Particle surface temperature measurements with multicolor band pyrometry

Particle Surface Temperature Measurementswith Multicolor Band Pyrometry

Hong Lu, Leong-Teng Ip, Andrew Mackrory, Luke Werrett, Justin Scott, Dale Tree, and Larry BaxterChemical Engineering Dept., Brigham Young University, Provo, UT 84602

DOI 10.1002/aic.11677Published online November 10, 2008 in Wiley InterScience (www.interscience.wiley.com).

A noncontact, color-band pyrometer, based on widely available, inexpensive digitalimaging devices, such as commercial color cameras, and capable of pixel-by-pixel re-solution of particle-surface temperature and emissivity is demonstrated and described.This diagnostic instrument is ideally suited to many combustion environments. Thedevices used in this method include color charge-coupled device (CCD), or comple-mentary metal oxide semiconductor (CMOS) digital camera, or any other color-render-ing camera. The color camera provides spectrally resolved light intensity data of theimage, most commonly for three color bands (Red, Green, and Blue,), but in somecases for four or more bands or for a different set of colors. The CCD or CMOS sen-sor-mask combination has a specific spectral response curve for each of these colorbands that spans the visible and often near infrared spectral range. A theory is devel-oped, based on radiative heat transfer and camera responsivity that allows quantitativesurface temperature distribution calculation, based on a photograph of an object inemitted light. Particle surface temperature calculation is corrected by heat transferanalysis with reflection between the particle and reactor wall for particles located infurnace environments, but such corrections lead to useful results only when the parti-cle temperature is near or below the wall temperatures. Wood particle-surface temper-atures were measured with this color-band pyrometry during pyrolysis and combustionprocesses, which agree well with thermocouple measured data. Particle-surface tem-perature data simultaneously measured from three orthogonal directions were alsomapped onto the surface of a computer generated 3-D (three-dimensional) particlemodel. � 2008 American Institute of Chemical Engineers AIChE J, 55: 243–255, 2009

Keywords: color-band, pyrometry, particle, surface, temperature, measurement

Introduction

Particle temperature determines reaction rates and productdistributions and yields, and, hence, is a critical combustionparameter. Optical pyrometry1–6 is an established, noncontacttemperature measurement technique for solid fuel particles,soot particles, and flames. Traditional pyrometry consists of

isolating narrow spectral line intensities with narrow band fil-

ters and recording signal intensities on, for example, photon

multiplier tubes. Line intensity at a single wavelength can be

used to estimate temperature if the surface emissivity, emitting

area, and optical and system parameters are known. Line

intensities at two or more wavelengths can eliminate the need

to know the surface emissivities (so long as they are the same

at each wavelength), emitting areas, etc., in a technique com-

monly called two-color or multicolor pyrometry. A digital

color image offers the potential to extend this technique to

obtain pixel-by-pixel temperature and emissivity measure-

ments, with the rapid commercialization of such cameras

Correspondence concerning this article should be addressed to H. Lu [email protected].

� 2008 American Institute of Chemical Engineers

AIChE Journal January 2009 Vol. 55, No. 1 243

driving prices down and performance up such that inexpen-sive and highly capable systems can be built on current tech-nologies. Such digital images spectrally resolve light intobroad bands,7–8 as opposed to narrow lines more traditionallyassociated with pyrometry. However, mathematical algo-rithms that are conceptually similar to the multicolor (andone-color) techniques traditionally used in pyrometry rendersuch images highly accurate and relatively sensitive meas-ures of particle temperature. We call this technique and algo-rithm color-band pyrometry to distinguish it from traditionalpyrometry methods.

Color-Band Pyrometry Algorith Development

Color-band pyrometry principles

Blackbody spectral radiance expressed as Bbk 5 C1k

5/(eC2/kT

21), forms the basis of Planck’s law. For a gray body: Bk 5ek Bb

k, where e(k) is the spectral emissivity. The total radi-ance is B 5 $10 Bkdk. In the wavelength range between k1and k2, the total radiance is B 5 $k2k1 Bkdk. For a typical py-rometer illustrated in Figure 1 (where, the point light sourcedimension is much smaller than the distance d from thesource to the receptor - lens, and d � D is usually true formost applications), the solid angle is X 5 pD2/4d2, where Dis the lens (receptor) diameter, and d is the working distancebetween the lens and the point light source.

Assuming the total effective extended source area is A1,the spectral radiant power and total radiant power incidenton the lens are Fi,k 5 BkA1pD

2/4d2, and Fi 5 BA1pD2/4d2,

respectively.If the transmittance (which is the is the ratio of the total ra-

diant or luminous flux transmitted by a transparent object tothe incident flux, usually given for normal incidence) of theoptical system is sk, the spectral energy incident on the imag-ing sensor will be Fi,k,sensor 5 BkskA1pD

2/4d2. With an effec-tive image area of A2 on the imaging sensor, which corre-sponds to the effective extended light source area, and a pixel/cell area of a on the CCD or CMOS sensor, the spectral irradi-ance obtained by a specific pixel/cell is Ek 5 BkskaA1pD

2/4d2A2. Generally, A2/A1 is proportional to the magnificationfactor of the lens, and here we can use X to replace it.

Usually there are two methods to describe the spectral sen-sitivity/responsivity Sk, of an imaging sensor. One is quan-tum efficiency (QE, electrons/photon), which is the photon-to-electron conversion efficiency; the other one uses theenergy to electron conversion efficiency. In addition, the

electron to digital number (pixel intensity) conversion isrelated the gain value of the CCD or CMOS camera.

The energy of a photon as a function of frequency orwavelength is E 5 hc/k. If QE is used as the spectral sensi-tivity/responsivity, with an exposure time of Dt, the digitalnumber (DN) or pixel intensity of any pixel in the image iscalculated by Eq. 1 for an ideally performing (perfectly lin-ear) black-and-white CCD or CMOS camera based on theoptical system and camera characteristics illustrated earlier.

DN ¼ a � p4� D

d

8>:9>;2

�f ðgÞ � aX� Dt

�Zk2

k1

ekskSk � C1 � k�5

eC2=k�T � 1:

h � ck

8>:9>;�1

�dk; ð1Þ

where, a is a proportional factor that ensures the units con-sistent for the equation, and f (g) is a function of gain valueof the camera; all other variables are explained earlier and inthe notation section.

The spectral radiances of the object surface from thelowest wavelength to the highest wavelength that can bedetected by the image sensor all contribute to the pixel inten-sity, by contrast to only two narrow wavelengths in tradi-tional two-color pyrometry. This dramatically increased thesignal strength.

If the spectral responsivity uses the energy to electron con-version efficiency, (h � c/k)21 needs to be removed from theaforementioned equation, as shown in Eq. 2. This equationcan be simplified if spectral emissivity is independent ofwavelength

DN ¼ a � p4� D

d

8>:9>;2

�f ðgÞ � aX� Dt �

Zk2

k1

ekTkSk � C1 � k�5

eC2=k�T � 1dk

(2)

By far the most common method of rendering color imagesin commercial digital cameras involves creating a color filtermosaic array (CFA) on top of a light intensity (black-and-white) imager. In most cases, a 3-color, red-green-blue(RGB) pattern appears in the CFA, although there are otheroptions, including 3-color complementary YeMaCy arrays,mixed primary/complementary colors, and 4-color systemswhere the fourth color is white or a color with shifted spec-tral sensitivity. One manufacturer (Foveon) uses a systemthat separates color based on penetration depth of the signalin the silicon detector with no mosaic filter.

A Bayer filter mosaic, as shown in Figure 2, represents themost commonly used CFA for arranging RGB color filterson a square grid of photo sensors. This term comes from thename of its inventor, Bryce Bayer of Eastman Kodak, andrefers to the particular arrangement of color filters used inmost single-chip digital cameras. Bryce Bayer’s patent calledthe green photo sensors luminance-sensitive elements, andthe red and blue ones chrominance-sensitive elements. Heused twice as many green elements as red or blue to mimicthe human eye’s greater resolving power with green light. Hereferred to these elements as samples and after interpolationthey become pixels.

Figure 1. Schematic diagram of the color-band pyro-metry.

[Color figure can be viewed in the online issue, which isavailable at www.interscience.wiley.com.]

244 DOI 10.1002/aic Published on behalf of the AIChE January 2009 Vol. 55, No. 1 AIChE Journal

There are a number of different ways the pixels arearranged in practice, but the pattern shown in Figure 2 withalternating values of red (R), and green (G), for odd rowsand alternating values of green (G), and blue (B), for evenrows is very common. The raw output of Bayer-filter cam-eras is referred to as a Bayer Pattern image. Since each pixelrecords only one of the three colors, two-thirds of the colordata are missing from each, as illustrated in Figure 3. Thegreen is usually read out as two separate fields, or as onefield with twice as many points.

A demosaicing algorithm interpolates a complete RGB setfor each point and produces the RGB image. Many differentalgorithms exist, and they are one of the distinguishing fac-tors in commercial cameras that otherwise often use thesame sensors. The simplest is the bilinear interpolationmethod. In this method, the red value of a non red pixel iscomputed as the average of the adjacent red pixels, and simi-lar for blue and green.

With the Bayer filter in front of the CCD or CMOS sen-sor, sensor spectral responsivity can be measured for each

individual color. As a result, the pixel intensity (or digitalnumber) of the red color/channel is correlated with objecttemperature and red color/channel spectral responsivity asshown in Eq. 3, and similarly for the other two colors/channels

DNRed ¼ a � p4� D

d

8>:9>;2

�f ðgRedÞ � aX� Dt

�Zk2

k1

ekskSk;Red � C1 � k�5

eC2=k�T � 1dk ð3Þ

Occasionally, a manufacturer provides the color or black-and-white spectral responsivity of CCD or CMOS cameras.To calculate the object surface temperature, Eq. 3 applies toeach color (typically RGB) channels. Assuming both spectralemissivity and spectral transmission are independent ofwavelength and setting the gain value of each color to be thesame, a new equation with only one (implicit) unknown —the temperature — results, as shown in Eq. 4. This is the ba-sic equation for the color-band method. Any two of the threechannels/colors can be used to calculate the object surfacetemperature based on the pixel intensity of each color. Thistechnique is flexible. The camera does not have to befocused on the surface of the object for a reliable tempera-ture, although it does have to be focused for reliable spatialdistributions of temperature and, in all cases, only pixelswith light that originates from the surface are valid pixels fortemperature measurement. Images with poor focus containmany pixels with mixed particle-background light. In addi-tion, working distance, lens aperture size, and exposure timeall provide additional adjustable parameters that impact sig-nal level. It is not necessary to measure working distanceand aperture size. Only exposure time might be needed,which can be controlled and read through the camera controlsoftware

DNBlue

DNRed¼

R k2k1

Sk;Blue � C1�k�5

eC2=k�T�1dkR k2

k1Sk;Red � C1�k�5

eC2=k�T�1dk

(4)

Figure 2. Bayer filter color pattern.


Figure 3. Color image reconstruction from a Bayer filter.9

[Color figure can be viewed in the online issue, which is available at www.interscience.wiley.com.]

AIChE Journal January 2009 Vol. 55, No. 1 Published on behalf of the AIChE DOI 10.1002/aic 245

CMOS-camera-based color-band pyrometry

In this project, a CMOS camera from EPIX, Inc. (modelSV2112) was first used to measure blackbody temperature(Mikron Model M330). The camera uses a PixelCamTM

ZR32112 CMOS sensor. Spectral responsivities of eachcolor/channel, as well as the monochrome, appear in Figure4. The IR filter in front of the CMOS sensor in the camerawas removed to obtain maximum response from the camerasince near-infrared signal is critical at low-temperature meas-urements, as indicated by Planck’s law.

Using the spectral responsivity curves obtained from themanufacture and without calibrating the SV2112 CMOScamera, temperatures of a blackbody were measured by botha type-K thermocouple and the CMOS camera. The XCAPimage acquisition and process software provided Pixel inten-sity of each color. The junction-compensated-thermocoupledata compared with the camera measurements appear inFigure 5.

When the blackbody temperature was higher than 900 K,the differences between the camera-measured temperaturedata and the thermocouple results were less than 50 K. Dueto the spectral responsivity similarity of the three colors/channels in the near IR range (k [800 nm) where low-tem-perature radiation dominates, the camera measurements differby more than 100 K from the blackbody temperature and arescattered at temperatures below 900 K. Figure 6 illustratesthis issue, the data for which are based on Eq. 4. When theblackbody temperature is lower than 900 K, the pixel inten-sity ratio of any two colors/channels approaches 1.0 makingaccurate temperature measurement difficult. If the near infra-red spectral response of the camera were blocked by a nearIR filter, as is usually done in commercial cameras (weremoved the filter in this camera), the pixel intensity ratio asa function of temperature would remain high, but the totalsignal would decrease dramatically. In such cases, largerapertures, longer exposure times, or more sensitive detectorswould be required to obtain useful data at low-temperatures.

CMOS sensors reportedly exhibit a slightly higher spectralresponsivity than CCD sensors, especially at the near infrared(NIR) range. However, CCD sensors commonly perform bet-ter than CMOS sensors with respect to uniformity, signal-to-noise ratio, and dynamic range. The low to moderate uni-formity and signal-to-noise ratio may also contribute to theinaccuracy of temperature measurement of the SV2112COMS camera at the low-temperature range.

To obtain better temperature measurement results, a CCDcamera was also used in this project.

CCD-camera-based color-band pyrometry

The SVS285CLCS camera uses a Sony Exview HADCCD, which has very high-sensitivity and low-smear. Thissensor shows higher sensitivity at the NIR range than tradi-tional CCDs due to the Exview HAD CCD technology. Simi-lar to the CMOS camera, the IR filter was removed from thecamera to maximize signal. The available spectral responsiv-ity graph from the manufacturer appears in Figure 7, whichonly displays the visible spectral range.

Figure 4. Spectral responsivity of the ZR32112 CMOSsensor.10


Figure 5. Temperature comparison of thermocoupleand CMOS camera measurements.


Figure 6. CMOS camera pixel intensity ratios as func-tions of temperature.



The complete spectral responsivity of each color/channelwas measured with a blackbody as light source, and a mono-chromator (CornerStone 130) separating the broad-band lightinto narrow-wavelength signals with a resolution of 0.5 nm.The blackbody temperature was set to 1,600 8C to maximizethe signal/noise ratio. The measured wavelength range isfrom 300 nm to 1,150 nm. A high-pass filter (LPF750 LotNNB) placed between the camera and the outlet of themonochromator blocked the second- and higher-order diffrac-tions when measuring wavelengths longer than 750 nm. Boththe spectral efficiency of the monochromator and the trans-mission efficiency of the filter impact the calculation of thespectral responsivity of the CCD sensor. Energy carried by aspecific wavelength signal was calculated by Planck’s law.The mathematical calculation of spectral responsivity foreach color/channel appears as Eq. 5

Sk;color ¼ DNcolor;k

sfilter;k nmono;k � C1 k5

eC2=k TB � 1

(5)

where, color 5 Red, Green, and Blue. sfilter,k is the transmit-tance of the high-pass filter, nmono,k is the spectral efficiencyof the monochromator, DNcolor,k is the digital number orpixel intensity of each color at k.

A complete relative spectral responsivity of each color/channel normalized by the maximum value, which occurredin the red color/channel, appears in Figure 8. The measuredspectral responsivity curves of the CCD camera have similarshapes as those obtained from the manufacturer for the CCDsensor, but include the near IR region. The small differencesmay arise from sample-to-sample variations in the sensors(manufacturer’s data are typical, but not obtained on eachsensor), the camera characteristics, and the transmittance ofthe camera lens.

With the measured relative spectral responsivity curves,the pixel intensity ratios of any two colors/channels appearin Figure 9. The ratio of blue to red depends more stronglyon temperature than the other two ratio values, as would be

Figure 7. Manufacturer provided ICX285AQ CCD sen-sor spectral responsivity.11

Figure 8. Measured relative spectral responsivity of theSVS285CSCL camera.


Figure 9. CCD camera pixel intensity ratios as func-tions of temperature.


Figure 10. Temperature comparison of thermocoupleand CCD camera measurements.



expected since they differ the greatest in average wavelength.This ratio should result in the most sensitive/accurate temper-ature measurement.

Figure 9 shows that the CCD camera is able to measure awider temperature range compared with the CMOS camerasince the CCD camera pixel intensity ratios approach 1.0 atlower-temperature (;650 K). The blackbody temperatureswere then measured with this CCD camera, and calculatedusing Eq. 4 and the complete spectral responsivities. Thecamera-measured data compared with thermocouple measure-ments appear in Figure 10, again without calibration. Theresults show that the CCD camera measurements were within50 K of the thermocouple measured values when blackbodytemperature is lower than 1,050 K, but the sensor appears tobegin to saturate at higher-temperatures. Further calibrationis necessary for more accurate and wider temperature rangemeasurements.

To calibrate the CCD camera and make this method morerobust, four variables were studied: the square of lens aper-ture size divided by the square of the working distance D2/d2; exposure time Dt; blackbody temperature T; and cameragain value g.

The first investigation explored the linearity implied byEq. 2 between the pixel intensity and D2/d2. Aperture sizeadjustments produced image pixel intensity data as a functionof D2 at a variety of temperatures and exposure times.Results indicated an almost perfectly linearity in all cases(Figure 11). Here only two cases appear: low-temperatureand long exposure time data appear in Figure 11a, and high-temperature and short exposure time data appear in Figure11b. In the calibration, working distance remained constantand only aperture size changed since D2 and 1/d2 shouldhave identical effects.

The camera gain value function f (g), was calibrated foreach channel at different temperatures and exposure times.For each channel at any condition, the gain value function f(g), followed the form of ey�g, as shown in Figure 12. Theparameter in this function is almost constant, so an averagevalue was calculated for each channel: 3.424, 3.424, and3.428, respectively, for red, green, and blue channels. They

can be treated as the same for each channel to simplify thecolor-band method.

When blackbody temperatures exceeded about 850 K, theimage pixel intensity was nearly proportional to exposuretime, but both the slope and intercept of the straight linestarted to increase with increasing blackbody temperature, asshown in Figure 13. The green and blue channel behavedsimilarly, and only the red channel appears here. These datafollow a linear correlation, but with a nonzero intercept (Eq.6), between the pixel intensity and exposure time whenblackbody temperature is higher than 850 K. Both the slopeand intercept are functions of the energy received by theCCD camera sensor, increasing with blackbody temperatureincrease. These effects presumably arise from pixel saturationas the signal nears the upper limits of the pixel dynamicrange

DNRed ¼ aRedðERedÞDtþ bRedðERedÞ (6)

Figure 11. Pixel intensity vs. effective aperture area.


Figure 12. Camera gain value calibration.



where

ERed ¼Zk2

k1

Sk;Red � C1 � k�5

eC2=k�T � 1dk

To determine the exact relation between the energy andslope, as well as that between the energy and intercept,blackbody temperature was changed from about 850 K to1312 K in increments of about 80 K. The slope and interceptin Eq. 6 were calculated by adjusting the exposure time ateach temperature. Both slope and intercept fit power func-tions of the energy, as shown in Figure 14. The camera wasalso calibrated at higher temperature range ([1,273 K) witha high-power blackbody. The fitted functions were slightlydifferent from what fitted for moderate temperature range, asshown in Figure 15.

The final calibrations appear as Eq. 7 for moderate- andEq. 8 for high-temperature ranges. The simplest form, Eq. 3,reliably computes low-temperature results by the color-bandmethod

DNRed ¼ a � D

d

8>:9>;2

�e3:424:g � ðDt � 280 � ERedðTÞ0:958

þ 60:5 � ERedðTÞ0:80Þ

DNGreen ¼ a � D

d

8>:9>;2

�e3:424:g � ðDt � 272 � EGreenðTÞ0:959

þ 63:4 � EGreenðTÞ0:86Þ

DNBlue ¼ a � D

d

8>:9>;2

�e3:428:g � ðDt � 272 � EBlueðTÞ0:955

þ 65 � EBlueðTÞ0:87Þ ð7Þ

Figure 13. Red channel pixel intensity vs. exposure time at different temperature.


Figure 14. Slope and intercept vs. energy at moderate temperature range.



DNRed ¼ a � D

d

8>:9>;2

�e3:424:g � ðDt � 264 � ERedðTÞ1:013

þ 526 � ERedðTÞ0:977Þ

DNGreen ¼ a � D

d

8>:9>;2

�e3:424:g � ðDt � 254 � EGreenðTÞ1:022

þ 344 � EGreenðTÞ1:022Þ

DNBlue ¼ a � D

d

8>:9>;2

�e3:428:g � ðDt � 252 � EBlueðTÞ1:024

þ 307 � EBlueðTÞ1:032Þ ð8Þ

If the CCD sensor spectral responsivities were measuredaccurately enough, all three colors/channels of the camerashould have the same digital number vs. energy correlation.The discrepancy might be introduced by the errors in spectralresponsivity measurements.

Both Figures 14 and 15, as well as Eqs. 7 and 8, showthat the slope and intercept become more linear with respectto the received energy when blackbody temperatureincreases. With the equations obtained, the blackbody tem-perature was measured by the color-band method at randomworking distance and random exposure time. Results com-pared with thermocouple data appear in Figure 16. The erroris within 3%.

Particle temperature measurements correction forreflection effects in furnace

If the particle is heated in a furnace and the furnace walltemperature differs from the particle surface temperature, themeasurement has to be corrected by taking surface reflectioneffects into account.

To simplify the correction, the following assumptions weremade:

� The wall temperature and surface properties are uni-form;

� All energy is emitted and reflected diffusely;� Both the particle and the wall surfaces are gray;� The particle surface is convex;� The incident, and, hence, reflected energy flux is uni-

form over the surface area;The reactor is an enclosure and the particle is located in

the center of the reactor.Based on the aforementioned assumptions, a heat-transfer

diagram between the particle and the wall surface appears inFigure 17. The surface areas are A1, A2, and Av for particle,reactor, and the view port hole, respectively. Particle surfacetemperature is T1 and T2 designates the reactor wall. It isalso assumed that the surface areas of the particle and theview port are much less than that of the reactor wall: A1 �A2 and Av � A2.

Geometric configuration factors12 of the particle surfaceand the reactor surface are F1-1, F1-2, F2-1 and F2-2. By ana-lyzing the radiant energy interchange between the particlesurface and the reactor wall, the radiant energy flux leaving

Figure 15. Slope and intercept vs. energy at higher-temperature range (>1,273 K).


Figure 16. Blackbody temperature measured by thecalibrated CCD camera.



the particle surface qo,1, and that leaving the reactor wall sur-face qo,2 were developed, as shown in Eqs. 9 and 10

qo;1 ¼ e1rT41 þ ð1� e1ÞðF1�1qo;1 þ F1�2qo;2Þ (9)

qo;2 ¼ e2rT42 þ ð1� e2ÞðF2�1qo;1 þ F2�2qo;2Þ (10)

Substituting Eq. 10 into Eq. 9 results in Eq. 11

qo;1 ¼ ½1� ð1� e2ÞF2�2�e1rT41 þ ð1� e1Þe2rT4

2

½1� ð1� e2ÞF2�2� þ ½ð1� e1Þð1� e2ÞF2�1� (11)

The aforementioned equation can be simplified for this sys-tem. The assumption of convex particle surface leads to F1-1

5 0, and F1-2 5 since F1-1 1 F1-2 5 1. According to thereciprocity relation between any two radiant elements F1-2 �A1 5 F2-1 � A2, the radiant energy leaving the reactor wallthat arrives the particle surface is F2-1 5 A1/A2. Similarly, itis straightforward to get F2–2 5 1 2 A1/A2. So, F2-1 5 0 andF2-2 5 1 since A1 � A2. Equation 11 then simplifies to Eq.12. Equation 11 also simplifies to Equation 12 if the emissiv-ity of the reactor wall is 1.0.

qo;1 ¼ e1rT41 þ ð1� e1ÞrT4

2 (12)

Similarly, the spectral radiant energy flux leaving the parti-cle surface appears as Eq. 13

qo;1;k ¼ e1C1 � k�5

eC2=k�T1 � 1þ ð1� e1Þ C1 � k�5

eC2=k�T2 � 1(13)

For color-band pyrometry with Eq. 4, the corrected equationresults from replacing the Planck’s radiant energy with Eq.13. The result appears in Eq. 14.

DNBlue

DNRed¼

R k2k1

Sk;Blue � e1C1�k�5

eC2=k�T1�1þ ð1� e1Þ C1�k�5

eC2=k�T2�1

8: 9;dkR k2k1

Sk;Red � e1C1�k�5

eC2=k�T1�1þ ð1� e1Þ C1�k�5

eC2=k�T2�1

8: 9;dk

(14)

Similar equation can be established for any other two chan-nel combinations: blue/green or green/red, and both particleemissivity and particle surface temperature can be obtainedsimultaneously.

Results and Discussion

With the calibrated multicolor band pyrometer, solid-parti-cle-surface temperature and flame temperature can be mea-sured during particle devolatilization and char burning pro-cesses in a single-particle reactor, and an entrained-flow reac-tor. Examples of temperatures for several burning particles ofvarious fuel type and size appear below. In the single-particlereactor, 11 mm poplar particles suspended in the center ofthe single-particle reactor provide most of the data. Particle-surface temperatures were also measured with a thermocou-ple embedded in a shallow groove opened on the particlesurface. Small-size sawdust particles (;0.3 mm) were usedin the entrained flow reactor. Both the single-particle reactorand the entrained-flow reactor have optical access from threeorthogonal directions, so that the particle-surface temperaturecan be measured simultaneously from different angles withthree CCD camera color-band pyrometers.

Particle-surface temperature during devolatilization

When a wood particle is first inserted into the reactor, theparticle-surface temperature rises rapidly, but the reflectionfrom the reactor wall dominates the particle-surface radiance.It was observed that the particle surface was bright and thenbecame black very soon. With Eq. 14 established for blueand red channels, together with green and red channels, sur-face temperature and emissivity of poplar particles in the sin-gle-particle reactor were calculated after the particle becameblack using particle pyrolysis videos taken by three cameras.An average particle surface emissivity of about 0.99 wasobtained for the particle, even though a typical wood particleemissivity of 0.85 is widely used. This might be explainedby the surface composition change due to biomass devolatili-zation: char was formed on the surface almost immediatelyafter being exposed to the high-temperature furnace. Usuallychar has very high-emissivity. The camera-measured particle-surface temperatures as functions of time appear in Figure18, compared with thermocouple- results. Three cameraswere used to record images for the whole process since a sin-gle camera does not have enough dynamic range, with eachcamera set at different aperture size and exposure time. Anuncertainty of about 100 K existed for the thermocouplemeasurements due to measurement difficulty (contact resist-ance, heat conduction along wires, effect of the scored sur-face, etc.). A single-particle combustion model13 was alsoused to model the particle temperature, with predictionsshown in Figure 18.

Figure 17. Radiant heat interchange between particleand reactor wall.



Particle-surface temperature and flame temperatureduring combustion

Temperatures of a poplar particle surface and the sur-rounding flame measured dynamically when burning in thesingle-particle reactor appear in Figure 19. When calculatingthe flame temperature, soot absorption was ignored and graybody emission was assumed. Even though the flame (thelarge region to the left of the particle) reaches a fairly hightemperature (over 2,300 K) due to the volatile combustionaround the particle, the particle itself still stays at a low-tem-perature (1,400 – 1,700 K) mainly caused by the effects ofoutgassing and possibly endothermic decomposition reactionsoccurring in the particle. In the thermal images, some regionsare marked as error since the pixel was saturated and accu-rate temperatures could not be obtained.

The flame temperature as a function of time for a particleburning in the single-particle reactor compared with thermo-

couple measured data and model predictions appear in Figure20. In this case, the flame temperature measurement thermo-couple was placed right next to the particle surface (about 1-2 mm) instead of being embedded in the particle, and itfailed to measure the flame temperature after devolatilization.The transition point captured by the modeling results repre-sents the transition from devolatilization to char oxidation ofthe particle during combustion process.

When a sawdust particle is traveling and burning in anentrained-flow reactor, the particle is usually covered by the

Figure 18. Particle surface temperature comparisonsduring pyrolysis in nitrogen, (particle dia. 59.5 mm, Tw 5 1,373 K, Tg 5 1,050 K).


Figure 19. Char particle surface temperature and flametemperature map during particle devolatili-zation process, (particle diameter 5 9.5 mm,particle aspect ratio 5 4.0, Tw 5 1,273 K, Tg5 1,050 K).


Figure 20. Flame temperature comparison during anear-spherical particle combustion in air,(particle diameter 5 9.5 mm, Tw 5 1273 K,Tg 5 1050 K).



flame formed next to it. Figure 21 maps flame temperaturedistribution back onto the image taken for a particle burningin the entrained flow reactor. The average flame temperatureis about 2,040 K. This image is not as sharp as a stationaryparticle image and the edge area signal intensity is relativelylow, which could potentially cause measurement errors in theedge area.

Char particle surface temperature during char burning

A 3-D char particle-surface temperature map results fromsimultaneously taking three images of a single particle fromthree orthogonal directions. With the three images, 2-D parti-cle-surface temperature distributions were calculated in each

image. Then the temperature data were mapped to the sur-face of a 3-D particle-shape model reconstructed using thealgorithm developed by Lu et al.14 Results are shown in Fig-ure 22.

Results show that the particle surface temperature is notuniform and obvious gradients exist. Both the particle geom-etry and bulk-gas-flow direction contribute to the nonuniformparticle-surface temperature and gradients. Bulk gas flowsupward. Both plane XZ and plane ZY face to the flow direc-tion but plane XY is somehow parallel to the flow direction.This dramatically affects the surface-reaction rate and furtherinfluence the particle-surface-temperature distribution.

Char particle-surface temperature as function of time dur-ing the char burning process with the multicolor band camerapyrometer compared with thermocouple measured data andmodel predictions appears in Figure 23. Results show thatthe camera measured data are higher than predicted or ther-mocouple measured values. There is reason to believe thatmeasured values may be higher than predicted by models,which do not account for 2- and 3-D effects caused byflames observed during char burning.

The images from such cameras reveal details of particlecombustion that previously escaped quantitative measure-ment. For example, Figure 24 illustrates spatially resolvedtemperature and emissivity data from a burning black liquorparticle in the early stages of char burning in which gasesapproach the particle from the bottom. The temperature datashow large spatial differences in surface temperatures. Nota-bly, a blow hole evident in the particle has a very low-tem-perature, presumably attributable to minimal oxygen accessto the surface. However, the same blow hole shows a signifi-cantly higher emissivity than most of the particle surface,presumably because of the cavity radiation nature of such asurface depression. Black liquor contains high-concentrationsof sodium salts that have relatively low-emissivity (typicallyaround 0.15-0.2) whereas char emissivity’s are high (0.85–

Figure 21. Temperature map of sawdust particle in airin entrained-flow reactor, (particle diameter5 0.3 mm, Tw 5 1,312 K, Tg 5 1,124 K).


Figure 22. 3-D particle-surface temperature map of a burning char particle, (particle diameter 5 9.5 mm, Tw 51,273 K, Tg 5 1,050 K).



0.99). The images indicate low emissivities on the windwardedges of the particle and high emissivities on the leewardside of the image and along the left-side, which is protectedsomewhat from the oncoming gases by a surface protrusion.These differences presumably arise from more rapid charconversion and higher salt concentrations in the windwardregions, consistent with the presumably thinner boundarylayers and more rapid mass transfer. While the subject ofthis article is the development of this diagnostic, such imagesare rich in potential improved understanding of combustionbehavior that is difficult to obtain by other techniques.

Experimental Uncertainty

This section discusses the uncertainty of the experimentalmeasurements. Two different sources of experimental uncer-tainty are considered: (1) errors associated with the measure-ment of the underlying parameters required to calculate theobject surface temperature, (2) bias errors due to the viola-tion of one of the fundamental assumptions. The uncertain-

ties of the underlying measurements used to determine objectsurface temperature include three major components: sensorspectral responsivity, sensor uniformity, and calibration blackbody. The CCD sensor spectral responsivity curves weremeasured with monochromator and a high-pass filter. Theuncertainty of the curves is less than 62%. No data was col-lected to determine the sensor pixel uniformity, but theuncertainty was counted during the calibration. According tothe black body spec, the uncertainty of the blackbody tem-perature is 6 1.5 C8. The total relative uncertainty caused byinstruments (CCD camera and the blackbody) is 63%, asshown in Figure 16.

Violation of one of the fundamental assumptions underly-ing the experimental technique represents the final source ofexperimental error. One potential violation is discussed: graybody emission. The color-band pyrometry assumes the parti-cle surface emissivity is constant and independent of wave-length in the measurement range (400 – 1,100 nm). In thiswavelength range, the char particle emissivity variation isusually less than 10%, which would results in a temperaturemeasurement relative error of 63% for single-color method.With the color-band technique discussed in this article, thetemperature relative error will be less than 63%. We use theworst case 63% error in our overall uncertainty analysis.

The overall uncertainty of the color-band pyrometry tem-perature measurement technique is determined by combiningin quadrature the relative uncertainty of the individual meas-urements (63%), and the estimate of the experimental bias(63%). The maximum overall relative uncertainty of thecolor-band pyrometry technique is 64.5%.

If the spectral emissivity is known, the measurement accu-racy of this technique can be dramatically increased. A CCDcamera with high-sensitivity in near infrared (NIR) rangewill also increase the measurement accuracy.

Conclusions

A multicolor band pyrometry algorithm, based on CMOSand CCD digital camera technology demonstrates dynamic,spatially resolved, temperature and emissivity measurements.Particle-surface temperature and flame temperature measure-ment results agree with thermocouple measurements andmodel predictions reasonably well. This new diagnostic pro-

Figure 23. Char particle-surface temperature as func-tion of time during combustion in air, (parti-cle diameter 5 9.5 mm, Tw 5 1,273 K, Tg 51,050 K).


Figure 24. Black liquor char particle images, temperature, and emissivity.



vides pixel-by-pixel particle temperature and emissivity datafrom widely available and relatively inexpensive cameras.Depending on magnification, shutter speed, etc., useful tem-perature and emissivity data are obtainable at temperaturesgreater than about 750 K, with the lower limit based on sig-nal intensity. Cameras more sensitive in the infrared produceuseful data at lower-temperatures.

These noncontact particle temperature data compare rea-sonably well with both predictions and thermocouple mea-surements for both suspended and entrained-flow systems.The images allow 3-D reconstruction of particle temperatureprofiles that reveal large-surface temperature and emissivitydifferences.

Acknowledgments

This investigation is supported by US Department of Energy (DOE)/EE Office of Industrial Technologies.

Notation

a5pixel/cell area of CCD or CMOS sensor, m2

B5 radiance, W.Sr21.m22.nm21

C1, C25 constants in Planck’s lawd5working distance, mD5Lens diameter, m

DN5digital number/pixel intensity, photon energy, JE5 irradiance on sensor, W.nm21

F5geometric configuration factor,g5 camera gain valueh5Planck’s constant, 6.626 3 10234 m2.kg.s21

q5 radiant energy flux, J.m22

S5 sensor sensitivity,Tg5gas temperature, KTw5 reactor-wall temperature, KX5 area ratio of effective sensor over light source,

Greek letters

k5wavelength, nmr5Stefan-Boltzman constant, W.m22.K24

e5 emissivitys5 transmitterX5 solid angle, Srt5 camera exposure time, s

Literature Cited

1. Mitchell RE, Niksa S. Temperature measurements of single pulver-ized-fuel particles by two-color pyrometry: II. particle burningbehavior. In: Extended Abstracts, Spring Meeting - ElectrochemicalSociety. San Francisco, CA: Electrochem Soc. Pennington, NJ; 1983.

2. Hernberg R, Stenberg J. Simultaneous in situ measurement of tem-perature and size of burning char particles in a fluidized bed furnaceby means of fiberoptic pyrometry. Combust Flame. 1993;95:191–205.

3. Joutsenoja TS, Hernberg R. Pyrometric measurement of the tempera-ture and size of individual combusting fuel particles. Applied Optics.1997;36(7):1525–1535.

4. Murphy JJ, Shaddix CR. Influence of scattering and probe-volumeheterogeneity on soot measurements using optical pyrometry. Com-bust Flame. 2005;143(1–2):1–10.

5. Fletcher TH. Time-resolved temperature measurements of individualcoal particles during devolatilization. Combust Sci Tech. 1989;63:89–105.

6. Tichenor DA, Mitchell RE, Hencken KR, Niksa S. Simultaneous insitu measurement of the size, temperature and velocity of particlesin a combustion environment. In: 20th Symposium (International) onCombustion. The Combustion Institute; 1984.

7. Larsson A. Optical Studies in a DI Diesel Engine. SAE Trans.1999;108(4):2137–2155.

8. Vattulainen J, Nummela V, Hemberg R, Kytola J. A system forquantitative imaging diagnostics and its application to pyrometric in-cylinder flame-temperature measurements in large diesel engines.Meas Sci Tech. 2000;11:103–119.

9. Bockaert V. Color Filter Array Sensor. 2003. [cited; Available from:http://www.dpreview.com/learn/?/key5color1filter1array.

10. Epix, Inc. Spectral responsivity of ZR32112 CMOS sensor. [cited;Available from: http://www.epixinc.com/products/ZR32112spectral-responsivity.pdf.

11. Sony, Inc. ICX285AQ CCD sensor spectral responsivity. 2005. [cited;Available from: http: //products.sel.sony.com/semi/PDF/ICX285AQ.pdf.

12. Siegel R, Howell J. Chapter Six - Radiation exchange in enclosurescomposed of black and/or diffuse-gray surfaces. In: Thermal radia-tion heat transfer. Taylor & Francis; 2002:207–267.

13. Lu H, Baxter LL. Effects of particle shape and size on biomass com-bustion. In: Science in Thermal and Chemical Biomass Conversion.Victoria, Vancouver Island, BC, Canada. 2004.

14. Lu H, Roberts W, Baxter LL. Effects of particle shape and size onbiomass reactivity. In: Advanced Combustion Engineering ResearchCenter Conference. Provo, UT; 2005.

Manuscript received Jan. 18, 2008, and revision received Aug. 19, 2008.


Particle surface temperature measurements with multicolor band pyrometry

Documents

Transcript of Particle surface temperature measurements with multicolor band pyrometry