Sebastian DUDZIK1, Waldemar MINKINA1

Czestochowa University of Technology, Institute of Optoelectronics and Measurement Systems(1)

doi:10.15199/48.2018.12.26

Examples of uncertainty calculations in thermographic

measurements
Streszczenie. We współczesnej metrologii istnieją dwie główne teorie - teoria błędów pomiarowych i teoria niepewności pomiaru. Celem niniejszego
artykułu jest przedstawienie praktycznych przykładów estymacji rozszerzonej niepewności standardowej i jej składników występujących w
bezdotykowych pomiarach temperatury dokonywanych za pomocą kamer termowizyjnych. W niniejszej pracy przedstawiono wyniki obliczeń dla
najbardziej typowych warunków występujących w praktyce. W każdym przykładzie obliczono 95% poziom ufności. Umożliwiło to uniwersalną ocenę
dokładności zgodnie z wytycznymi podanymi przez międzynarodowe organizacje metrologiczne. Przykłady obliczeń niepewności przy pomiarach
termograficznych

Abstract. In the contemporary metrological sciences there are two main theories - the theory of measurement errors and the measurement
uncertainty theory. The purpose of this paper is to show the examples of estimation of the combined standard uncertainty and its components
occurred in the non-contact temperature measurements made by means the infrared cameras. In this work the calculations conducted at the most
common conditions existed in practical situations were presented. In each example the 95% coverage interval was calculated. It enabled the
universal accuracy assessment in accordance with the guidelines of the international metrological organizations.
.

Słowa kluczowe: niepewności pomiarów termograficznych, termografia w podczerwieni, termowizja, bezstykowy pomiar temperatury.
Keywords: uncertainties in thermographic measurements, infrared thermography, thermovision, non-contact temperature measurements.

Introduction
1.1. Analysis of the measurements accuracy in the (5) U  ku c ( y )
presence of the random interactions where: k – the expansion coefficient.
In this work the static model of thermographic Unfortunately, in most cases, the probability distribution
measurement is presented. The static model (i.e. the model of the output variable of measurement model remains
in which a detector integration time is long enough unknown. Therefore, the Joint Committee for Guides in
compared with a time constant of infrared detector) can be Metrology (JCGM) developed the document [4], in which,
expressed as a function of five input variables: the method of the propagation of distributions is featured.
This method allows for correct estimation of accuracy
(1) Tob  f ( , T0 , Tatm ,  , d ) including the following cases:
• the partial derivatives are countless,
where: Tob – object temperature (K), ε – band emissivity of • the probability distribution of the output variable is not
the object surface, To – ambient temperature (K), Tatm – Gaussian,
temperature of atmosphere (K), ω – relative humidity, d – • the measurement model is strongly nonlinear.
camera-to-object distance (m). The evaluation of the measurement uncertainty with the
In equation (1) the model of the atmosphere method for the propagation of distributions can be divided
transmission can be calculated using a complex into the following stages [4]:
approximated functions delivered by a FLIR company [7]. • define of the measurand (the output quantity of the
Then, the model of atmospheric transmission can be measurement),
described as the following simplified formula: • define of the input variables of the measurement model,
• define of the input variables of the measurement model,
(2) Patm  f (, d , Tatm ) • design of the measurement model,
• determination of the probability distributions of input
where: Patm – an atmospheric transmission coefficient.
variables,
In practice, the measurement uncertainty can be defined
• evaluation of the probability distribution of the output
as [4]:
variable (measurand),
1 N • estimation of the statistical parameters of the probability
(3) u ( xi )   ( x  x )2
N  1 i 1
distribution of the output variable (e.g. 95% coverage
interval).
where: xi – a result of a single measurement, N – the 1.2. The software used in the numerical calculations of
number of measurements. the accuracy
Additionally, when we will assume a lack of the In the examples of the numerical calculations presented
correlations between the input variables of measurement below, the programs created in the MATLAB environment
model, we can define the combined standard uncertainty, were used. The basic functions of the software are [1]:
as a positive value of the square root from the combined • generating of the random variable realizations with the
variance, described as: parameters and distributions defined by the user,
2 • reading of the calibration parameters and the reference
N
 f  2
(4) u ( y )   
2
 u ( xi ) values from the Agema File Format files,
i 1  xi
c
 • simulation of the data processing algorithm based on
the mathematical model of the infrared camera
where f(x1, x2, …, xn) is a measurement model.
measurement,
The expanded uncertainty define the limits of the
• estimation of the combined standard uncertainty and its
coverage interval at the given level of confidence, the most
components,
commonly at 95%. The expanded uncertainty can be
• estimation of the 95% coverage intervals,
expressed as the following equation:

124 PRZEGLĄD ELEKTROTECHNICZNY, ISSN 0033-2097, R. 94 NR 12/2018

• graphical presentation of the simulation results.
The screenshot of main window of the program used in the
above-mentioned calculations is presented in Fig. 1. The
source code of this software was attached to the work [1].

Figure 2: Component of the relative combined standard uncertainty

associated with emissivity εob (results of simulations for data
collected in table 1)

Example 2
To solve the problem, the above mentioned program
was used. The simulation results (i.e. the components of
the combined standard uncertainty associated with the
object emissivity εob) were presented in Fig. 3. The value of
Figure 1: The main window screenshot of the program used for the component of the relative combined standard uncertainty
calculations of the components of combined standard uncertainty in associated with the object emissivity, for the relative
the measurements with the FLIR ThermaCAM PM595 infrared standard uncertainty of the object emissivity u(εob) = 15%
camera and value of object emissivity εob = 0.6, can be read from
2. Exemplary numerical calculations of the components Fig. 3. For the object temperature Tob = 323 K (50 oC), it is
of combined standard uncertainty in temperature equal to u(Tob)  1.3%.
measurements carried out with the FLIR infrared Example 3
cameras The temperature measurement was conducted using
2.1. Components of the relative standard uncertainty the FLIR ThermaCAM PM595 infrared camera. The
associated with the object emissivity emissivity of the investigated object was equal to εob = 0.9.
Example 1 Calculate the component of the relative combined standard
The temperature measurement was conducted using uncertainty u(Tob), associated with the object emissivity εob,
the FLIR ThermaCAM PM595 infrared camera. The assuming that Tob = 363 K (90 oC), if we know that the
emissivity of the investigated object was equal to εob = 0.4. relative standard uncertainty u(εob) = 20%. The estimates of
Calculate the component of the relative combined standard the remaining input variables of the model (1) should be
uncertainty u(Tob), associated with the object emissivity εob, assumed in accordance with the table 1. The component of
o
assuming that Tob = 303 K (30 C), if we know that the the relative combined standard uncertainty associated with
relative standard uncertainty u(εob) = 10%. The estimates of the object emissivity εob were presented in Fig. 4. They
the remaining input variables of the model (1) should be were obtained from simulations conducted for data from
assumed according to the table 1. table 1. As you can see in Fig. 4, the value of the
Table 1: The values of the input variables of the model (1) assumed component of relative combined standard uncertainty for
for the calculations of the accuracy u(εob) = 20%, εob = 0.9 and Tob = 363 K (90 oC), equals to
ε To [K] Tatm [K] ω d [m] u(Tob)  1.3%.
0.4; 0.9 293 293 0.5 1

To solve the problem, the above mentioned program

was used. The simulation results (i.e. the components of
the combined standard uncertainty associated with the
object emissivity εob) were presented in Fig. 2. The value of
component of the relative combined standard uncertainty
associated with the object emissivity, for the relative
standard uncertainty of the object emissivity u(εob) = 10%
and value of object emissvity ob = 0.4, can be read from
o
Fig. 2. For the object temperature Tob = 303 K (30 C), it is
equal to u(Tob)  0.4%.
The temperature measurement was conducted using
the FLIR ThermaCAM PM595 infrared camera. The
emissivity of the investigated object was equal to εob = 0.6.
Calculate the component of the relative combined standard
uncertainty u(Tob), associated with the object emissivity εob, Figure 3: Component of the relative combined standard uncertainty
o
assuming that Tob = 323 K (50 C), if we know that the associated with emissivity εob (results of simulations for data
relative standard uncertainty u(εob) = 15%. The estimates of collected in table 1
the remaining input variables of the model (1) should be
assumed according to the table 1.

PRZEGLĄD ELEKTROTECHNICZNY, ISSN 0033-2097, R. 94 NR 12/2018 125

 standard uncertainty associated with the ambient
temperature To obtained from the simulations were
presented in Fig. 6. The value of the component of the
relative combined standard uncertainty associated with the
ambient temperature To, for the relative standard
uncertainty of the ambient temperature u(To) = 1.5%, and
the object emissivity εob = 0.6 can be read from Fig. 6. For
o
the object temperature Tob = 323 K (50 C), the value of the
component u(Tob)  0.75%.

Figure 4: Component of the relative combined standard uncertainty

associated with emissivity εob (results of simulations for data
collected in table 1

2.2. Components of the relative standard uncertainty

associated with the ambient temperature
Example 4
In the example the emissivity of the investigated object
was equal to εob = 0.4. We should calculate the component
of the relative combined standard uncertainty u(Tob),
associated with the ambient temperature To, assuming that
o
Tob = 303 K (30 C), if we know that the relative standard Figure 6: Component of the relative combined standard uncertainty
uncertainty u(To) = 0.5%. The values of the remaining input associated with the ambient temperature To (results of simulations
variables of the model 1 should be assumed according to for data collected in table 1
the table 1. As above, the simulations with data from table 1
were carried out. The component of the relative combined Example 6
standard uncertainty associated with the ambient The temperature measurement was conducted using
temperature To obtained from the simulations were the FLIR ThermaCAM PM595 infrared camera. The
presented in Fig. 5. The value of the component of the emissivity of the investigated object was equal to εob = 0.9.
relative combined standard uncertainty associated with the Calculate the uncertainty component associated with the
ambient temperature To, for the relative standard ambient temperature To, assuming that Tob = 323 K (50 oC),
uncertainty of the ambient temperature u(To) = 0.5%, and if we know that the relative standard uncertainty u(To) =
the object emissvity εob = 0.4 can be read from Fig. 5. For 2.0%. The values of the remaining input variables of the
the object temperature Tob = 303 K (30 oC), the value of the model (1) should be assumed according to the table 1. The
task was resolved using the simulations and data from table
component u(Tob)  0.75%.
1. The components of the combined standard uncertainty
associated with the ambient temperature To obtained from
the simulations were presented in Fig. 7. The value of the
component of the relative combined standard uncertainty
associated with the ambient temperature To, for the relative
standard uncertainty of the ambient temperature u(To) =
2.0%, and the object emissivity εob = 0.9 can be read from
Fig. 7. For the object temperature Tob = 323 K (50 oC), the
value of the component u(Tob)  0.11%.

Figure 5: Component of the relative combined standard uncertainty

associated with the ambient temperature To (results of simulations
for data collected in table 1

Example 5
In this example the emissivity of the investigated object
was equal to εob = 0.6. We should calculate the component
of the relative combined standard uncertainty u(Tob),
associated with the ambient temperature To, assuming that
Tob = 323 K (50 oC), if we know that the relative standard Figure 7: Component of the relative combined standard uncertainty
uncertainty u(To) = 1.5%. The values of the remaining input associated with the ambient temperature To (results of simulations
variables of the model 1 should be assumed according to for data collected in table 1
the table 1. As above, the simulations with data from table 1
were carried out. The component of the relative combined

126 PRZEGLĄD ELEKTROTECHNICZNY, ISSN 0033-2097, R. 94 NR 12/2018

2.3. Components of the relative standard uncertainty uncertainty u(Tob), associated with the camera-to-object
o
associated with the camera-to-object distance distance d, assuming that Tob = 363 K (90 C), if we know
Example 7 that the relative standard uncertainty u(d) = 20%. The task
The temperature measurement was conducted using from the example was resolved using the simulations with
the FLIR ThermaCAM PM595 infrared camera. The model (1) and data from table 1. The component of the
emissivity of the investigated object was equal to εob = 0.4. relative combined standard uncertainty associated with the
Calculate the component of the relative combined standard camera-to-object distance d obtained from the simulations
uncertainty u(Tob), associated with the camera-to-object were presented in Fig. 10. The value of the component of
distance d, assuming that Tob = 303 K (30 oC), if we know the relative combined standard uncertainty associated with
that the relative standard uncertainty u(d) = 10%. The the camera-to-object distance d, for the relative standard
values of the remaining input variables of the model (1) uncertainty of the camera-to-object distance u(d) = 20% and
should be assumed according to the table 1. To solve the the object emissivity εob = 0.9 can be read from Fig. 10. For
task from this example, the simulations with data from table the object temperature Tob = 363 K (90 oC), the value of the
1 were carried out. The component of the relative combined component u(Tob) ≈ 0,0095%.
standard uncertainty associated with the camera-to-object
distance d obtained from the simulations were presented in
Fig. 8. The value of the component of the relative combined
standard uncertainty associated with the camera-to-object
distance d, for the relative standard uncertainty of the
camera-to-object distance u(d) = 10%, and the object
emissivity εob = 0.4 can be read from Fig. 8. For the object
o
temperature Tob = 303 K (30 C), the value of the
component u(Tob)  0.0010%.

Figure 9: Component of the relative combined standard uncertainty

associated with camera-to-object-distance d (results of simulations
for data collected in table 1

Figure 8: Component of the relative combined standard uncertainty

associated with camera-to-object-distance d (results of simulations
for data collected in table 1

Example 8
The temperature was measured using the FLIR
ThermaCAM PM595 infrared camera. The emissivity of the
investigated object was equal to εob = 0.6. Calculate the
component of the relative combined standard uncertainty
u(Tob), associated with the camera-to-object distance d,
assuming that Tob = 323 K (50 oC), if we know that the
relative standard uncertainty u(d) = 15%. The values of the
remaining input variables of the model (1) should be Figure 10: Component of the relative combined standard
uncertainty associated with camera-to-object-distance d (results of
assumed according to the table 1. To solve the task from simulations for data collected in table 1)
this example, the simulations with data from table 1 were
carried out. The component of the relative combined 3. Exemplary numerical calculations of the combined
standard uncertainty associated with the camera-to-object standard uncertainty and 95% coverage interval in
distance d obtained from the simulations were presented in temperature measurements carried out with the FLIR
Fig. 9. The value of the component of the relative combined infrared cameras
standard uncertainty associated with the camera-to-object Example 10
distance d, for the relative standard uncertainty of the Temperature measurement was conducted using the
camera-to-object distance u(d)=15%, and the object FLIR ThermaCAM PM595 infrared camera. The emissivity
emissivity εob = 0.6 can be read from Fig. 9. For the object of the investigated object was equal to εob = 0.4. Calculate
o
temperature Tob = 323 K (50 C), the value of the the combined standard uncertainty uc(Tob) and evaluate the
component u(Tob)  0.0038%. 95% coverage interval I95% assuming that the object
temperature Tob = 323 K (50 oC). The values of the input
Example 9 variables of the model (1) should be assumed according to
The temperature measurement was conducted using the table 2. The standard uncertainties of the input variables
the FLIR ThermaCAM PM595 infrared camera. The of the model (1) should be assumed according to the table
emissivity of the investigated object was equal to εob = 0.9. 3.
Calculate the component of the relative combined standard

PRZEGLĄD ELEKTROTECHNICZNY, ISSN 0033-2097, R. 94 NR 12/2018 127

Table 2: The values of the input variables of the model (1) input variables of the model (1) should be assumed
assumed for the calculations of the combined standard uncertainty according to the table 2. The standard uncertainties of the
ε To [K] Tatm [K] ω d [m] input variables of the model (1) should be assumed
0.4; 0.9 293 293 0.5 10 according to the table 3.
Table 3: Standard uncertainties of the input variables of the model
assumed for the calculations of the combined standard uncertainty
u(εob) [%] u(To) [%] u(Tatm) [%] u(ω) [%] u(d) [%]
10 10 10 10 10

To solve this problem, the simulations using data from

tables 2 and 3 were carried out. The calculations were
made with the program for simulations of the combined
standard uncertainty. The screenshot of the program is
presented in Fig. 11. The normalized histogram of the
probability density function g(Tob) of output variable Tob of
model (1) is presented in Fig. 12. The ends of the 95%
coverage interval are marked by vertical lines. The function
g(Tob) was evaluated with the method for the propagation of
distributions and Monte Carlo simulations described in
detail in [4]. In simulations the uniform distributions of the
input variables were assumed. The value of the combined Figure 12: The normalized histogram of the probability density
standard uncertainty uc(Tob) of the object temperature for function g(Tob) of output variable Tob of model (1).
the input variable estimates collected in table 2 and the Results of simulations for data collected in tab. 1 and 3 (example 7
standard uncertainties collected in table 3, was equal to
uc(Tob) = 11 K (3.4% of the expected value). To solve the task from this example, the simulations
using data from tables 2 and 3 were carried out. The
normalized histogram of the probability density function
g(Tob) of output variable Tob of model (1) is presented in
Fig. 13. The ends of the 95% coverage interval are marked
by vertical lines. In simulations the uniform distributions of
the input variables were assumed. The value of the
combined standard uncertainty uc(Tob) of the object
temperature for the input variable estimates collected in
table 2 and the standard uncertainties collected in table 3,
was equal to uc(Tob) = 6 K (1.7% of the expected value).
Calculations were carried out for εob = 0.6 and Tob = 343 K
o
(70 C). The width of the 95% coverage interval can be read
from Fig. 14. The shortest width is equal to 22 K. The limits
of the 95% coverage intervals read from Fig. 11 is I95% =
[333, 355] K.

Figure 11: The main window screen shot of the program used for
the calculations of the combined standard uncertainty in the
temperature measurements with the FLIR ThermaCAM PM595
infrared camera

Calculations were carried out for εob = 0.4 and Tob = 323 Figure 13: The normalized histogram of the probability density
o
K (50 C). The width of the 95% coverage interval for data function g(Tob) of output variable Tob of model (1).
Results of simulations for data collected in tab. 2 and 3 (example 8)
from this example can be read from Fig. 12. The shortest
width is equal to 39 K. The limits of the 95% coverage Example 12
intervals read from Fig. 12 is I95% = [302, 341] K. The temperature measurement was conducted using
the FLIR ThermaCAM PM595 infrared camera. The
Example 11
emissivity of the investigated object was equal to εob = 0.9.
The temperature measurement was conducted using
Calculate the combined standard uncertainty uc(Tob) and
the FLIR ThermaCAM PM595 infrared camera. The
evaluate the 95% coverage interval I95% assuming that the
emissivity of the investigated object was equal to εob = 0.6.
object temperature Tob = 323 K (50 oC). The values of the
Calculate the combined standard uncertainty uc(Tob) and
input variables of the model (1) should be assumed
evaluate the 95% coverage interval I95% assuming that the
according to the table 2. The standard uncertainties of the
object temperature Tob = 343 K (70 oC). The values of the

128 PRZEGLĄD ELEKTROTECHNICZNY, ISSN 0033-2097, R. 94 NR 12/2018

input variables of the model (1) should be assumed of the numerical calculations conducted in the examples,
according to the table 3. the following conclusions can be drawn:
To solve the task from example 11, the simulations • The greatest impact on the combined standard
using data from tables 2 and 3 were carried out. The uncertainty in thermographic measurements has a
calculations were made with the program for simulations of component associated with the object emissivity εob
the combined standard uncertainty. The screenshot of the • The component associated with the object emissivity εob
program is presented in Fig. 11 The normalized histogram strongly depends on the temperature of investigated
of the probability density function g(Tob) of output variable object Tob
Tob of model (1) is presented in Fig. 14. The ends of the • The component of the relative combined standard
95% coverage interval are marked by vertical lines. The uncertainty associated with the object emissivity εob
function g(Tob) was evaluated with the method for the does not depend on the object emissivity value (cf. Figs
propagation of distributions and Monte Carlo simulations 2–10)
described in detail in [4]. In simulations the uniform • The value of combined standard uncertainty increases
distributions of the input variables were assumed. The value strongly as the object’s emissivity decreases.
of the combined standard uncertainty uc(Tob) of the object Additionally, it can be seen that the increase in the
temperature for the input variable estimates collected in uncertainty value is the faster the lower the object
table 2 and the standard uncertainties collected in table 3, temperature.
was equal to uc(Tob) = 2.9 K (0.9% of the expected value) – By analyzing the results obtained in Examples 1 to 12, it
see Fig. 11. is clear that due to the complexity of the model (1), the best
Calculations were carried out for εob = 0.9 and Tob = 323 method for estimating complex standard uncertainty and
K (50 oC). The width of the 95% coverage interval for data 95% coverage interval appears to be presented in this
from this example can be read from Fig. 14. The shortest paper. In addition, according to the authors of this paper, it
width is equal to 10 K. The limits of the 95% coverage is appropriate to use specialized software to accurately
intervals read from Fig. 11 is I95% = [319, 329] K. estimate the uncertainty of temperature measurements
based on model (1), allowing numerical prediction of this
uncertainty for specific measurement situations.

Authors: dr hab. inż. Sebastian Dudzik, Prof. PCz., Czestochowa

University of Technology, Institute Optoelectronics and
Measurement Systems, Al. Armii Krajowej 17, 42-200
Częstochowa, e-mail: [email protected]; Prof. dr hab. inż.
Waldemar Minkina, Czestochowa University of Technology,
Institute Optoelectronics and Measurement Systems, Al. Armii
Krajowej 17, 42-200 Częstochowa, e-mail: [email protected]

REFERENCES
[1] Minkina W. Dudzik S. Infrared Thermography – Errors and
Uncertainties. John Wiley & Sons Ltd, Chichester, 2009, ISBN
978-0-470-74718-6
[2] Minkina W., Dudzik S., Gryś S. Errors of thermographic
th
measurements – examples. Proceedings of 10 International
Figure 14: The normalized histogram of the probability density Conference on Quantitative Infrared Thermography
function g(Tob) of output variable Tob of model (1). (QIRT’2010) 27-30.07.2010, Quebec (Canada), pp. 503-509,
Results of simulations for data collected in tab. 2 and 3 (example 8) ISBN 978-2-9809199-1-6
[3] Guide to the Expression of uncertainty in Measurement.
4. Conclusions ISO/TAG, editions: 1995, 1999 r.
In this work, the twelve examples of numerical [4] Evaluation of measurement data – Supplement 1 to the Guide
calculations were presented. The purpose of calculations to the expression of uncertainty in measurement – Propagation
was an estimation of the thermographic measurement of distributions using a Monte Carlo method, 2008, Joint
accuracy in the presence of the random interactions. The Committee for Guides in Metrology
model of the infrared temperature measurement was [5] Dudzik S., Minkina W. Application of the numerical method for
the propagation of distributions to the calculation of coverage
presented as well. Furthermore, the methodology of the th
intervals in the thermovision measurements. 9 International
95% coverage in thermographic measurements interval was Conference on Quantitative Infrared Thermography QiRT,
described. In addition, the methodology for calculation of Krakow, July 2-5, 2008, pp. 179-184
uncertainty and the 95% coverage interval was discussed, [6] Minkina W., Dudzik S. Simulation analysis of uncertainty of
and the latest recommendations of international metrology infrared camera measurement and processing path.
organizations were presented. Numerical calculations were Measurement, Vol. 39 (2006) No 8, pp. 758-763
carried out using author software. Summing up the results [7] Toolkit IC2, Dig 16 Developers Guide 1.00, AGEMA 550/570

PRZEGLĄD ELEKTROTECHNICZNY, ISSN 0033-2097, R. 94 NR 12/2018 129