High Power Phased EMAT Arrays for Nondestructive Testing of As-Cast Steel

A new high-power electromagnetic acoustic transducer (EMAT) solid state pulser system has been developed that is capable of driving up to 4 EMAT coils with programmable phase delays, allowing for focusing and steering of the acoustic field. Each channel is capable of supplying an excitation current of up to 1.75 kA for a pulse with a rise time of 1 μs. Finite element and experimental data are presented which demonstrate a signal enhancement by a factor of 3.5 (compared to a single EMAT coil) when using the system to transmit a longitudinal ultrasound pulse through a 22.5 cm thick as-cast steel slab sample. Further signal enhancement is demonstrated through the use of an array of detection EMATs, and a demonstration of artificial internal defect detection is presented on a thick steel sample. The design of this system is such that it has the potential to be employed at elevated temperatures for diagnostic measurements of steel during the continuous casting process.


Introduction
Diagnostic assessment of internal product quality during the continuous casting of steel is currently limited to offline and largely destructive methods, such as acid etching followed by sulphur printing [1], chemical analysis of drilled core samples [2] and optical emission spectroscopy methods [3]. There is a 5 M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT requirement from industry to perform product quality tests non-destructively and continuously during the casting process to allow feedback to the casting operators. This could, in principle, mitigate the development of internal defects, which both reduce the steel's sale value and in some cases present safety concerns [2,4,5]. 10 Detection of internal defects during the casting process presents a number of difficulties for conventional non-destructive evaluation (NDE) techniques; the high operating temperatures, surface roughness and continuous movement of the sample necessitate the consideration of a non-contacting approach. The thickness of a cast steel slab lies in a range from 12 -30 cm, which is sufficient 15 to preclude the consideration of practical radiographic measurements, and to perform active thermography through such a sample thickness would be impractical, due to the variable and uncontrolled ambient temperatures of the casting environment and the likelihood of false indications arising from surface oxide scale. Ultrasound measurements have been identified as a realistic 20 prospect of probing the surface and bulk of a cast slab and are the subject of previous studies on cast steel diagnostics [6,7,8,9], but there still exist a number of challenges when attempting to use acoustics. Namely, the slab itself is relatively thick (up to 30 cm) and contains inhomogeneous and relatively large grain structures when compared to the expected dimensions of a casting defect. 25 Hence attenuation of ultrasound signals, in particular the higher-frequency signals that have scattered from defects, will reduce detected signal amplitudes significantly. Additionally, previous studies have demonstrated that ultrasonic attenuation in metallic samples increases at high temperatures [7,10,11].
Non-contacting methods of ultrasound generation are well-established [12, 30 13, 14, 15], but the problem of non-contact measurements during continuous casting requires special considerations. The high sample temperatures of up to 1100 • C potentially make water jet coupling of piezoelectric transducers impractical [16], and the large impedance mismatch between the air and the steel sample precludes the use of air-coupled transducers [14]. Ablative laser gener- 35 ation of ultrasound in steel billets during the casting process has already been M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT demonstrated, and generates sufficient ultrasound wave amplitudes for both surface defect characterisation and possibly bulk wave measurements [8,17,18].
However, laser sources are relatively expensive, high-power laser beams present implications for the steel mill's safety regulations, the surface ablation pits can 40 interfere with other visual inspection systems in place at the steel mill and large surface coverage interferometric detection of ultrasound waves using lasers is difficult in optically-rough and moving samples [19]. Electromagnetic acoustic transducers (EMATs) have been used as ultrasonic detectors in conjunction with ablative laser generation sources for surface measurements of continuously cast 45 steel billets [8,20], and so represent one possibility for performing bulk diagnostic tests. The low cost and minimal requirement for adaptations to the steel mill's safety protocols makes an entirely EMAT-based system attractive, but their poor transduction efficiency presents challenges in obtaining a practicable signal-to-noise ratio [21,22]. The work presented here concerns the develop-50 ment of an EMAT phased array concept to overcome this inherent drawback of EMATs.

EMAT Generation and Detection
The EMAT generator devices presented in this work consist of an inductor 55 coil driven with a high amplitude (kA) dynamic current. Such devices have been demonstrated in previous studies to be relatively efficient bulk wave generation sources [23,24,25,26], and should in principle be more industrially-robust than conventional EMAT designs, since there is no requirement for an electromagnet or for active cooling of a permanent magnetic material to maintain a sensor 60 temperature lower than the Curie point.
When a coil-only EMAT above an electrically-conducting sample is driven with a large transient current pulse, the resulting time-varying magnetic field induces an eddy current density profile in the sample. Under the plane wave approximation for the magnetic field, the magnitude and phase of this current density profile decays exponentially with depth into the sample with a characteristic length scale known as the electromagnetic skin depth [27, 28,29]: where J 0 is the magnitude of the current density, J, at the surface, z is the depth into the sample and δ is the electromagnetic skin depth. The total induced current, as calculated from an integration of the current density profile over 70 depth, can be shown to be equivalent to a surface image current with magnitude J0δ 2 and a phase lag with respect to the driving voltage of −π 4 [27], allowing the eddy current distribution to be modeled as a current sheet as shown in figure 1.
The eddy currents interact with the EMAT's dynamic field and induce mechanical forces in the sample's surface through the Lorentz force (F L ), which is 75 a vector cross product of the eddy current density and the magnetic field density (B) [29]: rents and the dynamic field lines will reverse when the current in the driving coil is reversed, leading to exclusively repulsive mechanical forces normal to the 80 sample surface at twice the frequency of the applied driving current [29].
EMAT generation relies on the scattering of conduction band electrons from metal atoms to impart momentum into the metallic lattice; this is an inefficient process, due to the small electron-atom mass ratio. This contrasts with EMAT detection, which is a more efficient process, since sample motion is inherent to 85 the incidence of an acoustic wave. The motion of the conducting sample in an applied magnetic field induces dynamic currents in the sample, which themselves induce a measurable potential difference in the detection coil. In detection, a static bias magnetic field is always required, usually supplied by a permanent magnet [29]. This usually means that a coil-only EMAT cannot act as a detector 90 (work has been published which demonstrates the use of a specialised driving circuit for coil-only devices to detect bulk ultrasonic modes [24], but due to the added complexity, such a setup is not considered here).
The inherent inefficiency of electromagnetic ultrasound generation means that EMAT measurements typically suffer from poor signal-to-noise ratios. This 95 issue is compounded by the expected low signal amplitudes arising from the cast steel sample grain coarseness and high temperatures discussed in section 1, and hence design considerations are required to improve the signal amplitude of an EMAT-based system.

100
One approach that can be taken to improve the signal-to-noise ratio of a measurement is to utilise a phased array to increase the signal amplitude through linear superposition; if the ultrasound signals are summed coherently, the resulting total signal amplitude increases, whilst any stochastic noise in the measurement sums incoherently. Enhancement of EMAT sensitivity by the geometric 105 focusing of shear waves has been reported previously, however the approach taken relied on toneburst current excitations, which are more limited in power than the pulsed currents described in this work, and the dependence on geo-M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT metric focusing prevented dynamic beamforming [30,21]. The novelty of the work described here is the development of a high power EMAT phased array, 110 designed specifically for the inspection of thick, attenuative industrial samples.
Phased array generation and detection of ultrasound is a well-established technique for focusing and steering acoustic waves in both medical diagnostics and NDE [31]. By applying appropriate phase delays to each element, a point in space can be chosen such that the wavefront from each element will arrive 115 simultaneously so that the acoustic beam is locally intense. In order to perform beam focusing, the phase delays are calculated by first calculating the propagation time from each element to the chosen focus, then subtracting the maximum propagation time from each element. The applied time delay, φ, can be expressed as: where the subscript i refers to the i th element in the array, c L is the longitudinal wave propagation speed, x is the displacement in x of the element from the focus, y is the displacement in y from the focus and the subscript max refers to the element that lies at the greatest distance from the focus.
The work presented here describes the development of a phased EMAT array 125 generation system to enhance signals transmitted through the full thickness of as-cast steel slab samples.

EMAT Phased Array Driving Electronics
As discussed in section 2.1, coil-only EMAT designs require large dynamic 130 currents for efficient ultrasound generation, and hence a bespoke excitation circuit. The driving electronics for the experimental tests of the EMAT generation array consist of a capacitor bank discharged through a solid state switching device for each channel (see figure 2). The phase delays are applied by a fieldprogrammable gate array (FPGA) unit and have a temporal resolution of 2.5 ns.

135
The temporal current profile of the excitation pulse was measured by placing a  The driving circuit is similar to the driving electronics described by previous studies describing the development of a coil-only send-receive EMAT [32], but is capable of achieving much higher current amplitudes and hence more intense ultrasound generation, since the magnitude of the self-field Lorentz force scales 145 with the square of the excitation current [29]. Even higher current amplitudes have been reported for a single coil using a spark-gap discharge driving circuit, although this is not as practical as a solid state switching method [23,24,32].
The solid state switching for each channel allows for accurate and reliable appli- cation of phase delays, which is essential for control of the phased array beam 150 characteristics.

Finite Element Analysis
The commercial software package PZFlex was used for all following finite element calculations. PZFlex implements an explicit time domain integration algorithm for solving dynamic elastic and acoustic fields. Further details relating 155 specifically to the finite element solver can be found in reference [33].

Pulsed EMAT Array Optimisation
A high power EMAT pulser system consisting of four independent channels pulser's driving electronics is low. Typically when designing a phased array 165 of any kind, it is beneficial to adhere to the diffraction limit and maintain an element separation equal to, or less than, a half-wavelength. With such a limited number of elements, however, the aperture would be small when adhering to the diffraction limit and hence the expected beam characteristics would be poor. Moreover, the inherent inefficiency of EMATs necessitates relatively large 170 transducer footprints for practicable signal-to-noise ratios, making adherence to the diffraction limit difficult. A finite element study was therefore conducted to ascertain the best array parameters (element separation and element width) to achieve both a narrow beamwidth and sufficient sidelobe suppression.
Analysis of the self-field Lorentz generation mechanism indicates that the 175 coil-only design can be approximated as a rectangular piston source [26]. Each A finite element grid was meshed with an element density of 16 elements per wavelength in order to avoid numerical artifacts arising from coarse meshing relative to the wavelength [33]. 195 Internal defects of interest in cast steel, such as segregation defects and associated cracking, are likely to lie along the centreline, which in a 22.5 cm thick slab is at a depth of approximately 11 cm below the sample surface [34,35]. As the array's aperture is increased, the beamwidth is reduced at the expense of increased signal content outside of the 3 dB beamwidth.
Phase delays were therefore applied in accordance with equation 3 to model the focusing of an incident longitudinal ultrasound pulse at a depth of 11 cm (see 200 figure 4). The EMAT pulsing system available for experimental use has four channels, and so for meaningful comparison of the model with experimental results, a four-element EMAT generator was modeled in this way.
The EMAT generation array can be characterised in terms of two defining parameters; the element separation (the distance between the centre nodes 205 of adjacent elements) and the element width (the width of the active element region). The aim of the study was to obtain the optimal values for these parameters to achieve a narrow beam width and high directivity in the generated beam.
A series of simulations were run in which the element width was kept con- The total variation in the beamwidth is quite small over the whole range of element widths, and the point-wise variation arises due to truncation errors associated with grid discretisation.
Increasing the element width leads to a negligible effect on the beamwidth, whilst improving the main lobe power ratio.
terms of the 3 dB beam width (the angular range in which the beam amplitude is greater than, or equal to, half of the maximum amplitude, see figure 5) and in terms of the logarithmic ratio of integrated beam amplitude within the 3 dB beam width to integrated amplitude without of the 3 dB beam width. The first 220 parameter serves as a metric for comparing the directivity of the main beam lobe; a narrower beam width gives a more localised high pressure region, which is beneficial when aiming to separate defect indications that lie laterally close to each other. The second parameter serves as an indication of the relative amplitude of side lobes; if the ratio is low, then more of the beam energy is directed 225 outside of the main beam width and in separate lobes that are directed away from the intended target region, leading to regions of high localised amplitudes other than the intended focus, and therefore potentially confusing attempts at defect localisation using a focused beam.
The results from this series of simulations are displayed in figure 6. The 230 overall observed trend is that an increased element separation reduces the 3 dB beam width, which is desirable, though at the expense of increasing power distributed through side lobes. This is to be expected; the larger array aperture

Phased Array Signal Enhancement
Using the optimised array parameters obtained in section 3. An EMAT detector is sensitive to surface particle velocity, and hence in order to model the signal as detected using an EMAT, the velocity vector history was recorded for each node in the simulation grid. Nodes on the lower surface were chosen which corresponded to an EMAT detector, with a footprint described by induced currents in the detection coil canceling each other. In-plane particle velocities at the coil surface nodes were therefore summed over each half of the coil and then subtracted to account for this cancellation effect.
The resulting values give a measure of the calculated relative amplitude of the expected EMAT signal, however the numbers are not directly comparable 285 to experimental measurements without a full model of the EMAT detection device. This is an unnecessary complication due to the non-trivial field geometries arising from the permanent bias field and its interaction with the steel sample, the dependence of eddy current densities on sample properties and the degree of mutual inductance between the detection coil and the sample. Instead, it 290 is sufficient to compare the difference in amplitude between similar models of a single EMAT element and a phased array to determine the expected signal enhancement from using the phased array approach.
The resulting velocity histories, summed over the appropriate nodes, are shown for the cases of a single EMAT generation element and generation by a 295 phased array in figure 9. The difference in the peak-to-peak amplitude of the incident longitudinal pulse is a factor of 3.7. This result is to be expected, since it is approximately equal to the number of extra elements applied (though it is expected to be lower than 4, since attenuation losses at the focal point from the outer elements will be greater than for an element positioned directly above the 300 focus, due to the increased path length).

EMAT detector design
A coil-only EMAT generator predominantly excites mechanical forces that lie out of the sample's surface plane, and hence lead primarily to longitudinal wave generation (see section 2.1). Efficient detection of these transmitted 305 longitudinal signals therefore requires an EMAT design that is sensitive to outof-plane particle motion, and hence requires a static bias field with significant in-plane components. This is relatively difficult to achieve, since the permanent magnet supplying the bias field must lie above the sample surface and because the in-plane magnetic flux density falls rapidly with distance from the magnet's advantage of these designs is that they reduce the parasitic inductance in the coil and expose more of the coil's length to the sample, and so lead to more efficient detection of longitudinal ultrasound waves. Since the coils are still placed at the magnet edges, there is still in-plane particle motion sensitivity, and hence these designs are also suitable for detection of shear wave modes. wire of 0.14 mm diameter was wound into kapton tape that was pressed into 3D printed coil templates with these dimensions. The magnets used in the detection sensor were NdFeB magnets with a height (out of the page) of 2.5mm, stacked three high. Separation gaps between the magnets ensure that in-plane static magnetic field components are large enough for out-of-plane particle displacement detection. magnets with alternating polarities at a separation of 0.3 mm, and to align the inductor coil correctly in the resulting gaps between the magnets. Copper wire with a 0.14 mm diameter was used to wind a racetrack coil, with track width 3 330 mm, into the plastic template grooves beneath the magnets such that its edges lay under the magnet's edges. The coil was encased in kapton tape (see figure   10). wire of 0.14 mm diameter was wound into kapton tape that was pressed into 3D printed coil templates with these dimensions.

Experimental Validation of Signal Enhancement
to ensure tight control over element width and spacing. The EMAT array's generation coils were wound into a 3D printed template using 0.14 mm diameter copper wire enclosed in kapton tape. The parameters of the array (element 340 spacing and width) were chosen on the basis of the finite element study presented in section 3.2.1, and so the width of each individual racetrack coil element was 4 mm, with the distance between the centres of adjacent elements being 6 mm (see figure 11).
Phase delays were applied in accordance with equation 3 to the four-element 345 generation array to focus an incident pulse of longitudinal waves on the opposing face of a 22.5 cm thick as-cast steel slab sample. A single edge-field detection EMAT (constructed as described in section 3.3) was placed directly opposite; this was connected to an amplifier, which was then connected to an oscilloscope to measure the time-dependent voltage across the detection coil (see figure 12).  figure 2 drives the EMAT generation array (described in figure 11), which generates a focused longitudinal ultrasound wave. This is detected on the opposing side of the as-cast steel slab sample by the detection EMAT (described in figure 10). The signal is passed through an amplifier before being recorded on an oscilloscope.
surements were taken with no coherent averaging, but were digitally filtered using a Butterworth bandpass filter with low and high pass bands of 0.1 and

Enhancement Using a Detection Array
The amplitude enhancement demonstrated by the use of a four-channel generation array can be further improved through the coherent addition of the transmitted signal as detected using an array of detection EMATs. Using the phased EMAT generation array. The signal amplitude is improved by a factor of 3.5 when compared to a single element, which is in good agreement with the factor of 3.7 improvement predicted by finite element analysis (see figure 9) and focusing at a depth of 11 cm was recorded independently on each detection channel. The transmitted longitudinal pulse signal was identified in the A-scan 370 trace recorded by the central element in the detection array and cross-correlated with the data from each channel to determine the phase separation of the signal as recorded by each element. These phase delays were then applied to the Ascan data from each channel, before summing to produce a single A-scan data set with enhanced amplitude in the longitudinal signal.

375
The principle of noise reduction through this delay-and-sum method is that any genuine ultrasonic signals arriving in the expected time intervals should be coherent and add constructively, whereas any noise due to stochastic processes should sum to zero if enough independent measurements are considered. This where SN R is the value of the signal-to-noise ratio in decibels, n sig and n noise

Side-Drilled Hole Detection
With sufficient signal-to-noise ratio on detected ultrasound pulses propagated through the full thickness of a cast slab sample, it is possible to begin looking at detection experiments for internal defects. A four-element phased 420 generation array was placed on the upper surface of a 32 cm thick steel sample with a 6 mm diameter side-drilled hole centred at a depth of 16 cm (see figure   15). Phase delays were applied in accordance with equation 3 to focus the incident longitudinal beam on the defect. A detection EMAT was placed adjacent to the generation array to record any backscattered ultrasound signals. Using the coil arrangement described in figure 15, a pulse-echo ultrasound A-scan was recorded on the 32 cm thick steel sample (shown in figure 16).
Although the chosen coil orientation prevents amplifier saturation, the close proximity of the detection coil to the generation coils gives a dead time of 20 445 µs. The data were therefore processed, firstly by windowing away the generation noise before 20 µs, before fitting the A-scan trace with a 7 th order polynomial and subtracting the fit function to de-trend the low frequency generation noise from the signal. High frequency noise was then removed using a Butterworth bandpass filter between 0.1 and 3.5 MHz. The small reflected signal at 53 µs 450 (b) corresponds to a back-scattered longitudinal wave from the defect (corresponding to path 1 in figure 15). The larger pulses observed at 109 µs (c) and 132 µs (d) correspond to a longitudinal reflection off the sample's back wall and a forward-scattered mode-converted shear wave from the defect (corresponding to path 2 in figure 15) respectively. This interpretation of the A-scan trace in 455 figure 16 has been corroborated with finite element analysis.
The results of this experiment suggest that for detection of small defects, the largest indications are provided by forward-scattered mode-converted signals.
Although the sample used in this experiment is not as-cast, the signals in figure   16 that constitute the defect indication have traveled through 64 cm of steel, 460 and so the prospect of detecting internal defects in a 22.5 cm thick as-cast slab sample remains promising.

Summary and Conclusions
This work has discussed the development of a compact, low cost, high current four-channel phased array EMAT pulsing system that can drive coil-only