analysis. Fig. 7 shows that the low frequency component of CH* chemiluminescence have a much higher energy percentage at Î¦= 0.75 and 0.77 than that at Î¦ = 0.81. It was observed that the fraction of low frequency content increases rapidly as the combustor approached towards LBO. Thus the proximity to LBO can be sensed by quantifying the relative amount of low frequency content in the signal. Employing two filters, a band pass filter for measuring the low frequency content and high pass filter for removing the DC signal can fulfill the practical implementation of this approach. However, monitoring the lower frequency content require longer times, which will hamper the time response of the control system.
Fig.6. Fourier power spectrum of CH emission signal for different Î¦ for port1
Fig.7.Percentage of cumulative energy with frequency for port 1 test
In order to asses the effect of fuel unmixedness on the behavior of intensified low frequency content which decides the feasibility of implementing this approach for sensing the proximity of lean blowout in different configuration of combustor obtained so far. Tests were carried on using different ports for injecting fuel which give different mixing distances for fuel and air. Figure 8 shows the different plots for percentage of the cumulative energy for port 2, port 3, port 4 and port 5. From Fig. 8, it is observed that for port 2 and port 3 the trend of increasing fractional energy in low frequency component is more or less similar to port 1 results. For port 4 and port 5 the overall fractional energy of CH* chemiluminiscence increased towards lean blowout limit (Î¦LBO) but low frequency component is not so intensified. The reason behind this may be, recall the results of high speed flame images where we observed that towards LBO limit for port 4 and port 5 the flame does not oscillate significantly in the entire combustor rather the oscillations are limited to dump plane only. For port 1 and port 2 conditions towards LBO limit, the significant oscillation of flame in entire combustor creates low frequency flame extinction and reignition events which subsequently give intensified low frequency component. The above observation is similar to prior studies. For example Nicholson and Field noted large scale, low frequency flame oscillations as well as periodic detachment of the flame from its flame holder near the LBO limit. Muruganandam et.al.  observed low frequency power content of both acoustic and OH spectra were higher in a lean flame.
The above analysis reveled that spectral approach for predicting the proximity of lean blowout is more suitable in more premixed configuration combustor like port 1, port 2 and port 3 where the distances provided for mixing fuel and air is more.
Fig.8. Percentage of cumulative energy with frequency for port 2, port3, port4 and port5 position (read left to right)
As an alternative to frequency domain analysis, here the time series data of CH* variation captured on PMT is proved to be useful to probe the near-LBO combustion dynamics. Fig. 4 shows a typical CH* chemiluminiscene variations of flame for the one stable equivalence ratio far from LBO i.e. Î¦= 0.81, and one close to LBO i.e. Î¦ = 0.75. As discussed earlier, the mean emission signal decreases with reduced Î¦, as the fuel is reduced. To quantify the high amplitude, short duration bursts (unsteady events), which matches with the flame extinction and reignition events near lean blow off condition which present in the CH* signal shown in Fig. 4, a statistical analysis is used. In the past Murugandam et. al.  sucessesfully used the statistical moments like standard deviation, second moment, fourth moment and kurtosis for predicting the incipient blowout in a premixed, swirl stabilized, methane fueled dump combustor in real time using OH* chemiluminiscence.
Similarly Yi and Gutmark  used normalized root mean square deviation for early prediction of lean blowout in a multiswirl stabilized, turpentine fueled atmospheric gas turbine combustor. The same approach has been explored here in more details so that to observed the effect of unmixedness in the following section.
The normalized chemiluminescence RMS which is called as coefficient of dispersion (C.D.) in statistics. Which is the ratio of standard deviation (RMS) to the
C.D. = RMS/ Î¼ (2)
chemiluminescence mean (Î¼) is plotted with the equivalence ratio normalized by its value at blowoff i.e. (Î¦/Î¦LBO). Fig. 9 compares C.D. for different flow conditions, correspondingly gives different Reynolds numbers when port 1 is used for injecting the fuel. C.D. increases slowly with decreasing equivalence ratio until Î¦/Î¦LBO>1.15. This trend has been confirmed by other researches [17, 18], who observed that the heat release RMS was roughly proportional to the mean heat release rate. C.D. changes little in the large range of Î¦ and shoot up as the combustor approached towards LBO. Fig.11 shows the variation of probability density function (PDF) using normal distribution calculated for chemiluminiscence which is normalized by its mean, it has been observed that the PDF is almost invariant for port1 until Î¦/Î¦ LBO > 1.15(i,e. rich condition) which makes C.D. almost constant for Î¦ / Î¦ LBO> 1.15.
Fig.9. Variation of C.D. for different flow conditions for port 1.
Fig.10. Effects of N on C.D.
Figure 11. Variation of PDF for normalized chemiluminiscence/ mean for port 1 and port 4 (L to R)
For the calculation of C.D we need sample N. Because of the chaotic and seemingly irregular feature of near LBO combustion dynamics, a larger N is preferred. However, a larger N will inevitably slow down the detecting speed. Fig.10 shows that even a smaller sample window is capable of achieving the reasonable balance between the accuracy and the detecting speed.. To initiate the LBO warning signal which feed to active control system for controlling the imminent blowout, one needs some quantitative indicator. After carefully observing the variation of C.D. for port 1 case, the following condition found to be more suitable. On satisfying the below condition the LBO warning should be initiated 
It is found that the value of Î± =1.75 is more suitable in this type of premixed combustor confrugation where port 1 is used providing maximum mixing length ( L fuel= 35 cm.).
Tests were carried on the different ports which subsequently gives different mixing distances mention in table1. The same experimental conditions were used only the Î¦ LBO value has been changed for port to port. The variation of C.D with Î¦ / Î¦ LBO has been plotted for port 2, port 3, port 4 and port 5 shown in Fig. 12. It has been seen that the trend of variation of C.D. in port1 case has not been repeated in other ports. The PDF of normalized chemiluminiscence which almost invariant in port1 condition which makes the C.D is almost constant until Î¦ / Î¦ LBO > 1.15 is no longer seen in other ports. Fig 11 shows the variation of PDF using normal distribution for normalized chemiluminiscence by its mean for port 4 where it is vary and the standard deviation is continuously increased in rich case i.e until Î¦ / Î¦ LBO> 1.15.
Hong et. al.  experimentally demonstrated that the mixing length is a crucial parameter for triggering dynamic instability and observed different modes of pressure fluctuation with respect to unmixedness of fuel which is controlled by mixing length. So mixing length is important parameter which affects the amplitude of pressure fluctuation and subsequently heat release rate. That may be the reason to obtained different behavior of C.D. for different ports. Please note the focus of our paper is mainly on the behavior of sensing technique rather than the study of dynamic instabilities triggered due to unmixed ness.
The condition which we mention for port 1 to initiate the LBO warning signal and the threshold value 1.15 up to which the C.D. is almost constant and the value of Î± =1.75 would not be suitable for other ports where the mixing distance available for fuel and air is less. In order to make this sensing technique suitable for partial premixed configuration, detail priori LBO mapping of the combustor should be needed so that to find the suitable value for threshold and Î± .
Fig.12.Variation of C.D. for different flow conditions for port 2, port3, port4 and port5 (Read L to R)
5.3. Direct signal analysis (Thresholding approach)
The major limitation of preceding statistical method was, it gives some average measure of deviation of time trace signal of chemiluminiscence from its mean. Here, the time localized excursions of CH* signal is used to identify the precursor events. The simple threshold approach is used . The precursor event is acknowledge when the CH* signal drops below some threshold value. For example 35% of CH* mean is taken for the current experiment. The reasoning behind this approach is that, the precursor signature is initiated by a local extinction event that temporally lowers the chemiluminiscence. Thus, the low threshold provides the easiest detection of the event. Selection of optimum value of threshold for detection can vary depending on the combustor design, the sensing volume and the required sensitivity of the technique.
Fig. 13 shows the variation of frequency of identified precursor events (average over sampling time.) over Î¦. It is seen that, this parameter has almost zero value for the higher Î¦ and increases as the LBO limit is approached. The event frequency parameter gives sensible quantitative blow out indicator for control system to sense the proximity to LBO.
As discussed earlier, these precursor events occurs randomly both in frequency and duration. To take in to account this two effect, Yi and Gutmark  computed and used normalized cumulative duration of LBO precursor events. The same approach has been used here.
Î˜ = Nt / N (4)
Where Nt denotes total precursor events observed in sample N. Fig. 14 shows the variation of Î˜ with Î¦/ Î¦LBO for different flow conditions for more premixed configuration (port1). Similarly to C.D shown in Fig..9, Î˜ increases very slowly until Î¦/ Î¦LBO > 1.15 and increases significantly as the LBO limit is approached., the reason behind this is that the C.D. and the Î˜ are mathematically interdependent, as we have seen that the variation of PDF for chemiluminescence /mean ((normalized) is invariant until Î¦ / Î¦ LBO > 1.15. The Î˜ is the cumulative probability for chemiluminiscence / mean < 0.35. For generating LBO warning signal, the same condition given below which used earlier in statistical method found to be more suitable.
The value of Î± = 1.75 is more suitable for port1. Similar approach has been tried for other ports in order to see the effect of fuel unmixedness on behavior Î˜. Fig. 15 shows the variation for Î˜ index for port 2, port 3, port 4 and port 5 for different flow conditions. As Î˜ and C.D. are mathematically interdependent. The variation of Î˜ matches with the variation of C.D. plotted earlier in statistical approach. The above condition mentioned for port 1 for triggering the LBO warning signal and the threshold value 1.15 up to which the Î˜ is almost constant and the value of Î± =1.75 could not be found suitable in other ports. Similar to statistical approach here also, in order to make this approach suitable for partial premixed configuration where mixing distances for fuel and air is less, we need a detailed prioro LBO mapping of the combustor in order to find out the new value for threshold and Î± .
Fig.13. Dependence of number of CH based events upon Î¦ for port1
Fig.14.Variation of Î˜ with Î¦/Î¦LBO for port 1
Fig.15.Variation of Î˜ with Î¦/Î¦LBO for port 2, port3, port4 and port5 (Read L to R)
5.4 Image processing technique (RGB analysis)
For determining the proximity of combustor to lean blowout. Here a new approach has been explored based on the image processing technique carried on the direct digital photography of the flame. Being a camera located outside the harsh environment of combustor, this approach also fulfils the requirement of non intrusive imagining. For exploring this sensing technique in more detail so that to enumerate the effect of fuel unmixedness on the feasibility of implementing this technique in various configuration of combustor. Similar to preceding discussed methods here also five different fuel injecting ports have been used which subsequently gives different mixing distances(Lfuel)for air and fuel.
Figure 16. Variation of Red & Blue intensity with flame modes for different Equivalence ratio (port 1)
For each port the experiment has been carried out for three different air flow rates (Q) correspondingly gives three different Reynolds nos. mentioned in table 1 of experimental condition. The equivalence ratio (Î¦) is varied stepwise by changing the fuel flow rate while keeping the air flow rate (Q) constant. A larger incremental reduction in Î¦ of about 0.05 is used when combustion is well above the LBO, whereas a smaller incremental reduction in Î¦ of about 0.02 is used when combustion is close to LBO. The sensing methodology developed here is based on the image processing technique. At each equivalence ratio (Î¦) condition four photographs of the flame modes has been taken. The algorithm was developed to read the images of flame at each variations of Î¦. The average intensity of Red and Blue colour is calculated for the corresponding equivalence ratio (Î¦). Figure16 shows the different modes of flame obtained on varying the Î¦ from rich to lean blowout condition. Here the air flow rate used is Q=80 LPM and port 1 is used for injecting the fuel which gives maximum mixing length (Lfuel= 35 cm.). In the same plot the variation of Red and Blue colour intensity is also plotted. From Fig.16, it is observed that the red intensity is gradually decreased while the blue intensity is increased as the combustor approached towards blowout from rich condition. The reason behind this is towards lean conditions the less fuel with fixed air flow gives more premixed flame. In premixed flame the blue colour is dominant, also soot percentage is almost reach to zero, which is indicated by red colour. The transition of colour phenomenology which precedes the blowout could be used to give the LBO precursor for sensing the proximity of blowout.
Fig.17 plots the variation of Red/Blue (R/B) colour intensity ratio with equivalence ratio (Î¦) for port 1 where Lfuel= 35 cm for three different air flow rates giving different Reynolds number at the inlet section of combustor (Re. no.= 6545, 7480, 7948). Again for one single air flow rates three different run has to be taken so that to estimate the percentage errors. Error bars are plotted for all the different air flow rates conditions seen in Fig.17. From the percentage error estimation, it is found that the error percentage for most of the readings lies in the range of + - 2 to 8 % for all the three different flow conditions. From Fig. 17, it is observed that, as the combustor approach towards LBO in all the three variable air flow rate conditions, this (R/B) colour intensity ratio drops significantly, which could gives a suitable quantitative indicator for sensing the proximity of combustor to LBO.
Fig.17. Variation of R/B ratio with different (Î¦) for port 1(Lfuel= 35 cm.)
To find out the drop in percentage of the R/B colour intensity ratio from rich (stochiometric) condition to lean blowout limit. The R/B ratio at each Î¦ is normalised by the maximum value of R/B ratio encounter in that run which is usually seen near to stochiometric condition (Î¦=1.0). Figure 18 plots such variation of normalized R/B ratio with normalized equivalence ratio by Î¦LBO i.e. (Î¦/ Î¦LBO) for port 1 case. The same experimental data has been considered which we used earlier for plotting Fig.17. From Fig.18, it is observed that when the combustor reach to near LBO limit the percentage drop in R/B ratio is 85 % or rather in other words we can say near LBO the R/B ratio acquire 15 % of the maximum value of R/B ratio encounter in that run. It is seen that in all the three different air flow conditions the R/B ratio acquire more or less same value i.e. 15 % of (R/B) Max.
The above observation could give a value of threshold level which can be used for sensing the proximity of combustor to lean blowout with active LBO controller. The more suitable threshold found in this case of port1 result is 20 % (R/B)MAX, which fulfil the criteria of neither it is very close to LBO limit nor it is too away from it. So predicting the incipient blowout in real time the following condition is recommended. The LBO warning be initiated on satisfying the below condition. For calculating the R/B colour intensity ratio very few computation is required so the real time requirement is satisfied.
As discussed earlier the most suitable value we found out for Threshold in port 1 case is 0.20, once the LBO warning signal is triggered, the active control system take appropriate action for enhancing the stabilization of flame in combustor so that the incipient blowout could avoided and LBO limit can be extended.
To elucidate the effect of unmixideness of air and fuel (premixing percentage) the technique is tested for different ports which gives different mixing distances for air and fuel. Figure 19, 20, 21 & 22 shows the normalised (R/B) intensity ratio verses the normalised Î¦/ Î¦LBO for port 2, port 3, port 4 & port 5 respectively. The mixing distances i.e (Lfuel) for the respective ports is mentioned in the table1 of experimental parameters.
Fig18. Variation of normalized R/B ratio for normalized (Î¦/ Î¦LBO) for port 1(Lfuel= 35 cm.)
From Fig. 19, where the variation of normalized R/B intensity is plotted for port 2, it is observed that similar to port 1 result, the R/B ratio drop significantly as the combustor approach near to LBO. The only variation from port1 result is in percentage drop of R/B intensity, here this value is slightly decreased. Near LBO limit (Î¦LBO), R/B ratio reaches the value in the range of 20 to 25 %. of the maximum value of R/B ratio encountered in that run which we found at stochiometric condition (Î¦=1.0) in all the three different flow conditions.
Similarly when we observed Fig. 20, 21 & 22 we found that the percentage drop in the R/B intensity ratio is slightly decreased as the mixing distances is reduced respectively. Near lean blowout limit the R/B intensity value obtained in the range of 20 to 25% of the maximum R/B ratio value for port 3. Similarly for port 4 it lies in the range of 30-35 % and port 5 where the mixing distance is minimum it lies in the range of 35-40 % for all the three different flow conditions which we tried for each port. In Fig. 23, 24 and 25 we tried to illustrate the variation of normalized (R/B) ratio for different ports with respect to fixed flow condition. The same experimental data has been used which we discussed earlier for different port condition.
From this analysis it is clear that the Threshold value = 0.20 we set in port 1 case is no longer useful in other ports as the R/B intensity ratio acquired grater value near lean blowout limit.
The same sensing technique could be used in other ports condition only one has to change the threshold value. Here we observed the best suitable threshold value in port 2 and port 3 is = 0.3, for port 4 = 0.4 and for port 5 it is = 0.45. So just by changing the value of threshold level for different configuration of combustor obtained by varying the available mixing length for air and fuel make this technique suitable for monitering the proximity of combustor to incipient lean blowout.
Fig.19 Variation of normalized R/B ratio for normalized (Î¦/ Î¦LBO) for port 2(Lfuel= 30 cm.)
Fig.20 Variation of normalized R/B ratio for normalized (Î¦/ Î¦LBO) for port 3(Lfuel= 25 cm.)
Fig.21 Variation of normalized R/B ratio for normalized (Î¦/ Î¦LBO) for port 4(Lfuel= 20 cm.)
Fig.22 Variation of normalized R/B ratio for normalized (Î¦/ Î¦LBO) for port 5(Lfuel= 15 cm.)
Fig.23.Variation of normalized R/B ratio for normalized (Î¦/ Î¦LBO) for different ports for Re. no.=6545
Fig.24.Variation of normalized R/B ratio for normalized (Î¦/ Î¦LBO) for different ports for Re. no.=7480
Fig.25.Variation of normalized R/B ratio for normalized (Î¦/ Î¦LBO) for different ports for Re. no.=7948.
Fig.26.Variation of normalized R/B ratio for normalized (Î¦/ Î¦LBO) for port 1 with one snap condition.
In all the above experiments for determining the R/B intensity variations, four photographs were taken for each mode of combustion obtained on changing the equivalence ratio (Î¦). The algorithm developed reads all the four photographs and calculate the mean intensity for red and blue color. To reduce the computation time, here instead of four photographs we tried to find out the R/B intensity variations for the different Î¦ with one snap shown in Fig. 26.
From fig. 26, it is observed that even, one photograph computation is capable to produce the desired results and accurately sense the proximity of combustor to incipient lean blowout.
In the preceding discussed sensing technique to produce the robust measure of the proximity to LBO, one has to provide the time window or sample length (N). For calculating the stastical parameters it takes some time which hampers the time response characterstic of sensing technique. This major disadvantage of those discussed methods is considerebly reduced in this image processing RGB technique as the computation time for simply reading the images is very less.
In all the above discussed sensing techniques, as soon as LBO warning signal is initiated, active control system should be put in to operation to enhance the flame stabilization in combustor so that the incipient blowout could avoided and LBO limit can be extended. This can be done by either by increasing the pilot fuel  or possibly by using small amplitude fuel modulation .
In this study, in order to elucidate the effects of fuel unmixedness on the dynamics of combustor near blowout limit and asses the behavior and effectiveness of different sensing techniques for the early prediction of blow out , an experimental study was conducted in a swirl stabilized, LPG-Air fueled, dump plane combustor.
The results of this study are as follows.
It is found that with respect to the different fuel injecting condition, two different modes of flame oscillations has been observed. In more premixed configuration like port 1, port2 and port3 the flame oscillates significantly in the entire combustor as the LBO limit is approached. These flame oscillation produces flame loss (extinction) and rapid consumption of unburned fuel (reignition) events which matches with the short duration high amplitude bursts (unstedy events) seen in CH* chemiluminiscence (optical) signal. In port 4 and port 5 case the flame osciilations are not so significant and it is limited to dump plane only subsequently the unsteady bursts seen in CH signal is not so intense. It is found that apart from the fuel injecting condition these unsteady events increases in frequency and duration in all cases and can be used as LBO precursors.
The effect of fuel unmixedness on the behavior and effectiveness of different sensing strategies based on frequency analysis, statistical analysis and direct signal thresholding has been presented. It is observed that the fraction energy in low frequency content increases dramatically as combustor approaches LBO in more premixed configuration like port 1, port 2and port 3 so that this frequency analysis based sensing technique is more suitable in more premixed configuration of combustor. While dealing with the stastical and direct thresholding based technique, two indices i.e. C.D and Î˜ found to be more usefull for sensing the proximity of combustor to incipient lean blowout. The C.D and Î˜ are mathematically interdependent so that their variation is similar with respect to different port condition. It has been observed that the C.D and Î˜ behaves differently from port to port condition. Before using either of this two indices for LBO prediction , a detail priori mapping of the LBO of combustor is needed. .
A new RGB analysis technique has been explored for sensing the signature of LBO, based on the image processing technique carried on the direct digital photography of the flame. It is observed that for all the different ports conditions which gives different mixing distances, the Red/ Blue (R/B) intensity ratio decreases significantly towards LBO, This ratio could be found more suitable quantitative indicator for sensing the incipient lean blowout.
This work was supported by a grant from Defense Research and Development Organization (DRDO) of the Government of India. The author also would like to thank the All India Council for Technical Education (AICTE) for providing opportunity and fellowship to carry the research.
1. Lefebvre, A.H., The role of fuel preparation in low-emission combustion. Journal of Engineering for Gas Turbines and Power-Transactions of the Asme, 1995. 117(4): p. 617-654.
2. Ducruix, S., et al., Combustion dynamics and instabilities: Elementary coupling and driving mechanisms. Journal of Propulsion and Power, 2003. 19(5): p. 722-734.
3. Cohen, J.M., J.H. Stufflebeam, and W. Proscia, The effect of fuel/air mixing on actuation authority in an active combustion instability control system. Journal of Engineering for Gas Turbines and Power-Transactions of the Asme, 2001. 123(3): p. 537-542.
4. Kelsall, G. and C. Troger, Prediction and control of combustion instabilities in industrial gas turbines. Applied Thermal Engineering, 2004. 24(11-12): p. 1571-1582.
5. Gutmark, E., et al., SIMULTANEOUS OH AND SCHLIEREN VISUALIZATION OF PREMIXED FLAMES AT THE LEAN BLOW-OUT LIMIT. Experiments in Fluids, 1991. 12(1-2): p. 10-16.
6. Bayliss, A., G.K. Leaf, and B.J. Matkowsky, Pulsating and chaotic dynamics near the extinction limit. Combustion Science and Technology, 1992. 84(1-6): p. 253-278.
7. De Zilwa, S.R.N., J.H. Uhm, and J.H. Whitelaw, Combustion oscillations close to the lean flammability limit. Combustion Science and Technology, 2000. 160: p. 231-258.
8. Muruganandam., T.M. and J.M. Seitzman., CHARACTERIZATION OF EXTINCTION EVENTS NEAR BLOWOUT IN SWIRL-DUMP COMBUSTORS. Aiaa Journal, 2005.
9. Kukuck, S. and M. Matalon, The onset of oscillations in diffusion flames. Combustion Theory and Modelling, 2001. 5(2): p. 217-240.
10. Joulin, G. and G.I. Sivashinsky, INSTABILITY-INDUCED QUENCHING OF PLANE NON-ADIABATIC FLAMES. Combustion Science and Technology, 1987. 55(1-3): p. 83-88.
11. Muruganandam, T.M., et al., Active control of lean blowout for turbine engine combustors. Journal of propulsion and power, 2005. 21: p. 807-814.
12. Yi, T.X. and E.J. Gutmark, Real-time prediction of incipient lean blowout in gas turbine combustors. Aiaa Journal, 2007. 45(7): p. 1734-1739.
13. Mukhopadhyay, A., S. Datta, and D. Sanyal, Effects of tailpipe friction on the nonlinear dynamics of a thermal pulse combustor. Journal of Engineering for Gas Turbines and Power-Transactions of the Asme, 2008. 130(1): p. 9.
14. Prakash., S., et al., Acoustic sensing and mitigation of lean blowout in premixed flames. Aiaa Journal, 2005.
15. Hong, J.G., et al., Effect of the Unmixedness of Unburned Gases on the Pressure Fluctuations in a Dump Combustor. Energy Fuels, 2008. 22(4): p. 2221-2228.
16. T.M.Muruganandam., et al., Optical and Acoustic sensing of Lean Blowout Precursors. Aiaa Journal, 2002.
17. Lawn, C.J., Distributions of instantaneous heat release by the cross-correlation of chemiluminescent emissions. Combustion and Flame, 2000. 123(1-2): p. 227-240.
18. Bloxsidge, G.J., A.P. Dowling, and P.J. Langhorne, Reheat buzz: an acoustically coupled combustion instability. Part 2. Theory Journal of Fluid Mechanics, 1988. 193: p. 445-473.