On-line tracking of pulverized coal and biomass fuels through flame spectrum analysis

第28卷第11期仪器仪表学报Vol 28 No. 11007年1l月Chinese Journal of Scientific InstrumentNov,2007On-line tracking of pulverized coal and biomass fuelsthrough flame spectrum analysisChi Tianyang(I State Key Laboratory of Industrial Control Technology, Zhejiang University, Hangzhou 310027, China;2 College of Information Science and Electronic Technology, Jiamusi Uninersity, Jiamusi 154007, China)Abstract: This paper presents a new approach to the on-line tracking of pulverized coal and biomass fuels throughanalysis. A flame detector containing four photodiodes is used to derive multiple signals covering aof the flame from visible, near-infrared and mid-infrared spectral bands as well as a part of far-infra-red band. Different features are extracted in time and frequency domains to identify the dynamic"fingerprints"of theflame. Fuzzy logic inference techniques are employed to combine typical features together and infer the type of fuelbeing burnt. Four types of pulverized coal and five types of biomass are burnt on a laboratory-scale combustion testrig. Results obtained demonstrate that this approach is capable of tracking the type of fuel under steady combustionconditionsey words: biomass; pulverized coal fuel tracking; flame spectrumfore, even if the measurement of the sample is quite accu-1 Introductionrate, it does not al ways represent the entire fuel. another re-search direction is utilizing flame monitors that are sensitivePulverized coal is the main fuel in electrical power genera- to the heat radiation of different fuels 56). But these flametion. The security and economy of the pulverized coal boilers monitors analyse the fuel flame from near-infrared(NIR )totant source of renewable energy as the most important fuel small part of the heat radiation spectrum . which is still aare closely related to coal types. Biomass is also an impor- ultraviolet(UV)through the visible bandworldwide after coal, oil and natural gas. As coal reserve isThe heat radiation spectrum covers a wide range of wave-finite and subject to depletion as it is consumed, biomass is lengths from 0. 38 um to 100 um. Most heat energy distrib-becoming the substitutes for pulverized coal in many power utes inlengths from0.76μmto20μm, while theplants. A power station can have a wide range of fuels in its visible band (0. 38 um to 0. 76 um)possess a very smallstock but what type of fuels being burnt at any moment is of. portion". Biomass and pulverized coal may be different inen unknown and even unpredictable. Therefore on-line fuel physical ingredients and chemical elements. Flames genera-tracking at a power station where a wide range of fuels isfrom different fueed would help to increase the combustion efficiency and (fuel flownmany/secondary air flow rate etc. )pos-afety of the boiler. On the other hand biomass reduces sess different features both in time and frequency domainscapable of simultaneously addressing the energy, environ- identify and track the type of fue//sy unique signatures togreenhouse gas emissions and is abundantly available, it is Such flame features can be extracted as unique signatures toecononmIcIn this paper a new flame detector covering a wider rangeline coal analyzers using radiation( gamma ray, neu- of wavelengths is presented to gain more information from thetrons etc.), microwave and infrared techniques are available visible band to the far-infrared through the near infrared andon the market 2. However, these systems are very expen- mid-infrared regions. Flame features extracted from the timesive and difficult to operate. In theand fr中国煤化工* the biomass ansample only a small fraction of the materialThere. coalHis being burntCNMHGDate:200708收稿F期:200708Foundation item Supported by the Key Program of the National Nalural Science Foundation of China(60534030)11 M Chi Tianyang et al: On-line tracking of pulverized coal and biomass fuels through flame spectrum analysis2009The skewness and kurtosis of signal x, indicate the degree2 Methodologyof asymmetry and steep of the signal probability density func-tion( PDF)(1)System descriptionsk1=1∑(-The whole system consists of a flame deteclor, signal pro-c::中6m=x()(5)components of the system. The flame detector has four photo- In frequency domain, the quantitative flicker F, is defineddiodes covering the visible, near-infrared and mid-infrared as the weighted averrage frequency over the entire frequencyspectral bands as well as a part of far-infrared band. The de- range ol. It has been suggested that the lower frequencytector derives flame signals containing the dynamic character- components of flame flicker could be due to flame shape flucistics of the flame and hence the fuel typetuations caused by aerodynamic or convective effects whilethe higher frequency components reflect vibrational and rota-umacetional energy transitions in the intermediate radicals or kinet-sectoric variations in the rate of energy emission of the reactingBoilerFig. l Block diagram of the fuel tracking systemF、SP(i=1,2,3,4(6)The wavelengths detected are from 0. 4 um to 7 um. Ac- Where n is the lotal number of the discreteforcording to the spectral response of each photodiode, we use the power spectral density( PSD)analysis, f.for theband I to band 4 to stand for the wavelengths covered by the j th discrete frequency of the ith signal and pfour photodiodedensity of the jth frequency component of the ith signal.Band 1, visible bandIn general, all the features described above can be usedBand 2 former half part of NIRidentify the type of fuel. However, not all features are equalBand 3 back half part of NiRly sensitive to the type of fuel. Biomass and pulverized coalBand 4: back part of NiR, MIR and a part of FIRare two different kinds of fuel. They have their own sensitive(2)Feature extractionfeatures. The features that are most sensitive to the differentIn time domain the mean and standard deviation charac- types of fuel will be selected based on experitest daenize the steady-state(DC)A(i=1, 2, 3 and 4 denoting tathe signals of band 1 to 4 respectively)and the dynamic al- (3)Fuzzy inferenceternating component(AC)o,(i=1, 2, 3 and 4 denoting As we cant judge the type of fuel directly according to thethe signals of band 1 to 4 respectively )of the flame radia- feature value calculated. A fuzzy-logic-based inference sys-tion. The DC component has been identified to be dependent tem has been developed to determine the type of fuel beingon the volatility of the fuel, the size of the flame and the bunt. The aforementioned features have been used as fuzzycontribution to the brightness from the hot surroundingsinputs. The mean and standard deviation of a feature are the1x(i=1,2,3,4),=/1∑(x-)2(:123,4)commonly used according to the minimum algorithm to define(2) fuzzy relatiThe output of the fuzzy inference sys-Where N is the sample length and x is the jth sample point of tem has a good response to the type of fuelthe ith signalMean square value D, reflects the signal energy changes in 3中国煤化工time domainNMHGDboiler buming pure biomass is differ-ent from that burning pulverized coal, we find a combustion2010仪器仪表学报第28卷test rig to substitute the boiler that can bum both biomass fuels can be tracked; in band I and 2, different coals can beand pulverized coal. A series of experimental tests was car- identifiedried out under steady combustion conditions( sufficient airsupply). The biomass fuels under test are wood, fire wood,straw,peanut shell and com leaf. The pulverized coal testedare Coal A, Coal B, Coal C and Coal D. Fig. 2 shows theimages of the biomass and pulverized coals tested(a)Wood(b) Fir wood(c)Stn,.Sampling spot(d)Peanut shell (e)Com leaf(n Coal A(a)Biomass(g)Coal B(h)Coal C(i)Coal DFig 2 Images of the biomass and pulverized coalTypical signals derived from the biomass and pulverizedplotted in fig 3frequency is 100 Hz with the sampling length of 600. Differ-ent fuels appear to have different dC and AC levels at band 1to 4. The four signals obtained from each fuel are correlatedwhich implies that the natural fluctuation of the flame affectsthe flame signal in each band but to a different extenThe averaged DC level and AC level of biomass and coalsignals in band 1 to 4 are compared in fig 4 and fig. 5 re-Sampling spoteraged DC level error bar of each(b) Coalindicates the uncertainty of each feature. It can be中国煤化工masthe DC level for biomass that there are significant differencesCNMHGin band 3 while for pulverized coal in band I and 2. On theother hand from AC level in band 1 and 3. different biomass11# Chi Tianyang et al: On-line tracking of pulverized coal and biomass fuels through flame spectrum ana2011Kurtosis and skewness of the biomass and coal signalsalso calculated. However, there is no significant differencein skewness among different types of coal. From fig. 6, kurosis in band 2 of biomass data can be a feature to trackferent fuels. while in band 2 and 3 of coalband(a)BiomassWave band(a)Biomassand 3band 4ig. 4 Average DC level of the fame signalsband 2Wave bandFig. 6 Kurtosis of the flame signalsIn frequency domain, different fuels have different flicker frequencies. Some fuels(. g. straw in band 1)flicker more thanothers but in small amplitude while other fuels(e. g. fir woodfound that there isdifference of coal in band 1 andand 32 From fig. 7, biomass average flicker frequency in band 2 and(a) Biomasscoal in band 4 can be a feature to track different fuels中国煤化工CNMHGFig 5 AC level of the flame signalsFig. 7 Average flicker frequency of the signal2012仪器仪表学报第28卷2 and average flicker frequency in band 2 are used as the in-puts to the fuzzy inference system, while for coal average DCin band 1. Ac level in band 2, kurtosis in band 2 and aver-8ggteaged flicker frequency in band 4 are the inputs. The meanand standard deviation are the parameters of Gaussian membership functionTable I and Table 2 list the means and standard deviations" couI Dof typical features of the flames for each fuel, which are usedto configure the corresponding Gaussian membership funcWave bandtion. The triangle membership function is chosen as the output of fuzzy logic inference. The inference process for a setFig 7 Average flicker frequency of the signalsof typical input features is shown in fig 8 where biomass typeOther features may overlap in value for different fuels and 1 to 5 represent wood, fir wood, straw, peanut shell andcause vagueness in fuel tracking. They can also be utilizedcom leaf respectively, for coal type I to 4 are the counter.as one component in fuel tracking by combining the features parts of Coal A, coal B, Coal C and Coal D respectively. AFurry logic inference is employed to combine typical fea- correct decision on fuel type can be seen clearly from thetures together and infer the type of the fuel. For biomass, fuzzy inference process and hence prove the efectiveness ofaverage DC in band 3, AC level in band 1, kurtosis in bandTable 1 Mean and standard deviation of the features for biomassFeaturesAverage DCKurtosisFlicker fraquencySTDMeanMeanWood3.8l250.l1060.46l10.02314.60860.23048.83250.54160.17760.38470.01924.34280.28718.54620.32730.43930.06430.00322.38630.11932.67130.1335Peanut shel7.45960.60310.14960.01753.40670.270312.1740.8863Com leal5.74030.19410.0730.00372.70960.13559.7642Table 2 Mean and standard devlation of the featuresAverage DCKurtosisFlicker frequencyoA2.71780.l890.20l0.01012.97610.14882.16350.1782B7.64960.l6280.061300l3l2.55430.12773.36520.1683aC9.7850.00030,0110.00261.53520.17684.83740.2419ColD2.31380.07030.03450.00172.09360.10471.83670.0918Are._DC-6. AC_level-0.06 Kurtowiss225 Ave-requenc23Are_ DC-7.5 AC_level-0.05Kurtosia=23Ave. frequebcy=3.3中国煤化工(a)Biomass furry inference processCNMHGFig 8 Inference process of the fuzzy systemA 11 # Chi Tianyang et al: On-line tracking of pulverized coal and biomass fuels through flame spectrum analysisdentification using digital signal processing and soft-com-4 Conclusionsputing techniques[C]. IMTC 2003[9] JONES AR. Flame failure detection and modem boileDifferent features in time and frequency dumains have[J]. J Phys. E: Sci. Instrum., 1988, 21: 921-928.been extracted from flame signals of five biomass fuels and [10] HUANG Y P, YAN Y, LU G, et al. On-line licker meas-four pulverized coals, Four features are found to be effectivurement of gaseous flames by image processing and specfor fuel tracking. Fuzzy logic inference techniques have beentral analysis [J]. Meas. Sci. Technol, 1999, 10successfully applied to combine different features together726-733and infer the type of fuel being bumt. Experimental results [11] WU Xl, LIN ZH H. MATLAB assist fuzzy system designobtained on a laboratory scale test rig have demonstrated thatM]. Xi' an: Xidian University Press, 2002.the system is capable of tracking the type of fuelBiographies迟天阳,女,博士研究生,现为佳木斯大学讲师。浙江大学工业控制技术国家重点实验室,主要研究方向为控制科学5 Acknowledgement与工程。地址:浙江大学工业控制技术国家重点实验室,杭州,中The authors wish to thank Prof. Y. Yan of the University E, 310027of Kent, UK for his advice on the work as reported in thisEE TE: 0571-8795225; E-mail: tyche iipc zju. edu.cnpapeChi Tianyang, female, is from Harbin in Heilongjiang provReferencesince and a doctoral student in State Key Laboratory of IndustrialControl Technology, Zhejiang University. She majors in control[i]htp://www.ecologycom/archived-ink/biomase/indscience and engineering. She is a lecturer in Jiamusi Universityhtml[ EB/Ol]Address: State Key Laboratory of Industrial Control Technology2]WOODWARD R C, EVANS M P, EMPTY E R. A major Zhejiang University, Hanghou 310027, Chinastepforwardforon-linecoalanalysis[EB/OL].ThermoTel:+86-571-87952253:E-mail:tychi@ipezju.edu.cnElectronCorporationfrOmhttp://www.thermocomv张宏建,男,工学博士,教授浙江大学工业控制技术国om/CMA/Files/ articles File_20692. pdf家重点实验室,主要研究方向为过程参数测量理论与技术[3]BBrM, VOURVOPOLLOS G, PASCHAL J. A coomer信号处理及算法。cial on-line coal analyzer using puised neutron[B0L-地址:浙江大学工业控制技术国家重点实验室,杭州,中Fromhtp//www.wledw/apv/pub/caaRizocoal.pdf国,310027[4] GANGULI R A critical review of on-line quality analyzers 4t ta: 0571-87952253: E-mail hezhang@ ipe zju.edu.en[J]. Mineral Resources Engineering. 2001, 10(4)Zhang Hongjian, male, received B. Sc. and M. Sc. degrees43544from Zhejiang University, China, in 1982 and 1985, respectiv[5] XU LI, YAN Y, CORNWELL S, et a. On-line fuel iden- ly. In 1988, He received Ph. D. degree from University otification using digital signal processing and fuzy infer- Shanghai for Science and Technology, China. Currently, he is aence techniques[ J]. IEEE Transaction on Instrumentation professor in Department of Control Science and Engineering, Zhe-and measurement,2004.53(4):1316-1320.jiang University. His research field includes measurement theory[ 6] XU L J,YAN Y. A new flame monitor with triple photo- and techniques for process parameters, signal processing and itsand Measurement. 2006, 55(4): 1416-1421Address: State Key Laboratory of Industrial Control Technology[7] YANG S M, TAO W Q. Heat transfer[M]. 3rd ed Bei- Zhejiang University,Hangzhou310027,Chinajing: High Education Press, 1998+86-571-87952253; E-mail hjzhang@ iipc. zju. edu. en[8 XU J L, YAN Y, CORNWELL S, et al. On-line fuel i-中国煤化工CNMHG