REAL TIME INFRA-RED IMAGE PROCESSING FOR THE DETECTION OF DELAMINATION IN COMPOSITE PLATES

L.GUILLAUMAT*, J.C. BATSALE** and D. MOURAND***

*LAMEFIP-ENSAM **LEPT-ENSAM UMR CNRS 8508 ***Cellule „Themicar“ of LEPT-ENSAM Esplanade des Arts et Métiers 33405 Talence cedex-France

E-mail : [email protected]

Thermal Non Destructive Evaluation principle

Heat excitation (halogen lamp) Heterogeneous composite sample IR camera Rear face observation Analysis of the transient images of

temperature responses

recorded with the camera and estimation of

thermophysical parameters cartographies.

Main features of such thermal NDE • Advantages

-No contact -Sensitive to delamination (layer of poor thermal conductivity)

• But

-such methods are time consuming and expensive -Infrared Cameras are noisy -the image processing is heavy

• It is here proposed to:

-consider the new generation of infrared cameras -consider suitable image processing methods

New Infrared cameras Some low cost devices

Raytheon Palm IR 250, Indigo alpha, Boeing U3000

Some high performance devices

CEDIP Jade III, FLIR, AMBER etc…

Thermal NDE- Illustration with the flash method

Metrology Laboratory

method

Imaging Industrial

method Sample

1 T(t) measurement

0.1

T (°C) Thermocouple

>10 000 pixels T(t) measurement

2.5

T (°C) Sample IRcamera 0 10 20 30 t (s) 40 0 10 20 30 t (s) 40 • • •

1 T(t) measurement measurement with contact very accurate measurement

• • •

>10000 T(t) measurements measurements without contact very noisy measurements

0.5

0 -0.5

0 1 1.5

Can we discern two very noisy thermograms ?

Temperature level 20 40 Time (s)

DATA PROCESSING STRATEGY Signal proportional to the temperature Reduction of the measurement noise influence Low excitations Linear transform Great amount of data Reduction of the amount of data Knowledge of the transfer model Parameters estimation

Assumptions about the measurement noise

explicative variable

T

=

f

(

t

, b 1 , b 2 , b 3 ...)

Finite number of parameters measure (random variable)

=

T

+

e T

real value ” measurement error” (ramdom variable) estimator (random variable)

b i = b i +

e

b i

real value estimation error (random variable)

Linear least squares (Maximum likelywhood theorem)

T = X

b

Hypothesis : -zero mean and additive errors

-b

constant and unknown before the estimation and X ij error known without -constant variance (

s

^

b

optimum known) and uncorrelated errors ^

Estimator Estimation error

b

^

=

(

X

t

X

) -1

X

t

^

T

cov(

e

b )

=

(

X

t

X

) -1 s 2

T

=

Linear processing of data

B

Estimation of a reduced number of parameters

X

* About 20 Mbytes in 20s Sensitivity

Advantages:

-Reduction of the amount of data -Decreasing of the noise influence -Possibility of sequential processing without memory storage

But: -How to do the determination of the sensitivity matrix X ???

-What kind of linear transform (X t *T) can be chosen?

ESTIMATION AND LOCALLY 1D TRANSFER

Expression of the temperature response from a Flash experiment

T



z

 0 ,

t

) 

Q



cL

   1  2 

n

   1 exp   -

n

²  

L

 ² 

a



t

2     

Q



cL f



at

/

L

2 ) The delamination in a composite material act as a small thermal conductivity variation on the temperature response of each pixel.

Two kinds of asymptotic expansions can be considered:

T

(

L

,

t i

) 

Q



cL

   

f

  0

t i

/ 

cL

2 )     0      0 

f

 (  0

t i

/ 

cL

2 )         or

T



X

  b b 1 2   In this case

X

or the sensitivity vectors are calculated theoretically with the knowledge of nominal parameters.

T

( 0 ,

t i

) 

Q



cL f

  0

t i

/ 

cL

2 ) 

Q



cL

  0 

t



f



t

or

T



X

  b b 1 2   (  0

t i

/ 

cL

2 ) In this case signal

X

or the sensitivity vectors are calculated with a reference

f(t). Such reference signal can be obtained here with the spatial

average of the images of temperature.

Experimental Thermogram with Raytheon Palm IR 250

10 0 -10 -20 Temperature Level 50 40 30 20 0 10

One pixel thermogram Average of the image thermogram

20 Time (s) 30 40 50

Application of this method with a CEDIP camera to the study of a delaminated sample

Video image of a 5mm thick transparent delaminated fiber-glass–epoxy plate Delamination cartography estimation by thermal method of the previous plate.

3D representation of the delamination (equivalent air thickness)

Relation between thermophysical properties and structural properties:

e air

    2 

air e

It can be noted that in the centre of the damaged zone induced by impact exists an undamaged area observed also by de-ply technique.

Conclusion

• The main features of such NDE method are: • The experiment is simple and contactless. • The processing consists in computing weighted sums of the images or of the pixels. This can be done in real time (20s).

• The method can be implemented with low cost or high performance cameras • The main points for the study of damaged samples : • Such a device provides a 3D representation of the delamination in good agreement with physical destructive observations.

• Some future works will consist in observing the evolution of the delamination structure during fatigue experiments.

Bibliography about similar image processing methods (based on linear transform of the data and some physical knowledge about the heat transfer) •

Time Fourier Transform-Periodic excitation

D. Wu, C. Y. Wu, G. Busse

, Septembre 1996.

Investigation of resolution in lock-in thermography: Theory and experiment,

Eurotherm Quantitaive Infrared Thermography QIRT’96,Stuttgart 2-5 •

Flash method and asymptotic expansions estimations methods

Mourand D., Batsale J.C.

:(2000) Real time processing with low cost uncooled plane array IR camera-Application to flash non-destructive evaluation, QIRT 2000, Eurotherm seminar 64, Reims .

Mourand D., Batsale J.C., Gounot J.

:(1998) New sequential method to process noisy yemperature response from flash experiment measured by infrared camera, Review of Scientific Instrument, vol 69 n3, pp 1437-1441

Goetz C., Batsale JC, Mourand D

– Fast processing methods for thermal non-destructive evaluation of thin plates with low cost infared cameras. Image Analysis & Stereologie 20, (2) Suppl 1, 227-232, 2001.

•

Space Fourier transform

Philippi I., Batsale J.C., Maillet D. et Degiovanni A.

Instru., 66(1), pp182-192.

: (1995) Measurement of thermal diffusivity through processing of infrared images processing, Rev. Sci.

Krapez J.C

., 1999 Mesure de diffusivité longitudinale de plaques minces par méthode de grille-Journée SFT:”Thermographie IR quantitative” ONERA Mars 1999.

•

Homogenization:

Batsale J.C.., Gobbé C., and Quintard M

., 1996

, Local non-equilibrium heat transfer in porous media.

Recent Res. Devel. in Heat, Mass & Momentum Transfer 1.

Poncet E., Bereziat D., Grangeot G., Batsale J.C.

(1998) Experimental estimation of the heat exchange coefficient of a non-equilibrium model by infrared measurement temperature on a stratified system- 11 Int.Heat Transfer Conference Kyong Ju Korea.

Varenne M., Batsale J.C., Gobbé C., Varenne M., Batsale J.C., Gobbé C.,

(2000) Estimation of local thermophysical properties of a 1D periodic heterogeneous medium by infrared image processing and volume averaging method- Journal of Heat Transfer-ASME. February 2000, vol 122, pp21-26 (2000) Estimation of a local 1D or 2D thermal conductivity field with infrared images processing and volume averaging method, QIRT 2000, Eurotherm seminar 64, Reims .

2D thermal intercorrelation study:

Guillaumat L. Davy L. Bouquet J. Batsale JC

., (2003) A new thermal method for the crack detection in damaged composite plates-application of flash method and infrared thermography- Comp test 2003 communication (poster).