Liftoff Correction Based on the Spatial Spectral Behavior of Eddy-Current Images

1362 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO. 5, MAY 2010

Liftoff Correction Based on the Spatial SpectralBehavior of Eddy-Current Images

A. Lopes Ribeiro, Member, IEEE, Francisco Alegria, Octavian A. Postolache, Senior Member, IEEE, andHelena Maria Geirinhas Ramos, Senior Member, IEEE

Abstract—In this paper, a new processing scheme for compen-sating the liftoff effect is proposed to increase the signal-to-noiseratio of measurements carried out when a duralumin plate withartificially manufactured defects was scanned with eddy-currentprobes. The mathematical algorithm is based on the spatial fre-quency behavior of the output signals with the liftoff distance (thedistance between the probe used for inspection and the sample)and its equivalent form (deconvolution) in the spatial domain. Re-sults are presented for measurements performed at three differentdistances for the same defect of the duralumin plate.

Index Terms—Eddy currents, giant magnetoresistance, liftoffeffect, nondestructive testing, spectral analysis.

I. INTRODUCTION

ONE METHOD for determining the volume distribution ofmaterial conductivity consists of inducing eddy currents

inside the conductor under test and measuring the magneticfield produced by those currents. This method has differ-ent technological applications. Nondestructive testing [1], [2],thickness measurement [3], surface treatment inspection ofconductive materials, material property characterization [4],and biomedical induction tomography [5] can be mentioned.Single-frequency, multiple-frequency, and pulsed eddy-currentmeasurement systems [6]–[8] are reported as robust and accu-rate solutions with wide application in nondestructive inspec-tion of metallic structures of aircrafts.

In parallel with the experimental work, a large number ofarticles on numerical methods have been published, whereas,recently, the modeling tools for dealing with eddy currents havegreatly been enhanced [9], [10].

By applying eddy currents to determine the nature of thebreak of homogeneity inside a conductive material, due tothe presence of material defects, we intend to answer threequestions: Is there a defect present in a given area? What isthe location of the defect? What is the geometry of the defect?

Manuscript received June 30, 2009; revised December 12, 2009. Firstpublished March 18, 2010; current version published April 7, 2010. This workwas supported in part by Portuguese Science and Technology Foundationunder Project PTDC/EEA-ELC/67719/ 2006 and in part by the Instituto deTelecomunicações under Project CLASSE. The Associate Editor coordinatingthe review process for this paper was Dr. Mark Blodgett.

A. Lopes Ribeiro is with the Instituto de Telecomunicações, TechnicalUniversity of Lisbon, 1049-001 Lisbon, Portugal (e-mail: [email protected]).

F. Alegria and O. A. Postolache are with the Instituto Superior Técnico andthe Instituto de Telecomunicações, Technical University of Lisbon, 1049-001Lisbon, Portugal.

H. M. G. Ramos is with the Instituto Superior Técnico, Technical Universityof Lisbon, 1049-001 Lisbon, Portugal.

Color versions of one or more of the figures in this paper are available onlineat http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TIM.2010.2043981

These problems have been addressed by many authors, anddifferent techniques have been used to solve them.

The methods used can be divided according to the time-varying fields that are used for excitation or according to thetype of probe used. The most used waveforms to generatethe excitation field are of sinusoidal type. With this kind ofexcitation, it is possible to obtain sets of images mapping theamplitude or phase variations along the material under test. Itis still possible to change the operation frequency to control thefield penetration depth into the material to get some informationabout the material levels at different depths. Another type ofexcitation uses pulsed fields that try to get information fromtheir richer frequency spectra.

As far as the probes are concerned, they are classifiedaccording to the type of detection element. They are eitherinductive or use high-sensitivity detection components, suchas magnetoresistances [11] or fluxgates [12]. The inductive-type probes must be used with excitation frequencies that arehigh enough to produce measurable voltages in the sensingcoils. When high-sensitivity components are used, the detectingsensors must have high sensitivity and are oriented to sensethe field in directions that are perpendicular to the excitationto avoid saturation.

One of the problems that affect all kinds of probes is theliftoff effect [13], which originates from the nonconstant gapbetween the probe and the metallic specimen under test. Thispaper correlates the measurements taken at different distances.A processing scheme for compensating the liftoff effect isproposed based on the effect of liftoff distance in the Fourierdomain and its deconvolution in the spatial domain.

II. EXPERIMENTAL SETUP

The block diagram of our measuring system is shownin Figs. 1 and 2 shows the experimental setup. It includesone eddy-current probe mounted on an xy-positioning systemthat scans a duralumin alloy plate. The scanned area can be(40 × 35) cm2, and the permissible speed of each single axismodule is 810 mm/s.

The magnetic fields of the induced currents are measuredusing a multifunction data acquisition board NI PXI-6120 in-cluded in a PXI-1036 6-slot chassis. The xy-positioning systemis controlled via an RS232 interface. This motion system andthe data acquisition are under the control of a LabVIEW pro-gram installed in a personal computer [14]. The excitation cur-rent is also measured by the data acquisition board using a 10-Ωsampling resistor. This board has four 16-bit simultaneously

0018-9456/$26.00 © 2010 IEEE

LOPES RIBEIRO et al.: LIFTOFF CORRECTION BASED ON THE SPATIAL SPECTRAL BEHAVIOR OF EDDY-CURRENT IMAGES 1363

Fig. 1. Block diagram of the experimental setup.

Fig. 2. Measuring system.

TABLE ISPECIFICATIONS OF THE INDUCTIVE PROBE

sampled analog input channels and eight input ranges from±200 mV to ±43 V. The maximum sampling rate is 1 MS/sper channel. The PXI system also includes an arbitrary functiongenerator NI PXI-5406, followed by a transadmittance powergenerator that delivers the necessary current to the excitationcoil [14], [15].

The specifications of the inductive probe are summarizedin Table I. The three coils are coaxial. The output voltage ofeach sensing coil is amplified with an instrumentation amplifierINA-118 using an adjustable gain. The differential voltageis then amplified with another INA-118 and fed to the dataacquisition board NI PXI-6120.

Before taking measurements with the inductive probe, apreliminary calibration process is carried out. It consists of theequalization of the voltages that originated from the two sensingcoils by changing the gains on the two primary instrumentationamplifiers. This process must be accomplished far from the

TABLE IISPECIFICATIONS OF THE GMR PROBE

Fig. 3. AA002 GMR sensor experimental characteristics for a dc supplyvoltage of 3 V.

metallic plate, because the field arising from the eddy currentsinfluences mainly the lower sensing coil and is not in phase withthe primary excitation field.

The giant magnetoresistance (GMR) probe also designed atour laboratory is summarized in Table II.

The sensor includes four 5-kΩ GMRs configured as a Wheat-stone bridge. Two of these resistors are sensing elements, andthe other two, which are shielded from the applied magneticfield, act as reference elements. The imbalance leads to thebridge output that presents a sensitivity of 3.7 mV/Oe for eachvolt supplied by the bridge power supply. The transfer char-acteristic Vout = Vout (Ha) (Vout is the GMR sensor output,and Ha is the applied magnetic field) is shown in Fig. 3. Thesaturation field of the sensor is 15 Oe ≈ 1200 A/m.

To put the sensor to work in its linear range, biasing with apermanent magnet is carried out. The sensor is mounted to putits plane parallel to the aluminum plate under inspection. Thus,the excitation magnetic field on the coil axis, being perpendic-ular to the sensing axis of the GMR sensor package, shouldhave no effect on the sensor. The magnetic fields of the inducedcurrents are measured using the multifunction data acquisitionboard NI PXI-6120 included in a PXI-1036 6-slot chassis.

III. DETECTION AND LOCALIZATION OF A LINEAR CRACK

A duralumin 2024 T3 plate with 2-mm thickness and a longcrack at the center with dimensions of 20-mm length, 0.5-mmwidth, and 1-mm depth was examined. The scanned area was a30 × 20 mm rectangle. The scan resolution was 1 mm, and theresults were stored in 30 × 20 matrices.

The probe excitation current was set to 100 mA at differentfrequencies and distances to the plate. Figs. 4, 5, and 7 show


Fig. 4. Amplitude of the output voltage from the eddy-current sensor placed1 mm from the plate.


the raw data obtained for the output voltage with the GMR’ssensing axis perpendicular to the direction of the crack whenthe frequency was set to 5 kHz at different distances betweenthe probe and the plate, i.e., 1, 2, and 5 mm, respectively.

From these figures, one can conclude that the voltage am-plitudes are a function of the distance between the probe andthe conducting plate. Their values are higher when the sensoris placed near the sample. For the examples depicted, the peaksmeasured with reference to the base plane (defect-free zones)are 58 mV when the probe is 1 mm from the plate and decreasesto 17 or 2 mV when the probe stands at a distance of 2 or5 mm. The asymmetry observed on the peak values is due tomisalignments of the sensor. The situation presented in Fig. 5is shown in Fig. 6 as a contour plot with the exact location ofthe artificially manufactured crack. From this map, it is clearthat the peaks are located on each size of the crack and can beused to conclude about its geometry. Another effect that is also

Fig. 6. Output from the eddy-current sensor scanning the crack at 5 kHz withat a distance of 2 mm from the plate.


evident for the three cases shown is that the variation is sharperwhen the probe is placed very close to the duralumin plate andbecomes smoother and more spread over the scanned area whenthe probe is removed from the plate. The measurements shownin Fig. 7 were made for the same scanned area as in the casesin Figs. 4 and 5, but the scanned area, in this case, is not largeenough to attain the zone where no perturbation is detected.

These results denote that the nonzero liftoff acts as a low-passfilter on the surface crack signal.

IV. SIGNAL ATTENUATION WITH DISTANCE

The importance of using two different kinds of probes isjustified, because they are complementary. The inductive probesenses the eddy-current field component parallel to the ex-citation field, and the GMR probe senses a field componentperpendicular to the excitation field.

An experimental investigation of the signal attenuation withdistance was carried out for both probes. In the case of the


Fig. 8. Peak voltage extracted from results obtained at different distances fora GMR-based probe.

GMR-based probe, the crack shown in Fig. 6 was scannedat seven different distances from the plate d = 0 to 6 mm.The peak voltage from each resulting image was graphicallyrepresented as a function of distance, as shown in Fig. 8.

The measured values show an exponential tendency, whichcorresponds, for the taken experimental conditions, to an atten-uation of 8.69 dB/mm. It is important to note that the exponen-tial curve does not vanish for large distances. This effect is dueto a small part of the excitation field that is sensed by the GMR.With this background field being difficult to reduce by a realign-ment of the GMR, a correction has to be made by software.

The same experimental investigation of the signal was car-ried out for the inductive probe as well [16]. To estimate the fil-ter function associated to the liftoff effect, measurements wereperformed over the duralumin alloy plate on a defect-free area.The voltage amplitudes on the sensing coils are a function of thedistance between the probe and the conducting plate. It is notpossible to calibrate the probe in the presence of the plate, be-cause the two voltages in the two sensing coils are not in phase.Thus, the calibration procedure takes place with no plate, withthe two voltages being equalized. This means that, when themeasurements are made in the presence of the plate, in a zonewith no defects, a nonzero background voltage must be sensed.This background voltage decays toward zero at large distancesbetween the probe and the plate. Fig. 9 shows the normalizedvoltage variation obtained with the inductive probe, as a func-tion of distance between probe and plate. We see that an expo-nential decay fits well in the acquired values. Thus, we concludethat, for the conditions of this experiment (type of plate andfrequency), the background voltage drop is about 3.5 dB/mm.

The signal part that conveys information about the defects ismeasured relative to that background. Thus, a small variation onthe plate-to-probe distance produces a large variation in relationto the useful part of the signal. Fig. 10 shows that, with a back-ground signal of about 210 mV, we get a useful signal (maxi-mum deviation to the background voltage) of about 30 mV.

Another effect related to the distance between the plate andthe probe has to do with the spatial frequency spectrum of the2-D images. It is known, from the magnetic recording theory,

Fig. 9. Background voltage as a function of probe-to-plate distance.

Fig. 10. Resulting curves for three runs with the inductive probe over one8-mm hole in an aluminum plate. Lower curves correspond to increaseddistance.

that increasing distances d between the magnetic reading headsand the magnetized media negatively affects the discriminationof the smallest recorded wavelengths [17]. The transfer charac-teristic H(k) in the spatial frequency domain considering onlythe signal pick-up process could be represented as

H(k) = H0(d) exp (−|k|d) . (1)

Thus, the attenuation with distance has the effect of a low-pass filter, clearing the fine details in the images. It is possible todraw some conclusions about the signal-to-noise ratios (SNRs)adequate to obtain good images.

To test the attenuation law expressed by (1), we did thefollowing experiment: The same plate used to obtain the resultshown in Fig. 9 was perforated by a hole with a diameter of8 mm. The plate was scanned (for one single run), using theinductive probe, for three different distances between the probeand the plate, i.e., d = 0, d = 1, and d = 2 mm. Fig. 10 showsthe resulting three curves. The characteristic form of thesecurves is a depression with an elevation at the middle, whenthe probe is right over the hole, and the eddy currents fit wellon its circular geometry.


Fig. 11. Curves represented in Fig. 10, with the background levels subtracted.

Fig. 12. Amplitude spectra relative to the data in Fig. 11.

The voltage amplitudes obtained on regions far from the holeare compatible with the distances and can be used to estimatethat distance.

The three curves were overlapped, subtracting the back-ground levels from each one of those curves (Fig. 11). Thepurpose of this action is to separate the signal part related tothe hole to extract the respective spectra.

To determine the spectra of the previous data, we used afast Fourier transform. The three spectra are represented as afunction of the spatial wavenumber k = 2π/λ, with λ beingthe spatial period associated to each spectral component. Theamplitude spectra are shown in Fig. 12.

On the next step, we try to convert the spectra of two signalsinto the spectrum of the third signal using a transformation ofthe type (1). The following equation was used:

H01(k) = 0.22 exp(−|k| × 1.0−3

)

H02(k) = 0.44 exp(−|k| × 2.0−3

)(2)

where H01 transforms the signal at d = 0 into the signal takenat d = 1, and H02 has an equivalent meaning. Fig. 13 shows theresult after the application of the fast inverse Fourier transform.

The analysis of Fig. 13 shows that the two reconstructedimages appear spatially contracted. This means that (2) did not

Fig. 13. Reconstructed signal images after the application of the attenuationtransformations (2).

Fig. 14. Transposition of curves taken at d = 0 and d = 1 mm to the curvetaken at d = 2 mm.

take into account the image spreading with distance. It wasverified that, for the experimental conditions under test, therewas a 12.5% spatial compression that is compatible with aspectral expansion of the same amount. The image in Fig. 13was corrected, and the final result was shown in Fig. 14.

In the previous example, the data acquired at smaller dis-tances were transformed to fit the data acquired at a largerdistance. The inverse transformation would be more interesting,but it amplifies the high-frequency noise. Thus, the denoisingof the input signals is absolutely necessary. The design of therequired filters is very complicated, because its performancedepends on the possibility of correct defect classification, using,e.g., neural networks.

V. CONCLUSION

In this paper, we have presented the results obtained when aduralumin plate with a linear crack at the center was scanned.The attenuation effect with distance has been studied, and


measurements have been made to obtain its variation as afunction of distance. An exponential amplitude decrease withdistance has been measured for a GMR-based probe and for aninductive probe. It is known that the distance between the probeand the plate acts like a low-pass filter. Thus, the spatial spectralcharacteristic of such filter has been investigated.

The obtained results show that the reverse deconvolutionof those filtering effects is possible, i.e., the data measuredfor greater distances can be transposed to closer distances.However, this transposition is dependent on the quality of theacquired signals, which is expressed by a high SNR. Thisfact implies that denoising filters must be designed and tested.Special attention must be taken to the probe signal preprocess-ing to guarantee good SNRs. The final performance of thewhole system depends on those issues that are measured bythe capability of correct characterization of the defects insidethe metallic material.

REFERENCES

[1] N. V. Nair, R. V. Melapudi, H. Jimenez, X. Liu, Y. Deng, Z. Zeng,and L. Udpa, “A GMR-based eddy current system for NDE of aircraftstructures,” IEEE Trans. Instrum. Meas., vol. 42, no. 10, pp. 3312–3314,Oct. 2006.

[2] A. Bernieri, L. Ferrigno, M. Laracca, and M. Molinara, “Crack shapereconstruction in eddy current testing using machine learning systems forregression,” IEEE Trans. Instrum. Meas., vol. 57, no. 9, pp. 1958–1968,Sep. 2008.

[3] J. Moulder, C. E. Uzal, and J. H. Rose, “Thickness and conductivityof metallic layers from eddy current measurements,” Rev. Sci. Instrum.,vol. 63, no. 6, pp. 3455–3465, Jun. 1992.

[4] S. Konoplyuk, T. Abe, T. Uchimoto, T. Takagi, and M. Kurosawa, “Char-acterization of ductile cast iron by eddy current method,” NDT, E Int.,vol. 38, no. 8, pp. 623–626, Dec. 2005.

[5] N. B. Brás, A. Pestana, R. C. Martins, and A. C. Serra, “Dimensioningand characterization of a new magnetic induction tomography experi-mental setup,” in Proc. IEEE I2MTC, Victoria, BC, Canada, May 2008,pp. 1803–1808.

[6] O. Postolache, M. D. Pereira, H. G. Ramos, and A. L. Ribeiro, “NDTon aluminum aircraft plates based on eddy current sensing and imageprocessing,” in Proc. IEEE I2MTC, Victoria, BC, Canada, May 2008,vol. I, pp. 1803–1808.

[7] V. Sundararaghavana, K. Balasubramaniama, N. Ramesh Babub, andN. Rajeshb, “A multi-frequency eddy current inversion method forcharacterizing conductivity gradients on water jet peened components,”NDT, E Int., vol. 38, no. 7, pp. 541–547, Oct. 2005.

[8] A. Sophian, G. Tian, D. Taylor, and J. Rudlin, “Pulsed eddy current ‘De-sign of a pulsed eddy current sensor for detection of defects in aircraft lap-joint’,” Sens. Actuators A, Phys., vol. 101, no. 1/2, pp. 92–98, Sep. 2002.

[9] N. Ida, K. Betzold, and W. Lord, “Finite element modeling of absoluteeddy current probe signals,” J. Nondestruct. Eval., vol. 3, no. 3, pp. 147–154, Sep. 1982.

[10] M. Tanaka and H. Tsuboi, “Finite element model of natural crack in eddycurrent testing problem,” IEEE Trans. Magn., vol. 37, no. 5, pp. 3125–3128, Sep. 2001.

[11] T. Dogaru and S. T. Smith, “Giant magnetoresistance-based eddy-currentsensor,” IEEE Trans. Magn., vol. 37, no. 5, pp. 3831–3838, Sep. 2001.

[12] M. V. Kreutzbruck, K. Allweins, and Cheiden, “Fluxgate-magnetometerfor the detection of deep lying defects,” in Proc. 15th ECNDT, 2000.[Online]. Available: http://www.ndt.net/article/wcndt00/papers/idn291/idn291.htm

[13] B. Abu-Nabah and P. Nagy, “Lift-off effect in high-frequency eddy currentconductivity spectroscopy,” NDT, E Int., vol. 40, no. 8, pp. 555–565,Dec. 2007.

[14] H. Geirinhas Ramos, A. Lopes Ribeiro, F. Alegria, and O. Postolache,“Virtual instrument to detect defects in conductive materials,” in Proc.7th Conf. Telecommun., Santa Maria da Feira, Portugal, May 3–5, 2009,pp. 137–140.

[15] H. M. Geirinhas Ramos, O. Postolache, F. Corrêa Alegria, and A. LopesRibeiro, “Using the skin effect to estimate cracks depths in metallic struc-tures,” in Proc. IEEE I2MTC, Singapore, May 5–7, 2009, pp. 1361–1366.

[16] A. Lopes Ribeiro, F. Alegria, O. Postolache, and H. Geirinhas Ramos,“Eddy current inspection of a duralumin plate,” in Proc. IEEE I2MTC,Singapore, May 5–7, 2009, pp. 1367–1371.

[17] C. D. Mee and E. D. Daniel, Magnetic Recording Technology.New York: McGraw-Hill, 1996.

A. Lopes Ribeiro (M’90) was born in Lisbon,Portugal, on April 8, 1950. He received the Diplomadegree in electrical engineering and the Ph.D. de-gree in electrical and computer engineering from theInstituto Superior Técnico (IST), Technical Univer-sity of Lisbon (UTL), Lisbon, in 1973 and 1990,respectively.

In 1977, he joined the Department of Electri-cal and Computer Engineering, IST/UTL, where hehas been a Member of Teaching Staff. In 1991, hejoined the Instituto de Telecomunicações, UTL. His

research interests include the instrumentation and electric measurement andnumerical modeling of electrical and optoelectronic components.

Dr. Lopes Ribeiro is a member of the International Compumag Society.

Francisco Alegria was born in Lisbon, Portugal,on July 2, 1972. He received the Diploma, M.S.,and Ph.D. degrees in electrical engineering and com-puters from the Instituto Superior Técnico (IST),Technical University of Lisbon, Lisbon, in 1995,1997, and 2002, respectively.

Since 1997, he has been a Member of Teachingand Research Staff with the IST. Since 1994, hehas been a member of the instrumentation and mea-surement research line with the Instituto de Tele-comunicações, Technical University of Lisbon. His

current research interests include analog-to-digital-converter characterizationtechniques, automatic measurement systems, sensors, and signal processing.

Octavian A. Postolache (M’00–SM’06) was bornin Piatra Neamt, Romania, on July 29, 1967. Hereceived the Ph.D. degree in electrical engineeringfrom “Gh. Asachi” Technical University of Iasi, Iasi,Romania, in 1999.

In 1992, he joined the Faculty of Electrical Engi-neering, Department of Electrical Measurements andElectrical Materials, Technical University of Iasi,where he worked for nine years as a Lecturer and anAssistant Professor. In 2000, he became a Ph.D. Re-searcher with the Instituto Superior Técnico and In-

stituto de Telecomunicações, Technical University of Lisbon, Lisbon, Portugal,where he has been involved in different projects in the area of instrumentation.His research interests include smart sensors for environmental and biomedicalapplications, sensor and algorithm implementation for nondestructive testing,distributed instrumentation, sensor networks, and computational intelligenceimplementation in automated measurement systems.

Helena Maria Geirinhas Ramos (SM’05) was bornin Lisbon, Portugal, in October 1957. She receivedthe Eng. degree in 1980 and the M.Sc., Ph.D., andAggregation degrees from Instituto Superior Técnico(IST), Technical University of Lisbon (UTL), in1987, 1995, and 2006, respectively, all in electricaland computer engineering.

Since 1980, he has been an Associate Professorwith IST, where she is a Member of Teaching Staff.She is also a member of the Instrumentation andMeasurement Group, Instituto de Telecomunicações.

Her research interests include nondestructive testing of conductive materials,personal-computer-based instrumentation, measurement interfaces, analog-to-digital converter testing, transducers, and ferromagnetic materials modeling.She was and still is involved in many research national and internationalprojects in these areas.

Liftoff Correction Based on the Spatial Spectral Behavior of Eddy-Current Images

Documents

Transcript of Liftoff Correction Based on the Spatial Spectral Behavior of Eddy-Current Images