Transient Eddy-current NDE for Aging Aircraft Capabilities and Limitations.

Transcription

1 Transient Eddy-current NDE for Aging Aircraft Capabilities and Limitations. Mr R A Smith Structural Materials Centre Defence Evaluation and Research Agency Farnborough GU1 LX UK RASmith@dera.gov.uk Dr G R Hugo Defence Science and Technology Organisation Aeronautical and Maritime Research Laboratory GPO Box 331 Melbourne Victoria 31 AUSTRALIA Geoff.Hugo@dsto.defence.gov.au ABSTRACT Transient eddy-current NDE is currently the subject of widespread interest because of its key potential applications for detection of corrosion and cracks in aging aircraft fleets. There are significant benefits to be realised from the use of transient eddy-currents in terms of inspection time and ease of acquisition and analysis of the data. Large areas of structure incorporating multiple variations in thickness can be scanned without the need for any probe or setup changes. In addition, the use of a Hall sensor rather than a coil as a field detector improves the spatial resolution and the detectability of deep defects. As with any NDE technique, transient eddy-current NDE will not be a panacea and there is currently some uncertainty about the most appropriate applications for which it should be considered. For example, it would be wrong to expect the use of a transient system to necessarily improve absolute sensitivity when compared with conventional methods, since the signal-to-noise ratio in a wide-band pulsed system is rarely as high as in a narrow-band one. However, the use of wide-band time-domain signals obtained with transient eddy-currents can enhance the overall NDE capability for detection and classification of defects when compared with conventional eddy-current systems, provided that appropriate analysis methods are applied to the transient data. This paper presents an objective evaluation of the capabilities and limitations of transient eddycurrent NDE. Most of the work reported here is concentrated on detection and characterisation of corrosion but application to cracks is also discussed. Both current and potential capabilities are considered, differentiating between inherent limitations of the transient method and present limitations of either the analysis techniques or the signal acquisition instrumentation. INTRODUCTION A continuing requirement to detect and characterise both corrosion and fatigue cracks in aging aircraft structures has resulted in considerable interest in the capabilities of transient (or pulsed) eddy-current methods. Whereas conventional eddy-current instruments employ harmonic

2 continuous-wave excitation at one or more frequencies, transient eddy-current systems apply a broad-band pulse or step excitation to a coil to generate a magnetic field pulse. This magnetic field pulse propagates into the material generating transient eddy-currents which oppose the changing incident field. The net field, which is the sum of the incident field and the field reflected by the specimen due to the induced eddy-currents, can be detected by a coil (generally the same as the transmitting coil) or by a field sensor (ie a sensor that measures the magnetic field directly) such as a Hall-effect device or a giant magneto-resistive (GMR) sensor. The merits of these different sensors, in terms of signal-to-noise and spatial resolution as a function of depth, are referred to in this paper. Broad-band pulsed excitation allows the acquisition of a data set covering all of the frequencies required for the expected structural variations. The enhanced analysis capabilities that are possible with such a data set allow improved differentiation between defects and structural changes, and also between different defect types. This data set can be complete enough to remove the need for any setup changes or analysis at scan time. Transient (or pulsed ) eddy currents have been used for several years in non-destructive testing (NDT), primarily for the detection of small cracks emanating from fastener holes deep in the structure [1]. However, that implementation looks for asymmetry in the signal from around the fastener and requires centering of the probe over each fastener. The technique is therefore timeconsuming for use on the large numbers of fasteners on transport aircraft. Conventional continuous-wave eddy-current techniques are also able to detect cracks around fasteners using symmetry effects with similar centering requirements. For corrosion detection, past attention has focused on conventional eddy-currents to detect general thinning in multi-layered structures. In order to exclude structural effects such as plate separation it is necessary to use multiple frequencies and determine an empirical method for combining them. As the structure changes the required frequencies change, and so does the method of combination. Hence, for large area scanning, the need to frequently change settings means that these scans are also very time-consuming. Clearly the ease of data capture and analysis with transient eddy-current methods results in greatly reduced overheads associated with scanning large areas of structure using eddy-currents. However, there is still an urgent need for a clear presentation of the capabilities and limitations of transient eddy-currents in terms of various inspection scenarios. Potential users need to know where transient methods can be applied and when there will be an advantage over conventional methods. A key issue at present is to assess the capabilities of transient eddy-current methods for the detection and characterisation of corrosion: how much can be detected and at what depth? The other main requirement is for the detection of sub-surface cracks deep in the structure, often initiated from corrosion sites. If transients can detect cracks, then users need to know what size and at what depth. This paper is the result of a collaborative programme between the Defence Evaluation and Research Agency (DERA), Farnborough, UK and the Aeronautical and Maritime Research Laboratory (AMRL), Melbourne, Australia. Measurements were made using the transient eddycurrent scanning system (TRECSCAN ) developed at DERA Farnborough [2 ]. Area scans were performed using an ANDSCAN manual scanning arm [5] to digitise the probe position (see Figure 1). One of the features of the DERA and AMRL systems is that they use Hall-effect sensors to measure directly the magnetic field

3 Other transient eddy-current systems have been developed at Iowa State University in the USA [6,7] and by workers in Canada [8]. Both of these systems use a coil to sense the magnetic field and therefore measure the rate of change of field, rather than the field itself. This results in a sensitivity to defects which is related to frequency-squared for a coil, rather than to frequency for a field sensor, giving a relatively poorer sensitivity to deep defects which are detected preferentially by the lowest frequency components of the transient signal. In this paper a theoretical comparison is made between the capabilities of the Hall-effect device and a coil sensor. THE TRANSIENT EDDY-CURRENT METHOD The essential elements of the field generation and reception components of the TRECSCAN system are shown diagrammatically in Figure. In order to induce a broad-bandwidth pulse in the specimen, a coil is driven with an exponentially-damped square-wave drive current i(t). This produces a periodic reversal of the magnetic field which induces a pulse of eddy-currents that propagates down into the specimen each time. The response to this process at the receiver is dependent upon the original field that was propagated and on the structure into which it travels. As a magnetic field pulse propagates through the material it is attenuated and broadened. This attenuation is dependent upon both the temporal frequency and the spatial distribution of the field. A spatially narrow field will decay more quickly than a broader field. Hence the wideband transient field becomes both temporally and spatially broader with depth into the material as the higher temporal and spatial frequencies decay away first. This means that the spatial resolution inevitably decreases with depth. Although the coil can also be used as the field sensor, TRECSCAN uses a Hall-effect device. Hall sensors produce an output voltage that is proportional to the component of the magnetic field that is perpendicular to the device averaged over the area of the sensor. This remains true for a very wide frequency range (DC to 1 khz for the devices used in TRECSCAN ). The more commonly used coil sensor responds to the rate of change of magnetic field (see Figure 2) and therefore has a frequency-dependent response that reduces to zero at DC. This good response at low frequencies is the main reason for using Hall sensors because it is the low frequencies that penetrate deep into structures. A further advantage of a Hall device is its small size compared with the dimensional characteristics of the received field. This will always be better than the spatial resolution which would be obtained using a coil sensor which inevitably averages over the cross-sectional area of the coil. As the received field distribution from a defect becomes progressively broader with defect depth, both sensitivity and spatial defect resolution are reduced. However, for a defect at any given depth, the use of a point field sensor gives the best spatial resolution that is possible. In order to obtain better penetration into a structure the spatial characteristics of the field can be changed. A larger probe will produce a broader incident field distribution. Also, an air-cored coil will produce a different spatial profile to a ferrite-cored coil. A broader incident field should give better sensitivity to deep defects, but at the expense of poorer sensitivity to any shallower defects which are smaller than the probe size (this being an inherent limitation of all eddycurrent methods, both conventional and transient). Two ferrite-cored probes and a large aircored pancake coil were used for the experiments in this paper, as detailed in Table 1. The ferrite cores tend to enhance the narrower field characteristics while the air-cored pancake coil

4 produces a field with a wide range of spatial characteristics as there are some small turns on the coil and some large ones. These different probes are compared in the results shown in this paper. The relative capabilities of Hall devices and coils has been assessed using calculations based on the work of Dodd and Deeds [9] to compare the magnitude of the defect signals from each.. Figure 3 shows a theoretical comparison for both absolute voltages and relative sensitivities between a Hall sensor and a coil sensor for a range of coil dimensions similar to those used in this paper. The type of defect used for this theoretical calculation was a thin layer of air over a large region at a given depth in a thick metal specimen. The coils appear initially to give higher absolute voltages, suggesting better sensitivity. However, this is misleading as the defect signals are measured as perturbations on a carrier signal which has a much larger amplitude for the coil sensor than for the Hall devices. Hence, when the practical sensitivity is computed, by dividing the defect voltage by the amplitude of the carrier signal on which it is superimposed, the practical sensitivity for the coil is considerably less than for a Hall device. The reduction in sensitivity of coils compared with Hall devices at increasing defect depth can also be observed in Figure 3. In the TRECSCAN system, the Hall effect device is oriented to measure the component of magnetic field perpendicular to the surface, H z (t) (see Figure ). Typical signals measured using the Hall effect device are illustrated in Figure 5. With no specimen in front of the probe, the measured field H z (air) is simply proportional to the drive current i(t). The field H z (specimen), measured when the probe is on a metallic specimen, has a much longer rise time due to the eddy currents induced within the specimen which oppose the change in incident field. In order to view the effect of the specimen, the transient field reflected by the specimen H R z (ie. the field due to the induced eddy-currents within the specimen) can be calculated as the difference between the H z (specimen) and H z (air). This is the type of transient that is normally viewed by the operator, and is similar to an A-scan in ultrasonic NDT. Rather than using the H z (air) signal as the reference, it is normal to balance the signal on a good part of the structure in order to view variations in the transient response relative to this good structure. The balancing operation can be performed at acquisition time, or when analysing the scan, and involves pressing a button in the software having identified a good region of structure. Balancing generates a relative signal H z (t) that is nominally zero unless the structure or its properties change. Hence a defect will be seen as a perturbation of the transient from this zero response. This balancing operation can be performed at any time during acquisition or subsequent analysis of the data as it does not affect the raw transient data H z (t) that is stored for subsequent off-line analysis. As the transient response is a well-behaved and slowly varying curve it is not necessary to store all the digitised points in the way that an ultrasonic waveform would be stored. Instead, the values at a relatively few selected points in time (typically twelve time points) can be selected for storage as a function of probe position. The TRECSCAN system stores the transient response at an exponentially-distributed sequence of time points, see Figure 5, to exploit the fact that the transient response varies much more slowly at longer times than at earlier times. Unfortunately, with eddy-currents the defect information is smeared out over the timescale of the returning transient. For any one type of defect the response is delayed in time as a function of depth of the defect. However, defects of a different type will not have the same time-to-depth dependence (see Figure 6). One of the challenges remaining with transient eddy-currents is to be

5 able to unwrap that smeared and overlapping information to determine the depth of each defect indication. POTENTIAL ADVANTAGES OF TRANSIENT EDDY CURRENTS It is important to realise that, whilst some advantages come from using transient excitation of the coil, others come from the use of a Hall sensor due to its flat frequency response and optimal spatial resolution, as discussed above. In addition, Hall sensors are not restricted to transient eddy-current systems and have been used with conventional continuous-wave systems at DERA. The main advantages of transients are 1. the ease of scanning large areas of complex structure without the need to change any setup parameters 2. the ease of analysis of the data and ability to distinguish between structural changes and defects. (Analysis of multi-frequency data is always empirically based but for transient eddy-currents there is a theoretical basis to guide analysis of data for discrimination and quantification.) 3. the ability to compensate during post-processing for lift-off and edge effects.. the scope for off-line post-processing, rather than real-time processing of the transient data. 5. the speed of acquisition - a transient system gives equivalent information to a sweptfrequency measurement, but in about 1ms compared with a minimum of several seconds for a swept-frequency measurement. (This makes area scanning feasible for transient eddycurrents, whereas it would not be feasible for swept-frequency eddy-currents.) 6. Instrumentation costs may potentially be lower than for multi-frequency conventional eddycurrents, for which instrumentation costs increase with the number of channels required. Transient eddy-current instrumentation by comparison gives the equivalent of hundreds of frequency channels from almost DC to the upper bandwidth limit - currently 5kHz but it could be higher. It is important to realise that the most important issue with transient eddy-currents is the processing to untangle the different contributions to the transient response. There are commercially-available systems that just record and image the amplitude of a transient eddycurrent signal at a particular time. However, these systems do not exploit the potential of transient eddy-currents for discrimination of defects from structural changes. At present, there are some processing techniques available for analysis that produce unambiguous defect discrimination and quantitative measurements of material thinning. Ultimately it may be possible to invert the transient eddy-currents response and produce a full map of the actual structure and defects, although this will be very difficult. Before proceeding to describe an experimental determination of the capabilities and limitations of transient eddy-currents, we present, for illustration, the results obtained by applying the TRECSCAN system to inspect a section of Al alloy lap-joint cut from a retired B727 airframe, Figure 8. The section of lap-joint was approximately mm in length by 68 mm wide, with three rows of counter-sunk rivets, skin thickness 1. mm, painted, and backed by a top-hat section stringer running along the centre row of rivets.

6 The images shown in Figure 8 were derived by mapping the measured field values H z measured using TRECSCAN at a particular time point t i to a 16-bit colour scale within the ANDSCAN system. Figure 8(a) images the raw field data, at a relatively early time point (t 1 =.9 ms) directly as acquired by the TRECSCAN system. Figure 8(b), by comparison, images the same data after application of an algorithm which compensates for variations in probe lift-off. (This algorithm has been described previously [1].) Both Figure 8(a) and Figure 8(b) apply the same scaling for the mapping to the 16-colour palette. This mapping was chosen so that in Figure 8(b) the colour palette just spans the observed range of H z values from the thinnest single sheet regions to the thickest part of the lap joint. Application of the lift-off compensation algorithm greatly improves the images. Without lift-off compensation - Figure 8(a) - the presence of the corrosion is obscured by large signals due to variations in probe lift-off (caused by pillowing within the lap joint and ridges of excess paint along the bottom edge of the lap joint). However, regions of corrosion are clearly evident the lift-off-compensated image, Figure 8(b), with the most severe corrosion being located in the top half of the joint surrounding the 5 th and 1 th rivets from the left-hand end of the joint. There is some contrast to indicate that the corrosion has penetrated into the bottom half of the joint, primarily between the th and 12 th rivets from the left. A difference in signal level (colour) is observed between the single-sheet regions above and below the joint. This implies that there is a difference in the thickness and/or conductivity between the top and bottom sheets. At this relatively early time (t 1 =.9 ms), the field H z is only sensitive to changes occurring within the first two layers and there is no contrast observed to indicate the presence of the stringer along the centre row of rivets. A useful side-effect of the lift-off compensation algorithm is to greatly reduce the contrast due to the rivets in Figure 8(b) compared to Figure 8(a). It is noted that a region of very strong contrast (labelled A ) is evident within the single-sheet region immediately below the lower edge of the lap joint in Figure 8(a). This region corresponds to that part of the bottom sheet which could only be scanned with the probe resting partly on the top sheet and partly on the bottom sheet. This introduced a large amount of lift-off (up to 1. mm), resulting in lift-off signals which were a factor of 1 times greater than the signals due to corrosion. These large lift-off signals are almost completely eliminated by the application of the lift-off compensation algorithm, Figure 8(b). Figure 8(c) shows an image produced from the field H z at a later time point, t =.26 ms, with lift-off compensation applied. The band of contrast (colours towards the bottom of the palette) observed along the centre row of rivets is due to the presence of the stringer and indicates that the signal at this time is sensitive to changes occurring through the full thickness of both the first and second layers and into the stringer (third layer). Contrast due to corrosion is observed at the same locations as for Figure 8(b). However, since this contrast is superposed on the contrast due to the stringer, interpretation is more difficult than for Figure 8(b). Edge effects are also becoming more significant in Figure 8(c), causing a more gradual variation in the signal across structural changes. Figure 8(d) shows the effect of applying an edge-subtraction algorithm which removes the signals due to edges and other systematic variations in substructure. The algorithm effectively subtracts from the image a reference taken along a vertical line between the two right-most columns of rivets. All contrast between the single-sheet regions, the two-sheet regions and the stringer is removed, resulting in a uniform background with only the corrosion highlighted. This

7 permits a simple interpretation as for time slice 1, Figure 8(b), whilst allowing data from a later time slice to be used. This has the advantages of assuring sensitivity to changes through the full thickness of the structure, together with an optimum signal-to-noise ratio, since this time point t =.26 ms corresponds to the peak signal H z for corrosion in this specimen. Measurements of the percentage change in total thickness, computed using an algorithm described previously [,1], are plotted in Figure 8(e) for a line scan taken along a horizontal line between the top and middle rows of rivets. The measurements of percentage loss are relative to a reference taken between the two right-most columns of rivets. The estimated thickness change exhibits two distinct minima, corresponding to the two most severe regions of corrosion identified in Figure 8(b),(c),(d). The change in thickness is expressed as a percentage of the total thickness of the lap-joint at the reference location (ie. the top and bottom sheets, excluding the stringer). The maximum material loss is measured to be approximately 7% of the total thickness. The graph in Figure 9 shows the accuracy of the thickness change algorithm in measuring changes in total thickness. Measurements were made using the specimens described below and the data sets described in Table 2. The accuracy of the measurements for the largest diameter defects is reasonable given that the algorithm is completely uncalibrated. The algorithm assumes the region of thinning extends beyond the interrogating field and this explains why the values decrease in accuracy as the defect diameter decreases. Another application of this algorithm is its ability to distinguish between thickness change and other structural effects such as plate separation (see Figure 1). Once defects have been detected and their type distinguished it is also desirable to know their depth within the structure. The method, mentioned above, for measuring the time to the peak of transient can be used to produce simple time-of-flight scans, Figure 7, which can be related to depth in the structure via a previous calibration, Figure 6. From this information, it can be determined which layer the defect is in. EXPERIMENTAL DETERMINATION OF SENSITIVITY A set of versatile specimens were designed for determining the defect detection and characterisation capabilities of transients for both corrosion and cracks at a range of depths in the structure (see Figure 11 and Figure 12). These specimens did not attempt simulate exfoliation corrosion, although this is also a major aging-aircraft problem. The specimens were used to produce two sets of results for corrosion detection (Sets A and B in Figure 12 and Table 2) and two for crack detection (Sets C and D). Two data sets (Sets A and C) used a minimum stack thickness with the defective layer (see Figure 11) near the back of a variable-thickness stack. The other two data sets (Set B and D) used a stack of layers that was always 8.5 mm thick and the defective layer was positioned at different depths within the stack. Table 2 shows all the specimen permutations tested. For the current work three different probes were used, as detailed in Table 1. CORROSION DETECTION The spatial and depth resolution with which corrosion-induced metal loss within a multilayer structure can be characterised depends on the spatial characteristics of the eddy-current field and the location of the metal loss in the field. For large defects, such as material thinning, that tend

8 to reflect the incident field, the spatial distribution of the field reflected from the defect is similar to that which is incident on the defect. This in turn is governed by the probe characteristics and the coupling into the structure. Thus, in order to determine the capabilities and limitations of a transient eddy-current system it is necessary to investigate the dependence of the corrosion detectability on four variables: defect depth, defect size, metal loss, and the probe characteristics. Thus a large number of scans were obtained and analysed. Typical scans are illustrated in Figure 1 and Figure 13. In order to determine the detectability of metal loss it is necessary to set a criterion by which the scans will be assessed. This criterion was chosen to be a signal-to-noise ratio of 3:1. Hence the difference in signal between defective and good material should be greater than three times the noise level for that defect to be deemed detectable. For each scan (ie. for each specimen configuration from Table 2), the signal-to-noise was determined for each defect and plotted as a function of defect size and metal loss (see Figure 1). Lift-off compensation was enabled and spatial averaging was applied over a 5 mm diameter region to reduce noise. The signal level was measured as the difference in mean levels between the adjacent background and the defect, and the noise level was taken as the average standard deviation over all the reference regions on the panel (to avoid being dependent on local changes in the noise level.). Then a linear least-squares fit was used to interpolate or extrapolate the threshold detectability at a signal-to-noise level of 3:1. If all measurements for a particular defect diameter were below a signal-to-noise ratio of 2:1 then that defect diameter was excluded as being undetectable (rather than extrapolating up to a 3:1 signal-to-noise ratio). Resultant graphs of the detectability are shown in Figure 15 as a function of defect diameter and in Figure 16 as a function of defect depth.. Figure 15 shows the expected reduction in detectability of smaller or deeper defects. For defects larger than the field distribution at that depth, the defect size should not matter. Only the reduction in metal thickness, the depth, and the field distribution are then important. Hence for defects within mm of the top surface, the minimum detectable metal loss is independent of diameter above 2 mm diameter. This is also shown in Figure 16 where the 2 and 3 mm diameter defects are indistinguishable within mm of the surface, but give different responses beyond mm when the field broadens out to be larger than the 2 mm diameter defect. It would be interesting to know whether the total thickness of the structure has any effect on the detectability of the defects as a function of depth. Figure 17 plots the ratio of the minimum detectable metal loss for the Set A case (i) in which the total thickness varies such that the metal loss is always at the front face of the back-most sheet, to the Set B case (ii) in which the total thickness is a constant 8.5 mm and the layer containing the defects is moved within this constant total thickness (see Figure 12). Figure 17 illustrates the fact that there is a different effect for the ferrite-cored probes compared to the air-cored pancake coil. The ferrite-cored probes seem to be more sensitive when there is additional material behind the layer containing the defect, whereas the pancake coil s sensitivity is reduced by the presence of extra layers. It is reasonable to consider the results for defect detectability of metal thinning in two categories: defects larger than the interrogating field, and defects smaller than the interrogating field. This is because the eddy-current probe, situated directly over the defect, will not be able to distinguish between a large defect and an infinite defect if they are both larger than the field. In that situation the eddy-current system would be sensitive to the amount of thinning but

9 insensitive to how large the region of thinning is. The influence of smaller defects, however, is likely be dependent on both the amount and the extent of the region of thinning. Hence either the volume of the defect or its cross-sectional area (diameter multiplied by metal thinning) may be more useful for defining defect detectability. Firstly, a comparison of minimum defect detectability is shown in Figure 18 for the three probes as a function of defect depth in an 8.5 mm thick stack (Set B), for the 3 mm defect diameters. If the field is smaller than 3 mm then this should hold for all defects larger than 3 mm. The thick horizontal line in Figure 18 indicates a metal loss of 1% of the total thickness. The medium ferrite-cored probe appears to be the best of the three over the whole depth range measured. However, the small ferrite-cored probe has poor sensitivity at greater depths, as predicted above, whilst the larger air-cored pancake probe is poorer near the surface. Using the medium ferritecored probe,.9 mm (3.6 mils) of metal loss could be detected at depths down to 7 mm and.1 mm (. mils) at 1.5 mm depth. The strange shape of the curve for the air-cored probe does not appear to be an artefact of the measurements, although the analysis was carefully checked. It is thought to be a consequence of the particular field distribution from a pancake coil. The graphs in Figure 19 and Figure 2 investigate whether the detectability of smaller defects can be usefully quantified in terms of the volume and/or the cross-sectional area of metal loss respectively. In Figure 19 the minimum detectable volume is almost identical for all the defects smaller than 3 mm diameter at shallow defect depths, although the curves diverge significantly at greater depths. In Figure 2 the minimum detectable cross-sectional area (defect thickness multiplied by diameter) is plotted for each defect diameter, and for all three probes. The curves for each probe show greater scatter at shallow defect depths than those observed in Figure 19. However, for each probe the curves for different defect diameters cluster together fairly consistently across the full range of defect depths. This suggests that the mean of the minimum detectable cross-sectional area for each probe is a useful parameter to characterise probe performance across a range of defect diameters and depths. The means of the minimum detectable defect cross-sections are compared in Figure 21 for the three probes. The medium ferrite probe gives the best sensitivity over the whole depth range measured. Whilst the small ferrite probe has similar sensitivity to the medium ferrite probe for shallow defects, its performance deteriorates rapidly for defects deeper than mm. The large air-cored probe is relatively poorer for for shallow defects, but falls between the two ferritecored probes for deeper defects. This behaviour is generally consistent with that shown in Figure 18 for the 3 mm diameter defects. For areas of metal loss smaller than the probe, defect crosssectional areas of 3.3 mm 2 at 7 mm depth and. mm 2 at 1.5 mm depth could be detected. Previous work [,1] has shown that the time evolution of the transient EC signals can be used to discriminate effectively between signals due to metal loss from those confounding signals due to variations in interlayer gap within the structure (see Figure 1). CRACK DETECTION Small defects oriented perpendicular to the surface of the structure, such as cracks, tend to act as scatterers of the incident field. Thus the received field has a spatial distribution that is largely independent of the incident field. This is a significant difference to the situation with large planar defects such as metal thinning. Therefore different probe types are required to optimise detection of cracks compared with corrosion. In addition, cracks emanating from fastener holes pose a different problem to those which are isolated from other features.

10 Optimum crack detection capability may be obtained when the magnetic field sensor is place at a null point in the field returned from non-defective material. A crack should then perturb this field and register a signal. Previous work by Dr David Harrison at DERA has shown that the use of a C-core probe provides good crack detection characteristics but is most sensitive to cracks in one orientation and sensitivity decreases for cracks that are oriented differently. This is a major disadvantage for cracks growing from fastener holes and these C-core probes are not optimum for that scenario. Optimum sensitivity to cracks from fastener holes is obtained by looking for symmetry variations around a fastener. A probe with a symmetrical response is therefore needed. The Eddyscan system, also developed at DERA and now marketed by Staveley NDT, uses transient excitation to do just that [1] but it requires centering by hand over each fastener. A faster scanning capability is now required for cracks around fasteners in large areas of structure due to the massive number of fasteners that must be inspected on large transport aircraft. An approach taken by DERA, with a view to the use of array probes in the future, is to produce a scan of a region containing fasteners, using TRECSCAN with a symmetrical probe, and capture data at a.5 mm resolution around the region of each fastener. Then it is possible to investigate the symmetry variations by post-processing the scanned image. At present, with single-element probes, this process takes longer than with Eddyscan. However, if the scan is being performed anyway, for corrosion detection or mapping, then it would be possible to look for cracks on the same scan. Eventually, the use of array probes should speed up the scanning and make this a viable technique. Figure 22 shows an example of such a scan alongside the processed version that uses the rotational symmetry of the fastener to find its center and then plots any deviations from rotational symmetry as large colour swings. Hence cracks show up as dark regions on one side of the fastener. Measurements were made on the specimens described above. Before processing the data, lift-off compensation was used and the optimum time point chosen. Subsurface crack signals are accentuated when lift-off compensation is turned on because the signals due to uncracked rivets are somewhat similar to lift-off signals and the lift-off compensation algorithm is therefore reasonably effective at removing them. The display was produced by averaging the data over a 2.5 mm spot size. This reduced the noise in the image and improved the effectiveness of the rotational-symmetry processing. The results of the experiments are summarised in Figure 23 and Figure 2. Firstly, Figure 23 shows the relative performance of the three probes. For crack detection the large pancake coil gives the best defect detectability. This probe was originally designed for crack detection. Although it is large, this is not a disadvantage because the spatial resolution is not defined by the coil size. This is because cracks are scatterers and define their own scattered field spatial distribution. Hence the probe size is less critical and the pancake coil provides a field that can penetrate to a greater depth without loss of spatial resolution for cracks. This probe detected three out of the four 1 mm long simulated cracks in the second layer. A 1.5 mm long throughthickness crack in the second 1.5 mm layer could be detected reliably using this probe. At 6 mm depth only 6 mm long 1.5 mm deep cracks could be detected emanating from fastener holes. Figure 2 shows the presence of additional metal below the cracked layer reduces the sensitivity compared to a crack in the back layer at the same depth. This may be because the metal below

11 the cracked layer provides an additional path for the eddy-currents to be diverted into since, for the present specimens, the layers were all in electrical contact. This would not be the case in most aircraft multi-layer structures where the layers would be painted, sealed or adhesively bonded. When the cracks were in the top layer there was a blank plate as a second layer. This could partly explain the decreased defect detectability for top-skin cracks in the measurements. The use of the lift-off compensation for early time-slices unfortunately results in a reduction in the signal from a top-layer crack, even though it may be surface-breaking. This is because the signal due to a surface breaking feature is in some respects similar to a lift-off signal and therefore suppressed by the lift-off compensation. However, without the lift-off compensation the scan suffers from greater noise and variability, thus making the crack indications harder to distinguish. A compromise was to use the lift-off compensation and view the image of the second or third time points with the processing for rotational symmetry. This probably contributes to the effect shown in Figure 23 and Figure 2 where the best crack detection capability is for second-layer cracks. Another contributory factor is that the countersink of the fastener hole has removed a large proportion of the cross-sectional area of the 1 mm and 2 mm crack. In fact, due to the countersink, the 1 mm crack in the top layer probably has remaining less than one-third of the area of a 1 mm crack in the second layer. It is possible to use the time-of-flight to the peak to image the depth of cracks but, as Figure 25 shows, the signal is dominated near the crack root by the near-surface effects of the fastener. Finally, Figure 26 shows the possibility of using a B-scan cross-sectional slice through the fasteners. Cracked fastener holes can be distinguished from their different profile. CONCLUSIONS The many advantages of transient eddy-currents have been discussed and illustrated. Some of these have already been demonstrated using the DERA/AMRL TRECSCAN system, whilst others still require research and development before they can be implemented. Any transient eddy-current system will benefit from an improved sensitivity to deep defects by the use of a field sensor that measures the field directly. The Hall sensor used by TRECSCAN is a suitable field sensor and gives a considerable advantage over a coil sensor for depth penetration structures and for spatial resolution, particularly of cracks. Acquisition of transient eddy-current data is relatively straightforward. Little interpretation is required whilst scanning. Parameters do not need to be changed at any stage over a large area of structure, provided the maximum thickness penetration of the probe is not exceeded. A crucial factor in the successful application of the transient eddy-current method is the data analysis techniques employed. In this paper some examples of the potential for defect detection, discrimination, sizing and positioning have been given. The potential will only be realised if the best analysis methods are used. Reductions in total thickness of metal can be distinguished from plate separation and changes in the total thickness can be measured. The effects of lift-off and fasteners can be considerably reduced, and edge effects can also be removed from the images to help in detecting the defects. Minimum detectable defect sizes have been determined as a function of defect depth within a multilayer structure for the cases of (i) metal loss simulating corrosion and (ii) saw-cuts from fastener holes simulating cracks occurring within a layer. For metal thinning, the best sensitivity

12 was obtained using a ferrite-cored probe. However, cracks were detected best using a large aircored pancake coil. Using the best ferrite-cored probe available for the measurements,.9 mm (3.6 mils) metal loss could be detected at depths down to 7 mm and.1 mm (. mils) at 1.5 mm depth, provided it extended over an area larger than the probe. For areas of metal loss smaller than the probe, defect cross-sectional areas of 3.3 mm 2 at 7 mm depth and. mm 2 at 1.5 mm depth could be detected. Cracks from fastener holes were detected best using a large pancake coil with processing to detect rotational asymmetries at each fastener location on the scan. A 1.5 mm long throughthickness crack in the second 1.5 mm layer could be detected reliably using this probe. At 6 mm depth only 6 mm long 1.5 mm deep cracks could be detected emanating from fastener holes. ACKNOWLEDGEMENTS The authors gratefully acknowledge the considerable contributions to this work from Dr David Harrison of DERA (for the original development of TRECSCAN and the theoretical modelling of eddy-current fields) and Dr Steve Burke of AMRL (for his theoretical work leading to the thickness-mode analysis methods). The authors are also grateful to Miss Cayt Harding and Mr Mark Taylor of AMRL for performing much of the experimental work, and Mr Lyn Jones of DERA for specimen and probe manufacture. This work was funded by the Department of Trade and Industry and the Ministry of Defence through the TG Corporate Research Programme. REFERENCES 1. D. J. Harrison, Progress in the detection of cracks under installed fasteners using eddy currents. AGARD Conference Proceedings No 62, Impact of Emerging NDE/NDI Methods on Aircraft Design Manufacture and Maintenance, Brussels, (1989). 2. D. J. Harrison, Eddy-current inspection using Hall sensors and transient excitation, Defence Research Agency Technical Report DRA/SMC/TR918, DRA Farnborough, UK, (199). 3. D. J. Harrison, in Nondestructive Testing of Materials, Studies in Applied Electromagnetics and Mechanics, Vol 8, eds. R. Collins, W. D. Dover, J.R. Bowler and K. Miya, (IOS Press, Amsterdam, 1995), pp S.K. Burke, G.R. Hugo, and D.J. Harrison, in Review of Progress in QNDE, Vol 17A, eds. D. O. Thompson and D. E. Chimenti, (Plenum, New York, 1998), pp R.A. Smith, Insight - Journal of the British Institute of NDT, Vol 37, pp and (1995). 6. W.W. Ward III and J.C. Moulder, in Review of Progress in QNDE, Vol 17A,. (1998), pp J. A. Bieber, C.C. Tai and J. C. Moulder, in Review of Progress in QNDE, Vol 17A, op. cit. (1998), pp S. Giguère, B.A. Lepine and J.M.S. Dubois, Pulsed eddy-current (PEC) characterization of material loss in multi-layer structures. Presented at the 7 th Annual Conf of the Canadian Aeronautics and Space Institute. To be published in the Canadian Aeronautics and Space Journal. 9. C.V. Dodd and W.E.Deeds, Analytical solutions to eddy-current probe-coil problems. J. Applied Physics, Vol 19, No 6, (1968) pp G. R. Hugo and D. J. Harrison, in Review of Progress in QNDE, Vol 18B, eds. D. O. Thompson and D. E. Chimenti, (Kluwer Academic/Plenum Publishers, 1999), pp British Crown Copyright. Published with the permission of the Defence Evaluation and Research Agency and the Australian Defence Science and Technology Organisation on behalf of the controller of Her Britannic Majesty s Stationery Office, 2.

13 ANDSCAN and TRECSCAN are Registered Trademarks of DERA. TABLES Probe: Table 1. Parameters for probes Small Ferrite Cup-core Medium Ferrite Cup-core Large Pancake Coil Inner diameter of coil 5.9 mm 9.2 mm 5. mm Outer diameter of coil 11.9 mm 18. mm 2.7 mm Height of coil 2 mm 3.9 mm 2.8 mm Outer diameter of ferrite 13.9 mm 21. mm Not applicable cup-core Height of ferrite cup-core.2 mm 6.7 mm Not applicable Number of turns 2 of.12 mm wire 26 of.19 mm wire 1 of.12 mm wire Measurement Set (A B C or D) Table 2. Specimen permutations tested to determine defect detectabilities Total Thickness (mm) Depth of corrosion from surface (mm) Depth of cracks from surface (mm) Thickness of plates top to bottom (mm). () = countersunk = defect layer C (1.5),1.5 A & C (1.5),1.5 A & C (1.5),1.5,1.5 A & C (),1.5 A & C (),1.5,1.5 A, B, C & D (1.5),,1.5,1.5 B & D (1.5),,1.5,1.5 B & D (),1.5,1.5,(1.5) B & D (1.5),1.5,1.5, B & D (1.5),1.5,,1.5 D (1.5),1.5,,(1.5)

14 FIGURES Figure 1. TRECSCAN being used with an ANDSCAN arm and software. Figure 2. Comparison between coil and Hall sensor transient responses. The absolute voltage levels obtained depend on many different factors, as explained in this paper. Note that the zerocrossing of the coil response corresponds to the maximum of the magnetic field measurement.

15 Ratio of Hall / Coil Voltages Small Intermediate Medium Large Pancake Defect Depth (mm) Ratio of Hall / Coil Sensitivities Small Intermediate Medium Large Pancake Defect Depth (mm) Figure 3. Theoretical comparison of defect signal levels for a Hall sensor relative to a coil sensor as a function of defect depth, for a range of coil dimensions similar to those used in this paper. The type of defect used for this theoretical calculation was a large area of metal loss at a given depth in a thick metal specimen. Ratios are plotted for both the absolute defect signal voltage (left) and for a normalised defect sensitivity (right). The normalised sensitivity was computed as the absolute defect voltage divided by the amplitude of the baseline or carrier signal, on which the defect signal appears as a small perturbation. This baseline signal is much larger for the coil sensor than for a Hall device. Coil Current i input current i (t) time t Magnetic Field H z H z (air) H z (specimen) H z (defect) time t i(t) H z (t) R H z = H z (specimen) - H z (air) Ferrite cup-core Hall effect probe Hidden corrosion Specimen Figure. Schematic showing the probe on a multilayer specimen with hidden corrosion between layers, including typical input current i(t) and probe responses Hz(t).

16 Figure 5. Typical transient responses showing the sample times that increase exponentially (left) and responses from the back surface of aluminium plate of between 1 and 8 mm depth (compared with a reference half-space)..7 Time to Peak (ms) Gap Loss Poly. (Gap) Poly. (Loss) Defect Depth (mm) Figure 6. Graphs showing the dependence on defect depth of the time to the peak of the transient measured using the small ferrite-cored probe. Two different defect types are illustrated, metal thinning (Loss) and plate separation (Gap). It is evident that plate separation causes an earlier transient. The fitted lines assume a quadratic dependence although the theoretical justification for this has yet to be established. These measurements used a total stack thickness of 1.5 mm more than the defect depth (Set A) and can therefore be used as a calibration curve to determine defect depth using the same probe and conductivity of material.

17 Figure 7. Scan showing time-of-flight to the peak of the transient for machined thinning at the front of the second layer (depth 1.5 mm) within multi-layered structure 8.5 mm in total thickness. The time-of-flight values should be related to depth in the structure via the calibration in Figure 6. Measurements were made using the small ferrite-cored probe.

18 (a) A (b) (c) (d) (e) % change in thickness mm Figure 8. Results of TRECSCAN inspection of corroded B727 lap-joint using the small ferritecored probe. (a) (d) images formed by mapping H z (t i ) to a sixteen-colour palette: (a) t 1 =.9 ms without liftoff compensation, using a wrapped colour scale; (b) t 1 =.9 ms with liftoff compensation; (c) t =.26 ms with liftoff compensation; (d) t =.26 ms with liftoff compensation and edge subtraction. (e) Line scan of percentage change in total thickness, taken along a horizontal line between the top and middle rows of rivets. A reference for (d) and (e) was taken along a vertical line between the two right-most columns of rivets.

19 Measured Percentage Metal Loss Defect Diameter (mm) Linear (3) Linear (2) Linear (1) Actual Percentage Metal Loss Figure 9. Graph showing the ability of the thickness change algorithm to measure changes in total thickness relative to a reference point on the structure. Measurements were taken using the small ferrite probe. The algorithm assumes the region of thinning extends beyond the interrogating field. The lines are generated by a linear least-squares fit to the data sets for each defect size. Figure 1. Example of how signals due to plate separation (left) can obscure potential thinning defects which can be revealed using the thickness change mode (right). The image on the left is a simple mapping of the transient field H z at a particular time point. The image on the right plots the percentage change in total thickness computed (post-processed) from the same set of transient data.

20 Figure 11. Diagram of specimen with five different amounts of machined thinning, each in five different diameters, and six lengths of simulated cracks (piercing-saw cuts) in two different orientations, emanating from the fastener holes. This was fastened to other blank plates in various combinations using 36 aluminium fasteners. A similar specimen with countersunk fastener holes was made for use as a top plate for top-skin simulated cracks and thinning. Note that the countersink engulfed most of the 1 mm long cracks. All dimensions are in millimetres. Figure 12. Lay-up of multi-layered specimens for Set A (left) and Set B (right). Dimensions are in mm.

21 Figure 13. Set B scans. 1.5 mm (left) and mm (right) depth with the medium ferrite probe. Signal-to-Noise Ratio Defect Size 3 Defect Size 2 Defect Size 1 Defect Size 5 Defect Size Actual Metal Loss (mm) Figure 1. Signal-to-noise ratio for medium ferrite probe as a function of metal loss at a depth of 1.5 mm in a specimen of thickness 8.5 mm. Each curve represents a defect diameter, given in millimetres.

22 .6 Minimum Detectable Metal Loss (mm) Defect Depth (mm) % Loss Defect Diameter (mm) Figure 15. Comparison of defect detectability for the Medium Ferrite probe as a function of defect diameter, for different defect depths in an 8.5 mm thickness stack (Set B). The thick horizontal line indicates a metal loss of 1% of the total thickness..6 Minimum Detectable Metal Loss (mm) Defect Diameter (mm) % Loss Defect Depth (mm) Figure 16. Minimum detectable thinning using the medium ferrite probe as function of depth of defect for a constant 8.5 mm total stack thickness (Set B) different curves are for different defect diameters. Note that the 2 and 3 mm diameter defects are both larger than the field when closer than mm to the 18. mm-diameter drive coil. The thick horizontal line indicates a metal loss of 1% of the total thickness.

23 Sensitivity Improvement With Additional Meta 2.5 Small Ferrite Medium Ferrite 2 Large Pancake Additional Metal Beyond Defect (mm) Figure 17. This graph illustrates the dependence of sensitivity on total metal thickness. Values were obtained by comparing Set B (constant 8.5 mm thickness) with Set A. There is an improvement in sensitivity due to the presence of additional metal beyond the defect layer for the ferrite-cored probes but not for the air-cored pancake-coil probe..25 Minimum Detectable Metal Loss (mm) Small Ferrite Medium Ferrite Large Pancake 1% Loss Defect Depth (mm) Figure 18. Comparison of defect detectability for the three probes as a function of depth in an 8.5 mm thick stack (Set B), for the 3 mm defect diameters. If 3 mm is larger than the field then this should hold for all defects larger than the interrogating field. The thick horizontal line indicates a metal loss of 1% of the total thickness.

24 Minimum Detectable Metal Loss Volume (mm 3 ) Defect Depth (mm) Defect Diameter (mm) Mean Figure 19. This graph illustrates the validity of quantifying defect detectability in terms of the volume of metal lost. The data is for the medium ferrite probe. Minimum Detectable Metal Loss Cross-sectional Area (mm 2 ) Small Large Medium Defect Depth (mm) Defect Diameter (mm) Mean Mean Mean Figure 2. This graph illustrates the validity of quantifying detectability in terms of the crosssectional area of metal lost. The data is for all defect diameters and all three probes. The data naturally groups itself into the three different probe responses which suggests this is a superior method of quantifying detectability.

25 12 Minimum Detectable Defect Cross-sectional Area (mm 2 ) Small Ferrite Medium Ferrite 1 Large Pancake Defect Depth (mm) Figure 21. As a result of deciding to use defect cross-sectional area for defect detectability of metal loss, this is a comparison of minimum detectability for the three probes as a function of depth, after combining the effects for small defects as shown above. Hence, this should correspond to the detectability for defects smaller than the interrogating field Figure 22. Images of fasteners with cracks in the second 1.5 mm-thick layer of an 8.5 mmthick set of specimens (Set D), of lengths as shown. The right-hand image is the result of processing the fastener regions to identify rotational asymmetries and the cracks are shown as darker regions. Crack lengths are shown in millimetres.

26 7 Comparison of probes, regardless of thickness Detectable Crack Length (mm) Small Ferrite Medium Ferrite Large Pancake Crack Depth (mm) Figure 23. Crack length detectability as a function of crack depth. The error bars indicate the actual extent of the crack in depth. The performance of the different probes is also illustrated and the large pancake coil is noticeably more sensitive. Values were averaged from Sets C and D. 7 Detectable Crack Length (mm) Surface-breaking cracks Constant 8.5 mm Thickness Crack Depth (mm) Figure 2. Comparison of mean values from all probes for surface-breaking cracks in the back or front layers (Set C) versus cracks buried in the structure (Set D). It is evident that surfacebreaking cracks are more detectable than cracks embedded at the same depth within a greater total thickness. This is thought to be due to the electrical contact between plates in the stack, allowing eddy-currents to pass around the crack into the plate below. This is unrealistic for aircraft lap-joints which would be sealed, painted, or adhesively bonded.

27 Figure 25. Time-of-flight to the peak of the transient signal for cracks from fastener holes. The depth at the crack tip is greater than at the root because of the influence of the countersink of the fastener which always makes the actual fastener appear to be a near-surface effect. Figure 26. B-scan slices showing the transient signal amplitude on a colour scale. Note that the more severely cracked fastener holes have larger bright lobes and the uncracked holes have a dark central region.