Ultrasonics 54 (2014) 2072–2080

Contents lists available at ScienceDirect

Ultrasonics journal homepage: www.elsevier.com/locate/ultras

A row–column addressed micromachined ultrasonic transducer array for surface scanning applications Lawrence L.P. Wong, Albert I.H. Chen, Zhenhao Li, Andrew S. Logan, John T.W. Yeow ⇑ Advanced Micro-/Nano-Devices Lab., Department of Systems Design Engineering and Waterloo Institute for Nanotechnology, University of Waterloo, 200 University Avenue West, Waterloo, ON N2L 3G1, Canada

a r t i c l e

i n f o

Article history: Received 4 April 2014 Received in revised form 24 June 2014 Accepted 3 July 2014 Available online 10 July 2014 Keywords: Ultrasonic arrays CMUT Row–column addressed C-scan

a b s t r a c t Row–column addressed arrays for ultrasonic non-destructive testing (NDT) applications are analyzed and demonstrated in this paper. Simulation and experimental results of a row–column addressed 32 by 32 capacitive micromachined ultrasonic transducer (CMUT) array are presented. The CMUT array, which was designed for medical imaging applications, has a center frequency of 5.3 MHz. The CMUT array was used to perform C-scans on test objects with holes that have diameters of 1.0 mm and 0.5 mm. The array transducer has an aperture size of 4.8 mm by 4.8 mm, and it was used to scan an area of 4.0 mm by 4.0 mm. Compared to an N by N fully addressed 2-D array, a row–column addressed array of the same number of elements requires fewer (N instead of N2) pairs of interconnection and supporting electronic components such as pulsers and amplifiers. Even though the resulting field of view is limit by the aperture size, row–column addressed arrays and the row–column addressing scheme can be an alternative option of 2-D arrays for NDT applications. Ó 2014 Elsevier B.V. All rights reserved.

1. Introduction The objective of this paper is to propose an alternative 2-D array structure and control scheme, namely the row–column addressed array and the row–column addressing scheme, for non-destructive testing (NDT) applications. A capacitive micromachined ultrasonic transducer (CMUT) 2-D array was used to demonstrate the surface scanning ability of the row–column addressing scheme. The 32 by 32 CMUT array [1], with a center frequency of 5.3 MHz, was originally designed for medical imaging applications, more specifically for external ultrasound applications such as abdominal imaging. In this paper, the row–column addressed CMUT array is used to detect holes on a flat surface. Ultrasonic arrays are widely used to steer and focus sound beams in NDT applications. Many NDT applications use 1-D arrays to improve the scan flexibility and reduce the need of transducer movement [2]. For example, a 1-D array was used to inspect objects with complex geometry in an immersion setup [3], where the time delay of each element was adjusted according to the surface geometry. In two other examples, 64-element 1-D arrays were employed to increase the inspection speed of surfaces on aircraft [4,5]; in both designs, the arrays were immersed in a fluid-filled ⇑ Corresponding author. Tel.: +1 519 888 4567x32152. E-mail address: [email protected] (J.T.W. Yeow). http://dx.doi.org/10.1016/j.ultras.2014.07.002 0041-624X/Ó 2014 Elsevier B.V. All rights reserved.

probe, and C-scans were performed with the probe moving in one direction. It was concluded in [4] that the scan speed was limited by the time it took to maintain good contact between the probe and the scanned surface. For these examples, employing 2-D arrays can be beneficial because 2-D arrays reduce the frequency of transducer movement and further enhance the scan flexibility by providing an addition dimension where the sound beam can be steered and focused. However, the adoption of 2-D arrays for NDT has been slow [1]. The main obstacle faced by 2-D arrays, which scan volumes or surfaces, is the complexity of the imaging, or scanning, systems. For a system using a fully-populated N by N array, the best performance and flexibility can be achieved if each element in the array can be controlled individually. However, such a transducer requires the number of elements, as well as the number of connections to the array, to increase quadratically as the size of the array goes up. For example, a modestly sized 32 by 32 array requires over 1000 array controller channels, resulting in a complex design and making the control difficult. As a result, different 2-D array configurations and driving strategies have been proposed [2]. For instance, the Mills cross configuration (elements arranged in the shape of a cross) and the circular array (elements arranged in a circle) were investigated and compared with the fully-populated array by Mondal et al. [6]. It was concluded that given the same number of elements, circular arrays outperformed fully-populated

L.L.P. Wong et al. / Ultrasonics 54 (2014) 2072–2080

arrays in terms of beam directivity, and cross arrays in terms of side lobes. A 64-element segmented annular array was compared with a 68-element square array of similar size in [7] through simulation. It was found that the segmented annular array produced grating lobes that were on average 20 dB lower. Other NDT 2-D array examples include a 64-element ring array designed for farfield NDT imaging applications [8] and a sparse array, designed using conformal map theory, with an element count of 97 [9]. Thus far, the approach to adopt 2-D arrays for NDT systems has been reducing the number of active elements. However, the number of interconnects can still be reduced without sacrificing the element count, if multiple elements share the same connection in such a way that each element can still be addressed; it can be achieved using row–columned addressed 2-D arrays. Row–column addressed arrays were first proposed by Morton and Lockwood [10], who called the configuration a cross-electrode array. In 2009, a 256 by 256 row–column addressed 2-D array that was made from a 1-3 PZT composite was reported by Seo and Yen for rectilinear imaging [11]. More recently, CMUT Top Orthogonal to Bottom Electrode (TOBE) arrays, which is another name for row– column addressed arrays, were proposed for photoacoustic imaging [12]. Thus far, row–column addressed arrays have not received much attention in NDT research. Therefore, the goals of this paper are (1) to investigate the use of row–column addressed arrays as a practical and efficient solution for NDT, especially on surface scanning applications, and identify the limitations of these arrays; (2) to demonstrate surface scanning using a row–column addressed capacitive micromachined ultrasonic transducer (CMUT) array. This paper is organized as follows. The CMUT array and the operation of row–column addressed arrays are described in Section 2. The modeling of row–column arrays is presented in Section 3. Experiments and results are described and analyzed in Section 4. Finally, Section 5 discusses the potential of the row–column addressing scheme for NDT applications. While row–column addressed arrays can be implemented in piezoelectric technology [10,11], CMUTs are used in this work because large-area highdensity arrays can be more easily manufactured using the CMUT technology; however, the analysis presented is applicable to any row–column addressed arrays.

2. Row–column addressed CMUT arrays CMUTs, electrostatic transducers that are fabricated using micromachining techniques, have garnered a lot of research interest in the past two decades [13,14]. The operation of a CMUT is based on the vibration of a membrane. A voltage pulse applied across the CMUT electrodes causes the membrane to vibrate and generates sound pulses, while incoming sound waves displace the membrane and change the capacitance of the device, which can be detected as an output current. Because CMUTs are manufactured using micro-fabrication techniques, array elements with size in the order of micrometers can be realized. The CMUT array used in the experiment section was fabricated with a fusion bonding process that was reported in [15]. The fabrication process was similar to [16] except that silicon-on-insulator (SOI) wafers were not required. Silicon nitride, chosen as the membrane material, was deposited on two silicon wafers using low-pressure chemical vapor deposition (LPCVD). After the CMUT cavities were etched on the bottom wafer, the two wafers were bonded in a vacuum. Aluminum and polysilicon were used as the materials for the top and bottom electrodes, respectively. A detailed description of the fabrication process of the 2-D CMUT array can be found in [17]. Fig. 1 shows an image of the array. The image is a view of multiple array elements, with each element consisting of 30 CMUT cells. The squares in the image are bonding

2073

Fig. 1. Micrograph showing a section of the CMUT array.

pads. The bonding pads of adjacent columns or rows are located on opposite sides of the array. The characterization of the CMUT array was presented in [1]. The CMUTs have a center frequency of 5.9 MHz and a 6 dB bandwidth of 110% in immersion. The array consists of 32 columns and 32 rows, with a pitch of 150 lm in both directions. The array aperture is 4.8 mm by 4.8 mm. There are different ways to control a row–column addressed array transducer. The row–column addressing scheme, which was described in [1,10], uses the entire array for both transmit and receive operations. It delivers maximum acoustic energy, because the entire array is used, and is straightforward to implement because the operation is based on 1-D arrays. In [11], a subset of rows and columns were used for either transmit or receive beamforming. The elements always focus at the center, and beam steering is achieved by shifting the activated elements. This rectilinear imaging approach has the advantage that the signals received are highly uniformed, but it is only feasible for arrays of high element count (>100 in each direction) because the number of pixels in the resulting image cannot exceed the number of element in the array. A new addressing scheme that involves retroactive transmit focusing was proposed in [18]. This new approach provides an identical resolution as fully-addressed 2-D arrays in one direction (lateral in [18]), but the resolution in another direction is the same as the row–column addressing scheme. The new scheme requires selectively disabling certain rows and columns, thus it can only be implemented with CMUTs. In order for the discussion to be applicable also to piezoceramic transducers, and to keep the size of the array managable, the row–column addressing scheme is presented in this paper. The operation of the row–column addressing scheme was explained in [1,10,11] and it is further illustrated in Fig. 2. All the elements in the same column are connected through the top electrodes, and the bottom electrodes are connected in rows. If electrical pulses are applied to the columns when all the rows are connected to a constant bias voltage, the array becomes a 1-D array that generates a vertical line of focus, as shown in Fig. 2(a). On the other hand, if all the columns are connected together and each row is addressed individually, a rotated 1-D array that generates a horizontal focal line, as shown in Fig. 2(b), is produced. Instead of transmitting, the rotated array is in receiving mode; however, due to the principle of reciprocity, the effects on the beam profile can be considered the same regardless of whether the aperture is transmitting or receiving. As a result, if a row–column addressed array is configured such that a 1-D array is used to transmit and a rotated 1-D array is used to receive, the response is the convolution of two, vertical and horizontal, focal lines, resulting in a focal spot. Changing the location of the focal spot can then be achieved by adjusting the focal line locations of both the transmitting and the receiving operations. In summary, the row–column addressing scheme involves transmit beam-forming on one direction, for

2074

L.L.P. Wong et al. / Ultrasonics 54 (2014) 2072–2080

Fig. 2. Operating principle of row–column addressing scheme. (a) Transmit mode and (b) receive mode.

example the azimuth, and receive beam-forming on the other direction, for example the elevation. Row–column addressed 2-D arrays offer a few advantages over their individual-element addressed counterpart. First, the number of transmit and receive channels is greatly reduced. For example, a 32 by 32 array requires only 32 transmit-receive pairs instead of 1024. As a result, the number of connections going in and out of the transducer, the amount of analog electronic circuits, and the complexity of the digital signal processing algorithm are all reduced. In addition, row–column addressed arrays are easier to fabricate because of the reduction in connection pad counts. A CMUT implementation of individually-addressed 2-D array likely requires through wafer vias because the connection pads have to come out at the bottom of the wafer. On the other hand, the fabrication process of row–column addressed CMUT arrays is less challenging because the pads can come out on the sides of the array, as shown in Fig. 1. Row–column addressed arrays do have their limitations. It was shown through simulations in [18] that row–column addressed arrays have a signal-to-noise ratio (SNR) over 60 times, or 35 dB, worse than fully-addressed 2-D arrays of the same number of elements. However, the poor SNR can be improved by averaging, so it is not a major concern for NDT applications. The lateral resolution of row–column addressed arrays is also worse, in theory, by a factor of two because the transmitting and the receiving processes contribute to the lateral resolution on orthogonal directions; therefore the lateral resolution on either the azimuth or the elevation direction is affected by either transmit or receive beamforming, but not both. Finally, the field of view of row–column addressed arrays is also a disadvantage. The height, or length, of the focal lines in Fig. 2 defines the width of the field-of-view because a focal spot cannot be created at locations that cannot be reached by both focal lines. Moreover, the height of the focal line is limited to the aperture size within the near-field. Therefore, the field of view is limited to the size of the aperture, as long as the array lies on a flat surface. It was demonstrated in [19] that situating the array on a hemispherically shaped surface can increase the image sector angle from 16° to 60°. 3. Modeling A series of simulations was performed using Field II [20,21] so that the performance of the row–column addressing scheme can be visualized and better understood. The transmit aperture was defined as a 1-D array of 32 elements, with element height, element width and kerf set to 4.8 mm (19.8 k), 0.13 mm (0.54 k), and 0.02 mm (0.08 k) respectively. The receive aperture was defined similar to the transmit aperture but with a 90 degree

rotation. A 5.9 MHz sinusoid pulse in a Hanning window was used as the impulse response for each element. The modeling procedure described in this section is not specific for any kind of transducer technology. Even though Field II was designed for piezoelectric transducers, it has been proved that valid simulation results of CMUTs can achieved using the program [22]. The experiments described in the next section were performed in vegetable oil; therefore, a speed of sound of 1430 m/s is used. Vegetable oil was used to protect the transducer because water could create a short circuit between the exposed electrodes and pads of the CMUT array; however, sound travels at roughly the same speed in both water and vegetable oil, so the simulation results are still valid if water is used as the coupling medium instead. No apodization was applied to any of the simulations. The operating principle of row–column addressing scheme deserves to be revisited through modeling because the exercise will also provide insights on picking a suitable distance, or depth, between the transducer and the test object. The near-field to farfield transition of the 4.8 mm by 4.8 mm transducer is 32.2 mm in both the azimuth and the elevation directions. It is essential to place the test object in the near-field so that focusing can be done. Fig. 3(a) shows the transmit beam pressure profile 20 mm (82.5 k) away from the transducer when the row–column addressed array is focused at that distance. The focal line is at the center of the plot. Therefore, the center point, when both azimuth and elevation are at 0 mm, represents maximum pressure and has a value of 0 dB. Each line in the contour represents a 6 dB step. The 6 dB beam width and height are 1.6 mm and 4.7 mm respectively. The receive profile in Fig. 3(b) is just a 90-degree rotation of the Fig. 3(a) because of the principle of reciprocity; it is obtained from the transpose of the transmit data matrix. Combining the transmit and receive beam profiles results in a focal spot, as illustrated in Fig. 3(c). This third contour plot is the product, or the sum in dB, of the first two plots. The focal spot in Fig. 3(c) has a 6 dB beam width of 1.6 mm, consistent with the beam width in the transmit beam profile. The same simulation was repeated but with the focal depth set at 10 mm (41.2 k). The results are shown in Fig. 4. The transmit 6 dB beam width and height are 0.8 mm and 4.9 mm respectively. Changing the depth from 20 mm to 10 mm reduces the f-number of half, thus resulting in better focusing. The smaller, 0.8 mm, beam width translates into an overall focal spot of the same size. While it is beneficial to reduce the focal depth, configuration of the test system and the shape of the test object often dictate the minimum distance between the transducer and the test object. Moreover, image quality can be affected by a focal depth that is too small relative to the aperture. The minimum f-number depends on the medium, but as will be shown later, image artifacts start to appear

L.L.P. Wong et al. / Ultrasonics 54 (2014) 2072–2080

2075

Fig. 3. Simulated beam pressure profile at depth = 20 mm due to (a) transmit beamforming, (b) receive beamforming, and (c) a combination of transmit and receive beamforming; each line represents a 6 dB step.

Fig. 4. Simulated beam pressure profile at depth = 10 mm due to (a) transmit beamforming, (b) receive beamforming, and (c) a combination of transmit and receive beamforming; each line represents a 6 dB step.

when the f-number reaches 0.5. A focal depth of 10 mm is chosen as a reasonable distance to demonstrate the row–column addressed CMUT array in this work. A common modeling method for ultrasonic arrays is the computation of a point spread function (PSF) [1]. The PSF can be calculated in Field II by defining a point scatterer and performing a scan to determine how big the point scatterer appears. PSF modeling of row–column addressed arrays was reported before and the operation of the row–column addressing scheme had been proven [11,19]. Therefore, Field II simulation involving single point scatterer is not repeated here. However, for NDT applications such as the scanning of defects on a plate, a more intuitive way of modeling is to look at the reflection from different points on a plane when the focusing is fixed. Instead of just modeling the one-way pressure profile, two-way scans were performed with both the transmit and receive apertures focused at center and 10 mm away, or at (x, y, z) = (0, 0, 10) mm. The process was repeated with the point scatterer set at different points on the z = 10 mm plane. The maximum amplitudes of the received envelopes are plotted in Fig. 5. The received signals have higher amplitudes near the focal spot as shown in Fig. 5(a). When a surface scan, or a C-scan, is performed on a plate, the resulting signal will be the sum of all points in the contour plot. The contribution from each point is determined by the amplitude at that location. Therefore, a peak, as shown in Fig. 5(b), with a smaller top area and a steeper roll-off will result in a better lateral resolution. The 6 dB width of the peak in Fig. 5 is 0.8 mm. The simulation of moving point scatterer was repeated with the row–column addressed array focusing at other locations. Fig. 6(a)

shows a contour plot with the focal spot at (x, y, z) = (2, 2, 10) mm. The focal spot or the peak, as expected, is at the correct location. The 6 dB width is 0.8 mm, comparable with the case when the focal spot is at the center. The array was then set to focus at (x, y, z) = (4, 4, 10) mm, as shown in Fig. 6(b). The focal spot is again at the correct location, but the 6 dB width is much larger and the peak is not as sharp, as evidenced by the large distance between contour lines. What happens here is that the focal spot is outside the field of view of the aperture, thus the focusing power of the array gets worse. Comparing the row–column addressing scheme with regular 2-D array focusing shows the trade off between system complexity and image quality. Fig. 7(a) and (b) are contour plots when a regular 2-D array is used to focus at (x, y, z) = (0, 0, 10) mm and (4, 4, 10) mm respectively. The 6 dB width in the center-focused case is 0.6 mm, which is a slight improvement over the 0.8 mm achieved by the row–column addressed array. In addition, compared to Fig. 6(b), the regular array shows perfect focusing in area outside the size of the aperture. There is a 33% increase in beam width within the focal area, but the main trade off of using row– column addressing is the limited field of view. Using a larger array seems to be a logical solution to overcome the limited field of view, but a large aperture could affect image quality. Fig. 8 shows 3-D amplitude plots when row–column addressed arrays of various sizes are used to focus at (x, y, z) = (0, 0, 10) mm. When the aperture size is 10 mm by 10 mm, a single peak is still visible at the correct location. However, when the aperture size is increased to 20 mm by 20 mm and 40 mm by 40 mm, unwanted peaks, which will result in image artifacts, appear. These

2076

L.L.P. Wong et al. / Ultrasonics 54 (2014) 2072–2080

Fig. 5. Simulated amplitudes of received signals from scatterers on a plane 10 mm from the array, the focal spot is set at (0, 0); (a) contour plot with lines representing 6 dB steps, and (b) 3D plot.

Fig. 6. Contour plots of simulated maximum received signals from scatterers on a plane at depth = 10 mm, the focal spot is set at (a) (2, 2) and (b) (4, 4); each line represents a 6 dB step.

peaks are caused by diffraction because focusing is only done in one direction (azimuth for transmit and elevation for receive), while the focal point is well in the near field in the other direction. This problem can be avoided by increasing the focal depth, as illustrated in Fig. 9, or by applying defocusing to the array as suggested by [19]. 4. Experimental results The row–column addressed CMUT array was used to detect flaws on surfaces. The test objects for the experiment are made of polymethyl methacrylate (PMMA), which is also known as acrylic glass. The defect on each test object was mimicked by a hole with a diameter d, which was formed using a laser cutter. A diagram of the test object is illustrated in Fig. 10(a). Fig. 10(b) shows a piece of PMMA with a 1 mm-diameter hole (d = 1 mm). The larger hole on the right is for attaching the test object to a translation stage. The piece of acrylic glass has a thickness of 3 mm. The experimental setup is shown in Fig. 11. The 32 by 32 CMUT array was placed on a custom designed printed circuit board (PCB),

along with the required front end electronic circuits. A vegetable oil container was built on the PCB. Again, vegetable oil was used to prevent damaging the transducer because the conductive top electrodes of the CMUT array are exposed. This problem can be avoided in the future when an insulating protective layer is put on top of the CMUT array. The PMMA test object, supported by a translation stage, was placed in the vegetable oil, 10 mm away from the transducer. The 10 mm distance corresponds to 41.2 k in vegetable oil. It can be translated into different distances depending on the applications. For example, if water is used as the coupling medium in an application similar to [4], the height of the water column should be close to 10 mm. If a row–column addressed pizeoceramic array is used for direct contact inspection of steel, assuming a speed of sound of 6000 m/s in steel, the distance between the transducer and the defects should be around 40 mm for the same transducer aperture size. The transmit beamformer was implemented with a Spartan-3 FPGA (Xilinx Inc., San Jose, CA) that ran on a synthesized clock frequency of 250 MHz. The array was programmed to focus on a plane 10 mm away. The focal line was set to 2.0 mm to 2.0 mm in the

L.L.P. Wong et al. / Ultrasonics 54 (2014) 2072–2080

2077

Fig. 7. Similar contour plots using a regular individual-element addressed 2-D array, the focal spot is set at (a) (0, 0) and (b) (4, 4); each line represents a 6 dB step.

Fig. 8. Simulation results when a row–column addressed array is focused at (x, y, z) = (2, 2, 10) mm, size of the aperture is (a) 10  10, (b) 20  20, and (c) 40  40.

Fig. 9. Simulation results when a row–column addressed array of size 40 mm by 40 mm is focused at (x, y) = (2, 2)mm, with depth at (a) 20 mm (b) 40 mm and (c) 80 mm.

azimuth direction, in 0.1 mm steps. The CMUT array outputs were connected to transimpedance amplifiers (based on OPA657, Texas Instruments, Dallas TX) with a gain of 10 kX, and the output voltages were recorded using a digital oscilloscope (DSO7104B, Agilent

Technologies, Santa Clara CA). Each waveform was averaged 32 times and was stored with a time-step of 5 ns. Receive beamforming that involves delaying and summing and the subsequent steps were done in Matlab (The Mathworks Inc., Natick, MA). The receive

2078

L.L.P. Wong et al. / Ultrasonics 54 (2014) 2072–2080

Fig. 10. (a) A 3-D illustration and (b) an image of the test object. The hole diameter, d, is 1 mm in the image; the larger hole on the right is for fastening purpose.

Fig. 12. Measured image of a 1 mm hole on a piece of PMMA 10 mm away from the transducer. Fig. 11. A photograph of the experimental setup.

focus was also set to 2.0 mm to 2.0 mm in 0.1 mm steps, but in the elevation direction. Hilbert transform was then applied to the resulting waveforms. Finally, the maximum amplitudes of the envelopes were plotted in a logarithmic scale. No apodization was applied. The total processing time in Matlab was less than 1 min using a system with a duo-core 2.9 GHz CPU and 3 GB of RAM. Fig. 12 shows the scanned image of the PMMA with a 1 mm hole, 10 mm from the transducer. Again, transmit beamforming was done in the azimuth direction (left to right), and the receive beamforming was done in the elevation direction (bottom to top). The dark circle, pointed by an arrow, is clearly visible in the middle, centering at around azimuth = 0.25 mm and elevation = 0.5 mm, but there are some dark areas along the sides of the image. These dark areas appear for two reasons. First, because points in the beam pressure profile, as shown in Fig. 4(a), do not have constant amplitudes, the maximum received signal will be different as the scan angle changes. Focusing at the center will result in a larger signal compared to focusing at the corner, because the beam focal line has a maximum pressure at the center. However, this effect only accounts for a maximum difference of less than 10 dB, given the chosen scan area of 2 mm by 2 mm. A second factor that makes a bigger contribution to the dark areas is the acoustic reflectivity of the test object. When the sound waves hit the test object on an angle, a large portion of the wave reflects away from the source. The further away the focal spot is from the center, the more likely that the reflected wave cannot get back to the transducer surface, thus resulting in a smaller signal. This effect is more pronounced in the azimuth direction, which is the direction of the transmit beam steering.

The dark areas do not affect flaw detection accuracy if the hole is at the center of the aperture. However, if the hole is located close to the edge of the aperture, as shown in Fig. 13(a), identification of the hole can become difficult. Fortunately, it is possible to model the two effects and apply compensations to the image. Because the reflectivity of the test object is highly dependent on the surface roughness, which cannot be predicted accurately and consistently for different materials, the simplest solution to compensate for the uneven received signal amplitudes is to use a reference object. While the use of a reference may not be feasible in medical imaging, it is reasonable to assume that reference objects are available in NDT applications. A perfect piece of PMMA, with no holes, was imaged and the received signals were stored as the reference levels, as shown in Fig. 13(b). Fig. 13(c) shows the result of adjusting the raw data with the reference levels. The adjusted data was taken as the difference between the raw data and the reference data. It was then linearly shifted such that the maximum value stayed at 0 dB. The compensated image shows a better defined hole shape, thus proving that the CMUT array can capture holes that are offcenterd. The experiment was repeated with another piece of PMMA with a smaller hole. The hole has a diameter of 0.5 mm, which is smaller than the 6 dB focal spot size of the array (0.8 mm). Fig. 14(a) and (b) show the raw data and compensated images of the 0.5 mm hole respectively. The hole location can still be identified (centering at around azimuth = 0.5 mm and elevation = 1.0 mm), but the circular shape of the hole is not as well defined compared to Fig. 13(c) and the hole size in the image is greater than 0.5 mm. When the hole diameter is smaller than the 6 dB focal spot size, a significant amount of acoustic energy still gets reflected even if the focal spot

L.L.P. Wong et al. / Ultrasonics 54 (2014) 2072–2080

2079

Fig. 13. Measured (a) raw data, (b) reference data, and (c) compensated images (difference of (a) and (b)) of a 1 mm hole on a piece of PMMA 10 mm away from the transducer.

is near the hole, thus the contrast between the hole and normal region is reduced. The 6 dB focal spot size, or the 6 dB beam width when only one-way operation is considered, can be used as a good rule of thumb in determining the minimum hole size that the array can characterize.

5. Discussion CMUT is chosen as the transducer technology for the row– column addressed array in this paper. It is a good candidate because of the need of large-area arrays row–column addressed arrays. A big limitation of row–column addressing is its small field of view: the array can only inspect an area of roughly the size of the aperture. Therefore, it is beneficial to use a large-area array to increase the field of view. Moreover, in order to avoid grating lobes, a high-density array is required for high frequency applications. CMUTs seem to be a good match with this requirement because they are manufactured using micromachining processes, thus suitable for the fabricating large-area and high-density arrays. The CMUT array can only be used in immersion or with a coupling medium between the transducer and the test object. For direct contact NDT applications, other technologies such as piezoceramic transducers can be used. Nevertheless, the analysis presented in this paper still provides insight for the design of non-CMUT row–column addressed arrays. As mentioned above, the biggest shortcoming of the row– column addressing scheme is its limited field of view, which is much smaller than regular 2-D focusing. However, if the surface of the test object is smooth and highly reflective, area outside of the aperture cannot be imaged even with a regular 2-D array.

While row–column addressed arrays are not going to replace regular 2-D arrays, they can be useful in certain applications. The control of row–column addressed arrays is not limited to using the entire aperture to steer the sound beam, as described in this paper. Seo and Yen demonstrated a 256 by 256 row– columned addressed array that used rectilinear scanning to move the sound beam [11]. Rectilinear scanning eliminates the uneven signal strength problem that was encountered in the experiment section. As shown in the modeling section, while a large array is preferred to increase the field of view, a large aperture dimension compared to the focal depth will affect the image quality. If moving the object away from the array is not possible, one can divide the array into subsections, perform scanning in each of them, and combine the results to form a bigger image. In addition, the amount of data that needed to be processed is greatly reduced. Finally, row–column addressed CMUT arrays have the potential to enhance current NDT systems. A popular signal processing technique in commercial ultrasonic NDT systems is the combination of full matrix capture (FMC) and total focusing method (TFM) [23]. FMC and TFM involve transmitting with one element at a time and then capturing the received signals of all elements due to that one element. When the cycle involving the transmission of all elements is complete, a matrix that contains all possible transmitreceive element pairs can be used to focus at every point. Using CMUTs, any one element in the array can be activated because a CMUT requires both a bias voltage and a pulsing signal to generate sound pulses. As a result, regular 2-D arrays can be replaced by row–column addressed CMUT arrays for the sole purpose of FMC and TFM. While the acquisition time will be increased, combining row–column addressed CMUT arrays with FMC and TFM reduces the number of interconnects and the hardware system complexity.

Fig. 14. Measured (a) raw image and (b) compensated image of a 0.5 mm hole on a piece of PMMA 10 mm away from the transducer.

2080

L.L.P. Wong et al. / Ultrasonics 54 (2014) 2072–2080

6. Conclusion In this paper, row–column addressed arrays and the row– column addressing scheme are proposed for surface scanning NDT applications. Row–column addressing scheme performs transmit beamforming in one direction and receive beamforming in an orthogonal direction; doing so reduces the number of interconnects and the amount of support circuits significantly. The operation of row–column addressing scheme is illustrated through modeling. The biggest limitation of a row–column addressing scheme, compared to a regular 2-D focusing, is its limited field of view, which is about the size of the aperture. Using a larger array will increase the field of view, but a large aperture to depth ratio could affect the image quality. In addition, a row–column addressed CMUT array was used to obtain measurement results in surface scans, or C-scans, of defects on pieces of PMMA. The main advantage of using the CMUT technology is that large-area and high-density arrays can easily be made, thus mitigating the limited field of view shortcoming of row–column addressed arrays. In addition, elements of a CMUT row–column addressed array can be activated individually, allowing implementation of different control strategies such as FMC and TFM. Implemented using the CMUT technology or any other technologies, row–column addressed arrays offer a unique set of trade-off compared to regular 2-D arrays. While row–column addressed arrays are not meant to replace regular 2-D arrays, they provide another option for designers when simpler systems are preferred. Acknowledgments This work is supported by the Natural Sciences and Engineering Research Council of Canada, the Canadian Institutes of Health Research, the Waterloo Institute for Nanotechnology, and the University of Waterloo. Fabrication of the CMUTs was performed at the Cornell NanoScale Facility, in Ithaca NY, a member of the National Nanotechnology Infrastructure Network, which is supported by the National Science Foundation (Grant ECS-0335765). Lawrence Wong gratefully acknowledges funding support from the Waterloo Institute for Nanotechnology in the form of Nanofellowship. References [1] A.S. Logan, L.L.P. Wong, A.I.H. Chen, J.T.W. Yeow, A 32  32 element row– column addressed capacitive micromachined ultrasonic transducer, IEEE Trans. Ultrason., Ferroelect, Freq., Contr. 58 (6) (2011) 1266–1271. [2] B.W. Drinkwater, P.D. Wilcox, Ultrasonic arrays for non-destructive evaluation: a review, Nondestr. Test Eval. Int. 39 (2006) 525–541.

[3] S. Mahaut, O. Roy, C. Beroni, B. Rotter, Development of phased array techniques to improve characterization of defect located in a component of complex geometry, Ultrasonics 40 (2002) 165–169. [4] R.A. Smith, J.M. Bending, L.D. Jones, T.R.C. Jarman, D.I.A. Lines, Rapid ultrasonic inspection of ageing aircraft, Insight 45 (3) (2003) 174–177. [5] C.J. Brotherhood, B.W. Drinkwater, R.J. Freemantle, An ultrasonic wheel-array sensor and its application to aerospace structures, Insight 45 (11) (2003) 729– 734. [6] S.C. Mondal, P.D. Wilcox, B.W. Drinkwater, Design of two-dimensional ultrasonic phased array transducers, J. Pressure Vessel Technol. 127 (2005) 336–344. [7] O. Martinez, M. Akhnak, L.G. Ullate, F.M. de Espinosa, A small 2D ultrasonic array for NDT applications, Nondestr. Test Eval. Int. 36 (2003) 57–63. [8] S.N. Ramadas, J. Dziewierz, R.L. O’Leary, A. Gachagan, A. Velichko, P.D. Wilcox, An annular array with fiber composite microstructure for far field NDT imaging applications, Proc. IEEE Ultrason., Symp. (2009) 2557–2560. [9] S.N. Ramadas, J.C. Jackson, J. Dziewierz, R. O’Leary, A. Gachagan, Application of conformal map theory for design of 2-D ultrasonic array structure for NDT imaging application: a feasibility Study, IEEE Trans. Ultrason., Ferroelectr., Freq. Control 61 (3) (2014) 496–504. [10] C.E. Morton, G.R. Lockwood, Theoretical assessment of a crossed electrode 2-D array for 3-D imaging, Proc. IEEE Ultrason. Symp. (2003) 968–971. [11] C.H. Seo, J.T. Yen, A 256  256 2-D array transducer with row–column addressing for 3-D rectilinear imaging, IEEE Trans. Ultrason., Ferroelectr., Freq. Control 56 (2009) 837–847. [12] R.J. Zemp, R. Chee, A. Sampaleanu, D. Rishi, A. Forbrich, S-sequence biasencoded photoacoustic imaging with top Orthogonal to Bottom Electrode (TOBE) CMUT arrays, Proc. IEEE Ultrason. Symp. (2013) 1197–1200. [13] D.M. Mills, Medical imaging with capacitive micromachined ultrasound transducer (CMUT) Arrays, Proc. IEEE Ultrason. Symp. (2004) 384–390. [14] O. Oralkan, A.S. Ergun, J.A. Johnson, M. Karaman, U. Demirci, K. Kaviani, T.H. Lee, B.T. Khuri-Yakub, Capacitive micromachined ultrasonic transducers: nextgeneration arrays for acoustic imaging?, IEEE Trans Ultrason., Ferroelectr., Freq. Control 49 (2002) 1596–1610. [15] A. Logan, J.T.W. Yeow, Fabricating capacitive micromachined ultrasonic transducers with a novel silicon-nitride based wafer bonding process, IEEE Trans. Ultrason., Ferroelectr., Freq. Control 56 (2009) 1074–1084. [16] Y. Huang, A.S. Ergun, E. Haggstrom, M.H. Badi, B.T. Khuri-Yakub, Fabricating capacitive micromachined ultrasonic transducers with wafer-bonding technology, IEEE J. Microelectromech. Syst. 12 (2003) 128–137. [17] A.S. Logan, L.L. Wong, J.T.W. Yeow, 2-D CMUT wafer bonded imaging arrays with a row–column addressing scheme, Proc. IEEE Ultrason. Symp. (2009) 984–987. [18] A. Sampaleanu, P. Zhang, A. Kshirsagar, W. Moussa, R.J. Zemp, Top Orthogonal to Bottom Electrode (TOBE) CMUT arrays for 3D ultrasound imaging, IEEE Trans. Ultrason. Ferroelect. Freq. Control 61 (2) (2014) 266–276. [19] C.E.M. Démoré, A. Joyce, K. Wall, G.R. Lockwood, ‘‘Real-time volume imaging using a crossed-electrode array, IEEE Trans. Ultrason. Ferroelect. Freq. Control 56 (6) (2009) 1252–1261. [20] J.A. Jensen, N.B. Svendsen, Calculation of pressure fields from arbitrarily shaped, apodized, and excited ultrasound transducers, IEEE Trans. Ultrason., Ferroelect., Freq. Control 39 (6) (1992) 262–267. [21] J.A. Jensen. Field: a program for simulating ultrasound systems. Med. Biol. Eng. Comp., in: 10th Nordic-Baltic Conference on Biomedical Imaging vol. 4(1). 1996b, pp. 351–353. [22] D. Baek, O. Oralkan, M. Kupnik, M. Willatzen, B.T. Khuri-Yakub, J.A. Jensen, Simulating capacitive micromachined ultrasonic transducers (CMUTs) using field II, Proc. IEEE Ultrason. Symp. (2010) 439–442. [23] C. Holmes, B.W. Drinkwater, P.D. Wilcox, Post-processing of the full matrix of ultrasonic transmit-receive array data for non-destructive evaluation, Nondestr. Test Eval. Int. 38 (2005) 701–711.