Вы находитесь на странице: 1из 4

278

Measurement Automation Monitoring, Jun. 2015, vol. 61, no. 06

Krzysztof
PACHOLSKI, Boguslaw WIECEK
1
2

INSTITUTE OF ELECTRICAL ENGINEERING SYSTEMS, LODZ UNIVERSITY OF TECHNOLOGY, 90-924 d, Stefanowskiego 18/22 St.
INSTITUTE OF ELECTRONICS, LODZ UNIVERSITY OF TECHNOLOGY, 90-924 d, Wlczaska 211/215 St.

Practical assessment of accuracy of thermographic


indirect measurements
Abstract
In this paper, the basis of error and uncertainty assessment for
thermographic measurement of temperature is presented. An engineering
simplified analytical approach of determining the coverage coefficient is
described for different number of thermal images taken into analysis and
different contributions of A and B-type of uncertainties. In the form of an
example, the entire procedure of uncertainty assessment was used for
emissivity. The exemplary calculus are presented for emissivity in LWIR
spectral range.
Keywords: errors and uncertainty, infra-red (IR) thermography, emissivity
metallurgy, slag.

1. Introduction
The users of thermographic cameras are facing with the problem
of accuracy assessment of the IR measurements. In the reports
from the thermal investigations using IR cameras, one has to
present the accuracy analysis of the results. In the literature, there
is no simple engineering procedure to perform this analysis [1, 2,
3, 4]. On the other hand, such analysis is required in every
quantitative measurement. The thermographic measurement of
temperature as an indirect one. Its accuracy depends of the
environmental parameters. It is rather difficult to determine the
accuracy of assessment of these parameters, and typically, this
accuracy is poor.
Not all users of the thermal camera have the access to the
advanced software tools, to calculate the accuracy of the
measurement, using e.g. Monte Carlo method [1, 3, 5, 6, 7, 8].
Many users do not distinguish the error and the uncertainty of the
measurement. The main aim of this paper is to present a compact,
simple procedure of calculating the uncertainty of the
thermographic measurement.

2. Errors and uncertainty of thermographic


measurements

X%

T f , To , Tatm , RH , d

(2)

Each parameter of temperature measurement using IR camera is


evaluated with its accuracy. The user does not know the random
components of the measurement errors of these parameters.
Because of that, the best approximation of the temperature
measured by the camera is the average value calculated in
a certain region of interest (ROI) for a given series of
measurement.
N

Many of thermal camera users use the terms error and


uncertainty alternatively. It is a consequence of bad
interpretation of these words that are well defined in the official
document of Joint Committee for Guides in Metrology [9].
The term error, one can interpret either qualitatively or
quantitatively. In the first meaning, many users of thermal
cameras understand that it is a difference between the measured
and the exact values of a quantity. The absolute and relative errors
defined as in the eqns. (1) [5, 9].

X X X o
X Xo
100%

Xo

(exact) temperature of an object with very high emissivity and


varying temperature. Before calibration, Non-Uniformity
Correction (NUC) has to be performed [4].
In order to estimate the maximum permissible systematic error
of a thermal camera, one uses the mean value (in time and area) of
the signal expressed in Isothermal Units (IU). The manufactures of
the thermal cameras use the maximum permissible absolute or
relative errors with reference to the measurement range. The errors
for most of the cameras are e.g. 2C or 2%, and the user of
thermal camera must take the higher value. One should take into
account, that the values of the errors given by the manufacturers
are obtained in laboratory conditions with the use of a static model
of the measurement process with the error equal to zero. During
the temperature measurements in the real conditions, the error of
the system can be higher due to the errors of the calibration
method or/and errors caused by the external conditions, such as
object emissivity, transmission of the atmosphere, etc. [1, 4].
In the practical thermographic measurements, the camera errors
are not the only ones. One should remember that the radiative
measurement of temperature is the indirect one. The static model
of the measurement process depends on many parameters, such
a emissivity , ambient temperature Ta, temperature of atmosphere
Tatm, relative humidity of atmosphere RH and distance between the
camera and object [1, 2, 4].

i 1

(3)

where N in a number of measurements [5, 9].


In order to get the reliable thermographic measurement, one has
to determine the supreme of the error with the probability p,
named the confidence level. This value is known as the expanded
uncertainty U. The extended uncertainty gives a possible range of
variation of the measurement results caused by the random
fluctuations of the measured data [5, 9].

U k p u c T

(1)

where X and X0 denote the result of the measurement and exact


value of the measured quantity.
The paradox is that the error can determined if one knows the
exact value X0 only. In practice no one knows the exact value of
the thermographic measurement. In consequence, the errors are
useful in the modelling of fluctuation and disturbances of the
measuring process. Using errors one can define the accuracy of
a thermal camera and other measurement systems. In order to
calibrate IR cameras one can use the maximum permissible
systematic error. One uses a black body to define the reference

Ti

(4)

The quantity uc is called the combined standard uncertainty and


kp the coverage factor dependent on the number of measurements.
For uncorrelated values of the measurements parameters xi, the
combined uncertainty uc takes a form [5, 9].
u c T

2
ci u xi

j 1

ui T

j 1

(5)

Measurement Automation Monitoring, Jun. 2015, vol. 61, no. 06

T
denotes the sensitivity of the measurement
xi
process defined by the mathematical model (2) to a variation of
the parameters value xi, u(xi) is the standard uncertainty of the
parameter xi and ui(T) is so called contribution of i-th parameter.
In case of correlated parameters xi, one has to take into account
the covariance [2,5,9]. There are 2 standard uncertainties: of type
A and B. The A-type standard uncertainty uA(T) refers to
a deviation of the set of N measurements {Ti}, while B-type
uncertainty uB(T) presents the systematic error of the temperature
measurement of IR camera. Finally, the combined uncertainty (4)
can be presented as

where: ci

2
T uB2 T .
uc T u A

(6)

Typically, the component uA(T) of determining the estimate T is


expressed by the standard deviation calculated for N
measurements. The component uB(T) is defined by the absolute
error T of a single measurement [5, 9].

uA T

s T
N

uB T

confidence level [5, 6, 7].


Using (8), the expanded uncertainty U of the measurement is
presented as below [5, 6, 7].

N 1

random variable T.
The component uB(T) of the uncertainty is assumed to have the
continuous uniform distribution of the probability [5]. The
coverage coefficient kp in (4) depends on the probability
distribution function of random variable T. In case of normal
distribution kp2 for confidence level p=0.95. For p=0.99, kp is
equal to 3. Practically, one assumes the normal probability
distribution of the temperature measurement results when N30,
and in addition, A-type uncertainty component uA(T) is
dominating (uA(T) >> uB(T), uc(T) >> uA(T),) [5,6,7]. In
consequence, for N<30, one assumes the t-Student distribution of
random variable T, and coverage factor kp is a function of socalled effective number of degrees of freedom veff [5,6,7].
If B-type uncertainty component uB(T) dominates (uB(T) >>
uA(T)), the uniform probability distribution of a random variable is
assumed, and kp takes the constant value (6).

kp 3

j 1

kN

ui T ,

(9)

where t() is quintile of t-Student distribution for degrees of


freedom.
If it is impossible to define the number of degrees of freedom
the eq. (9) can be replaced by eq. (10) [5, 6, 7]:
n

2
ui T .

(10)

j 1

The proposed algorithm for estimation the extended uncertainty


of the objects measured temperature using infrared camera is
presented in Fig. 1.

accidental uncertainty
type A

systematical uncertainty
type B

series of N measurement results

rectangular distribution

standard deviation calculated for N


measurements, eqn. (7)

standard uncertainties of parameters


of measurement static model

contributions of typu A uncertainties,


eqn. (5)

contributions of typu B uncertainties,


eqn. (5)

combined standard uncertainty,


eqn. (5)

(6)

There is a problem of getting the right value of kp if uA(T) and


uA(T) are comparable, i.e. uA(T) uB(T), [5]. In such a case, the
probability density function and combined uncertainty uc(T) can
be calculated using Fourier transform or Monte-Carlo method
[5,6,7,8]. In order to perform this analysis, a dedicated software is
needed. Both methods allow to determine uncertainty uc(T).
Thermographic measurement of temperature is the indirect one
with relatively low accuracy. In practice, the user of the infrared
camera does not know all external sources that disturb the
measurement. This one of the reason that instead of using the
more precise numerical methods, a simplified analytical approach
is available and recommended for practitioners [5, 6, 7]. This
approach can be used if the expanded uncertainty of temperature
assessment is dominated by B-type component with the rectangle
probability distribution. In such a case one can define a parameter
ru [5, 6, 7]

U k RN

is the standard deviation of the

(7)

(7)

where kN denotes coverage coefficient for the normal


distribution (kN=2 for p=0.95 or kN=3 for p=0.99), and
3
k p 3 p , p is the
kT
1 ru 2 ru 1 p ,
ru2 1

i 1

u 2c T ui2 T

U k RN

N N 1

ui T

Next, the coverage coefficient kRN is calculated analytically


using eqns. (8):
k RN k N for 0 ru 1
(8)
k RN kT for 1 ru 10 ,
k RN k p for ru 10

Ti T
N

where: s T

Ti T

i 1

ru

279

YES

u A T uB T

NO

YES

k RN , eqns. (7) and (8)

modification uncertainty contributions,


eqn. (9)

YES

NO

N 30

k p 2 p 0,95
k p 3 p 0,99

u A T uB T NO

k p t eff

the extended uncertainty eqn. (4)

Fig. 1. The algorithm of estimation of extended uncertainty

kp 3 p

280

According to the algorithm shown in Fig. 1, it is possible to


estimate the uncertainties of other thermal parameters of the object
investigated, e.g. emissivity.

3. Uncertainty of emissivity measurement


The results of measurements of the temperature using contact
and contactless methods can be applied for emissivity evaluation.
If the measured temperature in significantly higher than the
ambient one, the emissivity can be measured as [4]


Tc

(11)

where both the thermographically measured temperature T and the


temperature measured by the contact thermometer Tc are in K.
In practice, emissivity is determined using single temperature
measurements of T and Tc or by averaging the N results of
contactless measurements using IR camera and a single
measurement by a contact thermometer. The authors of this paper
used the emissivity evaluation to measure the temperature of the
slag and steel during the discharge the oxidation converter in
steelworks [10]. In metallurgy, one uses a long lance with high
temperature nonrecurring one-time thermocouple.
For the single temperature measurements of T and Tc, the
combined standard uncertainty is expressed by eqn. (12) [5, 9].
2


2
u T
u T
T
Tc

where:

,
T
T

u Tc 2 ,

(12)

T g
3

u Tc

Tcg
3

(13)

(14)

where: Tg and Tcg denote the maximum errors of measuring


temperature using contactless and contact methods.
Using eqns. (13) and (14), one can get the combined uncertainty
of emissivity in the form:
u T

4
3

1 Tc
2
.
Tg2 Tcg
Tc T

standard deviation s(T) = 1.67 K [10].


The camera accuracy declared by the manufacturer was 2C or
2% of the operating range of the camera. The distance between
the camera and the object in the steelworks was estimated at d=30
m. A moderate pollution and low humidity were during the
measurements. The reason why the humidity was low, was the
high ambient temperature and the possible vapor condensation far
away from the convertor due to the size of the convertor and the
high convection.
The reference temperature for emissivity evaluation was
Tc=1932 K (1659C). It was measured using the one-time
thermocouple probe made by Hereause Electro-Nite. The
measurement error of the thermocouple was 6C.
Using eqn. (11), one can get
4

T 1912 4

0,95 .
Tc 1932

2
Tcg

(17)

Combined standard uncertainty for the indirect measurement of


emissivity is given by eqn. (16). Neglecting the humidity impact
on the temperature measurement, the supreme error of temperature
measurement Tg takes a form
(18)

where ais an error caused by the air pollution. This error is


estimated for industrial condition at the level 0.7C/m [1]. For
d=30 m one can get a=21C. The second component of eqn. (18)
2000 2
40 K . Finally, the error is
is Tk
100

Tg a Tk 21 38 59 K .

(19)

Because Tg is almost 10 times higher that Tcg=6 K and 35


times higher than the standard deviation s(T), the expanded
uncertainty of the indirect measurement of the emissivity can be
evaluated using eqn. (20).

(15)

For multiply temperature measurements using IR camera the


uncertainty can be presented as:
2
2
1 T s 2 T Tg

u T 4 c
3
Tc T N

In order to evaluate the emissivity of the stream of the slag


during the discharge of the oxidation convertor in steelmakers, 15
measurements were performed. LWIR microbolometer IR camera
was used in this research. The measurement lasted 1 s. The mean
temperature T 1912 K was measured by the camera with the

U 3u T 4

4. An example

Tg a Tk .

4
.

Tc
Tc

Because of the determined character of u T and u Tc for onetime measurement, B-type uncertainties are considered only.
u T

Measurement Automation Monitoring, Jun. 2015, vol. 61, no. 06

Tg 4

0,95
59 0,06
1912

(20)

Taking the real values of the parameters of the measurement,


one can get
0,95 0,06 ,
but never grater than 1, 1.

(16)

where s(T) is a standard deviation of the series of N measurement


results.
In order to evaluate the expanded uncertainty U one needs to
get the value of the coverage factor kp. For the single
measurements of T and Tc, coverage coefficient kp is defined by
(6), while for multiply measurements, the value of kp depends on
the most contributive component of u [5, 6. 7].

5. Conclusions
The uncertainty analysis is a powerful tool to assess the
measurement accuracy in many domains. In addition, it is a very
good complementary element for the measurement modelling. It
allows to evaluate the calibration procedures of thermographic
cameras not only in laboratory conditions. Thanks to the
uncertainty, it is possible to use the precise measurement system
in noisy industrial environments, where there is a large level of
possible random and determined disturbances. As an example, the
authors used a themography system in steelworks to evaluate to

Measurement Automation Monitoring, Jun. 2015, vol. 61, no. 06

content of FeO in the slag during discharging the oxidation


convertor. For such a case the uncertainty depends mainly on the
maximum systematic errors of IR camera. The difficulties to
obtain the high accuracy of thermographic indirect measurement
of temperature and other derivative quantities was caused by the
pollution of the environment. The indirect measurement of the
slag emissivity was disturbed by the high pollution as well.
In this paper we tried to present an engineering simple approach
for evaluation the thermographic measurement accuracy.

6. References
[1] Minkina W., Dudzik S.: Infrared thermography errors and
uncertainties. John Wiley & Sons, New York, Chichester 2009.
[2] Minkina W., Dudzik S.: Termografia w podczerwieni bdy
i niepewnoci. Measurement Automation and Monitoring, 11, pp. 868873, 2009.
[3] Minkina W., Dudzik S.: Simulation analysis of uncertainty of infrared
camera measurement and processing path. Measurement 8, pp. 758763, 2006.
[4] Wicek B., De Mey G.: Termowizja w podczerwieni. Podstawy
i zastosowania. Publishing House PAK, Warsaw 2011.
[5] Ziba A.: Analiza danych w naukach cisych i technice.
Wydawnictwo PWN, Warszawa 2014.
[6] Korczyski J., Fotowicz P., Hetman A., Gozdur R., Hloba A.:
Metody obliczania niepewnoci pomiaru. Measurement Automation
and Monitoring, 2, pp. 13- 16, 2005.
[7] Fotowicz P., Korczyski J., Hetman A.: Zastosowanie
probabilistycznego modelu obliczania niepewnoci pomiaru przy
wzorcowaniu omomierza i woltomierza. Measurement Automation
and Monitoring, 11, pp. 20- 24, 2006.
[8] Guide: Guide to the Expression of Uncertainty in Measurement.
Supplement 1. Numerical Method for the Propagation of
Distributions. Dokument Midzynarodowego Biura Miar, 2004.
[9] Guide: Guide to the Expression of Uncertainty in Measurement,
ISO/TAG, 1995. Wydanie polskie 1999 r. opracowane przez J.
Jaworskiego.

281

[10] Strkowski R., Pacholski K., Wicek B., Olbrycht R., Wittchen W.,
Borecki M.: Radiative parameters of steel slag for FeO content
estimation using multispectral thermography system. Quantitative
InfraRed Thermography Journal, 2014, vol. 11, no. 2, p. 222-232.

_____________________________________________________
Received: 26.03.2015

Paper reviewed

D.Sc. Ph.D. Krzysztof PACHOLSKI


Krzysztof Pacholski is an Associate Professor of
electrical measurements with the Institute of Electrical
Engineering Systems, d University of Technology.
His current research interest includes the uncertainties
estimation of IR thermography measurements and the
use of methods of the artificial intelligence in the
measuring techniques.

e-mail: krzysztof.pacholski@p.lodz.pl
Prof. Bogusaw WICEK
Bogusaw Wicek is the head of Electronic Circuit and
Thermography Division in Institute of Electronics where
he has worked for more than 35 years. Actually, he is
the co-author of 68 papers on JCR list. He was the
supervisor of 9 finished Ph.D. processes. His scientific
interests are: industrial and biomedical applications of
IR thermography, heat transfer modelling and advanced
IR analog and digital system developments.

e-mail: wiecek@p.lodz.pl

Accepted: 05.05.2015

Вам также может понравиться