Академический Документы
Профессиональный Документы
Культура Документы
Corrosion Science
journal homepage: www.elsevier.com/locate/corsci
a r t i c l e i n f o a b s t r a c t
Article history: Principles of infrared thermographic non-destructive testing for determining hidden corrosion in met-
Received 30 June 2009 als are summarized and discussed. 1D, 2D and 3D heat conduction models are introduced to simulate
Accepted 6 November 2009 thermal processes in corroded areas. In thick metals, the lateral heat dissipation is mainly responsible
Available online 11 November 2009
for significantly smoothing the temperature contrast curves. Modelling small-size corroded sites
requires 3D numerical models. Numerous defects have been simulated and the inversion formulas
Keywords: for determining material loss have been modelled for both flash and square-pulse heating. It has been
A: Steel
shown that corrosion characterization inaccuracy is lower than 20% in cases representing a practical
B: IR thermography
B: Modelling studies
interest.
Ó 2009 Elsevier Ltd. All rights reserved.
1. Introduction 2. Theoretical
0010-938X/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.corsci.2009.11.005
866 S. Marinetti, V. Vavilov / Corrosion Science 52 (2010) 865–872
10
(a) ΔL/L=90%
T/W'
4 ΔL/L=75%
ΔL/L=50%
2
ΔL/L=25%
ΔL/L=10%
ΔL/L=0%
0
0 0.2 0.4 0.6 0.8
Fo
10
(b)
ΔL/L=90%
Fig. 1. (a) 1D and (b) 3D sample models for corrosion detection. 8
Δ T/W'
Crun
(c) both the front and rear sample surfaces exchange heat with 6
the ambient by the Newton’s law, however, in most cases
adiabatic conditions can be accepted,
4
(d) boundaries between a host material and gas-filled defects
can be regarded adiabatic (possible presence of corrosion ΔL/L=75%
products deserves a special treatment), 2
(e) defects are regarded as flat bottom holes or grooves to sim-
ΔL/L=50%
ulate uniform, crevice or pitting corrosion, ΔL/L=25%
(f) variations of material thermal properties because of chemi- 0 ΔL/L=10%
0 0.2 0.4 0.6 0.8
cal reactions are neglected.
Fo
A grade of corrosion expressed in relative material loss d = DL/L Fig. 2. 1D corrosion analysis results: (a) normalized front surface temperature
is characterized by the transient differential temperature signal evolution (Dirac-pulse heating), (b) normalized differential temperature signal DT/
DT(s) = Td(s) Tnd(s), where s is time, Tnd and Td are the surface W0 and running contrast Crun evolution.
1 h
P i
1þ2
2 2 2 2 2 2
en p ðFoh FoÞ FoFo en p Fo en p Foh 1 garded 1D, thus requiring their lateral dimensions to be much
DT=T h
larger than L.For objects of a cylindrical and spherical shape, the
SSP
L ¼ ¼ 1 2 Foh n¼1
1
P
DL=L corresponding 1D inversion formulas have been proposed as fol-
Foh þ p22 1
n2
en2 p2 Fo en2 p2 Foh 1
n¼1 lows [8]:
ð4Þ ( )
1 T nd 1=2
2
where the superscripts Dirac and SP refers to the Dirac and square- dC ¼ 1 1 1 1 ð1 rÞ ; cylinder ð6Þ
r Td
pulse heating functions.
The corresponding plots are presented in Fig. 3a. The negative ( )
1 T nd 1=3
values of sensitivity mean that a decrease in thickness will result dS ¼ 1 1 1 1 ð1 rÞ3 ; sphere ð7Þ
in an increase in temperature. It is clearly seen that there is a par- r Td
ticular time period when corrosion cannot be detected because
both the sound and defect areas behave as the semi-infinite body where r is the ratio between wall thickness and external radius. It
(SL = 0). The Dirac pulse estimate for this ‘dead time’ is was shown that taking into account the shape of a tested sample
Fo1% might improve the accuracy of data inversion by up to 15% [8].
sens ¼ 0:124 (again, the 1% threshold is assumed).
Detecting corrosion is a typical inversion problem of transient In this study, 3D heat diffusion phenomena will be further ex-
heat conduction which allows, in the 1D case, to evaluate relative plored in application to defects of different geometry of which
material loss d by the Tnd(s)/Td(s) ratio (or Crun) through the follow- detection requires choosing carefully both heat source parameters
ing inverse formula proposed in [2]: and observation times.
T nd ðsÞ C run
d1D
P ðsÞ ¼ 1 ¼ ð5Þ 2.2. 3D analysis
T d ðsÞ 1 þ C run
where the subscript P specifies a plane geometry (plate), and the 2.2.1. 3D model verification
superscript 1D means that the formula holds in the 1D case. Eq. 1D models explain qualitative features of thermal NDT, but in
(5) is valid for uniform heating of adiabatic samples with extended practice quantitative estimates are needed thus justifying imple-
defects [2], as well as in the case of a moving heat source [6,7]. Cor- mentation of 3D models. It was shown elsewhere that, in case of
rosion estimates by Eq. (5) depend on observation time s. The accu- disk-shape defects, signal reduction due to lateral heat diffusion
racy of such estimates can be expressed with the coefficient can be described with a coefficient which tends to unity in the case
l ¼ d1D of extended defects (see Fig. 4) [2–5,8,12].
P =ðDL=LÞ, of which evolution vs. Fo is presented in Fig. 3b.
If the process can be assumed adiabatic, the sensitivity to rear- Such a coefficient g(D0 ), with D0 = D/L, was obtained for round-
side corrosion is increasing in time reaching a particular maxi- shape defects using Eq. (5) at Fo = 0.68 in the following form [5]:
mum. The accuracy in evaluating d is high if defects can be re- 3
ðD0 Þ4 1:1102 ðD0 Þ3 101 ðD0 Þ2 6:1102 ðD0 Þþ3:9102
gðD0 Þ ¼ 1 e1:810 ð8Þ
-0.6
2D case (axial symmetric geometry).
A more general approach to defining a correction coefficient as a
function of time s, material loss DL/L, D/L ratio and sample thick-
-0.8
ness L will be discussed below.
(a) The 3D transient heat conduction problem formulated as re-
-1 Dirac pulse
ported above has been solved by using the ThermoCalc-6L program
1%
0 Fosens 0.5 1 1.5 2
Fo
1
1 0.8
0.8
0.6
ΔL/L=10%
η
0.6 ΔL/L=25%
μ
ΔL/L=50% 0.4
0.4 ΔL/L=75%
ΔL/L=90% 0.2
0.2
(b)
0
0 0 1 2 3 4 5 6 7 8
0 0.2 0.4 0.6 0.8 D/L
Fo
Fig. 4. 3D heat diffusion correction coefficient adapted from [11] for round-shape
Fig. 3. (a) Sensitivity to sample thickness vs. Fo, (b) accuracy of Eq. (3) vs. Fo. defects (Dirac-pulse heating, Fo = 0.68).
868 S. Marinetti, V. Vavilov / Corrosion Science 52 (2010) 865–872
from Innovation, Ltd., Russia [10], which implements a finite-dif- absolute temperatures but, as shown by Cases 6 and 4, did not af-
ference method and an implicit computation scheme. Some results fect the running contrast and, hence, corrosion estimates.
have been also obtained by using the COMSOL MultiphysicsÒ pack- An example of the computed 3D corrosion model is presented in
age involving a finite elements method. The computation accuracy Fig. 6a along with the temperature distributions observed at two
was first checked in a sound area by comparing the numerical re- different times (Fig. 6b and c). It is clearly seen that heat diffusion
sults with the classical solution for heating an adiabatic plate with significantly smoothes temperature patterns over hidden defects
a square pulse and proved to be better than 1%. and thus prevents detecting small defects.
‘‘Defect” temperatures have been checked by comparing two The time evolution of temperature signals in a sound and defect
50%-material loss models: a disk-shaped defect with a 2.8 mm ra- area (15 15 mm) is shown in Fig. 7a for Case 1. Other plots in
dius (2D axial symmetric geometry, COMSOL Multiphysics pro- Fig. 7 illustrate how the 1D estimate of material loss dest calculated
gram) and a 5 5 mm square defect (3D Cartesian geometry, by Eq. (5) evolves in time in regard to the true material loss value
ThermoCalc-6L program). The temperature profiles and their rela- dtrue. It is seen that dest is always smaller than dtrue, and the best
tive difference are shown in Fig. 5a. The observed divergence, estimates of corrosion appear at particular times sm which depend
which is partially due to the difference in the shape of the two de- on both material loss dtrue and pulse duration sh. It can be easily
fects having the same in-plane area, is less than 1% in the time demonstrated that sm is the time when the corresponding maxi-
interval of interest. Also, the maximums of Crun (Fig. 5b) calculated mum running contrast Crun, or minimum Tnd/Td ratio, occurs. Note
with the two programs matched well producing errors under 2.5%. that, with longer heat pulses, contrast maximums could appear
also within heating. In some cases, there is a local maximum of dest
which takes place immediately after a heat pulse (see Fig. 7c). In
2.2.2. Corrosion detection peculiarities the case of short heat pulses, optimum detection times appear after
The analysis below has been done on rectangular-shaped de- the pulse as single maximums (Fig. 7d) thus approaching the case
fects (Fig. 1b) with different ratios between lateral dimensions of Dirac-pulse heating.
and material loss. In total, over 100 situations differing by defect In accordance with the conclusions following the 1D analysis
lateral dimensions (Dmin, Dmax), material loss (d), thermal proper- [2], the results above prove that flash heating provides better esti-
ties (K, a), sample thickness (L) and heating time (sh) have been mates of corrosion. However, short heat pulses available in practice
computed and afterwards used to derive inversion formulas (some may be not powerful enough to produce detectable temperature
cases are presented in Table 2 as illustration). Most cases were as- signals DT. Therefore, thick metallic samples should be heated long
sumed to be adiabatic. The correctness of such assumption is con- enough to absorb a necessary amount of thermal energy. In this
firmed by Case 7 where the presence of convection on both sample case, ensuring a maximum accuracy of corrosion estimates re-
surfaces (h = 10 Wm2 K1) did not significantly change detection quires that heating times sh should not typically exceed predicted
parameters in regard to Case 4 (see Table 2). Also, variations in sm values. If the energy absorbed by a sample by that time will not
material thermal conductivity used to lead only to variations in build up temperature signals higher than a temperature resolution
of a used IR camera, heating should be continued with progressing
loss of the sensitivity to material loss.
15 1
The limitation of the 1D approach is illustrated in Fig. 8 where
difference [%] Eq. (5) has been applied to evaluate both a 1D (extended) and 3D
(a) 0.5
(finite-size) defects representing 25% corrosion. In the 1D case,
0 the accuracy of corrosion evaluation increases in time and reaches
-0.5 the true value (25%) after the pulse due to the lack of considerable
10
difference, %
-1 lateral heat diffusion. In its turn, the best estimate of the 3D case
T, °C
-1.5 occurs within the heat pulse to result in only 17% of material loss.
-2
Comsol Multiphysics (r=2.8mm)
5
ThermoCalc-6L (l=5.0mm) -2.5
3. Corrosion characterization
-3
-3.5 Practical application of the methodology described above
0 -4 should result in quantitative estimation of material loss. By numer-
0 0.1 0.2 0.3 0.4 0.5 0.6
ical modelling, optimum inspection parameters, such as type of
Fo
heating, as well as heat pulse energy and duration, should be cho-
sen. Heating duration should be as short as possible and powerful
0.3
enough to increase sample temperature up to a detectable level at
(b) Comsol Multiphysics (2D)
time of observation. Experimental data is to be typically enhanced
0.25 ThermoCalc-6L (3D)
to provide an acceptable value of signal-to-noise ratio, then sample
0.2
areas, where corrosion is being suspected, are to be segmented.
Relative material losses in corroded sites are to be estimated by
Crun
0
s
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi !
0 0.1 0.2 0.3 0.4 0.5 0.6 Dmin Dmin
d3Dest ¼ A exp B þC þD ð10Þ
Fo L apparent L apparent
Fig. 5. Comparing numerical results in computing (a) defect temperature and (b)
running contrast. where
S. Marinetti, V. Vavilov / Corrosion Science 52 (2010) 865–872 869
Table 2
3D corrosion detection models numerically calculated with ThermoCalc-6L.
Fig. 6. Modelling 25% corrosion in a 5 mm-thick steel AISI 1010 plate heated with 5 s square pulse (Case 2 in Table 2): (a) scheme of defects, (b) temperature distribution at
s = 0.1 s and (c) s = 5 s.
A ¼ 0:2652
C max
0:036
2 þ 0:0759 In Eq. (10), the term (Dmin/L)apparent depends on the aspect ratio R of
run ðCmax
run Þ
the defect, defined as Dmax/Dmin. For square defects (R = 1)
B ¼ 1:7537 e 6:6779C max
run 1:1901 ð11Þ
pffiffiffiffiffiffiffi Dmin Dmin
C ¼ 1:3449 10:1429 e 3:6099 C max
run
¼ : ð12Þ
L apparent L
870 S. Marinetti, V. Vavilov / Corrosion Science 52 (2010) 865–872
5 1
4.5 (a) 0.9 (b) 5s
4 Defect 0.8
3.5 0.7 5 mm
3 0.6 50 %
δest\δtrue
T, °C
1 1
1s
0.9 (c) 0.9 (d) 0.01 s
0.8 0.8 2 mm
50 %
0.7 5 mm 0.7
50 %
0.6 0.6
δest\δtrue
δest\δtrue
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
time, s time, s
Fig. 7. Temporal evolution of corrosion detection parameters (Table 2): Case 1 (a) temperature signals in defect #4 (15 15 mm) and non-defect area, (b) dest/dtrue for the
same defect, (c) dest/dtrue for Case 3 defect #4 (15 15 mm) and (d) dest/dtrue for Case 5 defect #4 (6 6 mm).
where
δest
0.15 3D
0:1638
0:5980
Dmax Dmin
0.1 k ¼ 0:4543 Fo0:2512
h Fo0:4025
max
L L
0:2286
0.05 C max
run ð17Þ
Fig. 8. Estimates of 25% of corrosion in a 5 mm-thick steel AISI 1010 sample heated
with a 5 s square pulse: 1D (large defect) and 3D (Case 2 defect 25 25 mm).
4. Experimental
Dmin Dmin The experimental set-up included an IR imager, a flash heater
¼ 1:225 ð13Þ (two tubes 4.8 kJ each delivered in 5 ms) (Fig. 10) and a tubular
L apparent L
quartz lamp heater (30 kW continuous energy). IR image se-
otherwise quences were captured with the acquisition rate varying from 1
"
2 # to 30 Hz to be further processed on a computer by using the inver-
Dmin Dmin Dmin Dmin sion algorithms described above, as well as applying some image
¼ a þb þc ð14Þ
L apparent L L L processing tools available in thermal NDT.
where
4.2. Flash heating
a ¼ 0:0785 0:2754 e0:7817R
b ¼ 0:4635 þ 1:4809 e0:7737R ð15Þ An AISI 1045 steel sample (L = 3 mm) containing nine bottom-
0:7754R hole defects, of which scheme is shown in Fig. 11 along with some
c ¼ 1:8798 2:0579 e
raw IR images, has been inspected by using the set-up from Fig. 10.
This procedure can be easily implemented in any programming lan- The front surface of the sample was black-painted, that is a typical
guage, and the block diagram is depicted in Fig. 9.In case of square- procedure in IR thermography enhancing material absorptivity/
S. Marinetti, V. Vavilov / Corrosion Science 52 (2010) 865–872 871
Table 3
Identifying material loss in a 3 mm-thick AISI 1045 steel sample, with thermal diffusivity a = 13.5 106 m2/s (Fig. 11).
References
[1] K.E. Cramer, P.A. Howell, H.I. Syed, Quantitative thermal imaging of aircraft
structures, Proc. SPIE ‘‘Thermosense XVII” 2473 (1995) 226–232.
[2] V. Vavilov, E. Grinzato, P.G. Bison, S. Marinetti, M. Bales, Inversion for hidden
corrosion characterization: theory and applications, Int. J. Heat Mass Transfer
(39) (1996) 355–371.
Fig. 12. Inspecting a 10 mm-thick AISI 1045 steel sample by square-pulse heating:
[3] E. Grinzato, V. Vavilov, Corrosion evaluation by thermal image processing and
(a) defect scheme and contrast features, (b) raw images.
3D modelling, Rev. Generale Termique 8 (37) (1998) 669–679.
[4] V. Vavilov, V. Shiryaev, E. Grinzato, Detection of hidden corrosion in metals by
using transient infrared thermography, Insight 6 (40) (1988) 408–410.
[5] S. Marinetti, P.G. Bison, E. Grinzato, 3D heat flux effects in the experimental
evaluation of corrosion by IR thermography, in: Intern. Conf. Proc. ‘‘Quant. IR
Thermography 6 – QIRT’02” Dubrovnik, Croatia, 2002, pp. 92–98.
[6] K.E. Cramer, R. Jacobstein, T. Reilly, Boiler tube corrosion characterization with
a scanning thermal line, Proc. SPIE ‘‘Thermosense XXIII” 4360 (2001) 594–605.
[7] D.F. Woolard, K.E. Cramer, The thermal photocopier: a new concept for thermal
NDT, Proc. SPIE ‘‘Thermosense XXVI” 5405 (2004) 366–373.
[8] E. Grinzato, V. Vavilov, P. Bison, S. Marinetti, Hidden corrosion detection in
thick metallic components by transient IR thermography, Infrared Phys.
Technol. (49) (2007) 234–238.
[9] J.-S. Han, J.-H. Park, Detection of corrosion steel under an organic coating by
infrared photography, Corros. Sci. (46) (2004) 787–793.
[10] ThermoCalc-6L Operation Manual, Innovation, Ltd., Russia, 2007.
[11] S. Marinetti, V. Vavilov, Sensitivity analysis of classical heat conduction
Fig. 13. Advanced data treatment in the inspection of a 10 mm-thick AISI 1045 solutions applied to materials characterization, Heat Transfer Eng. 9 (26)
steel sample: (a) raw image at the best observation time, (b) phasegram. (2005) 50–60.
[12] E. Grinzato, V. Vavilov, P.G. Bison, S. Marinetti, Methodology of processing
experimental data in transient thermal NDT, Proc. SPIE ‘‘Thermosense-XVII”
2473 (1995) 167–178.
noticeable differential temperature signals in corroded sites. [13] D. Almond, P. Patel, Photothermal Science and Techniques, Chapman & Hall,
Therefore, the compromise between producing necessary temper- London, UK, 1996.
[14] W.J. Parker, R.J. Jenkins, C.P. Butler, G.L. Abbot, Flash method of determining
ature contrasts and ensuring detectable temperature signals is to
thermal diffusivity, heat capacity and thermal conductivity, J. Appl. Phys. (32)
be achieved in each particular test case. (1961) 1679–1684.
Simple estimates for corrosion detection limits have been found [15] X. Maldague, S. Marinetti, Pulse phase infrared thermography, J. Appl. Phys. 79
as a function of defect size and noise parameters. Flash heating is (5) (1996) 2694–2698.