You are on page 1of 9

Emissivity measurement and temperature correction accuracy

considerations
Robert P. Madding
Inframetrics, Inc.
North Billerica, MA

ABSTRACT

Extraction of temperatures or temperature differences with thermography is not possible without knowledge of the target
emissivity. As the technology of thermography evolves, many applications from predictive maintenance through R&D
projects have increasingly stringent requirements for quality temperature measurement. Today’s IR cameras and software
can correct for target emissivity variations on a point-by-point basis or over the entire image. One problem is how to
measure emissivity, and how emissivity measurement uncertainties propagate to temperature uncertainties. This paper
discusses emissivity measurement techniques, why table values are often not valid for a particular IR camera (why you must
measure emissivity) and how emissivity measurement accuracy affects temperature measurement accuracy (error budget for
opaque targets).

1. INTRODUCTION

Virtually all IR cameras in use today for predictive maintenance and most research and development applications are
radiometric. That is, you can get target temperatures with proper use. Typically, one enters the target emissivity and a
“background” temperature. The camera cursor is moved around the image and temperatures are displayed on the screen.
Other measurement tools include boxes, color coded isotherms and line graphs.

No IR camera reads temperature directly. They read the IR energy coming from the target which includes emitted, reflected
and sometimes transmitted IR energy. Today’s cameras correct for emissivity only for opaque targets whose transmissivity is
zero. For brevity and clarity, and in keeping with the preponderance of applications, this paper will do likewise. Note that
assuming a target is opaque does not guarantee opaqueness. The user should always be cautious about painted metals and
tape applied to metals. Paint type and thickness determine opacity1. The type of tape used is also key to opacity, especially
in the short wavelength region (3-5 µm). Many tapes are semi-opaque, an undesirable quality for this work.

Other caveats are the wavelength sensitivity and linearity of your camera. This paper is based on a camera with wavelength
sensitivity from 3.4 to 5.0 µm. Error analysis results will vary with camera wavelength due to the increasing sensitivity to
the background as wavelength increases. Camera linearity is assumed for all analyses in following sections. The validity of
this assumption depends on the quality of the camera. Quantifying these effects is beyond the scope of this paper.

In a perfect world, emissivity is defined as the ratio of the energy emitted by a flat, opaque, optically polished object at a
given temperature and wavelength to that of a blackbody at the same temperature and wavelength. Emittance is the same as
emissivity with the restrictions of flatness, opaqueness and optically polishedness removed. So, emittance is the real-world
target, emissivity represents a material property. We do not live in a perfect world. One of our imperfections is the IR
community mixes emissivity and emittance with little regard for the above restrictions imposed on emissivity. In this paper, I
will follow form and use emissivity with the broader definition that incorporates real-world targets, without the above
restrictions. That is, emissivity is the same as emittance.

A graybody emitter is one whose emissivity is constant with wavelength. This approximation must be used with IR cameras
as they ordinarily measure over a wavelength region or band, e.g. 3.4 to 5.0 µm (micrometers) or 8 to 12 µm. The graybody
approximation causes problems when the target emissivity varies with wavelength. The most direct solution to this problem
is to measure the emissivity of a non-gray target at the same (or close to the same) temperature that occurs in use. In this
paper, emissivity will refer to graybody emissivity unless specifically noted. The measurement techniques employ IR
cameras, hence they assume graybody validity.
2. RADIANCE EQUATION AND LINEAR IR SYSTEMS

For an opaque target at a temperature Tt and emissivity, ε, with a background at temperature Tb, the radiance measured by an
IR camera, Lm, is2:

L m = εL(Tt ) + (1 − ε )L(Tb ) Eqn. 1

where εL(Tt) is the radiance emitted by the target and (1-ε)L(Tb) is the radiance reflected off the target. Units of radiance are
Watts/(m2sr), where sr denotes steradians, the solid angle. The target radiance depends on the target emissivity and the target
temperature. IR camera detectors measure watts. The surface area and solid angles are defined by target size, optics
aperture, detector size and distance between the target and the camera. For most users all you need to know is the target is an
extended source. That is for the optics, detector and distance, the target is large enough for a good measurement. Many
manufacturers use a spot-size ratio calculation as a rule-of-thumb for determining whether a target is an extended source. For
this paper, extended source is assumed.

The IR detector converts the photon watts captured by the camera optics to an electrical signal. Modern IR cameras use
digital electronics, so the output signal is typically a digital value or level value. Quality IR camera design ensures linearity
between the digital value output for a given radiance input. Camera linearity allows us to interpret the radiance equation “L”
as radiance or level values. So, another way to read the radiance equation is: The measured digital level equals the target
emissivity times the digital level corresponding to the target temperature, Tt, plus the reflectivity times the digital level
corresponding to the background temperature Tb.

3. HOW AN IR CAMERA GETS TEMPERATURE

Equation 1 can be solved for the target level, L(Tt):

Lm − (1 − ε )L(Tb )
L(Tt ) = Eqn. 2
ε
Knowledge of the measured level (Lm), target emissivity (ε) and background level at temperature Tb gives the level for the
target at temperature Tt. The camera measures the first, the user inputs the other two from prior knowledge or measurement.
Camera calibration quantifies digital camera levels to temperature for various camera ranges. Here’s the procedure:

1. Operator enters emissivity.


2. Operator measures and enters background temperature.
3. Camera looks up the digital level corresponding to the operator entered background temperature.
4. Operator takes measurement.
5. Camera calculates the digital value for the target based on equation 2.
6. Camera finds (usually from a digitally stored calibration look-up table) corresponding temperature for the value
calculated in step 5 and prints it on the display.

The camera portion of this happens in a few milliseconds. You get a temperature reading, but how accurate is it? You need
to know how uncertainties in the values on the right hand side of equation two affect the uncertainty in the final result, the
target temperature. This will give you an overall temperature measurement accuracy. This paper deals with a significant
contributor to the error budget, the target emissivity, ε.
4. EMISSIVITY MEASUREMENT CONSIDERATIONS

To understand the contribution of emissivity to the error budget, we must first understand how emissivity is measured, and
with what accuracy. Measurement errors in emissivity will propagate through the calculation of temperature via equation 2.
As the target radiance is not linear with temperature, camera digital level values are also not linear with temperature. Figure
1 gives a typical calibration curve for a modern IR camera. Note that at low temperatures, a small temperature change
produces a small level change. At higher
temperatures a small temperature change
produces a large level change. 16000

14000
Conversely, at the low end of the
calibration curve, small level changes 12000

produce large temperature changes and at


Camera Digital Level
10000
the upper end of the curve, large level
changes produce small temperature 8000

changes. System noise produces finite 6000


uncertainties in the level that are fairly
constant over the calibration range. Thus, 4000

to minimize errors due to camera noise, it 2000


is best to work at the higher end of a
calibration range. 0
70 90 110 130 150 170 190 210 230 250
Temperature in deg F
There are several techniques for
measuring emissivity of opaque targets: Figure 1. IR camera digital level vs. temperature—typical
calibration curve.
1. Reference emitter—A known
emissivity material is applied to the
target so both are at the same temperature. The target temperature must be different from the background, usually
higher:
Lmt − Lmb
εt = Lmr − Lmb
εr Eqn. 3
where L denotes camera level and subscripts mt, mb and mr denote target, background and reference emitter
measurements, respectively.

2. Reference temperature—The temperature of the target is found using a thermocouple or other measurement device.
Again the target must be at a temperature different from the background. Sometimes 1 and 2 are combined and the
camera is used to find the target temperature using the reference emitter:
Lmt − Lmb
εt = Lt − Lmb Eqn. 4
where subscripts are as above and Lt is the camera level corresponding to the measured temperature.

3. Two temperature—The target is raised to two different, known temperatures.


Lm1 − Lm 2
εt = Lt 1 − Lt 2 Eqn. 5
where Lm1 and Lm2 are the camera level measurements at the two temperatures, and Lt1 and Lt2 are the camera levels
corresponding to the known target temperatures.

4. 1- reflectivity—Reflectivity is measured for an opaque target. ε=1−ρ.


Numbers one through three are readily derived from equation 1 and the relevant measurement scenario. For all techniques
we are dealing with the ratio of differences of camera digital level values. Error budgets will evolve in a similar manner.
Everything else being equal, the more variables, the greater the resulting uncertainty in the result. For example, if the
uncertainty in Lt in technique number 2 equals that of Lmr in technique number 1, we expect the overall accuracy of technique
2 to be greater than that of 1 due to the uncertainty in εr. Other techniques are given in reference 3.

In the above equations, uncertainty in levels with subscript m are due to camera noise. Temperature measurement accuracies
are converted to equivalent camera noise levels for all subscripts without m. Their accuracy depends on the type of measuring
device used.

5. EMISSIVITY MEASUREMENT ERROR BUDGET

Calculating the uncertainty in emissivity measurement is straightforward using the root-sum-square error method. The
general equation is4:

2 2 2
 ∂f   ∂f   ∂f 
E rss =  ∆u1  +  ∆u2  + • • • +  ∆un 
∂un 
Eqn. 6
 ∂u1   ∂u2  
As an example, let’s apply this concept to the reference emitter technique:

2 2 2 2
 ∂ε t   ∂ε t   ∂ε t   ∂ε 
∆ε t =  ∆Lmt  +  ∆Lmr  +  ∆Lmb  +  ∆ε r t  Eqn. 7
 ∂Lmt   ∂Lmr   ∂Lmb   ∂ε r 

where ∆εt is the error in the calculated target emissivity due to the accumulation of measurement errors of the target level,
reference level, background level, and reference emissivity measurements. The ∆L for all camera level measurements will be
the same. Using this, taking the partial derivatives and rearranging to show the fractional error yields:

2
∆ε t  ∆ε   1 1 
=  r  + 2 ∆L2  + 

εt  εr  (Lmt − Lmb ) (Lmr − Lmb )2
2 (ref emitter) Eqn. 8
 
Multiplying equation 8 by 100 gives the per cent error in the measurement. Analyzing equation 8 tells us the following:

• The bigger the difference between the background level and the target level, the lower the error. Bigger difference
means bigger denominator which translates to smaller numerator.
• The bigger the difference between the background level and the reference level, the lower the error. Bigger difference
means bigger denominator which translates to smaller numerator.
• The higher the reference emissivity, the lower the error. Big denominator effect.
• The less uncertainty in both the level measurement error and the reference emissivity error, the lower the error.

Similar analysis for techniques 2 and 3 yields:

∆ε t
=
(∆L mt
2
+ ∆Lmb
+
2
∆Lt + ∆Lmb) ( 2 2
)
εt (Lmt − Lmb )2
(Lt − Lmb )2 (reference temperature) Eqn. 9
∆ε t
=
(∆L m1
2
+ ∆Lm 2
+
2
) (
∆Lt1 + ∆Lt 2
2 2
)
εt (Lm1 − Lm 2 )2
(Lt1 − Lt 2 )2 (two temperature) Eqn. 10

6. EMISSIVITY MEASUREMENT ERROR SENSITIVITY ANALYSIS

Illustrating the sensitivity of *>17,424IU in f r a m e t r ic s


emissivity measurement to the
measurement variables can be 16,000
done using a combination of 14,000
experimental data and computer
12,000
analysis. Knowledge of level
uncertainties (∆L) for camera and 10,000
15,160 15,424

thermometer, if applicable, 8,000


together with the individual 6,000
measured levels will give us our 6,776

*<5,648IU
resulting target emissivity
accuracy estimate (∆εt/εt).
A straightforward method to get
∆L for your camera is to copy 50%
Legend gives target temperature
pixel level values from a uniform
45% rise above ambient of 71F
region and paste the them into a
40% 10 deg rise
spreadsheet. The uniform region
35% 20 deg rise
optimally would be a blackbody 30 deg rise
30%
source, but any uniform target will
% Error

40 deg rise
do. It should have level values in 25%
60 deg rise
the mid-range of the camera range 20%

you want to test. In the 15%


spreadsheet, use a standard 10%
function to get the standard 5%
deviation of the camera data. In 0%
Excel it is STDEV( ). For the 0 0.2 0.4 0.6 0.8 1
Emissivity value
camera used for this paper
∆L(Range1) = 3*17=51 and Figure 2: Emissivity measurement error vs. emissivity for various target
∆L(Range2) = 3*11=33. We temperature rises above background.
assume all errors are random.
Systematic errors are zero or
negligible. We multiply the
standard deviation by three to get
50%
the 3σ (3-sigma) error as the error Legend gives thermal noise,
45%
bound. For random errors, this 3NE∆T (3σ standard deviation)
value gives us confidence that 40%
DT =0.28
99.7% of all measurements will 35%
DT =0.55
fall within these limits. 30% DT =1.0
% Error

Establishing initial values of the 25% DT =2.0


relevant parameters by 20%
experiment, then changing the 15%
proper level variable to cause the
10%
emissivity to change, gives us a
5%
good indication of the expected
accuracy over a wide range of 0%

emissivity values. 0.0 0.2 0.4 0.6 0.8 1.0


Emissivity value

Figure 3: Emissivity measurement error for various camera noise equivalent


temperature differences (NE∆T)
As an example, let’s use the reference emitter technique. The photo/thermogram pair above shows the target—an electric
iron with 3M Scotch Super 88 vinyl electrical tape (054007-06143) applied as the reference emitter, ε=0.95±0.05. [The
number following the ± (plus or minus) symbol is the estimated uncertainty.] As the iron heats non-uniformly, there are
many locations from which to choose the reference emitter and target levels relative to the background level. The setup was
configured so the background level equaled the level selected in the lower right hand corner of the thermogram.

Figure 2 shows emissivity measurement error vs. emissivity for various target and reference emitter temperature rises above
the background. The lower limit at the highest emissivity values is limited by the reference emitter uncertainty, 5.3% in this
example. As emissivity values decrease, the measurement per cent errors increase. Also, as the temperature difference
between the target and the background decreases, measurement error increases. For higher emissivities a temperature rise
above 30 F° is adequate to minimize errors to that of the reference emitter uncertainty. Measurement of lower values
demands higher temperature differences. The 60 F° rise gives good results down to ε=0.2 and fair results to ε=0.1.
The quality of the infrared camera plays a major role in the emissivity measurement error. The camera used in this example
has a Noise Equivalent Temperature Difference (NE∆T) of 0.09 F°. With a good quality camera and reasonable care,
emissivities can be measured with good accuracy.

Figure 3 shows what happens when the camera thermal noise increases. For a 60 F° rise of target to background temperature
an NE∆T = 0.09 produces a three standard deviation DT of 0.28 F°. This lowest plot is the same as above. As the NE∆T is
increased through simulation, the emissivity measurement error increases. This example indicates the sensitivity of
emissivity measurement errors to camera noise.

7. TEMPERATURE MEASUREMENT ERROR BUDGET

For systems linear in radiance, temperature values can be functionally approximated as shown in equation 11 where A and B
are constants, and ln(Lt ) is the natural logarithm of the calculated digital target level.
T = A ln( L t ) + B Eqn. 11
A∆Lt
∆T = Eqn. 12
Lt
Equation 12 gives the error in temperature as a function of the error in target level. We find the target level uncertainty by
applying the root-sum-square error method given by equation 6 to the radiosity equation (2):
2
 ∆ε 
 ( Lm − Lb ) + ∆ Lb (ε − 1)
2 2
∆ Lm + 2 2

∆ Lt  ε 
=
Lt (Lm − (1 − ε ) Lb ) Eqn. 13

Combining equation 13 with equation 12 gives equation 14:


2
 ∆ε 
 ( Lm − Lb ) + ∆Lb (ε − 1)
2 2
∆Lm + 2 2

 ε 
∆Tt = A Eqn. 14
εLt
Equation 14 tells us a good deal about temperature measurement accuracy considerations:
• The error in measurement depends on the measurement errors in emissivity, measurement level and background level.
• As the emissivity decreases, temperature error increases (denominator effect).
• As the target level decreases, temperature error increases (denominator effect).

Here is how to find values for all these variables:


• Get an image that has values spanning at least the middle two-thirds of the range of interest.
• In image processing software, establish points across the range.
• Capture the temperature values and corresponding level values in a spreadsheet.
• Plot temperature vs. level as shown in Figure 4.
• Fit the temperature vs. level curve to equation 11. This determines A and B. In Excel, this is as simple as adding a
trendline to the plotted data.
• Use the emissivity error as
found above. 250
• Use ∆Ls as found when T = 83.486Ln(L2) - 575.61
R2 = 0.9994
measuring emissivity.
• Find the level value, Lb, 200
corresponding to the
background temperature.

Temperature in Deg F
150
Lm is the measured level value, Lt
is determined using equation 2.
[Note: Lt≡L(Tt), Lb≡L(Tb)] 100 T = 75.76Ln(L1) - 593.02 R2 Temp
Figure 4 gives an example of 2
R = 0.999 R1 Temp
Ln(L2)
finding A and B using plotted Ln(L1)
temperature vs. level data. 50

Temperature errors are often


more meaningful when stated in
terms of absolute error as 0
70 2070 4070 6070 8070 10070 12070 14070 16070 18070 20070
opposed to relative (per cent)
Level
error.

Figure 5 gives plots of


Figure 4: Temperature vs. Level with curve fit equations for camera ranges 1 and 2.
temperature error vs. emissivity R2 gives a measure of the “goodness of fit.”
for curves of various temperature
rises above background, using
equation 14. For high emissivity
targets the absolute error is
higher for the higher temperature 10.0
rise. However, the relative error 9.0
is less. For lower emissivity
targets, below about 0.4, the 8.0 Legend gives target temperature rise
above background in F.
absolute error is greater for the 7.0
10 deg rise
lower temperature rise targets.
DT in F degrees

6.0 30 deg rise


The curves were generated 45 deg rise
5.0 60 deg rise
assuming a 3NE∆T = 3*0.09 F°,
representative of a high quality 4.0

IR camera. Emissivity errors are 3.0

based on reference emitter 2.0


technique with a 5% uncertainty
1.0
in the emissivity value of the
reference. A lower uncertainty in 0.0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
the reference emissivity affects emissivity
the higher target emissivity
temperature measurement Figure 5: Plots of temperature error vs. emissivity for various temperature
uncertainty significantly, but has rises above background.
much less effect on temperature
measurement accuracy for
emissivity values less than 0.2 to 0.3.

For emissivity values above 0.5, the temperature error is almost constant for all the example temperature rises. At lower
emissivity values, the temperature error increases dramatically with decreasing emissivity, making measurement in this
region fraught with uncertainty.
Figure 6 is a plot of temperature measurement error in F° vs. per cent emissivity error. The example is for a target emissivity
of 0.4 with three graphs for different temperature rises above background. Large emissivity errors are shown to give an idea
of the magnitude of error resulting from guessing the emissivity, or for using table values incorrectly. There is often a wide
discrepancy between a
target’s emissivity and the 50
value in a table due to many
factors including: surface 45
Legend gives target temperature
cleanliness, surface 40
rise above background of 71 F.

Temperature measurement error in deg F


roughness, surface shape, DT= 60 F
DT= 45 F
surface oxidation or corrosion 35
DT= 35 F
and IR camera spectral 30
sensitivity.
25

For example, suppose you


20
guessed an emissivity of 0.8
when the actual emissivity 15
was 0.4. This represents a Emissivity = 0.4

100% error from the true 10

value. If the indicated 5


temperature rise was 60 F°,
Figure 6 shows the actual 0
0% 20% 40% 60% 80% 100% 120%
temperature rise was 40
% error in emissivity
degrees higher. In most
quantitative work, this error
Figure 6: Temperature error vs. per cent error in emissivity.
would be a serious problem.
You don’t need to carry this
plot with you. Your IR
camera or image processing software can give you the error. Simply enter a range of emissivity values with the correct
background temperature input, and observe the variation in temperature readout. Note the temperature error is not zero when
the emissivity error equals zero. This is due to intrinsic camera noise (∆L) as discussed above. Also note we assume the
background temperature is accurately known. In many cases the background is uncertain due to its wide variation and the
unknown angular reflective characteristics of the target.

8. MEASURING EMISSIVITY

Emissivity errors can be significant when using an IR camera for temperature measurement. To avoid the pitfalls of table
values, measurement is recommended. Emissivity can be measured relatively easily using one of several methods outlined
by equations 3, 4 and 5. For the reference emitter method, measuring emissivity is straightforward for many types of opaque
targets:

• Attach a piece of electrical tape (type given above) on the sample and place in an oven set to about 120 °F.
• Set up your IR camera in level units. Measure the background where you plan to place the target.
• After the target has heated, remove it from the oven and place in position.
• Be sure your range setting is OK and you are close enough. Focus on the target and freeze the image.
• On the frozen image or in image processing software, measure and record the target level, reference level and
background level.
• Use equation 3 to calculate the emissivity. If you use the tape referenced above, εr = 0.95.
• If you are using an IR camera of quality similar to the one used to generate the above data, you can use Figure 1 to
estimate the accuracy of your measurement.
9. SUMMARY AND CONCLUSIONS

The most reliable way to get good emissivity values for your IR camera and application is to measure the emissivity of
targets you will be using. Though you cannot do this readily in situ, you can get samples of the targets and measure as shown
above. If you suspect your target is not a graybody, measure the emissivity at several temperatures that span the expected
temperature during usage.
The better job you do in measuring emissivity and temperature, the more confidence you will have in your interpretations and
results. Often in predictive maintenance applications, many thermographers don’t worry about emissivity. This can be OK
for some applications, but it can be crucial for others. For example, for indirect measurement (where the hot spot is thermally
insulated from the measured surface as in oil-filled breakers, bus ducts, bushings, load tap changers, and so on) a very small
temperature rise can indicate big problems. A few degrees error in the estimated temperature rise can cause a critical
problem to be interpreted as a minor problem.

Mis-estimating emissivity by just guessing, or using table values incorrectly, can also result in significant measurement errors
as shown above. Today’s IR cameras can do a great job for you if you exploit their capabilities for both temperature and
emissivity measurement.

10. REFERENCES

1. Madding, Robert P.; “Science behind thermography”; pp.2-9; Thermal Infrared Sensing for Diagnostics and Control
(Thermosense V); Vol. 371; SPIE; 1982
2. Madding, Robert P.; Handbook of Applied Thermal Design; editor-in-chief, Guyer, Eric C.; Part 12—Thermal Sensors;
Chapter 8—Infrared Thermography; pp 12-57 to 12-70; McGraw Hill; 1989
3. Johnson, R. Barry et al; “On the validity and techniques of temperature and emissivity measurements”; pp. 202-206;
Thermal Infrared Sensing for Diagnostics and Control (Thermosense X); Vol. 934; SPIE; 1988
4. Measurement Systems—Application and Design; Ernest O. Doebelin; McGraw Hill; 1975

You might also like