Abstract
The paper focus on a new combined experimental methodology and signal processing for an EMAT based on a Halbach magnet system in order to increase smaller defect detection in thick metallic plates. The visualization of the EMAT signal using a novel dynamic C-scan procedure is based on a continuous movement of the transducer above plate and an off-line signal processing that is enhanced and optimized for detection of small defects (5%) located on the opposite side of metallic plates (50 mm). The feasibility and performance of the acquisition signal method, and several signal processing algorithms are validated using experimental measurements. The results present an optimized two-dimensional visualization technique for EMAT signal, which can be used for continuous surface scans of plates for detection of smaller defects that self-calibrate as the scan is conducted.
Introduction
For the in-service inspection (ISI) of the reactor vessel of sodium cooled fast reactors (SFR), EMATs (electro-magnetic acoustic transducers) has the advantage of being able to operate a long time at high temperature due to the lack of couplant between transducer and the metallic vessel [1–3]. EMAT operation is based a permanent magnet and a coil driven by a transient repetitive pulse current at a high frequency [4–6]. When EMAT is located above a metallic plate, the eddy currents induced in the material in the presence of the permanent magnetic field of the magnet generates ultrasonic waves due to Lorenz force oscillations of the material near-subsurface area where eddy currents are localized [7,8]. Traditionally, the EMAT signal is acquired while the EMAT is located in a fixed point above the material when transmission and reception of ultrasonic wave occurs without moving the EMAT. Also, the signal processing analysis is related to each measurement point without a clear way to evaluate the EMAT signal variation along a line or surface measurement. In a previous work [9], the authors reported a new methodology to acquire the EMAT signal using not a static (EMAT position fixed during measurement) but a dynamic C-scan procedure. The original signal processing algorithm builds a C-scan map of the material surface using only optimized selected data from EMAT signal at each point of scanning, with an output C-scan visualization mapping.
In a previous paper [9], the new way to build the EMAT signal characteristics as a C-scan map was based on a continuous dynamic scanning and acquisition/integration of EMAT signal. However, while previous data showed the feasibility of the new EMAT signal representation, the experimental measurements showed high signal to noise (S/N) ratio only for large defects (depths above 20% of plate thickness). Also, more important, the results of the previous algorithm were based on the assumption that we could separate the area without a defect from the defect area and build therefore the distribution of the EMAT noise by considering only the no-defect area signal. However, in a real inspection it is necessary to find the proper definition of EMAT noise distribution without known before the source of the EMAT signal (no information about defect location).
In the present paper, we focus on finding a suitable pair of algorithms: one algorithm to properly define the EMAT signal and another algorithm to define the EMAT noise distributions without any advanced knowledge about EMAT signal in order to acquire a C-scan visualization with higher S/N ratio for smaller defects (starting from at least 5% from plate thickness). The paper identify a visualization method of the EMAT signal as a C-scan map with an automatic self-calibrated noise distribution of both areas with and without defect to visualize small defects position (5%) above the background noise of no-defect areas. The method to achieve it is validated in the paper using experimental measurement on a test specimen.
Experimental set-up of the EMAT system, scanning procedure
The main parameters of EMAT system and test specimen used in the experimental measurements are described in Table 1 and 2. The EMAT (single emitter-receiver unit) shown in Fig. 1, above a metallic plate (400 mm × 180 mm) is shown in Fig. 2. Four slits 15 mm long and 0.5 mm wide were machined on the plate with depth of 5, 10, 20 and 50%. Experimental measurements were conducted using the RITEC RPR-4000 to generate the tone burst pulse. The position of the EMAT above plate, and in close contact with the metallic surface, is accurately controlled with an XY stage (0.1 mm location precision).
While in the previous report [9] the optimum excitation frequency of the EMAT coil was reported to be 680 kHz for detection of defects up to 20%, it was found experimentally that actually for smaller defect detections a higher frequency of 970 kHz is more suitable for the specific algorithm presented earlier (it is known both the location of defect and no-defect area). The optimized frequency for specific purposed of detecting smaller defects (up to 5%) was therefore conducted extensively experimentally due to the large number of parameters involved in the optimization. It was found out that higher S/N ratio for smaller defect detection can be obtained only when the operation point of EMAT is chosen to be close to the middle-low range of various EMAT parameters.
The scanning of the surface of the measurement plate is conducted with a lower speed (forward) and a higher speed (backward) on the same line and then advanced to the next line (1 mm or 2 mm apart), as shown in detail in Fig. 2. The EMAT is moving continuously with a constant speed controlled by the XY stage while the equipment acquired continuously the data. Due to particularities of EMAT signal acquisition, in order to obtain a low noise, the pulse is repeated in burst sequences which are integrated to enhance the signal and reduce the noise. However, if the EMAT moves continuously, for each pulse in the burst there is a different EMAT position, contributing to increase of noise as the transducer speed increases. Therefore the source of non-alignment of data, if movement of transducer backward and forward is conducted on different lines, is due to change of EMAT position while its signal is averaged. By recording in real time the oscilloscope (KEYSIGHT:DSOS204A) display image with a 30 images/sec the EMAT signal is captured continuously in a video file as it is also averaged for the last 256 pulses. The EMAT signal at each image of the acquired video files is extracted, calibrated and digitized. For example, if EMAT moves with a speed of 2.5 mm/s, then an EMAT signal can be acquired for each EMAT position step of 0.083 mm.
EMAT test parameters
EMAT test parameters
Parameters of test specimen
Schematic view of EMAT and signal acquisition method.
Test sample and scanning procedure.
Definition of algorithms to digitize EMAT signal and noise
In the previous work it was explained the new way to acquire the C-scan image based on a new spatial representation, associating an EMAT signal to each point in the 2D surface scan, starting from a previous definition based on the amplitude of the signal S(x, y) at a defined point (x, y) in the defined optimized time interval (t0, t0 + Δt). In the present paper further investigations explore over ways of defining the EMAT signal representation S(x, y) using not only amplitude but also areas (in the time interval Δt) of EMAT signal to evaluate S/N ratio improvements.
Figure 3 shows the definition of four algorithms to extract EMAT signal from the each frame of the video image (30 images/sec) capturing the EMAT signal from oscilloscope display. Because of the envelope of a signal in the oscilloscope image, the digitization of the image (as shown in Fig. 4c) extracts signals S1 and S2 corresponding to the upper and lower envelope edge in the digitized bitmap signal version. While Algorithm-1 and 2 for definition of EMAT signal in Fig. 3 is based on lines S1 and S2, the Algorithm-3 and 4 define the signal as a measure of the surface spanned in the interval (t0, t0 + Δt).
Figure 4 shows the definition of the EMAT noise. The Noise-1 is defined by analyzing the maximum amplitude of EMAT signal over a known area without defect as in Fig. 4(a). However, in real inspection conditions is not known the location of defect or no-defect area and this noise signal definition is not a suitable choice. In the present paper a new methodology is adopted in which the EMAT noise is considered to be at the very end of the EMAT pulse data, before the signal starts to repeat itself. This approximation is based on the assumption that for very large values of t0 there is little information of the defect signal in EMAT data at higher time intervals. If for example the burst is 100 Hz, then the length of EMAT signal is 10 ms and only that very end zone near the 10 ms time scale is selected as indicator of EMAT noise (shown in Fig. 4b), that is the area before the next EMAT burst starts. As EMAT moves along the material surface, the noise distribution changes accordingly and three different models of noise distributions are selected in the following way: Noise-2 is defined as maximum noise signal for each EMAT position, Noise-3 is defined as maximum noise signal along all points in a line scan, and Noise4 is defines as maximum noise signal along all scanned surface. Therefore, by adopting the new definition of EMAT noises (Noise-2, 3 and 4) is it possible to evaluate the performance in building an S/N ratio C-scan map that can be compared with the reference map given by the Noise-1 approach, when location of free defect area is known in advance.
Schematics of four algorithms to extract the EMAT signal from digitized oscilloscope image.
Definition of EMAT noise: (a) Noise-1 (scanned line) when it is known the location of defect and no-defect area, (b) Noise-2 (each point), Noise-3 (scanned line) and Noise-4 (scanned surface) defined using the end zone of EMAT signal.
In experimental measurements of EMAT signal an additional source of noise can arise from improper selection of voltage amplitude signal collected in the oscilloscope image that has a slightly non-linear effect on the overall signal amplification. In order to increase the S/N ratio for detecting small defects, the measurements were focused first on building the C-scan mapping of S/N ratio only for 5% and 10% defects. In the procedure of building the C-scan map in Fig. 5 was initially used the Algorithm-1 of EMAT signal representation and Noise-1. Figure 5 shows the C-scan map of material surface and peak S/N ratio for 5% and 10% defects when the input amplitude of oscilloscope scale (8 divisions) is limited to 12, 25, 50 and 100 mV/div. The measurements results, conducted with a speed of 2.5 mm/s of EMAT scanning, showed that the values of 25 and 50 mV/div are the most suitable in enhancing S/N ratio for small defects.
Therefore, in the next analyses it was chosen only the 50 mV/div scale due to missing artifacts in the C-scan map. The C-scan map is built in that way that the darker color indicates any S/N ratio larger than 2, using a non-linear function color to represent it, while the lighter color shows the S/N ratio close to 1. In order to have a clear indication of the defect, a larger than 2 of S/N ratio is adopted as the defect versus no-defect criteria.
If 30 images/sec of video image are digitized every second, when EMAT is moving at 2.5 mm/s, a digitized signal is available at each step X (in X-direction) of 0.0833 mm along scanning line. However, it was found that the S/N ratio for small defect (5%) does not change drastically for all noise source definitions (1 to 4). Figure 6a shows the variation of S/N ratio when Algorithm-1 is used as source for EMAT signal and C-scan maps were built with points apart from each other along scanning line from 0.0833 mm to up to 8 mm. However, the distance between two successive scanning lines was fixed to 1 mm apart.
Higher speed of EMAT is decreasing more drastically the S/N ratio for small defect as shown in Fig. 6b. For speeds higher than 4 mm/s, S/N ratio of smaller defect (5%) is below threshold of 2. Previous reported work [9] showed that speed effect could be reduced if less EMAT signal averaging (64 versus 256 as in the present analysis) is used, but as the noise increases, overall the S/N ratio will not decrease.
C-scan of EMAT signal (scanning speed of 2.5 mm/s) using Algorithm- 1 and Noise- 1 with oscilloscope voltage levels of 12, 25, 50 and 100 mV/div.
(a) Effect of step size between points extracted during scanning on S/N ratio using Algorithm-1 and Noise-1 to 4, (b) Effect of EMAT speed on S/N ratio using Algorithm-1 and Noise-1 for small defect 5% and 10%.
The influence of the noise model, when using the Algorithm-1, was analyzed to evaluate best the C-scan map of S/N ratio of small defect detection. In Fig. 6a, it was observed that Noise-3 results are very close to Noise-1 data while Noise-2 and Noise-4 were overestimate and underestimate the S/N ratio as compared with Noise-1. Figure 7 shows a C-scan map when using Noise-2 and Noise-4. Based on the results, it was concluded that Noise-3 is the most suitable noise model that compare well with Noise-1. However, while Noise-1 can be defined only if it is known the location of a no-defect area, the Noise-3 is independent of the knowledge of no-defect area and also as the EMAT moves along an area with changing material parameters incorporate also the noise level associate with the specific area being investigated.
Therefore, next analysis it was based only on a noise distribution using the Noise-3. Another parameter that has a large effect on S/N ratio is the value of the Δt that selects only the information in the EMAT signal around an optimized value time of t0, in all points as EMAT moves along surface. All algorithms (1 to 4) for definition of EMAT signal are impacted by the values of Δt as presented in Fig. 8. Results show that for detection of small defect (5%) a value of Δt = 1∕𝜈 (1 μs in our specific case) is more appropriate and provides the same range of S/N ratio when using both Noise-1 and 3. The analysis was focused on obtaining reliable results of S/N ratio that are close to the results obtained when using known position of no-defect area (defined by Noise-1) and therefore only the intersection points of the continuous line (Noise-1) and dashed-line (Noise-3) were taken into consideration. Among all algorithms (1 to 4) it can be seen that the best S/N ratio for small defect detection is obtained with Algorithm-1. While Algorithm-2 and 3 displays a higher S/N ratio with larger Δt, the C-scan map shows additional peaks signal in the map, therefore it was concluded that the signal representation is not enough reliable. The Algorithm-4 results are the most conservative ones with smallest S/N ratio for small defect detection.
C-scan of 5% and 10% depth slit in 50 mm thickness plate using Algorithm-1 with Noise-2 (left) and Noise-4 (right).
Effect of the Algorithm-1 to 4 using 50 mV data and speed of 2.5 mm/s for definition of EMAT signal while noise signal is defined either with Noise-1 (continuous line) or Noise-3 (dashed line).
The previous analyses showed that the best options to build the C-scan map of S/N ratio of EMAT data is based on the Algorithm-1 and Noise-3. The results of S/N ratio are in agreement with the data when the location of defect and no-defect area are known (to define Noise-1), therefore validating the robustness of the Noise-3 that can be selected without any advanced knowledge of defect or no-defect areas. Using Algorithm-1 and Noise-3, the C-scan map of EMAT data for the small defects detection (5% and 10%) are shown in Fig. 9. If the EMAT speed is 2.5 mm/s a high S/N ratio (above 3.5) even for detection of 5% defect is obtained. Even for a large EMAT step (2 mm step scanning each in X and Y direction) S/N ratio does not degrade significantly, therefore reducing the number of points required to build the map and increasing the speed of the visualization analysis.
A C-scan of the sample surface is presented in Fig. 10, in order to illustrate the capability and sensitivity of the methodology to visualize the data for small defects 5% and 10% as well as large defects 20% and 50% depth located on the opposite side of thick metal plate.
C-scan of 5% and 10% depth slit in 50 mm thickness plate using Algorithm-1, Noise-3 at speed of 2.5 mm/s.
C-scan of 5%, 10%, 20% and 50% depth slit in 50 mm thickness plate using Algorithm-1, Noise-3 (speed 2.5 mm/s and step X = 1 mm using 1 Mohm coupling at 50 mV/div scale).
It was shown the feasibility of a new dynamic C-scan procedure using EMAT signal based on a Halbach permanent magnet for detection of small defects (up to 5%) located on opposite side of thick (50 mm) metallic plates. The new optimized combination of algorithm to define both EMAT signal and EMAT noise distribution during continuous scanning of surface is based on no knowledge of the defect area, the signal being self-calibrated and building directly the S/N ratio mapping, with a defect indicated for a value larger than 2. The paper shows that by carefully optimization of measurement methodology combined with new signal processing analysis of EMAT signals, highly sensitive C-scan maps of S/N ratio can be built for detection of small defects, enabling easy identification of defects.