A cheap chemiluminescence flame sensor for central heating systems B.B.C.G. van Elten January 2008 WVT 2008.01 Supervisors: ir. J.D.B.J.
van den Boom prof. dr. L.P.H.
de Goey Eindhoven University of Technology Department of Mechanical Engineering Section of Combustion Technology 1 2 / w 2 Contents Abstract Samenvatting 1. Introduction 1.1.1 Motivation 1.1.2 Goals 1.2 Chemiluminescence 1.3 General structure 2. Simulations 2.1 Setup 2.2 Static results and discussion 2.3 Dynamic results and discussion 2.4 Conclusion 3.
Investigation of possible sensor layouts 3.1.1 Suitable sensors 3.1.1 Suitable optics 3.1.1 Supporting electronics 3.1.1 The prototype 3.2 The black-box 3.3 Conclusion 4. Experiments 4.1 Experimental setup 4.2 Static results and discussion 4.3 Dynamic results and discussion 4.4 Conclusion 5. Conclusion and recommendations Nomenclature References Appendixes A: Correcting the non-linear gain of the BC517 B: Costs of electronic parts 3 4 5 5 5 6 7 8 8 9 14 14 15 15 17 18 20 21 22 23 23 25 29 31 32 33 34 35 36 1 2 / w 3 Abstract In industry and home, methane or natural gas is commonly used for heating purposes, like in central heating systems.
The combustion process in these systems is relatively simple. To actively control combustion, to improve efficiency ... more. less.
while maintaining safety, a sensor that can monitor the equivalence ratio ( Æ ) is needed. In modern high efficiency boilers, pressure fluctuations, caused by flame instability, do not even out easily and can lead to resonances.<br><br> To actively reduce flame instability a sensor that can monitor the heat release (Q) is needed. Existing sensors are either expensive, inaccurate or too slow. Is the light emission of a flame, also called chemiluminescence and mainly caused by OH * , CH * , C 2 * and CO 2 * radicals, a good measure for the heat release and for the equivalence ratio?<br><br> Is it possible to build a cheap chemiluminescence flame sensor that can monitor both? In chapter 2, simulation software is used to study chemiluminescence, so less time will be needed to build a working prototype flame sensor. Chapter 3 investigates possible sensor layouts for the prototype sensor and outlines the electronics and optics needed.<br><br> Chapter 4 describes experiments done, on metal surface burners, with the prototype sensor and a photomultiplier. It is concluded that the OH * /CH * chemiluminescence quotient is a very good measure for the equivalence ratio. Of all examined photodetectors a Sharp PT100MC0MP phototransistor is the most suitable for the purpose of this research, but even with amplification by a darlington transistor, the prototype sensor is not able to detect OH * emission.<br><br> CH * emission is detected, but due to the non-linear amplification of the prototype, the OH * /CH * quotient is not independent of U u , when OH * emission is determined with the photomultiplier. Because the prototype sensor is too slow, the dynamic measurements are done with the photomultiplier. The dynamic measurements are suited for predicting instability, because the phase of either OH * or CH * is exactly -90° at the point of resonance.<br><br> Unfortunately an extra sensor, like a microphone, to monitor the fluctuations of U u is needed. At this time it is not possible to build a cheap chemiluminescence sensor that is able to monitor either the dynamic behaviour or the equivalence ratio of a flame, let alone both. 1 2 / w 4 Samenvatting Aardgas wordt, zowel in de industrie als thuis, gebruikt voor verwarmingsdoeleinden, zoals bijvoorbeeld in cv-ketels.<br><br> Het verbrandingsproces in deze systemen is relatief eenvoudig. Om het verbrandingsproces actief te kunnen sturen, zodat de efficiëntie verbeterd wordt terwijl de veiligheid gewaarborgd blijft, is een sensor nodig die de equivalentie verhouding ( Æ ) kan meten. In moderne hr-ketels kunnen drukfluctuaties, veroorzaakt door vlaminstabiliteit, zich niet eenvoudig verspreiden, wat tot resonanties kan leiden.<br><br> Om de vlaminstabiliteit actief te reduceren is een sensor nodig die de warmteafgifte (Q) kan meten. Bestaande sensoren zijn duur, onnauwkeurig of te langzaam. Is de lichtstraling van een vlam, ook wel chemieluminescentie genoemd en hoofdzakelijk veroorzaakt door OH * , CH * , C 2 * en CO 2 * radicalen, een goede maat voor de warmteafgifte en de equivalentie verhouding?<br><br> Is het mogelijk om een goedkope chemieluminescentie sensor te bouwen die beide kan meten? In hoofdstuk 2 is simulatiesoftware gebruikt om chemieluminescentie te bestuderen, zodat minder tijd nodig zal zijn om een prototype vlamsensor te construeren. In hoofdstuk 3 worden mogelijke sensorontwerpen onderzocht en een schets gegeven van de benodigde elektronica en optica.<br><br> Hoofdstuk 4 beschrijft de experimenten gedaan op metalen oppervlakte branders, met de prototype sensor en een fotomultiplicator. De conclusie is dat het OH * /CH * chemieluminescentie quotiënt een zeer goede maat is voor de equivalentie verhouding. Van alle onderzochte fotodetectoren, blijkt de Sharp PT100MC0MP fototransistor het meest geschikt voor dit onderzoek, maar zelfs met versterking door een darlington transistor, kan de prototype sensor geen OH * straling waarnemen.<br><br> CH * straling wordt wel waargenomen, maar door de niet lineaire versterking van het prototype, is het OH * /CH * quotiënt niet onafhankelijk van de snelheid, als de OH * straling door de fotomultiplicator wordt gemeten. Doordat de prototype sensor te langzaam is, zijn de dynamische metingen gedaan met de fotomultiplicator. De dynamische metingen zijn geschikt om instabiliteit te voorspellen, omdat het fase verschil tussen de fluctuaties van OH * en U u of CH * en U u , precies -90° is op het resonantiepunt.<br><br> Helaas, is dan een extra sensor nodig, zoals een microfoon, om de fluctuaties van U u te meten. Op dit moment is het nog niet mogelijk om een goedkope chemieluminescentie sensor te construeren die of het dynamische gedrag of de equivalentie verhouding kan meten, laat staan beiden. 1 2 / w 5 1.<br><br> Introduction 1.1.1 Motivation In industry and home, methane or natural gas is commonly used for heating purposes, like in central heating systems. The combustion process in these systems is relatively simple. By adjusting the flow of the methane/air mixture, the heat release is controlled, but the combustion process itself is uncontrolled.<br><br> A mixture is characterized by the fuel/air ratio, denoted as Æ , or the air/fuel ratio, denoted as » . Of course, for » applies » = 1 / Æ . Both ratios are equal to one if, to burn the mixture, the amount of fuel is just enough to consume all the oxygen in the air.<br><br> Such a mixture is called a stoichiometric mixture. A fuel/air ratio or equivalence ratio, larger than one, also called a rich mixture, will produce a lot of carbon monoxide, which is poisonous for people. As a safety measure, most of these combustion processes use a lean mixture with Æ H 0.75.<br><br> To improve efficiency and for environmental reasons it is better to use a mixture with a fuel/air ratio closer to one. To maintain safety, it is necessary to actively control the combustion. Therefore, a sensor that can monitor Æ is needed.<br><br> Modern central heating systems use high efficiency boilers. Because this type of boiler is a closed system, pressure fluctuations, caused by flame instability, do not even out easily and can lead to resonances. A commonly used method to detect flame instability is to monitor a flame for fluctuations in heat release, denoted as Q.<br><br> To actively reduce flame instability a sensor that can monitor Q is needed. 1.1.2 Goals Equivalence ratio sensors already exist, like » -sensors used in cars, but most sensors have a very small range of operation and are expensive. The types that are able to measure a wide range of equivalence ratios are even more expensive.<br><br> For a car, costing thousands of Euros, a price of ¬50 for a sensor is acceptable, but not for a central heating boiler, costing only a few hundred Euros. Monitoring the heat release of a flame is difficult. With typical frequencies up to 1 kHz, or even higher, a thermocouple is too slow to monitor temperature fluctuations and there is a risk that the flame fluctuates so violently, that it moves past the thermocouple.<br><br> The high temperatures and corrosiveness of combustion products, make monitoring pressure fluctuations with a microphone an unreliable option, apart from the fact that the high temperature gradient causes a non constant sound velocity. An optical sensor, monitoring a flame, does not have these disadvantages. The response to changes in the flame is instantaneous and neither temperature nor corrosion resistance is necessary, because an optical sensor can be positioned away from the flame, behind a sheet of glass.<br><br> However, is the light emission of a flame, also called chemiluminescence, a good measure for the dynamic behaviour and for the equivalence ratio? Is it possible to build a cheap chemiluminescence flame sensor that can monitor the dynamic behaviour and the equivalence ratio? Figure 1.1 shows the general concept.<br><br> An optical sensor monitoring a flame is coupled to a black box containing an electronic circuit with two outputs, Æ and Q. 1 2 / w 6 Figure 1.1: The general concept 1.2 Chemiluminescence Methane is the most basic hydrocarbon and its oxidation reaction is as follows: CH 4 + 2 O 2 ? CO 2 + 2 H 2 O The actual process is much more complicated as can be seen in figure 1.1.<br><br> Figure 1.1: Simplified reaction mechanism of methane combustion [ref. 1] The molecules circled (in red) are examples of so called free radicals, denoted with a * . These excited molecules are very unstable and act as a catalyst to keep the reaction going.<br><br> When returning to their basic state, radiation is emitted. This is called chemiluminescence, of which figure 1.2 shows the principle. 1 2 / w 7 Figure 1.2: Chemiluminescence principle The wavelength of the radiation is dependent on the type of molecule and its state.<br><br> The dominant radicals for hydrocarbon combustion are OH * , CH * , C 2 * and CO 2 * [ref. 1]. Figure 1.3 shows the chemiluminescence spectra for two different equivalence ratios.<br><br> Figure 1.3: Typical chemiluminescence spectra of premixed laminar flames [ref. 2] OH * , CH * and C 2 * emit in narrow bands at 308 nm (ultraviolet), 431 nm (indigo) and 513 nm (cyan/green) respectively, but CO 2 * emits over a broad range. Although OH * and CH * have side bands also, they are not nearly as visible as the second C 2 * band around 465 nm (blue).<br><br> Because C 2 * emission is absent below Æ = 1.0 [ref. 2], OH * and CH * emissions are best suited for the purpose of this study. OH * emission is used a lot in flame detectors, because 308 nm light is not present in daylight, so no shielding is necessary.<br><br> 1.3 General structure Chapter 2 will discus the computer simulations that are carried out to obtain a better understanding of chemiluminescence and its suitability for combustion control purposes. Chapter 3 will explain the choices to be made for the prototype sensor. Chapter 4 will describe the experiments to be done with this prototype sensor.<br><br> Units are put in square brackets, like [cm/s]. References to additional information are also put in square brackets, but starting with 8ref. 9, with numbers for books/papers and letters for internet links, like [ref. a].<br><br> excited state radiation basic state 1 2 / w 8 2. Simulations Simulation software is used to get a better picture of chemiluminescence in a flame, so less time will be needed to build a working prototype flame sensor. Paragraph 2.1 will describe the software, simulation models and used parameters.<br><br> The results of the simulations for static and externally exited dynamic flame behaviour will be presented in paragraphs 2.2 and 2.3. Paragraph 2.4 will contain the conclusions. 2.1 Setup In a simulation a simplified model is used to reduce calculation time.<br><br> In this case the model is reduced to a one-dimensional (1D) space. For flames on surface burners, this approach is very suitable, because only the outer edges of a flat flame have some irregularities, but the rest of the flame is homogeneous. The software used is CHEM1D, which is basically a solver for a system of differential equations, in combination with the GRI 2.11 mechanism.<br><br> This mechanism was developed by the Gas Research Institute and describes CH 4 /NO x chemistry for the combustion of natural gas. There are more variants of this combination, optimized for different types of burners. The one used here is optimized for burner-stabilized flames and modified in a way that the chemiluminescence of OH * and CH * is also calculated.<br><br> Matlab scripts are applied for postprocessing of the resulting data. The data is then displayed in comprehensible graphs. For examination of static flame behaviour the following conditions will be used: " 0.6 d Æ d 1.1 fuel/air ratio or equivalence ratio [-], in steps of 0.01 " T u = 298 unburnt mixture temperature [K], constant " P u = 1.01325·10 5 unburnt mixture pressure [Pa], constant " 7 d U u d 37 unburnt mixture velocity [cm/s], in steps of 1 cm/s For each Æ an fuel/air mixture has a maximum burning velocity, which is the velocity that the flame front can reach in a stationary mixture.<br><br> As the flame front reaches almost the adiabatic temperature, this is referred to as the adiabatic burning velocity (s L ). As obviously no calculation is possible above s L , these velocities are calculated first for all above equivalence ratios. Then all points of lower velocity are processed for each Æ and for each point the following parameters are calculated: " OH * emission per amount of flame surface [W/cm 2 ] " CH * emission per amount of flame surface [W/cm 2 ] " T b burnt mixture (flame) temperature [K] For examination of dynamic flame behaviour a certain static U u is used as a starting point and then a step of 10 -4 cm/s is applied to it.<br><br> CHEM1D is used to calculate the reaction of the flame to this step and the resulting data is processed with a Fast Fourier Transform (or FFT) routine in Matlab, to obtain the frequency and phase response of the flame. Most FFT routines work best when the amount of datapoints is a power of 2 and that is why calculations are done in 5·10 -5 s intervals for a total of 0.1024 s, which will produce 2048 or 2 11 datapoints. This implies that CHEM1D has to perform 2048 separate calculations for each unburnt mixture velocity and therefore the grid was reduced to these points: " Æ = 0.75 ; 0.90 ; 1.05 [-] " U u = 10 ; 20 ; 30 [cm/s] 1 2 / w 9 Because s L H 23 cm/s at Æ = 0.75, U u = 30 cm/s is discarded for this equivalence ratio.<br><br> This leaves 8 points to be calculated. Apart from OH * , CH * and T b used in the static calculations, the following additional parameter is calculated: " U b burnt mixture velocity [cm/s] 2.2 Static results and discussion Figures 2.1 and 2.2 show the simulated emissions of OH * and CH * respectively. The solid (red) lines in the graphs represent the points at adiabatic velocities.<br><br> 5 10 15 20 25 30 35 40 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1 2 3 4 5 x 10 23 U u [cm/s] OH * emission Æ [ 2] OH* emission [W/cm 2 ] Figure 2.1: Simulated OH * emission 5 10 15 20 25 30 35 40 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1 2 3 4 5 x 10 26 U u [cm/s] CH * emission Æ [ 2] CH* emission [W/cm 2 ] Figure 2.2: Simulated CH * emission 1 2 / w 10 The OH * emission is around 10 -3 W/cm 2 and CH * emission is around 10 -6 W/cm 2 . An intensity of approximately the same order is expected [ref. 2, 3].<br><br> Modifying CHEM1D to give the correct absolute values is possible, but for now the results are used as they are. It is displayed that both OH * and CH * emissions are dependent on Æ and on U u . This means that neither of them are suitable on their own as a measure for Æ .<br><br> However, the quotient of the emissions is almost independent of U u , as was also concluded in [ref. 3]. Figures 2.3 and 2.4 show the simulated quotients.<br><br> 5 10 15 20 25 30 35 40 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 0 1000 2000 3000 4000 5000 6000 7000 8000 U u [cm/s] OH * emission / CH * emission Æ [ 2] OH * emission / CH * emission [ 2] Figure 2.3: OH * /CH * emission quotient (displayed with reversed Æ -axis for better visibility) 5 10 15 20 25 30 35 40 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 0 0.2 0.4 0.6 0.8 1 1.2 x 10 23 U u [cm/s] CH * emission / OH * emission Æ [ 2] CH * emission / OH * emission [ 2] Figure 2.4: CH * /OH * emission quotient 1 2 / w 11 Only at low U u , irregularity occurs. Figures 2.5 and 2.6 show the above graphs from a different viewpoint. 5 10 15 20 25 30 35 40 0 1000 2000 3000 4000 5000 6000 7000 8000 OH * emission / CH * emission U u [cm/s] OH * emission / CH * emission [ 2] Figure 2.5: OH * /CH * emission quotient, each line represents points of equal Æ 5 10 15 20 25 30 35 40 0 0.2 0.4 0.6 0.8 1 1.2 x 10 23 CH * emission / OH * emission U u [cm/s] CH * emission / OH * emission [ 2] Figure 2.6: CH * /OH * emission quotient, each line represents points of equal Æ From this point of view it becomes even clearer that the lines in the graph are more or less parallel to the x-axis and are almost independent of U u .<br><br> The quotient to be used for the sensor depends primarily on the flame it needs to monitor. As figures 2.7 and 2.8 show, the OH * /CH * quotient is more linear over a wide area, but the CH * /OH * quotient 1 2 / w 12 has a higher resolution around Æ = 1.0 and is more suited to monitor flames which stay close to stoichiometric equivalence ratios. Because of the wide range of Æ generally used in central heating systems, for this research, the OH * /CH * quotient is the better choice.<br><br> 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 0 1000 2000 3000 4000 5000 6000 7000 8000 OH * emission / CH * emission Æ [ 2] OH * emission / CH * emission [ 2] Figure 2.7: OH * /CH * emission quotient, each line represents points of equal U u 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 0 0.2 0.4 0.6 0.8 1 1.2 x 10 23 CH * emission / OH * emission Æ [ 2] CH * emission / OH * emission [ 2] Figure 2.8: CH * /OH * emission quotient, each line represents points of equal U u Because the OH * /CH * emission quotient graph is very similar to the one found in [ref. 3], it is assumed that shape of the OH * and CH * emission graphs is correct. The absolute 1 2 / w 13 value is probably incorrect, but as this is of minor importance, no attempts were made to modify CHEM1D.<br><br> The flame temperature T b is shown in figure 2.9 and a different viewpoint in figure 2.10. For each U u , T b has a maximum around Æ = 1.0. Most central-heating boilers operate close to Æ = 0.75.<br><br> If operation close to Æ = 0.95 becomes possible, more heat is produced with the same burner or the same amount of heat generated by a smaller burner. 5 10 15 20 25 30 35 40 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1500 1600 1700 1800 1900 2000 2100 2200 2300 U u [cm/s] Flame temperature Æ [ 2] T b [K] Figure 2.9: Flame temperature 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1500 1600 1700 1800 1900 2000 2100 2200 2300 Flame temperature Æ [ 2] T b [K] Figure 2.10: Flame temperature 1 2 / w 14 2.3 Dynamic results and discussion With 2 11 datapoints the FFT graphs produced by Matlab are too coarse in the lower frequency range. The sample frequency is 20 kHz (1 / 5·10 -5 s) and the frequency range is divided into steps of 9,765625 Hz (20 kHz / 2 11 ).<br><br> This means that if a narrow band resonance occurs at 44 Hz, the nearest datapoints are at 39,0625 Hz and at 48,828125 Hz, which is too far apart. The amount of datapoints has to be increased to 2 13 or even 2 14 . Calculation of a single gridpoint with 2 11 datapoints already takes up to several days and due to a possible problem, CHEM1D fails to complete dynamic calculation at some of the chosen grid points.<br><br> It is decided to leave these calculations and move on to the building of the prototype sensor. Experiments with this sensor will be used to study dynamic flame behaviour in more detail. 2.4 Conclusion Neither the OH * chemiluminescence nor the CH * chemiluminescence can be used on its own as a measure for Æ .<br><br> Their quotient is very well suited as a measure for Æ . This means that two optical sensors are needed, instead of one, which will make the sensor and the electronics in the black box more complicated. No conclusive remarks about the examination of the dynamic flame behaviour have been made yet, because of the unfinished dynamic simulations.<br><br> Experiments later on, will be used to form a conclusion on this subject. 1 2 / w 15 3. Investigation of possible sensor layouts The simulations in chapter 2 show that the quotient of the OH * and CH * chemiluminescences is a good indicator for the equivalence ratio Æ .<br><br> This means that the single sensor concept presented in paragraph 1.1.2 needs the addition of an extra sensor, as shown in figure 3.1. Figure 3.1: The modified concept. The choice of sensors, optics and electronics directly involved with these sensors will be described in paragraph 3.1, the electronics in the black box will be briefly discussed in paragraph 3.2 and paragraph 3.3 will contain the conclusions.<br><br> 3.1.1 Suitable sensors There are many different photodetectors. Chemical, electrical and thermal. The commonly used electrical types are divided in semiconductors and tubes: " Semiconductors are cheap and small compared to tubes and unlike tubes do not require a high voltage to operate.<br><br> Common types are: - Photovoltaic cells: A photovoltaic cell or solar cell, like found in small electronic calculators, convert light energy directly into electrical energy. These cells are not very sensitive in the ultraviolet light area and have a low response speed. - Photoresistors: A photoresistor or light-dependent resistor (LDR) decreases its resistance with increasing light intensity.<br><br> The sensitivity peak for this device is mainly in the infrared and far infrared area and the response to change of light intensity is relatively low. - Photodiodes: A photodiode is a normal semiconductor diode in a transparent case, so photons can reach the P-N junction and create a photocurrent. Common P-N photodiodes are very responsive to changes in light but the generated photocurrent is very small and therefore these devices are not used to measure extremely low light intensities.<br><br> PIN photodiodes are much faster and more sensitive, but still the photocurrent is low. Avalanche photodiodes have an internal gain of up to 10 3 typically, but, similar to tubes, need a high voltage of up to 1.5 kV to operate. 1 2 / w 16 Photodiodes that use silicon as semiconductor have a spectral range of 190-1100 nm [ref.<br><br> a]. - Phototransistors: A phototransistor is a normal bipolar transistor in a transparent case, so photons can reach the P-N junction between base and collector. Compared to P-N photodiodes the photocurrent is much higher, depending on the characteristics of the transistor.<br><br> PIN phototransistors have an even higher amplification. Phototransistors that use silicon as semiconductor have a spectral range of 190-1100 nm [ref. a].<br><br> " Tubes are more expensive than semiconductors and much larger and require a high voltage of up to 2 kV to operate. However, because of their larger size these tubes usually have a much larger detector area making them more sensitive than semiconductors: - Phototubes: A phototube is a vacuum tube that is sensitive to light. Like all vacuum tubes, a high voltage is needed for operation and photocurrent is relatively low.<br><br> - Photomultiplier tubes: A photomultiplier tube is a phototube with a very high internal gain, of up to 10 8 typically. The photocurrent is very high. Response to change in light is very fast and noise is low.<br><br> Because of the required low cost, simplicity and long lifetime needed, a semiconductor sensor is chosen for building a prototype. An available photomultiplier is used to verify the measurements done with this semiconductor sensor. From the available semiconductor types, only photodiodes and phototransistors are suitable.<br><br> The irradiance of a flame is very low and because of the higher photocurrent a phototransistor is chosen. Almost all devices specifically developed for ultraviolet light are photodiodes and comparatively expensive, but most common phototransistors, although not designed for ultraviolet light, are reasonably sensitive in the ultraviolet range. After examining datasheets of possible options, the Sharp PT100MC0MP has the best specifications [ref.<br><br> b]. This is a small and cheap (only ¬0.36) SMD (Surface Mounted Device) phototransistor and although its peak sensitivity is at 900 nm (infrared), its relative sensitivity at 400 nm is over 30%, as shown in figure 3.2. This is by far the highest of all examined devices.<br><br> Figure 3.2: Sharp PT100MC0MP - relative sensitivity vs. wavelength. (graph taken from the datasheet [ref.<br><br> b]) Figure 3.3 shows that the response is very linear and because the dark current is approximately 1 to 10 nA, according to the datasheet, the graph will extend well below the 1 2 / w 17 depicted area. Dark current is the current that still flows through the transistor at no-light conditions. Figure 3.3: Sharp PT100MC0MP 3 collector current vs.<br><br> irradiance. (graph taken from the datasheet [ref. b]) 3.1.2 Suitable optics The Sharp PT100MC0MP has a built-in lens with a narrow half intensity angle of 15°.<br><br> This will help to focus both sensors on the same part of the flame. Each sensor should only work in a very narrow part of the frequency spectrum and, therefore, an optical bandpass filter is needed. This can be a coating applied to the built-in lens, but for just a few devices this is too expensive.<br><br> For a large batch of these devices, coated at once, the extra cost of the coating per device can be low. For testing purposes two available optical bandpass filters are used. The Andover 307FS10-25, with peak transmittance of 17% at 307 nm, half transmittance passband of only 10 nm and 25 mm diameter, and the Andover 430FS10-50, with peak transmittance of 49% at 432 nm, half transmittance passband of only 10 nm and 50 mm diameter.<br><br> The measured optical characteristics, as supplied with the filters, are shown in figures 3.4 and 3.5. Figure 3.4: Andover 307FS10-25 light transmittance (measured) 1 2 / w 18 Figure 3.5: Andover 430FS10-50 light transmittance (measured) Note that only 17% and 49% of the light passing through the filters is able to reach the phototransistor. 3.1.3 Supporting electronics For further processing a voltage output is preferred.<br><br> The Sharp PT100MC0MP, like all phototransistors and photodiodes, is basically a light controlled current source. As a consequence extra electronic components are needed. The simplest possible circuit is a combination with only one resistor, which is shown in figure 3.6.<br><br> Figure 3.6: Simplest possible sensor layout. A supply voltage is applied to the + and 3 points at the left side of the circuit and the output is at the right side. Output voltage is given by: V out = I photo ·R1 With: V out output voltage [V] I photo current through the phototransistor [A] R1 resistance R1 [ © ] 1 2 / w 19 R1 is typically 100 © to 1 k © .<br><br> A higher value for R1, increases amplification, but unfortunately output impedance and sensitivity to noise are increased also. Therefore, a value up to 100 k © or 1 M © is probably the maximum, making an amplification ratio of 10 5 to 10 6 possible. Capacitor C1 is only added for extra stability of the supply voltage.<br><br> In very low light situations, as may be expected from a flame, extra amplification may be needed. The most commonly known electronic amplifiers are transistors and operational amplifiers or opamps. The simplest option is to add a transistor as shown in figure 3.7.<br><br> Figure 3.7: Amplification by an added transistor. Most small-signal transistors have a DC current gain (h FE ) of 10 1 to 10 3 typically. A darlington transistor, which is basically two transistors combined in one case, has an h FE of up to 10 5 .<br><br> Output voltage is given by: V out = I photo ·R1·h FE With a darlington transistor and a value for R1 up to 100 k © , an amplification ratio up to 10 10 is possible. Unfortunately, h FE is not constant, but varies with the current through the transistor and with the frequency of the signal. An opamp has a more linear gain.<br><br> A basic opamp circuit is shown in figure 3.8. Figure 3.8: Amplification by an added ideal operational amplifier. Output voltage is given by: V out = I photo ·R1 1 2 / w 20 In this configuration, the opamp copies the voltage over R1 to the output, keeping output impedance low.<br><br> This means that values up to 1 G © are possible for R1, enabling an amplification ratio up to 10 9 . Because an opamp is a complex integrated circuit, it has a tendency to oscillate in this configuration. Capacitor C2 is needed to suppress these oscillations.<br><br> C2 and R1 will form a lowpass filter with a -3 dB point given by: f -3 = 1 / (2· À ·R1·C2) Although C2 is very small, 10 pF typically, the combination with a large value of R1, gives a low f -3 of 150 Hz or 15 Hz for R1 = 100M © or 1 G © respectively. This makes this circuit unsuitable for measuring the dynamic flame response, if a high amplification ratio is needed. Another problem with the circuit of figure 3.8 is that in this configuration, the phototransistor needs a negative bias, which means a more complex circuit as shown in figure 3.9 needs to be used, dropping the overall performance.<br><br> Figure 3.9: Amplification by an added real operational amplifier. 3.1.4 The prototype Figure 3.10: The prototype sensor. 1 2 / w 21 The prototype is constructed on a piece of stripboard using the circuit from figure 3.6.<br><br> An opaque PVC case, approximately Ø 80 x 80 mm, with a holder for the optical filters on the front and banana plug connectors at the back, completes the prototype. Figure 3.11 shows the electronics on more detail. Figure 3.11: Prototype electronics (Together, C1 and C1 9 form C1 of figure 3.6) 3.2 The black box The black box is basically an electronic circuit that processes the sensor signals.<br><br> The simplified layout is shown in figure 3.12. Figure 3.12: The simplified electronic circuit layout of the black box. The signals from the sensors are arriving at the inputs at the left of the layout and then buffered by B1 and B2.<br><br> This is necessary because the sensors have a high output impedance and, without buffering, their output signals would be affected by the electronic components in the black box. The buffered signals are then split up to postprocessors PP1 and PP2. These postprocessors can be either analogue or digital, but 1 2 / w 22 analogue systems are usually simpler, more robust and faster.<br><br> PP2 divides the signal from the 308 nm (OH * ) sensor by that of the 430 nm (CH * ) sensor. The signals are passing lowpass filters LP1 and LP2, before arriving at PP2. These filters average the signal by filtering out all high frequencies.<br><br> This is needed because small variations of the sensor signals can give large differences in the quotient, if the signals are not exactly in phase. Unfortunately, the dynamic simulations are inconclusive, so it is difficult to comment on the operation of PP1 yet. More discussion on this point will follow after the experiments, which will be described in chapter 4.<br><br> 3.3 Conclusion Of all photodetectors a phototransistor is the most suitable for the purpose of this research. A Sharp PT100MC0MP is chosen, because of its parameters [b] and built-in lens. Amplification by a transistor or an operational amplifier might be needed.<br><br> In spite of more linear amplification, an opamp is not used, because of its low bandwidth. 1 2 / w 23 4. Experiments All experiments carried out are described in this chapter.<br><br> Paragraph 4.1 will describe the experimental setup. Paragraph 4.2 will present the results of the measurements on static flames. Paragraph 4.3 will discuss measurements on externally excited flames and paragraph 4.4 will contain the conclusion.<br><br> 4.1 Experimental setup The setup used for static experiments is shown in figure 4.1. Figure 4.1: Static setup A metal surface burner (MB) is temperature stabilized by a surrounding water jacket (WJ). The coolant is kept at a temperature of 90 °C by the temperature control unit (TCU).<br><br> To control the mixture, 2 massflow controllers (MFC1, MFC2) provide the burner with methane and air. They are controlled by a computer, through a multilab (ML). After an input of Æ and U u , the necessary settings for the massflow controllers are calculated by software.<br><br> Sensor (S), which is either the prototype sensor or the photomultiplier, is connected to a digital voltmeter (DVM) and positioned at approximately 10 cm distance at a 45° angle. An optical bandpass filter (OF), in front of the sensor (S), completes the setup. Measurements are done for Æ = 0.7 ; 0.8 ; 0.9 ; 1.0 ; 1.1 and for U u = 10 ; 15 ; 20 ; 25 ; 30 ; 35 cm/s.<br><br> The resulting data is presented in graphs. For the dynamic experiments, the setup is slightly changed, as shown in figure 4.2. The waterjacket (WJ) and the temperature control unit (TCU) have been discarded, because the waterjacket does not fit the dynamic setup.<br><br> The volume behind the burner surface is connected to a loudspeaker (LS), enabling a controlled fluctuation on U u . The loudspeaker is driven by a power amplifier (PA), which is connected to a second computer through a connection block (CB). A hotwire device (HW) is installed just below the burner surface, for a very accurate measurement of U u .<br><br> This device is controlled by a hotwire controller (HWC), which is connected to the second computer. The digital voltmeter (DVM) has also been discarded and the sensor (S) is connected to the second computer. The computer drives the loudspeaker (LS) with a sinusoidal signal and samples the outputs from the hotwire and the sensor for a certain amount of time.<br><br> The resulting data 1 2 / w 24 is stored and an FFT is applied to it. Next, the frequency of the output signal to the loudspeaker is increased and a new measurement is done. The frequency ranges from 5 Hz up to 1 kHz.<br><br> From 5 Hz up to 220 Hz frequency is increased in steps of 5 Hz, from 220 Hz up to 700 Hz the step size is 10 Hz and from 700 Hz up to 1 kHz the step size is 15 Hz. Using Labview scripts, the measuring process is fully automated, except for the adjustment of Æ and U u , of which 5 combinations are chosen. At Æ = 0.8, U u = 15 ; 25 cm/s and at Æ = 1.0, U u = 15 ; 25 ; 35 cm/s.<br><br> The data is presented in gain and phase graphs. Figure 4.2: Dynamic setup For most experiments, the same flat metal surface burner is used. This is a 2 mm thick brass plate of 60 mm diameter (D1) with a hexagonal hole pattern over 50 mm diameter (D2), as shown in figure 4.3.<br><br> Figure 4.3: Flat metal surface burner The hole diameter (d) is 0.5 mm and the pitch (p) is 0.7 mm. The hole size is small enough that a flat methane/air flame stabilizes on top of it [ref. 4].<br><br> Because of the available massflow controllers U u d 33 cm/s for this burner. The measurements done with the photomultiplier, to verify the prototype sensor and the simulations from chapter 2, are therefore done on a burner of the same type, but the hexagonal hole pattern is only 30 mm in diameter. 1 2 / w 25 Initial testing is done with a prototype sensor, constructed using the circuit from figure 3.6, with C1 = 10 ¼ F and R1 = 1 k © .<br><br> This means an amplification ratio of 10 3 . A power supply of 5 V is connected to the red and black connectors on the sensor and a digital voltmeter is connected to the yellow and black connectors. Pointed towards a fluorescent light, the sensor output is in the order of 1 V, but pointed towards a flame the sensitivity is not even enough to detect this flame.<br><br> Increasing the value of R1 to 100 k © , which increases the amplification ratio to 10 5 , does not solve this problem, because the output is still below 0.1 mV. Therefore, extra amplification has to be used, so the prototype sensor is modified according to figure 3.7, with C1 = 10 ¼ F and R1 = 1 k © . T2 is a BC517 darlington-transistor with a very high DC current gain (h FE ) of over 10 5 depending on temperature [ref.<br><br> c]. Figure 4.4 shows the typical h FE of this transistor. Figure 4.4: BC517 typical DC current gain [ref.<br><br> c] Although the gain is clearly not linear, the non linearity will be the same for both OH * and CH * emission measurements. The prototype sensor is now able to detect the emission of CH * , but unfortunately no OH * emission is detected. This is probably caused by a combination of the lower sensitivity of the Sharp phototransistor in the ultraviolet range and only 15% transmittance of the optical bandpass filter.<br><br> Another probable contributing factor is the built-in lens of the Sharp phototransistor. It is made of plastic and most plastics are less transparent for ultraviolet light. An expensive sensor with much more detection surface and a lens made of quartz is needed.<br><br> To complete all tests, the photomultiplier will be used to measure the OH * emission. 4.2 Static results and discussion The measured emissions of CH * and OH * are shown in figures 4.5 and 4.6. At points Æ = 0.7, U u = 25 ; 30 ; 33 cm/s the flame is blown off, so no measurement is possible.<br><br> The graphs have a similar shape as the simulated graphs from chapter 2, but the OH * /CH * emission quotient, as shown in figures 4.7 and 4.8, is not independent of U u . This is probably caused by the non-linear gain of the BC517. Note that even though U u extends beyond s L , the graphs are still smooth and show no conspicuous irregularities.<br><br> This is possible because a real flame, contrary to a 1D simulation, changes shape when U u > s L . The flame will wrinkle and, as a consequence, the surface is enlarged. 1 2 / w 26 10 20 30 40 0.6 0.8 1 1.2 0 50 100 150 200 U u [cm/s] OH * emission Æ [ 2] OH * emission sensor output [mV] Figure 4.5: OH * emission measured by the photomultiplier 10 20 30 40 0.6 0.8 1 1.2 0 50 100 150 200 250 U u [cm/s] CH * emission Æ [ 2] CH * emission sensor output [mV] Figure 4.6: CH * emission measured by the prototype sensor 10 20 30 40 0.6 0.8 1 1.2 0 2 4 6 8 U u [cm/s] OH * emission / CH * emission Æ [ 2] OH * emission / CH * emission [ 2] Figure 4.7: OH * /CH * emission quotient, OH * measured by the photomultiplier, CH * measured by the prototype sensor 1 2 / w 27 5 10 15 20 25 30 35 40 0 1 2 3 4 5 6 7 8 OH * emission / CH * emission U u [cm/s] OH * emission / CH * emission [ 2] Figure 4.8: OH * /CH * emission quotient OH * measured by the photomultiplier, CH * measured by the prototype sensor The output of the prototype sensor ranges from 2 mV to 232 mV, which can be increased a factor 10, to a more practical value of up to 2.32 V, by increasing the value of R1 from 1 k © to 10 k © .<br><br> An opamp is probably more linear, but at this high amplification ratio, the bandwidth is certainly not sufficient. To verify that the cause is indeed the non-linearity of the BC517 and that the simulations are indeed valid, OH * and CH * are measured again, with the photomultiplier. At points Æ = 0.7, U u = 25 ; 30 ; 35 cm/s and Æ = 0.8, U u = 35 cm/s the flame is blown off, so no measurement is possible.<br><br> With both the emissions measured by the photomultiplier, the OH * /CH * emission quotient is practically independent of U u , as shown in figures 4.9 and 4.10. This means that the non-linearity of graph 4.7 and 4.8 is indeed caused by the BC517. Appendix A demonstrates a way to linearise the output of the sensor.<br><br> This also proves that the simulations from chapter 2 are generally correct, although there are dissimilarities. For OH * applies 10 mV d V out d 197 mV and for CH * applies 4 mV d V out d 135 mV. Assuming that the photomultiplier is linear, this means that both emissions are of the same order, like also found in [ref.<br><br> 2, 3], instead of a 10 3 difference. 10 20 30 40 0.6 0.8 1 1.2 0 0.5 1 1.5 2 2.5 3 U u [cm/s] OH * emission / CH * emission Æ [ 2] OH * emission / CH * emission [ 2] Figure 4.9: OH * /CH * emission quotient OH * and CH * measured by the photomultiplier 1 2 / w 28 5 10 15 20 25 30 35 40 0 0.5 1 1.5 2 2.5 3 OH * emission / CH * emission U u [cm/s] OH * emission / CH * emission [ 2] Figure 4.10: OH * /CH * emission quotient OH * and CH * measured by the photomultiplier 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 0 1 2 3 4 5 6 x 10 23 OH * emission Æ [ 2] OH * emission [W/cm²] Figure 4.11: normalised OH * emission measured by the photomultiplier (dashed) simulated OH * emission (solid), circle denotes normalisation point 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 x 10 26 CH * emission Æ [ 2] CH * emission [W/cm²] Figure 4.12: normalised CH * emission measured by the photomultiplier (dashed), simulated CH * emission (solid), circle denotes normalisation point 1 2 / w 29 If normalised to Æ = 0.9 and U u = 25, the measured data differs from the simulated data, as figures 4.11 and 4.12 show. The slope of the measured OH * emission graph (dashed) is slightly flatter than the simulated graph (solid) for U u = 10 ; 15 ; 20, practically an overlay for U u = 25 and the slope is steeper for U u = 30 ; 35.<br><br> For the measured CH* emission graph, the slope is clearly flatter at all U u . 4.3 Dynamic results and discussion A test with reveals that the prototype sensor unfortunately is too slow for doing dynamic measurements. This is because h FE of the BC517 decreases if frequency increases [ref.<br><br> c], so all dynamic measurements are done with the photomultiplier. At Æ = 0.8, U u = 25 cm/s and at Æ = 1.0, U u = 35 cm/s, the resulting data is not as expected [ref. 4, 5].<br><br> U u is very close to s L at these points, because s L H 27 cm/s at Æ = 0.8 and s L H 37 cm/s at Æ = 1.0. With a sinus signal superposed on U u , U u will periodically surpass s L , causing the erratic behaviour. The remaining graphs are presented in figures 4.13 to 4.18.<br><br> 10 1 10 2 10 3 22 0 2 4 6 8 10 Gain f [Hz] |H (s) | [dB] Figure 4.13: Normalised gain at Æ = 0.8, U u = 15 cm/s OH * (dashed) and CH * (solid) emissions and their difference (dotted) 10 1 10 2 10 3 2225 2180 2135 290 245 0 45 Phase f [Hz] ± [°] Figure 4.14: Phase at Æ = 0.8, U u = 15 cm/s OH * (dashed) and CH * (solid) emissions and their difference (dotted) 1 2 / w 30 10 1 10 2 10 3 22 0 2 4 6 8 10 Gain f [Hz] |H (s) | [dB] Figure 4.15: Normalised gain at Æ = 1.0, U u = 15 cm/s OH * (dashed) and CH * (solid) emissions and their difference (dotted) 10 1 10 2 10 3 2225 2180 2135 290 245 0 45 Phase f [Hz] ± [°] Figure 4.16: Phase at Æ = 1.0, U u = 15 cm/s OH * (dashed) and CH * (solid) emissions and their difference (dotted) 10 1 10 2 10 3 22 0 2 4 6 8 10 Gain f [Hz] |H (s) | [dB] Figure 4.17: Normalised gain at Æ = 1.0, U u = 25 cm/s OH * (dashed) and CH * (solid) emissions and their difference (dotted) 1 2 / w 31 10 1 10 2 10 3 2225 2180 2135 290 245 0 45 Phase f [Hz] ± [°] Figure 4.18: Phase at Æ = 1.0, U u = 25 cm/s OH * (dashed) and CH * (solid) emissions and their difference (dotted) Because only one photomultiplier is available, OH * and CH * emissions have to be measured separately. At low Æ and low U u , flames are less stable than near Æ = 1.0 and U u = s L , which is why with increasing Æ and U u , the OH * and CH * lines are closer together. If these measurements are done with two photomultipliers simultaneously, the OH * and CH * lines will be nearly identical.<br><br> Resonance peaks are at approximately 300 Hz, 650 Hz and 750 Hz respectively, which is higher than expected [ref. 4, 5]. Because the burner is not cooled during the dynamic experiments, it becomes very hot, like ceramic surface burners do.<br><br> As a result, the flame standoff distance, which is the distance between burner surface and flame, becomes very small. The phase corresponding to the resonance peaks is very close to ± = -90° in all cases, which is exactly as expected. This means that the phase difference between the fluctuations, on either OH * and U u or CH * and U u , is a very good measure for resonance.<br><br> PP1 from figure 3.12 needs to output approximately 0 V for ± e 0°, increase the output to a certain level for ± = 0° ? -90° , decrease the output for ± = -90° ? -180° and 0 V again for ± d -180°.<br><br> Unfortunately, a sensor to monitor the fluctuations on U u needs to be added. Although this is possible with a cheap microphone, this still adds more complexity and costs. 4.4 Conclusion Even with amplification, the prototype sensor is not able to detect OH * emission.<br><br> CH * emission is detected, but due to the non-linear amplification, the OH * /CH * quotient is not independent of U u . The prototype sensor is too slow to be used for dynamic measurements, so these measurements were done with a photomultiplier. The dynamic measurements are very suited for predicting instability, because the phase difference between the fluctuations, on either OH * and U u or CH * and U u , is exactly -90° at the point of resonance.<br><br> Unfortunately an extra sensor, like a microphone, to monitor the fluctuations of U u is needed. 1 2 / w 32 5. Conclusion and recommendations The chemiluminescence of a flame is a very good measure for its dynamic behaviour and for its equivalence ratio.<br><br> Although neither the OH * chemiluminescence nor the CH * chemiluminescence can be used on its own as a measure for Æ , their quotient is very well suited as a measure for Æ . This means that two optical sensors are needed, instead of one, which will make the sensor and the electronics in the black box more complicated. Of all photodetectors a phototransistor is the most suitable for the purpose of this research.<br><br> A Sharp PT100MC0MP is chosen, because of its parameters and built-in lens, but even with amplification by a darlington transistor, the prototype sensor is not able to detect OH * emission. CH * emission is detected, but due to the non-linear amplification of the prototype, the OH * /CH * quotient is not independent of U u determined with the photomultiplier. In spite of better linearity, an opamp is not used in the prototype, because of its low bandwidth.<br><br> Because the prototype sensor is too slow, the dynamic measurements are done with the photomultiplier. The dynamic measurements are suited for predicting instability, because the phase difference between the fluctuations, on either OH * and U u or CH * and U u , is exactly -90° at the point of resonance. Unfortunately an extra sensor, like a microphone, to monitor the fluctuations of U u is needed.<br><br> At this time it is not possible to build a cheap chemiluminescence sensor that is able to monitor either the dynamic behaviour or the equivalence ratio of a flame, let alone both. 1 2 / w 33 Nomenclature symbol definition unit ± phase angle ° Æ fuel/air ratio or equivalence ratio - |H (s) | gain of transfer function |OH * 9/U u 9| or |CH * 9/U u 9| - (the accent denotes a fluctuation, so U u 9 is a fluctuation of U u ) P u unburnt mixture pressure Pa T u unburnt mixture temperature K T b burnt mixture (flame) temperature K U u unburnt mixture velocity cm/s U b burnt mixture velocity cm/s s L adiabatic burning velocity cm/s OH * emission per amount of flame surface W/cm 2 CH * emission per amount of flame surface W/cm 2 h FE a transistors DC current gain - Cx capacitor x F Rx resistor x © V out output voltage V I photo current through the phototransistor A 1 2 / w 34 References Books / Papers  H.N. Najm, P.H.<br><br> Paul, C.J. Mueller and P.S. Wyckoff On the adequacy of certain experimental observables as measurements of flame burning rate Sandia National Laboratories, Livermore, CA 94551, USA  J.<br><br> Kojima, Y. Ikeda and T. Nakajima Spatially resolved measurement of OH * , CH * , and C * 2 chemiluminescence in the reaction zone of laminar methane/air premixed flames Department of Mechanical Engineering, Kobe University Rokkodai, Nada, Kobe 657-8501, Japan  Y.<br><br> Hardalupas and M. Orain Local measurements of the time-dependent heat release rate and equivalence ratio using chemiluminescent emission from a flame Mechanical Engineering Department, Imperial College London, Exhibition Road, London SW7 2BX, UK  R. Rook Acoustics in burner-stabilised flames Eindhoven University of Technology  L.P.H.<br><br> de Goey, J.H.M. ten Thije Boonkkamp and V.N. Kornilov Response of laminar flames to acoustic waves Department of Mechanical Engineering, Eindhoven University of Technology, P.O.<br><br> Box 513, 5600 MB Eindhoven, The Netherlands Websites [a] http://en.wikipedia.org/wiki/Photodiode [b] http://www.farnell.com/datasheets/83301.pdf or: http://www.farnell.com/datasheets/62633.pdf [c] http://www.onsemi.com/pub/Collateral/BC517-D.PDF [d] http://nl.farnell.com/ 1 2 / w 35 Appendix A Correcting the non-linear gain of the BC517 It is possible to correct the signal from the prototype sensor, to obtain a linear output, by using V corrected = V measured 0.7 . Figures A.1 and A.2 show the corrected versions of 4.7 and 4.8. However, with analogue black-box electronics this is very difficult to implement.<br><br> 10 20 30 40 0.6 0.8 1 1.2 0 0.5 1 1.5 2 2.5 U u [cm/s] OH * emission / CH * emission Æ [ 2] OH * emission / CH * emission [ 2] Figure A.1: Corrected OH * /CH * emission quotient, OH * measured by the photomultiplier, CH * measured by the corrected prototype sensor 5 10 15 20 25 30 35 40 0 0.5 1 1.5 2 2.5 OH * emission / CH * emission U u [cm/s] OH * emission / CH * emission [ 2] Figure A.2: Corrected OH * /CH * emission quotient, OH * measured by the photomultiplier, CH * measured by the corrected prototype sensor 1 2 / w 36 Appendix B Costs of electronic parts To give an insight in the total costs of the prototype sensor, the electronic components and their price are listed below. Prices are per unit and for large quantities (+100) [ref. d].<br><br> Actual costs for the prototype sensor are higher, because of the small quantity of parts bought. resistors (1% tolerance) ¬ 0.02 electrolytic capacitor (10 ¼ F) ¬ 0.03 ceramic capacitor (10 pF) ¬ 0.02 transistor (BC517) ¬ 0.06 phototransistor (PT100MCoMP) ¬ 0.36 suitable operational amplifier ¬ 2.50