NDT.net • February 2004 • Vol. 9 No.02
2nd MENDT Proceedings

An EMAT Array for the Rapid Inspection of Large Structures using Guided Waves

Peter Cawley and M. J. S. Lowe
Department of Mechanical Engineering, Imperial College, London SW7 2AZ, UK and
P. D. Wilcox
Department of Mechanical Engineering, Imperial College, Exhibition Road, London, SW7 2AZ, UK
Corresponding Author Contact:


A novel EMAT array system for the rapid inspection of steel plate-like structures using guided acoustic waves is described. Issues considered include the choice of guided wave mode, the design of suitable array elements, the layout of elements in the array and the instrumentation required. Results are presented from a variety of plate specimens that illustrate the sensitivity of the system to artificial defects, the effect of generalised corrosion on the test piece and the effect of plate thickness.


The use of guided waves in NDE applications is widespread [1] and can be grouped into three categories according to the propagation distances involved. Long-range applications typically involve one-dimensional waveguide structures such as pipes [2] where the propagation distances are in the order of tens or even hundreds of metres. Plate inspection, where the propagation distance is a few metres, falls into the category of medium-range applications. Guided wave plate inspection in the form of continuous line scanning between two transducers is ideally suited to continuous monitoring of large areas of, for example, steel plate in rolling mills [3]. However, there has been very little commercial exploitation of guided waves for inspecting large areas of plate-like structures from a single location in a manner analogous to that employed in pipe and rail inspection.

This may be because much of the research effort into guided waves in plates has developed from the classical Rayleigh-Lamb model [4] of guided wave propagation in a cross section through a plate in plane strain, hence forcing the plate to behave as a one-dimensional waveguide. Much work has been published on guided wave transducers [5] and modal selectivity [8] in this simplified system, but the practical development of medium range plate inspection systems is limited by the fact that real plates are two-dimensional waveguides. Hence, the issue of guided wave transducer directionality is at least as important as the issue of modal selectivity. For example, a defect free rectangular plate contains four major reflectors due to the edges (and, depending on the guided wave mode, a further four due to the corners). If the directionality of a guided wave transduction device is not understood, then spurious side-lobe reflections from any of these large reflectors can easily be mistaken for reflections from real defects.

This paper describes a novel electromagnetic acoustic transducer (EMAT) array system for inspecting large areas of a thick (5-25 mm) metallic plate structure from a single test position using guided acoustic waves. The target application for this device is testing the floors and walls of steel plate structures in the petrochemical industry, such as storage tanks and pressure vessels. The layout of the elements in the array is in a two-dimensional circular pattern and this enables a synthesized guided wave beam to be steered in any direction with well-controlled directionality. The first part of this paper deals with the development of the prototype system, results from which are presented in the second part. Finally, the limitations of the technique and the present prototype are discussed and the direction of future research is indicated.

The EMAT Array System

A photograph of the complete prototype system in operation is shown in Figure 1(a). The most important part of the system is the EMAT array. This is placed on the plate to be inspected and the test sequence is initiated from the controlling laptop PC. The array contains 16 transmitter EMATs and 32 receiver EMATs. A signal generator and special power amplifier is used to generate a high voltage, high current toneburst containing 5-10 cycles at a suitable centre frequency. This signal is automatically routed to one of the transmitting EMATs in the array. The signals that are detected at the 32 receivers are amplified, digitised and uploaded to the PC using a custom made 32 channel digitiser. During each test, time domain signals are sampled from all 512 combinations of transmitter-receiver pairs. The signal processing applied to the time-domain data obtained from the array provides beam steering and wavelength selectivity. The overall effect is to mimic the operation of a monolithic, wavelength selective guided wave transducer operating in pulse-echo mode placed at the test location and rotated though 360°. The result is an omni-directional B-scan of the surrounding area of the plate under test.

Fig 1: Diagrams of (a) the complete prototype system, (b) the operation of an EMAT element and (c) the overlapping of EMAT coils used in the actual array.

Mode Selection

For simplicity and to minimize the amount of modal selectivity required, it was decided relatively early in the development of this device to use either one of the fundamental Lamb wave modes, A0 and S0, or the fundamental shear-horizontal mode SH0. The problem with the A0 mode is that the large out-of-plane surface displacement means that its attenuation is very high in cases where a plate is liquid loaded. This makes it unsuitable for many of the proposed applications. The S0 mode at low frequencies and the SH0 mode have very little out-of-plane surface displacement and are therefore not significantly attenuated by liquid loading. The use of both of these modes was therefore considered. Ultimately, it was the ease with which the S0 mode could be excited and detected using omni-directional pancake coil EMATs, that caused this mode rather than the SH0 mode to be used, as described in the following section.

EMAT elements

It is thought that for a guided wave array to have omni-directional sensitivity, the individual elements within it must also have omni-directional sensitivity. For an array using the A0 mode this is readily achieved with piezoelectric point transducers that are sensitive to out-of-plane surface displacement and results from arrays of such devices have been presented [7]. The problem with the S0 and SH0 modes is to make an omni-directional array element that couples to in-plane surface displacement. For example, consider the case of a transmitting transducer that applies in-plane surface traction. If the device acts at a single point, then the force must be applied in a specific direction and hence the device cannot have omni-directional sensitivity. Instead the device must apply a pattern of axi-symmetric surface tractions to a finite area of the surface of the plate. For SH0 waves, the forces need to be in the tangential direction about the centre of the transducer (i.e. the transducer applies a twisting surface traction about an axis normal to the surface of the plate) and for S0, the forces need to be in the radial direction. Applying uniformly distributed surface tractions in this manner could in theory be realized by using specially polarized piezoelectric crystals, but practical problems were foreseen in achieving the required uniformity of coupling of such devices onto real plates.

Pancake coil EMATs can produce the necessary surface traction distribution to excite the S0 Lamb wave and a schematic diagram is shown in Figure 1(b). The magnetic field provided by the permanent magnet is in the direction normal to the plate surface and the current in the coil, and therefore the eddy currents in the plate, flow in a circular pattern. The interaction of the circular eddy currents with the magnetic field causes a radial pattern of surface tractions that are capable of exciting the S0 mode uniformly in all directions.

Array Layout

The array comprises 48 pancake coil EMATs arranged in two concentric circles. The 16 EMATs in the inner circle (52 mm pitch circle diameter) are transmitters and the 32 EMATs in the outer circle (136 mm pitch circle diameter) are receivers. In order for the array to function correctly the separation between the centres of adjacent elements has to be less than approximately one third of the wavelength of the excited guided waves. This conflicts with the EMAT diameter needed for efficient excitation and detection of guided waves, which requires the outer diameter of the pancake coil has to be more than half of the wavelength. This resulted in the scheme shown schematically in Figure 1(c), where the coils of adjacent elements overlap. In the actual array, the EMAT coil patterns for the complete array were printed on 4 layers of flexible, 100 mm thick polyamide PCB material in order to achieve the necessary inter-element spacing and coil diameter (26 mm).



The goal of processing is to transform the 512 time-domain signals obtained from the array into an omni-directional B-scan of the surrounding structure. The processing begins by transforming the time-domain signals to frequency spectra. These spectra are then transformed to the wavenumber domain by interpolation, making use of the dispersion relationship between frequency and wavenumber for the plate under inspection. The purpose of this operation is to compensate for the effect of dispersion [8], thereby increasing the frequency range of S0 over which the array can be operated without degradation in performance. The next phase of processing is to combine the 512 wavenumber spectra with various phase shifts into a number of new spectra, one for each steering direction in the final omni-directional B-scan image. Finally, these wavenumber spectra are transformed by inverse spatial Fourier transform to the distance domain to provide the data for the omni-directional B-scan. The most important part of the algorithm is the phased addition to go from the 512 raw wavenumber spectra to the wavenumber spectra in the steering directions.

Basic Phased Addition Algorithm

In the basic phased addition algorithm, the following procedure is followed to obtain the wavenumber spectra for one steering direction in the omni-directional B-scan. First each of the raw spectra is subjected to a wavenumber dependent phase shift. Physically, this phase shift can be regarded as corresponding to either advancing or delaying the time-domain signal to simulate the effect of the transmitter and receiver element being moved onto a line through the centre of the array and perpendicular to the steering direction. This is illustrated schematically in Figure 2(a). All 512 phase shifted wavenumber spectra are then added to yield the wavenumber spectra for that steering direction.

The operation is then repeated for every other steering direction to build up the complete omni-directional B-scan. The effect of applying the basic phased additional algorithm to simulated data from two array layouts is shown in Figure 2(b) and (c). In both cases, the raw data contains signals from a single location, and the overall diameter of the arrays is the same. It can be seen that the results from the second array are much worse, in that many side-lobes are present. However, because the second array is only populated around its perimeter, it contains significantly less elements than the first array, which was fully populated. This was the motivation for trying to improve on the performance of the basic phased addition algorithm.

Deconvolution Enhancement

A real structure contains multiple reflectors, and the omni-directional B-scan from an array can be thought of as a superposition of numerous signals from individual reflectors, such as those shown in Figures 2(b) and (c), in different locations and with different amplitudes. These leads to the concept of deconvolution of the omni-directional B-scan with the simulated signal from a reference reflector, in order to suppress side-lobes. It has been found that, with certain caveats, deconvolution in the angular domain enables the performance of an array with elements around its perimeter to match or exceed the performance of a fully populated array of the same diameter. It is this technique that has been used to obtain the B-scans shown in the following section from the EMAT array.

Fig 2: (a) Schematic diagram showing principle of phased addition algorithm. The omni-directional B-scans (b) and (c) show the result of applying the basic phased addition algorithm to simulated data from the two array layouts shown.


The prototype EMAT array has been tested on a variety of steel and aluminium plate samples. The results presented here are some representative examples that show the potential performance and limitations of the system.

6 mm Thick Steel Test Plate with Artificial Defects

This was a 2 m x 3 m x 6 mm thick steel plate sample that was designed as a test specimen for a variety of plate testing techniques and contained 44 documented artificial defects with a variety of morphologies. The condition of the plate (apart from the defects) was as-new, unpainted and with a good surface finish and square cut edges. The layout of the features in the plate is shown in Figure 3(a) and results obtained from the EMAT array at a centre frequency of 170 kHz are shown in Figure 3(b).

The dynamic range in Figure 3(b) is 32 dB. Large signals corresponding to the specular reflection from the four plate edges are clearly visible. Over most of the rest of the plate area, a signal to coherent noise ratio of at least 32 dB has been achieved. This sample was not specifically designed for testing with guided waves and many of the features are too small to be visible and too close together to be resolved. The key features that are visible are the 3 deepest flat bottom holes in the upper right quadrant of the plate, the deeper slots in the upper left quadrant and larger diameter through holes in the lower left. The type of feature in this plate that is most representative of an isolated corrosion path is arguably a flat bottom hole. The shallowest flat bottom hole that is visible on this 32 dB scale is 3 mm (i.e. half the plate thickness) deep and 20 mm (i.e. just over 3 times the plate thickness) in diameter. This gives an indication of the typical sensitivity of the EMAT array in its current form, although the precise relationship between the reflected signal amplitude and the defect shape, distance and size is the subject of ongoing research [9].

Fig 3: (a) Layout of artificial defects in 6 mm thick steel plate with defect dimensions indicated in mm and (b) results obtained from the EMAT array at 170 kHz.

10 and 20 mm Thick Steel Plates

The operation of any array is always constrained by the size of the array and the spacing of the elements to a limited range of wavelengths, and hence operating frequencies. The long wavelength (low frequency) limit is when the radial and angular resolution becomes unacceptably poor and the short wavelength (high frequency) limit is when wavelength aliasing begins to take place and causes grating lobe effects. The current EMAT prototype is also constrained in the frequency domain by severe mechanical resonances within the array preventing operation below around 150 kHz. For a guided wave array such as this, the long wavelength (low frequency) limit is arguably the most important, as this limits the maximum plate thickness on which the array can be used with a particular guided wave mode. This is because the frequency axis of the dispersion curves for a given plate material compresses as the plate thickness increases. The useful part of the S0 mode for this application is the low attenuation region below the A1 cut-off frequency. In order for the 150 kHz low frequency limit of the current EMAT array to remain on this part of the S0 mode, the plate thickness needs to be less than around 10 mm.

Results obtained on a defect-free 10 mm thick steel test plate are shown in Figure 4(a). This is close to the low frequency limit of the current prototype, and coherent noise reduces the usable dynamic range to 26 dB as shown in the figure. In order to further increase the maximum thickness of plate that can be inspected, there are two possibilities. Firstly, the dimensions of the array could all be scaled up. This allows the array to operate up to the same point on the S0 dispersion curve, which occurs at a proportionally lower frequency as the thickness increases. The problems with this approach are the increased size of the array and the worsening resolution in the radial direction due to the decreasing operating frequency. The second approach is to use a suitable higher order mode on thicker plates. The S1 mode for instance also has low attenuation close to its maximum in group velocity. In a 20 mm thick steel plate, this occurs at around 220 kHz, which is within the operating range of the current EMAT array. The result obtained from using the current array with the S1 mode on a 20 mm thick plate is shown in Figure 4(b). The signal to coherent noise ratio achieved is not as good as with the S0 mode on thinner plates, but it should be remembered that this array was not originally designed for operation under these conditions.


The feasibility of using a guided wave EMAT array for rapidly inspecting large areas of steel plate structures has been demonstrated and a signal to coherent noise ratio of 26 to 32 dB can be achieved. It should be noted that the lower limit of dynamic range is determined by coherent noise signals that are artifacts of signals from the largest real reflectors (typically the plate edges). If a plate is welded into a structure, then the absolute amplitude of the largest reflectors (now the welds) and the associated coherent noise becomes smaller; hence the absolute sensitivity of the array will be increased.

Fig 4: Results from (a) 10 mm and (b) 20 mm thick steel plates.


This work was supported by the UK Engineering and Physical Sciences Research Council, Shell, Chevron-Texaco, BP and Exxon-Mobil. The 6 mm steel test plate was provided by Röntgen Technische Dienst in Rotterdam and the 10 and 20 mm thick steel plates were provided by the Shell Global Solutions.


  1. Chimenti, D. (1997) Guided waves in plates and their use in materials characterization, Applied. Mechanics Review, 50(5), 247-284.
  2. Alleyne, D., Pavlakovic, B., Lowe, M. and Cawley, P. (2001) Rapid long-range inspection of chemical plant pipework using guided waves, Insight, 43, 93.
  3. Ball, D. and Shewring, D. (1973) Some problems in the use of Lamb waves for the inspection of cold-rolled steel sheet and coil, Non-destructive Testing, 6, 138-145.
  4. Viktorov, I. (1967) Rayleigh and Lamb waves: Physical theory and applications, Plenum Press, New York.
  5. Rose, J., Pelts, S. and Quarry, M. (1998) A comb transducer model for guided wave NDE, Ultrasonics, 36, 163-169.
  6. Wilcox, P., Lowe, M. and Cawley, P. (2001) Mode and transducer selection for long range lamb wave inspection, Journal of Intelligent Materials, Systems and Structures, 12, 553-565.
  7. Wilcox, P., Lowe, M. and Cawley, P. (2000) Lamb and SH wave transducer arrays for the inspection of large areas of thick plates, in D. Thompson and D. Chimenti (eds.), Review of Progress in Quantitative NDE, Vol. 19B, AIP, New York, pp. 1049-1056.
  8. Wilcox, P. (2001) A signal processing technique to remove the effect of dispersion from guided wave signals, in D. Thompson and D. Chimenti (eds.), Review of Progress in Quantitative NDE, Vol. 20A, AIP, New York, pp. 555-562.
  9. Diligent, O., Lowe, M., Cawley, P. (2002) Reflection and scattering of the S0 Lamb mode from 3-D circular defects in plates, in D. Thompson and D. Chimenti (eds.), Review of Progress in Quantitative NDE, Vol. 21A, AIP, New York, pp. 231-238.

© NDT.net |Top|