crack imaging by scanning pulsed laser spot thermography
DESCRIPTION
A new crack imaging technique is presented that is based on second derivative image processing of thermal images of laser heated spots. Experimental results are shown that compare well with those obtained by the dye penetrant inspection method. A 3D simulation has been developed to simulate heat flow from a laser heated spot in the proximity of a crack. A ‘ghost point’ method has been used to deal efficiently with cracks having openings in the micometre range. Results are presented showing the effects of crack geometry and system parameters on thermal images of laser heated spots.TRANSCRIPT
-
sty of
niq
eate
ant
pot
ving
and
on-contact, quick inspectionts such as delaminations andue that is not suitable for theking cwhich
e, typicd thaon aser splocalifromal sha
material in the proximity of a crack. Furthermore, a new image
of a crack.
Modelling has been used to gain an understanding of the optimumoperating parameters for the laser spot thermography technique andto assess the theoretical limits of its sensitivity for the detection of
ut of
TLTi1 TRTL TiTR
where Ti1, TL, TR, Ti and Ti+1 are, respectively, the temperature
Contents lists available at ScienceDirect
.els
NDT&E Inte
NDT&E International 44 (2011) 216225point. By dening a ghost point to equal the temperatureE-mail address: [email protected] (D.P. Almond).rise at the grid point i1, the left boundary of the crack, the rightboundary of the crack and the grid points of i and i+1; g is theheat ux.
Eq. (2) shows the calculation of temperature rise at the ghost
0963-8695/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ndteint.2010.08.006
n Corresponding author.processing technique has been developed to form a direct imageKs s Ka d Ks dsd g 1perpendicular cracks. Based on the previous studies, a full 3Dghost point heat transfer nite difference model has beendeveloped to predict the thermal behaviour of laser spot heated
point i1 into the crack, through the crack, and then ows othe crack to the grid point i givestions if the material is thermally isotropic. Perpendicular cracks,close to the heated spot, will perturb the round lateral heat owand the perturbation may be detected by a thermal camera toreveal the cracks. Preliminary studies [614] showed that theshape of the heat ow at certain times was deformed clearly by
The distance of the crack to the left grid point is s. Usually, thecrack is full of air and its conductivity Ka is much lower than theconductivity of the metal block Ks. Thus thermal gradient across thecrack will be larger than in other parts of the metal block.
Heat ux balance in the x direction when it ows from the grid1. Introduction
Pulsed thermography (PT) is a nmethod that detects in-plane defecimpact damage [14]. It is a techniqdetection of important surface breaby fatigue or creep. These cracks,perpendicular to the material surfacfew microns. It has been reporteopenings (widths) of below 0.5 mmnot be detected by PT [5]. In the latechnique, a laser provides a highlyow in the volume of the materialrelatively symmetrical half-sphericracks in metals causedgrow predominantly
ally have openings of at surface cracks withconcrete surface couldot thermography (LST)zed heating spot. Heatsuch a spot presents ape in the radial direc-
cracks. The effects of cracks may be simulated by ghost points in anumerical modelling grid that are generated by balancing thermaluxes owing into a crack and through a crack, with those owingout of the crack according to Fouriers law. They guarantee correctthermal gradients in the bulk material on either side of the crack.
The concept of the ghost point in a 1D nite difference heattransfer model is shown in Fig. 1. In this case, the crack is embeddedbetween the current grid point i and its left grid point i1. Thewidth of the crack d may be far smaller than the grid spacing d.2. Ghost point heat diffusion modelCrack imaging by scanning pulsed laser
Teng Li, Darryl P. Almond n, D.Andrew S. Rees
UK Research Centre in NDE (RCNDE), Department of Mechanical Engineering, Universi
a r t i c l e i n f o
Article history:
Received 9 December 2009
Received in revised form
18 August 2010
Accepted 19 August 2010Available online 25 August 2010
Keywords:
Infrared thermography
Laser
Thermal imaging
Simulation
Finite difference method
Cracks
Image processing
a b s t r a c t
A new crack imaging tech
thermal images of laser h
obtained by the dye penetr
ow from a laser heated s
efciently with cracks ha
effects of crack geometry
journal homepage: wwwpot thermography
Bath, Claverton Down, Bath BA2 7AY, UK
ue is presented that is based on second derivative image processing of
d spots. Experimental results are shown that compare well with those
inspection method. A 3D simulation has been developed to simulate heat
in the proximity of a crack. A ghost point method has been used to deal
openings in the micometre range. Results are presented showing the
system parameters on thermal images of laser heated spots.
& 2010 Elsevier Ltd. All rights reserved.
evier.com/locate/ndteint
rnational
-
By using the ghost points to balance the heat ux, the heattransfer model avoids the need of very ne mesh spacing that isnecessary to deal with real cracks that often have openings of onlya few micrometres [6,7]. In the following simulations, theboundary conditions for all metal samples are assumed to beinsulated. Table 1 shows the property parameters used in thesimulations for the air gap and two types of steel.
3. Experimental and modelling results
3.1. Modelling validation
Previous publications using the 2D ghost point method [6,7]gave an indication of the way in which defect opening and hostmaterial affect LST response. In this research, a full 3D ghost pointmodel was developed to enable a comprehensive investigation ofall factors that affect the response of the technique. A furtherdifference is that a much higher power laser was used in thecurrent work. The previous work [6,7] was completed using laserpowers of o1 W, whilst in the current work a 21 W Laservallindustrial bred diode welding/brazing laser has been used. Thelaser beam wavelength is 808 nm and its focal spot diameter isabout 1.8 mm. The laser is operated in a pulse mode. The pulsewidth can be set from 1 ms to 10 s.
Fig. 2 shows the experimental setup. A computer is used to
T. Li et al. / NDT&E International 44 (2011) 216225 217increase effect because of the crack, the isotropic heat transfermodel can still be applied; however, temperature rise at the gridpoint i1 should be replaced by the value of the ghost point TG:
KsTiTG
d g 2
After substituting Eq. (1) in Eq. (2), we can represent TG by Tiand Ti1:
TG x
xd Tid
xd Ti1 3
where x is related with Ks, Ka and d:
x KsKa1
d 4
The 1D heat diffusion equation with no internal heat genera-tion is
k@2T
@x2 @T
@t5
where k is the thermal diffusivity. Substituting Eq. (3) in Eq. (5),and representing Eq. (5) using nite difference elements, we canhave
Tm1i kDt2d
Tmi1aTmi bTmi1Tmi 6
where Dt is the time step, Tmi the temperature rise of the gridpoint i at time m and Tm1i the temperature rise at the next timestep. The values of a and b are
a x2dxd 7
b dxd 8
If the heat diffusion model is 2D, then the value of a becomes
TG
KaTi-1
Ks
TR
dx
Ghost point
TL
Ti
Ti+1
Fig. 1. 1D ghost point nite difference heat diffusion model. Heat ux is balancedwhen it ows into, through and out of the crack.a 3x4dxd 9
Similarly, in the 3D heat diffusion model, a has the value
a 5x6dxd 10
If the crack is embedded between current grid point i andpoint i+1, then grid point i+1 becomes the ghost point andtemperature rise at this point should multiply the coefcient blike the ghost point i1 in Eq. (6).
Furthermore, the temperature gradient across the crack canalso be derived from Eq. (1) as follows:
TRTLd
TiTi1Ka=Ksddd
11control the translation stage to move either the test-piece or thelaser head (Fig. 2 shows only one arrangement). The computeralso controls the laser output power and pulse duration. Theinfrared camera used to record the thermal images was a MerlinMID (Indigo Systems Cooperation) with a frame rate of 60 Hz.Thermal images are stored into a second computer for post-processing.
Fig. 3a shows a mild steel test-piece (painted with black paint onthe surface) with a fatigue crack developed vertically at the tip of aT-notch. The crack opening (width) is about 50 mm, length is about5 mm and depth is about 4 mm. Fig. 3b shows the resulting thermalcontour images obtained after subtracting the background thermalimage. The rst 9 frames are shown after a 13W laser spot heating
Table 1Parameters used in the simulations.
Material Thermal
conductivity
K (W/m K)
Specic heat
C (J/kg K)
Mass density
r (kg/m3)Thermal
diffusivity
kK/rC (m/s2)
Air 0.025 1000 1.205 2.0747105Mild steel 40 500 7850 1.0191105Stainless steel 13.5 485 7900 3.5234106Fig. 2. Experimental setup.
-
with a duration of 50 ms irradiated at the right hand side of thecrack (about 1 mm from the laser spot centre to the crack). The rst3 frames show the time when the laser was still on. From the 4thframe, the laser spot was switched off and the heat started todissipate across the test-piece. The effect of thermal resistance of theair-lled crack is to cause thermal ows, evident in the images, toform a D shape rather than a round shape. It can also be seen that
the full extent of the crack becomes clearer with elapsed time. Fig. 3cshows corresponding simulated thermal contour images using aGaussian shape pulse. Note here, temperature rise scales in Fig. 3bare in IR camera digital level units and in Fig. 3c are in units ofdegrees Celsius. The conversion factor is around 300. There is a verygood agreement between the sets of experimental and simulatedthermal images.
Crack
Laser spot
X / mm X / mm X / mm
X / mm X / mm X / mm
m
Frame no. = 1, Time 16.67ms
Y /
mm
Y /
mm
/ m
m
Y /
mm
Y /
mm
/ m
m
Y /
mm
Y /
mm
/ m
m
0 2 4 6 80
2
4
6
8
0
0.5
1
1.5
2
2.5x 104 x 104 x 104
Frame no. = 2, Time 33.34ms
0 2 4 6 80
2
4
6
8
0.5
1
1.5
2
2.5
Frame no. = 3, Time 50.01ms
0 2 4 6 80
2
4
6
8
0.5
1
1.5
2
Frame no. = 4, Time 66.68ms
0 2 4 6 80
2
4
6
8
0
5000
10000
Frame no. = 5, Time 83.35ms
0 2 4 6 80
2
4
6
8
0
2000
4000
6000
Frame no. = 6, Time 100.02ms
0 2 4 6 80
2
4
6
8
0
1000
2000
3000
4000
5000
4
6
8
2000
3000
4000
ime
6
4
6
8
2000
3000
4
6
8
1500
2000
2500
peri
T. Li et al. / NDT&E International 44 (2011) 216225218X / mm X / m
Y Y
Frame no. = 7, Time 116.69ms
0 2 4 6 80
2
0
1000
Frame no. = 8, T
0 2 40
2
Fig. 3. (a) Mild steel test-piece with a crack at the tip of a T-notch. (b) Resulting ex
(c) Corresponding simulated thermal contour images. Temperature rise scales are in dX / mm
Y
133.36ms
80
1000
Frame no. = 9, Time 150.03ms
0 2 4 6 80
2 500
1000
mental thermal contour images. Temperature rise scales are in digital level units.egrees Celsius.
-
mm
mm
Time
0
e
T. Li et al. / NDT&E International 44 (2011) 216225 219Frame no. = 1, Time 16.67ms
Y /
mm
Y / mm Y /
Y / mm Y /
Y /
mm
mm
Y /
mm
Y /
mm
mm
0 2 4 6 80
2
4
6
8
Frame no. = 2,
0 2 40
2
4
6
8
Frame no. = 4, Time 66.68ms
0 2 4 6 80
2
4
6
8
Frame no. = 5, Tim
0 2 40
2
4
6
8
6
8
6
8
10
20
30
40
50
51015202530
81012Eq. (12) shows the 3D analytical solution of temperaturedistribution using Duhamels theorem in a semi-innite metalblock after the irradiation of a laser spot [15]:
Tr,z,t Imaxra2k1=2
Kp1=2
Z t0
pttuexpz2=4ktur2=4ktura2tu1=24ktura2
dtu
12
where, (r,z) are cylindrical coordinates with the origin on thesurface at the centre of the irradiated spot. Imax is the maximumpower density of the laser pulse, p(t) the normalized temporalprole of the laser pulse at time t, k the diffusivity, K theconductivity and ra the laser beam radius.
Fig. 4 shows a comparison of the 3D analytical modelling resultsand the 3D ghost point modelling results. The laser pulse was ofsquare shape (50 ms) in the time domain and Gaussian shape in thespatial domain. The radius of the spot was taken to be 1 mm (1/e falland ra1mm). Laser output power was 20W; thus, ImaxE20/(pra2)E6.37106W/m2. The width of the crack d in the ghost pointmodel was set to 0 since Eq. (12) is for an isotropic metal block. Notehere that reectivity of the sample surface was not considered in bothmodels, that is, all 20W energywas assumed to be absorbed bymetalsamples. Lines with different colors in the gures correspond totemperature rises of the laser spot centre at different depths in themetal block. Differences between the two curves are very small,caused by several factors: boundary conditions of the 3D numericalmodel and accuracy of the 3D numerical model, such as time step,space step and different numerical methods. The numerical model
Y / mm Y / mm
Y /
Y /
Frame no. = 7, Time 116.69ms
0 2 4 6 80
2
4
Frame no. = 8, Time
0 2 40
2
4
246
Fig. 3. (ConY / mm
Y / mm
Y /
mm
mm
33.34ms
6 8
Frame no. = 3, Time 50.01ms
0 2 4 6 8
83.35ms
6 8
Frame no. = 6, Time 100.02ms
0 2 4 6 80
2
4
6
8
6
8
5
10
15
20
5
10
15
6
8
10
6
8Y /
mm
2
4
6
8
20
40
60
20
40
60
80here produces a good approximation to the analytical model, whichvalidates the model.
Fig. 5a shows the experimental results obtained using a laserspot, with radius of about 1 mm, irradiated on a mild steel surfacefor 50 ms with powers between 12 and 20 W, indicated in thegure. The reectivity of the sample surface was about 60%. Thegraph shows measurements obtained at 1/60 s time intervals,corresponding to the 60 Hz frame rate of the IR camera. Maximumtemperature rises (corresponding to temperature rises at the laserspot centre) are shown. Fig. 5b shows the 3D ghost point nitedifference modelling results. Temperature rises after cooling time50 ms under different laser power irradiations in experiments(Fig. 5a) are less distinguishable than theoretical curves (Fig. 5b)because of the sensitivity of the IR camera and thermal noise atroom temperature (20 1C). However, there is a relatively goodagreement between the measured and modelling results beforecooling time, which adds to the validation of the modellingprovided by the above comparison with the analytical solutions.
3.2. Crack metric
Fig. 6 is a line prole taken through the centre of the thermalimage of the laser spot shown in frame 4 of Fig. 3b. A step changein temperature (TRTL), temperature to the right minus tempera-ture to the left of the crack, occurs across the crack and itsmagnitude characterizes the thermal blocking power of the crack.It is also a measure of the effect that the crack has on thermal
Y / mm
Y /
133.36ms
6 8
Frame no. = 9, Time 150.03ms
0 2 4 6 80
2
4
2
42
4
tinued)
-
0 20 40 60 80 100 120 140 160 180 200
30
35
/ K
Temperature rise within mild steel
Laser power=12W
Laser power=14W
Laser power=16W
T. Li et al. / NDT&E International 44 (2011) 216225220TIME / ms
80
90Temperature Rise within Steel after an incident Square laser pulse
0 mm0.1 mm0.2 mm0.5 mm0
10
20
30
40
50
60
70
80
90
TEM
PE
RA
TUR
E R
ISE
/ K
Temperature Rise within Steel after an incident Square laser pulse
0 mm0.1 mm0.2 mm0.5 mm1 mmimage of the laser spot and, consequently, it may be used to assessthe sensitivity of the technique to cracks of varying dimensions.The step temperature change metric is also found to be afunction of system operating parameters, as detailed in Sections3.3 and 3.4.
3.3. Suitable laser power and pulse duration
Important operating parameters are laser power and pulseduration. For laser pulses with the same output energy it is foundthat the shorter the laser pulse, the higher the temperature differenceacross the crack. This is a result of the rapid thermal diffusion rate inmetals, which causes transient thermal perturbations to dissipateover a shorter time than the duration of the longer pulses. Balancedagainst this effect is a general increase in crack thermal metric withlaser pulse energy. Variations in spatial shape of the laser heating anddiameter of the spot can change the distribution of laser energy at thesample surface. However, they are also related with laser outputpower. In practice, it has been found that good results are obtained inseveral metal test pieces using a 50ms pulse duration and laserpower in the range 1021W, depending on reectivity of the metalsurface. These sample materials include mild steel, stainless steel,titanium alloy and Udimet.
0 50 100 150 2000
10
20
30
40
50
60
70
TIME / ms
TEM
PE
RA
TUR
E R
ISE
/ K 1 mm
Fig. 4. (a) 3D analytical modelling results. (b) 3D ghost point nite differencemodelling results.3.4. Optimum imaging distance
The optimum imaging distance of the laser spot centre fromthe crack was also found by both the experiment and the ghostpoint model. The optimum distance is where the temperature
0 50 100 150 2000
5
10
15
20
25
TIME / ms
TEM
PE
RA
TUR
E R
ISE
Laser power=18W
Laser power=20W
Fig. 5. (a) Experimental result. (b) 3D ghost point modelling result.
Fig. 6. Frame 4 of Fig. 3b and thermal line prole along line indicated.
-
20
0 20 40 60 80 1000
5
10
15
20
25
Tem
p. d
iff. a
cros
s th
e cr
ack
/ K
Crack opening / m
Crack depth: 2mm; crack depth: 2mmLaser to crack: -1mm; laser radius: 0.9mm
5 m
T. Li et al. / NDT&E International 44 (2011) 216225 22125
30
ck /
Kdifference across the crack has the largest value. Thus the thermalgradient across the crack reaches the largest value too. Fig. 7ashows the maximum temperature difference (near the laser spotcentre) across the crack when the spot centre has differentdistances to the crack (moving along a line through the spotcentre perpendicular to the crack). The same test-piece shown inFig. 3a was used in the experiment. The laser radius was about1.25 mm. It is shown in Fig. 7a that the maximum temperaturedifference across the crack occurs at the position when the laserspot centre is 1 beam radius (1/e fall) away from the crack. Fig. 7bshows the modelling result by using the 3D ghost point model.The laser radius was 0.9 mm (1/e fall). Simulated crack openingwas 10 mm, length was 2 mm and depth was 2 mm. Temperaturedifference across the crack reaches the maximum value when thedistance of the laser spot centre to the crack was 0.9 mm. Thereectivity of the sample surface was set at 50%. Again, themodelling result shows that the optimum imaging distance isone radius of the spot.
3.5. Variations in crack parameters
Fig. 8a shows modelling results of variations in temperaturedifference across a crack with different openings in a mild steelblock (10 mm10 mm5 mm). The laser spot radius was
-4 -2 0 2 40
5
10
15
20
Tem
p. d
iff. a
cros
s th
e cr
a
Distance of laser to crack / mm
-0.9mm
Fig. 7. Temperature differences across the crack when the spot centre hasdifferent distances to the crack: (a) experimental results, expressed in IR camera
digital levels and (b) modelling results.300.9 mm and the spot was centred at the position near theoptimum position, one beam radius from the crack. The laseroutput power and pulse duration were assumed, respectively, to
0 2 4 6 8 100
5
10
15
Crack length / mm
Tem
p. d
iff. a
cros
s th
e cr
ack
/ K Crack depth: 2mm; crack opening: 10 mLaser to crack: -1mm; laser radius: 0.9mm
0 1 2 3 4 50
5
10
15
20
25
Crack depth / mm
Tem
p. d
iff. a
cros
s th
e cr
ack
/ K
Crack depth: 2mm; crack opening: 10 mLaser to crack: -1mm; laser radius: 0.9mm
Fig. 8. Dependence of temperature differences across the crack on (a) crackopening, (b) crack length and (c) crack depth.
-
be 21 W and 50 ms. The reectivity of the sample surface was setat 50%. Temperature difference across the crack reaches a plateauvalue when the opening is larger than 5 mm. Similarly, whenchanging one of the other crack parameters, crack length or depth,temperature difference across the crack also reaches a plateauvalue, as shown in Fig. 8b and c. Note here that the residualtemperature difference caused by the Gaussian spot prole hasbeen removed. These gures show only the changes caused by thecrack. The plateaus occur for crack lengths 43 mm and crackdepths 42 mm. These results provide an indication of the limitsof laser spot imaging for detecting cracks with certain geometries.They also indicate that the technique, theoretically, has asensitivity adequate for many inspection requirements.
observation time happened at around 200 ms for a 3 mm long,10 mm wide half-penny crack in an Udimet test-piece. The fullextent of longer cracks cannot be satisfactorily imaged by a singlelaser spot heating but it will be revealed by combining the imagesobtained on executing a suitable scanning raster.
4. Crack imaging by scanning laser spot
The 3D ghost point model has helped establish the limits ofeffectiveness of LST in the detection of cracks with opening in themicrometre range. In addition, this 3D model has helped thedevelopment of a new crack imaging method. A common imageprocessing technique is to compute the rst spatial derivative,which reects the amplitude change rate in an image and thusextracts edges in images. The perturbation of a laser spot imagecaused by a crack can be regarded in the same way as an edgefeature. However, the background heat ow caused by thefocused laser spot is still strong and causes large thermalgradients that are mixed together with the crack effects whenthe spot is close to the crack. The second spatial derivative wasalso considered since it can further enhance the edge effect in an
T. Li et al. / NDT&E International 44 (2011) 2162252223.6. Effect of crack shape
In reality, cracks present different shapes to the rectilinearslots employed in the basic modelling. Very often, a half-pennyshaped crack is formed [16]. Fig. 9 shows an ideal square slotcrack and a half-penny crack. Simulation results show that thetemperature difference across the crack for a half-penny crack is1.26 1C smaller than that of a slot crack (lengths of both cracks are3 mm, openings are 10 mm, in mild steel pieces). The laser outputpower was 21 W, pulse duration was 50 ms and beam radius was1 mm. The results were calculated at 0.2 s. Thus, heat blockageeffect of the half-penny crack is smaller than that of an ideal slotcrack as might be expected. Furthermore, a pouched half-pennycrack (smaller openings at crack tips and bottom) can also besimulated, which has smaller temperature difference than that ofthe half-penny crack by about 0.5 1C under the above situations.
These modelling results show that crack shape can also affectthe acquired thermal images and consequent sensitivity of thetechnique.
3.7. Optimum observation time
There is an optimum observation time at the optimumdistance at which the maximum amount of a long crack isrevealed. It can be seen in Fig. 3 that the full length of the crackemerges in the later frames. However, if a crack in mild steel islonger than about 3 times of the laser diameter, the signal to noiseratio (SNR) will be poor by the time the heat diffuses to reach thecrack tip. Thus, the optimum observation time can be dened asthe time when part of the crack, usually with the length within 23 diameters of the laser spot (related with the material, thesurface situation, etc.), can be observed at the optimum distance.The term can be observed means that the maximum temperaturedifference across certain part of the crack is at least 6 dB abovethe noise. For example, in our experiments, the optimum
Depth
Side view Side view
Opening
Length
Opening
Length Depth Profile line Fig. 9. (a) An ideal slot crack. (b) A half-penny crack.X / mm-6 -4 -2 0 2 4 6
6
Fig. 10. Simulated thermal image of a stainless steel sample containing a 5 mmlong, 1 mm opening crack at time 0.15 s. The temperature prole across the crackimage. Fig. 10 shows a simulation of the surface temperatureimage of a laser heated spot on a stainless steel metal blockobtained from the 3D model. The geometry of metal block was12 mm12 mm5 mm in x, y and z directions and the model-ling grid spacing was 0.1 mm. A 5 mm long half-penny shapedcrack was embedded in the grids in the middle of the block. Thecrack opening was 1 mm. The thermal image, Fig. 10, was obtained0.15 s after the block was irradiated by a 21 W, 50 ms laser pulsewith radius of 1 mm (1/e fall). The laser spot was centred 1 mm tothe right of the crack. The temperature prole across the crackand the centre of the spot is also shown in the gure.
Fig. 11a shows the normalized rst (blue line) and second (redline) derivatives of the temperature prole line plotted in Fig. 10.The red line in Fig. 11a shows the much improved crackdiscrimination obtained from second derivative processing.Fig. 11b and c shows normalized 2D images obtained using therst and second derivative image processing of the data shown inFig. 10 (derivatives taken in the x direction perpendicular to thecrack). It is clear that the second derivative processing provides a
Y /
mm
-6
-4
-2
0
2
4 5
10
15
20
25and the centre of the spot is shown in the lower part of the gure.
-
20.2 4 0.2
pera
e re
T. Li et al. / NDT&E International 44 (2011) 216225 223-6 -4 -2 0 2 4 6-1.5
-1
-0.5
0
0.5
1
X / mm
First derivativeSecond derivative
X / mmY
/ m
m-6 -4 -2 0
-6
-4
-2
0
2
4
6
Fig. 11. (a) Normalized rst (blue line) and second (red line) derivatives of the tem(b) Second derivative image in x direction of Fig. 10 image. (For interpretation of th
this article.)
-6
-4 0.8
1better means of isolating the image of the crack. The contrast ofthe crack to the background heat ow (ratio of the maximumamplitude at crack positions to the maximum amplitude of thebackground heat ow) in Fig. 11b is about 3.21. In Fig. 11c, thecontrast becomes 9.79, which is 3.05 times higher than that inFig. 11b.
In practice, image acquisition time and noise need to beconsidered. The simulation shown in Fig. 10 corresponds to animage that might be collected by a thermal imaging camera, 0.15 safter the extinction of a laser heating pulse. In practice, morethermal images at different times can be used. This provides theopportunity to form a summed thermal image that will furtheremphasize the crack structure over the background heat ow.However, the second derivative method is known to be sensitive tonoise [17]; hence only thermal images with high signal to noise ratioshould be considered. The effects of image addition and noise wereinvestigated by adding 70.15 K (one standard deviation) randomnoise to the above model and summing rst and then secondderivative thermal images from 0.05 s after the extinction of a laserheating pulse to 0.3 s, when the remaining heating becamecomparable to the noise level. Images were computed at intervalsof 1/60 s matching the 60 Hz frame rate of the IR camera used in theexperimental work below. The normalized rst and secondderivative images are shown in Fig. 12a and b, respectively. Afterintegrating thermal images at different times, the contrast of thecrack to the background heat ow in Fig. 12a is 7.23. In Fig. 12b, the
X / mm
Y /
mm
-6 -4 -2 0 2 4 6
-2
0
2
4
6 0
0.2
0.4
0.6
Fig. 12. (a) Summed second derivative image in x direction from 0.05 to 0.34 60
X / mm-6 -4 -2 0 2 4 6
6 0
ture prole line in Fig. 10. (b). First derivative image in x direction of Fig. 10 image.
ferences to colour in this gure legend, the reader is referred to the web version of
-6
-4 0.8
1conaddin
Fig5 muncomsiz
50falprocam
1.
Y /
mm
-2
0
2
4
6
s. (c0.4
0.6
0.8
1
Y /
mm
-6
-4
-2
0
20.4
0.6
0.8
1trast is 27.76, which is 3.8 times higher than that in Fig. 12a. Inition, the integration method improves the contrast by 2.8 timesFig. 12b compared to Fig. 11c.The crack length indicated by the single spot images shown in. 10 is about 3.8 mm, which is somewhat shorter than the truem crack length. In practice the location of a crack might beknown and a raster scanning technique would be used. Abination of results obtained from such scans will improve the
ing of cracks, as shown in the experimental work below.A test-piece containing a crack was raster scanned using 21 W,ms laser heating pulses. The laser spot radius was 1 mm (1/el) and the IR camera eld of view was 26.420.8 mm2. Thecessing steps performed on the thermal images from the IRera are as follows:
Subtract the background IR image to produce a dark-eldimage showing only the transient heat diffusion (this can beeasily done by subtracting a thermal image obtained before thelaser was switched on), and then compute the secondderivative of each dark-eld image at each scanning step inboth x and y directions. Note here, the dark-eld image at eachposition can be the averaged image of several pre-imagestaken by the camera before the laser spot is switched onat each position on the specimen surface or any one of thepre-images.
X / mm-6 -4 -2 0 2 4 6
0
0.2
0.4
0.6
) Summed second derivative image in x direction from 0.05 to 0.3 s.
-
2. Integrate all second derivative images in x or y directioncollected from 0.05 to 0.3 s at each scanning position.
3. Form a composite image of all the integrated images obtainedat the different scanning positions. A nal sum image can beobtained by adding the squares of the two second derivativeimages in x and y directions to show the derivative informationin both x and y directions at the same time.
Fig. 13a shows a dye penetrant inspection ( DPI) image of an11 mm long, 3 mm deep crack with average opening of 24.5 mm ina test-piece (austenitic stainless steel) [18]. The images shown inFig. 13b and c were obtained using the above image processingmethods after scanning the test-piece. Fig. 13b shows the secondderivative image in x and Fig. 13c in the y direction. Since thecrack was orientated at an angle of approximately 451, signalamplitudes for the crack in Fig. 13b and c are similar. These crackimages are in good agreement with the DPI image of the crackshown in Fig. 13a. Information about crack shape and distributionof crack opening can also be seen in the image; it shows strongersignals in the middle part of the crack, where the crack is mostopen.
Raster scanning steps of 0.43 mm in the x direction and0.5 mm in the y direction were used to obtain the images shownin Fig. 13. Investigations were made on the images produced bycoarser scanning rasters. The images obtained after increasingscanning steps to 2.58 and 3.44 mm in the x direction are shownin Fig. 14. Most parts of the crack can still be seen using these
coarser scanning rasters. This indicates that the crack imagingprocess may be speeded up by using coarser scanning rasters butit should be recognized that size of the minimum detectable crackwill be an increasing function of raster step length.
An image of a smaller, 8 mm long crack in an Udimet test-piece with an opening of less than 1 mm is shown in Fig. 15. Thisprovides evidence of the remarkable sensitivity of the technique.
Y /
mm
0
5
10
2.5
mm
0 25
0
5
10
X / mm
Y /
mm
0 5 10 15 20 25
0
5
10
15
20
d x direction derivative image of the crack. (c) Summed second y direction derivative
vely.
Fig. 15. Summed second derivative image of an 8 mm long crack with openingless than 1 mm.
T. Li et al. / NDT&E International 44 (2011) 216225224Fig.
3.4X / mm
14. (a) Summed second derivative image in x direction with scanning step ofY /
mm
0 5 10 15 20 25
0
5
10
15
20
X / mm
Y /
0 5 10 15 2
15
20
Fig. 13. (a) Dye penetrant image of the crack in the test-piece. (b) Summed seconimage of the crack. Scanning steps 0.43 and 0.5 mm in x and y directions, respecti4 mm.X / mm0 5 10 15 20 25
15
20
8 mm. (b) Summed second derivative image in x direction with scanning step of
-
5. Discussion and conclusions
The modelling results presented in this paper indicate the laserspot thermography technique to have a sensitivity to cracks thatis competitive with most established NDE techniques. Thetechnique has the advantages of being non-contactive andrequiring no surface preparation. However, it is known that forthe technique to be successful, surfaces should be clean and freeof deep scratches or indentations that would perturb heat ow ina similar manner to a crack. To date, studies have been made onlyon surface breaking cracks. The ability of the technique to detectcracks beneath coatings is under investigation.
This paper also presents a novel second derivative imageprocessing method for extracting images of cracks after rasterscanning. The results obtained show scanned pulse laser spotthermography, incorporating second derivative image processing,to be a new effective nondestructive evaluation technique fordetecting and imaging surface breaking cracks with openings inthe range of micrometres. Crack images obtained by the newtechnique are at least comparable to those obtained by the longestablished DPI method. In addition, this new technique has theadvantages of eliminating the long preparation time of the dyepenetrant technique, eliminating the use of undesirable liquids,being deployable remotely and being suitable for automation. Thepractical sensitivity and reliability of the technique are under
References
[1] Reynolds WN. Thermographic methods applied to industrial materials. Can JPhys 1986;64:11504.
[2] Milne JM, ReynoldsWN. The non-destructive evaluation of composites and othermaterials by thermal pulse video thermography. Proc SPIE 1984;520:11922.
[3] Lau SK, Almond DP, Milne JM. A quantitative analysis of pulsed videothermography. NDT&E Int 1991;24:195202.
[4] Shepard SM, Lhota JR, Ahmed T. Flash thermography contrast model based onIR camera noise characteristics. Nondestr Test Eval 2007;22:11326.
[5] Sham FC, Chen N, Hong L. Surface crack detection by ash thermography onconcrete surface. Insight 2008;50(5):2403.
[6] Rashed A, Almond DP, Rees DAS, Burrows SE, Dixon S. Crack detection by laserspot imaging thermography. Rev Prog QNDE 2006;26:5006.
[7] Burrows SE, Rashed A, Almond DP, Dixon S. Combined laser spot imagingthermography and ultrasonic measurements for crack detection. NondestrTest Eval 2007;22:21727.
[8] Bantel T, Bowman D, Halase J, Kenue S, Krisher R, Sippel T. Automatedinfrared inspection of jet engine turbine blades. Proc SPIE 1985;581:1823.
[9] Devitt J.W., Bantel T.E., Sparks J.M., Kania J.S. Apparatus and method fordetecting fatigue cracks using infrared thermography. US Patent 5111048, 1992.
[10] Gruss C, Balageas D. Theoretical and experimental applications of the yingspot camera. In: . 1992; Proceedings of the 27th Eurotherm seminar onquantitative infrared thermography, QIRT. p. 1924.
[11] Hermosilla-Lara S, Joubert PY, Placko D, Lepoutre F, PiriouM. Enhancement ofopen-cracks detection using a principal component analysis/wavelet techni-que in photothermal nondestructive testing. In: Proceedings of thequantitative infrared thermography seminar, QIRT, 2002. p. 416 [ArchivesQIRT 2002-002].
[12] Hermosilla-Lara S, Joubert PY, Placko D. Identication of physical effects inying spot photothermal non-destructive testing. Eur Phys J Appl Phys2003;24:2239.
[13] Krapez JC, Gruss C, Huttner R, Lepoutre F, Legrandjacques L. La cameraphotothermiquepartie I: principe, modelisation, application a la detectionde ssures. Instrum Mes Metrol 2001;1:939. [in French].
[14] Krapez JC, Lepoutre F, Huttner R, Gruss C, Legrandjacques L, Piriou M, et al.
T. Li et al. / NDT&E International 44 (2011) 216225 225This research was funded as a targeted research project of theEngineering and Physical Science Research Council (EPSRC), UKResearch Centre in NDE (RCNDE). The work also received supportfrom Rolls Royce plc, RWE Npower and the National NuclearLaboratory. The authors are grateful to Ben Weekes of ourlaboratory at the University of Bath for the image shown in Fig. 15.damelioration par un nouveau traitement dimage. Instrum Mes Metrol2001;1:4167. [in French].
[15] Doyle PA. On epicentral waveforms for laser-generated ultrasound. J Phys D1986;19(9):161323.
[16] Tang Y, Yonezu A, Ogasawara N, Chiba N, Chen X. On radial crack and half-penny crack induced by Vickers indentation. Proc R Soc A 2008;464:296784.
[17] Gonzalez RC, Woods RE. Digital image processing. 3rd ed.. New Jersey:Prentice Hall; 2006. p. 7025.
[18] Trueaw Ltd., Espoo, Fl-02330, Finland.Acknowledgements La camera photothermiquepartie II: applications industrielles, perspectivesinvestigation.