An acoustic method of studying sequential explosions during gas combustion in bubbling fluidized bed
An Acoustic Method of Studying Sequential Explosions
During Gas Combustion in Bubbling Fluidized Beds
WITOLD˙ZUKOWSKI
Cracow University of Technology,Institute of Inorganic Chemistry and Technology,30-155Cracow,Ul.
Warszawska24,Poland
Dynamic phenomena associated with the combustion of gaseous mixtures of a fuel and air in a bubbling ?uidized bed have been analyzed.The multiphase hydrodynamic structure of a bubbling?uidized bed leads to combustion which can be either explosive in the bubbles or continuous in the emulsion phase.These two modes of combustion can be distinguished by analyzing the acoustic effects.An algorithm is proposed to determine the frequency of the explosions.This algorithm has been used to examine experimental results obtained in laboratory-scale experiments.It is shown that the frequency of the explosions decreases with rising temperature. This leads to continuous combustion above a certain“transition temperature.”?2001by The Combustion Institute
NOMENCLATURE
B L bit sequence obtained from the
PM L series,
b l value of B L series,
FBC?uidized bed combustion,
h j j-th value of symbol sequence
histogram SSH(n),
L length of the PM L series,
LPG liquid petroleum gas,
M number of experimental results,
n word length,
p i value of the acoustic signal at time t i,
P M experimental results,
PM L series of local maxima of P M
values,
SSH(n)symbol sequence histograms of2n
elements,
T bed temperature,
t i time of the discretization of the
analogue signal,
w value word in the W L series,
W L string of n-bit words obtained from
B L,
?SSH(n)differential symbol sequence
histograms of2n elements,
?t interval between two consecutive
data points in the raw data,
subscripts,
i i-th element in the experimental
record,j j-th of element in the SSH and
?SSH,
l l-th of element in the PM,B and
W strings
INTRODUCTION
Combustion in?uidized beds,operating in the bubbling or circulating mode,is normally used for solid fuels,with the fuel and oxidant in separate phases.However,a?uidized bed can also serve as a reactor for gaseous combustible mixtures,when the presence of the particles organizes and stabilizes the combustion process. Oxidation can take place either in interparticle spaces in the emulsion phase,or inside bubbles and the bed does not have to be catalytic,but it is a problem to distinguish between these two combustion modes.Previous experimental stud-ies[e.g.,1–5]examined different aspects of the combustion of gases in?uidized beds.Experi-ments at incipient?uidization have shown that under some conditions the particulate phase can inhibit the combustion of hydrocarbons[6]. Other work[4,7,8]has shown that the combus-tion of a gaseous fuel in a bubbling?uidized bed of inert material is accompanied by characteris-tic acoustic effects,associated with local pres-sure changes within the reactor and that these effects can be used to study the dynamic phe-nomena responsible for them.
Registration and analysis of acoustic emis-sions from a FBC running on gaseous fuel has
*Corresponding author.E-mail:pczukows@https://www.360docs.net/doc/1111579948.html,.pl.
COMBUSTION AND FLAME125:1075–1082(2001)
?2001by The Combustion Institute0010-2180/01/$–see front matter Published by Elsevier Science Inc.PII0010-2180(01)00229-2
demonstrated [8]that it is possible to describe the emission in terms of certain quantities of statistical signi?cance,such as the mean sound intensity and instantaneous signal minima and https://www.360docs.net/doc/1111579948.html,ing LPG gas as fuel it was found,that over the temperature range 750to 950°C,when the experimental reactor could operate under stable conditions,both the mean signal level and the signal dynamics,measured in terms of the difference between the maximum and minimum signals,decrease with rising tem-perature.On simple considerations,the oppo-site effect could have been expected,since chemical reaction rates and the volumetric gas ?ow rate increase with temperature.Examina-tion of the acoustic signals over short time intervals suggested the existence of different combustion regimes.Depending on the temper-ature range,the process can either be continu-ous and accompanied by a hissing sound,or some of the mixture can burn explosively,with characteristic loud reports.These are associated with explosions and can appear singly,but usu-ally form sequences,when a large signal is followed by a cascade of weaker ones,with diminishing amplitude.These deterministic sig-nals are superposed on a background of sto-chastic ones,associated with the hissing sound.Serial explosions are separated by calm periods;this makes it possible to pick out the series in the time records.The frequency of the explosive events depends on the bed’s temperature.
The aim of this work was to show how to distinguish between sound records for non-explosive combustion and those when se-quences of explosions with different durations and structures take place.Determining the fre-quency of the sequences can help to assess the proportions of the mixture burning under the two regimes and also supply quantitative infor-mation about the dynamics of the process.It has been noted that with low bed temperatures,when explosive combustion in bubbles domi-nates,a sequence of explosions can be acciden-tally interrupted,leading to extinction of com-bustion.This is why acoustic signals can be
used
Fig.1.The algorithm for computing the frequency of explosions series from acoustic signal records registered during the
combustion of a gaseous fuel in a bubbling ?uidized bed under different conditions.
1076
W.Z
˙UKOWSKI
to control combustion and steer a process on-line,but they must?rst be transformed in an appropriate manner.To determine the fre-quency of the explosive sequences an algorithm has been developed and is represented schemat-ically in Fig.1.It can be seen that three inde-pendent procedures can be followed.Signals obtained during combustion below1000°C are transformed and analyzed along the?rst path. Signals registered without explosive events,at over1000°C are transformed following the sec-ond https://www.360docs.net/doc/1111579948.html,parison of the results of these two transformations yield parameters propor-tional to the frequency of the events analyzed.A calibration is then needed to link these param-eters to the physical time scale.This is achieved using the third path in the algorithm,in which a calibration curve is obtained after transformat-ing an arti?cial signal,consisting of stochastic signals with added sequences of known fre-quency,simulating the explosive events.The detailed algorithm,consisting of a number of
steps is listed below.
DATA RECORDING
The experiments used a quartz?uidized bed reactor,96mm in diameter and400mm long, described in detail earlier[8].Quartz sand(300 g;size385to430?m;expanded height of bed?92mm at1000°C)was used as the inert bed material.Air(40%excess;?ow rate1.66dm3/s) and LPG gas were mixed in the plenum cham-ber.With the pressure inside the reactor slightly below ambient and the?ue gases drawn into the exhaust line,the reactor could be kept open at the top.This made it possible to record acoustic signals from the reactor relatively easily by placing a microphone centrally,on the axis of the reactor,over the reactor quartz tube,with-out overheating it.
The electrostatic microphone,adapted for use with computer equipment,produced an analogue signal which was ampli?ed and con-verted to digital form in an A/D converter. These data were then stored on a hard disc.The sampling frequency was high(44.1kHz)in order not to lose information on the dynamic acoustic pressure changes.(Cutting off frequen-cies above5kHz does not change the signal.)The experimental arrangement is shown sche-matically in Fig.2and the changes in the overall signal level in Fig.3.Samples of different signal patterns are illustrated in Fig.4.
DATA REDUCTION
The aim of the?rst step in converting the data are to isolate those elements of the signal which characterize the explosions best.Because se-quences of explosions are represented by groups of peaks,the?rst step for data reduction is an examination of the sizes of the local maxima.As can be seen in Fig.4,the intervals between individual peaks in a series are not equal;hence the times at which the maxima occur are
ig-Fig. 2.Schematic representation of the?uidized bed
reactor.
Fig.3.The characteristic of the acoustic signal obtained for different bed temperatures.A–Amplitude high,the level rises with increasing temperature.B–Amplitude and peak frequency decrease with increasing temperature.C–Hiss-ing sound,level independent of bed temperature.
1077
GAS COMBUSTION IN FLUIDIZED BEDS
nored.By using the local maximum signals and not,for example,values over a certain thresh-old,it is possible to make the analysis indepen-dent of the absolute level of the signals,deter-mined partly by,for example,the distance from the ?uidized bed to the microphone.Also,sequences of explosions differing in intensity can be grouped together.
The raw data consist of pairs of values:the ?rst is the time when probing is made by the A/D converter;the second is the value of the signal.Probing takes place at regular intervals for which the A/D converter is programmed.The digital signal can be described as:P M ???t i ,p i ?,t i ?1?t i
??t ?const.
i ?1...M }
After this step only the values of local maxima remain,all other information is discarded—this amounts to data reduction.The sequence of numbers obtained does not consist of the values collected at equal time intervals,so series of explosions with different intervals between indi-vidual events can be grouped together.After reduction,the series obtained can be described as:
PM L ??p i :p i ?PM L
N p i ?1?p i ∧p i ?1?p i
i ?1...M }
The size of the series PM is much smaller than that of the series P (L ??M ).The analysis step described is illustrated in Fig.5.The Construction of Binary Series
The relationship between the values of local maxima which belong to sequences can be used to distinguish between stochastic sound and the sound produced during a series of explosions.The hissing sound derives from a “background”of independent or weakly linked phenomena in the bed,so the level of a given peak does not depend on the height of the preceding one (or the dependence is weak).On the other hand,for a series of explosions the peaks depend on one another,because one explosion leads to the next.Sequences are attributed to explosions inside large bubbles initiating reaction within neighboring,smaller ones.Not the size of the local maxima,but the type of sequence they form is important here.The signal can contain repeating patterns of local maxima.Such se-quences can comprise as few as two maxima,but are often
longer.
Fig.4.Examples of local dynamic effects:A –continuous
combustion with no explosions,B –a single explosion,C-E –series of different length,E i –high-pressure
peaks.
Fig.5.Characteristic patterns of local maxima during serial explosions.A -860°C,B -885°C.
1078
W.Z
˙UKOWSKI
The second step in the analysis is an exami-nation of the series obtained after data reduc-tion.Only one bit is needed to stand for the difference between two values.A binary series is then de?ned:if a local maximum is followed by a higher one,it is assigned the value unity,otherwise the value is zero.B L ?
?
b l ?0N p l ?1?p l
b l ?1N p l ?1?p l
?
,l ?1...L
The construction of the bit sequences is pre-sented in Fig.6.
The Construction of n-bit Word Series A series of bits re?ects the deterministic char-acter of the acoustic signal on the local scale of two successive explosions or shorter,when be-tween the two explosions considered a local maximum of the acoustic signal occurs.To arrive at series representing typical sequences of explosions,successive groups of bits have to be combined into words.The numerical value of a given word is associated with the unique se-quence of bits,and thus also with a speci?c type of series of peaks in the original acoustic record.If n is the length of the word,then the construc-tion of a string of words can be formally de-scribed as:
W L ?n ?:??
w l ?n ?:w l ?
?j ?0
n ?1b
l ?j
?2j
?
The length of the words must be chosen to ?t the time scale of the phenomena analyzed.Further analysis was made here with n ?4and n ?5.The construction of strings of words is shown in Fig.7.The frequency of appearance of individual word values re?ects the frequency of the phenomena investigated.The hissing sound can also produce a characteristic pattern of word frequency.
Analysis of the Word Series—Symbol Sequence Histograms
The character of the original signal can be described in terms of a histogram of the relative frequency of occurrence of certain values in a word series.The symbol sequence histogram SSH(n)is a set of 2n values:SSH(n):??h j :h j ?p ?w ?j ?,j ?0...2??1}
where
p ?w ?j ?is the probability that w ?j .
Symbol Sequence Histograms (SSHs)can be obtained for acoustic records registered at
dif-Fig.6.Construction of series of bits,A -860°C,B -
885°C.
Fig.7.Construction of series of words,A -860°C,B -885°C.
1079
GAS COMBUSTION IN FLUIDIZED BEDS
ferent ?uidized bed temperatures,including those when explosions do not take place.For example,SSHs derived from different records at 1000°C are shown in Fig.8.It should be noted that acoustic records obtained under the same conditions,particularly of temperature,yield almost identical histograms.Also n ?4and n ?5produce similar histograms,but when n ?5the diagram is more detailed.
Analysis of Word Series—Differential Symbol Sequence Histograms
The SSHs change their shape when changes in deterministic sequences occur in the acoustic signal.To follow these,difference histograms (?SSHs)can be used,derived from subtracting the reference histograms from the actual one.When the intention is to determine the fre-quency of a series of explosions,the ?SSH must be obtained for differences between the fre-quency of occurrence of words obtained for signals from experimental runs with in-bed ex-plosions (?1000°C)and without them (?1000°C).In our case it can be written:?SSH(n)?SSH(n)?T ?SSH(n)?1000°C
The above de?nition leads to the important conclusion,that when there are no explosions ?SSH ?0.At progressively lower tempera-tures,below 1000°C,some values in ?SSH change only slightly,but others increase mono-tonically.From direct measurements of the loudness of the acoustic signal it is known,that with falling temperatures sequences of explo-sions become more frequent [8].This makes it possible to observe changes in the signal asso-ciated with sequences of explosions taking place with a certain mean frequency,as seen in Fig.9.The increase of occurrence sequences of certain peaks in the acoustic record leads to certain peaks in the difference histogram.More fre-quent explosions are re?ected by higher peaks (some positive values)in ?SSH.
The numbers 5,10and 10,21are character-istic and indicate series of explosions,when n ?4and n ?5,respectively.In binary form these values are equal to 0101,1010with n ?4and 10101,01010with n ?5.All of these values derive from one type of “root”binary sequence,which could have different length,according to the actual length of the series of explosions.This root sequence can also end with 0or 1,but inside the sequence 0is followed by 1and 1by
0.
Fig.8.Symbol sequence histograms at 1000°C,A –n ?4
bits,B –n ?5
bits.
Fig.9.Difference symbol sequence histograms (T ?1000°C),A –n ?4bits,B –n ?5bits.
1080
W.Z
˙UKOWSKI
This deterministic dependency in the signal is “detected”by the algorithm,independent (over a limited range)of the value of n.Similar results obtained with two values of n mean that all peaks in ?SSHs presented in Fig.9have the same origin,which is a series of explosions.With falling temperature these peaks become higher—one type of bit pattern appears more frequently in the PM string.This means that one type of deterministic event is more frequent during an experiment.Calibration of ?SSH
The bit sequence 101010can be used to simu-late a series of explosions.It can be added to the bit sequence derived from signals without explo-sions.This leads to a synthetic signal simulating an experimental record with known frequency of explosions.On the basis of this scaling,?SSH can be used to determine the relation between the relative frequency of occurrence of words with values 5and 10with n ?4or 10and 21with n ?5and the known frequency of simu-lated explosive events (see Fig.10).In this way a calibration curve can be obtained,as shown in Fig.11A,where peak height is related to the
mean value of the two most characteristic words in ?SSH’s.The calculations were made using the mean height of “5”and “10”peaks with n ?4or using the mean high of “10”and “21”peaks with n ?5.Thus,the height of peaks derived from experimental record (see Fig.9)can be related to the mean frequency of events,repre-sented as solid lines in Fig.11B.These results in Fig.11B qualitatively demonstrate changes in the dynamics of combustion when the temper-ature is varied.As the temperature falls,explo-sions become possible at 940°C.Then the fre-quency of these events increases until a second characteristic temperature,895°C,is reached,at which the frequency stabilizes.The maximum frequency of appearance of series of peaks is about 14to 16Hz.This means that the mean interval between series is 62to 70ms,which is approximately the same as the residence time of gas in the bed.Thus suggests that a series of explosions consumes almost all of the combus-tible mixture inside the bed and the second series appears when a fresh portion of gas ?lls the bed up to the surface and ignition
occurs.
Fig.10.Reference difference symbol sequence histograms.Short sequences,simulating explosions,added to a series obtained at 1000°C,A –n ?4bits,B –n ?5
bits.
Fig.11.A –calibration curve,obtained from peak height in Fig.10,B –frequency of explosions series and the determi-nation of the temperature when explosions disappear.
1081
GAS COMBUSTION IN FLUIDIZED BEDS
CONCLUSIONS
The changes in the frequency of explosions in the bed show that a gaseous mixture can burn in two ways.Some of the gas reacts periodically in bubbles,but the rest reacts continuously either in the emulsion phase or directly at the distrib-utor.The amounts of gas burning in each man-ner depend on the combustion conditions,such as temperature,gas velocity,particle diameter. It has been shown that it is possible to identify periods of time,when bubbles explode and a parametric relationship can be obtained be-tween the frequency of these periods and the conditions prevailing in the reactor.The method“recognizes”the characteristic patterns of peaks and does not use the absolute value of the acoustic signal,so it is independent of the bubble size,reactor construction or the absolute loudness of the signal and microphone’s posi-tion.Also the speci?c pattern of the bit se-quence associated with an explosive period dur-ing combustion is de?ned during inspection of the?SSH,so it need not be always equal to the formula proposed above.This should also make the method more general.
A frequency analysis of explosions also helps to understand the character of the whole pro-cess in the bed;for example,it shows when mainstream reaction shifts from the bubbles to the emulsion phase or to the distributor.The analysis presented can be used to locate the temperature range,over which the process changes its course.Under the conditions used, the transition range starts at about900°C and ends at940°C.It is characteristic that below 900°C the frequency does not increase with falling temperature and above940°C the explo-sive path for combustion can be neglected. During the experiments at low bed tempera-tures(?800°C),when combustion occurs in the bubbles,it was observed that unexpected extinc-tion of the process could take place.When the explosion frequency is high,combustion must propagate from one bubble to another and thus can lead to lack of continuity of the process. From the practical point of view,a lower fre-quency of explosions means more stable and safer operation.For this reason the algorithm presented could probably be used for online control,for example,to prevent combustion extinction.
REFERENCES
1.Cole,W.E.,and Essenhigh,R.H.,3rd International
Conference on Fluidised Combustion,Washington,
D.C.,1973,Pap.II-5,p.1.
2.Janata,J.,5th International Congress CHISA,Prague,
p.1,1975.
3.Dennis,J.S.,Hayhurst,A.N.,and Mackley,I.G.,19th
Symposium(International)on Combustion,The Com-bustion Institute,Pittsburgh,1982,p.1205.
4.Hayhurst,A.N.,Combust.Flame,85:155,(1991).
5.Van der Vaart,D.R.,and Davidson,J.F.,5th Engi-
neering Foundation Conference on Fluidization,Fluidi-zation V,Denmark,1986,p.539.
6.Hesketh,R.P.,and Davidson,J.F.,Combust.Flame
85:449(1991).
7:Van der Vaart,D.R.,Fuel67:1003(1988).
8:Z˙ukowski,W.,Combust.Flame117:629(1999).
Received22May,2000;revised6November,2000;accepted 7January2001
1082W.Z˙UKOWSKI