Development and validation of an ultrasound imaging software for nondestructive evaluation of concrete pavements

: A Synthetic Aperture Focusing Technique-based ultrasound imaging software was developed and validated in this work. The software development process was discussed in detail. For validation, Jointed Plain Concrete Pavements were imaged with the evaluated software and a software assumed to yield standard images. The quality of the standard and evaluated images were analyzed and compared. The backwalls and dowels were recognizable in the standard and evaluated images. The thickness estimates obtained with each software was the same. The characteristics of objects and artifacts were explained in terms of the technology’s fundamentals and its implementations’ characteristics. The evaluated software resulted, therefore, in images of similar quality to the standard software’s images from a practical standpoint. The herein developed software was then validated for research on nondestructive evaluation of concrete pavements and can be used for testing and development of performance-improving signal and image processing techniques.


INTRODUCTION
To select a method for nondestructive evaluation (NDE) of a given concrete structure, the penetration depth, resolution, signal-to-noise ratio (SNR), and application record of prospective methods should be taken into consideration [1]. Based on this, a good method for Jointed Plain Concrete Pavement (JPCP) NDE would be ultrasound imaging (UI) with the Synthetic Aperture Focusing Technique (SAFT).
The SAFT is an UI algorithm that tackles resolution and noise-related issues hindering conventional concrete UI and offers a better compromise among SNR, resolution, and penetration depth [2], [3]. SAFT UI has been successfully used to recognize steel bars, ducts, delamination, air voids, and cracks in concrete pavements [4]- [7] and other concrete structures [8]- [10].
Nevertheless, JPCP NDE with SAFT UI has been hindered by efficiency and accuracy-related issues [11]. It could be bolstered, therefore, with signal and image processing techniques that improve its efficiency and accuracy [12]- [14]. The development and application of such techniques both require, however, a validated and extensible SAFT UI software; that is, a software that outputs good-quality images within the SAFT's limitations at any time and where those techniques could eventually be plugged in.
A SAFT UI software with these characteristics was developed and validated in this work. The software development process was presented in detail. JPCPs were imaged with the herein developed software and the SAFT UI software accompanying the data acquisition device used in this work. The quality of their images was evaluated from a practical standpoint, and the similarities and differences in image quality were used to validate the herein developed software assuming that the other one yields standard images. The characteristics of the objects and artifacts in the images were discussed to explain image quality and differences between the software.
As previously mentioned, the development of such solutions is requisite for employing or developing performanceimproving techniques and therefore, to make the most of the technology for a given application. This work's discussion on image characteristics and their relationship to the algorithm's fundamentals and its implementation's characteristics should be relevant for developing and validation similar solutions for any application. Nevertheless, a general framework for validating such solutions is not presented in this work.
The SAFT is reviewed in the next section. The herein developed software is presented subsequently. The validation process and its results are presented and discussed next, followed by this work's conclusions.
The SAFT yields higher-quality images because the best achievable resolution is achieved everywhere in the image plane [15] and reflecting objects, e.g. backwall, steel bars, and defects, are enhanced, whereas noise, including structural noise, is reduced [2], [3]. In Equation 3, the amplitudes of different signals mapped to the same position are superimposed. If these amplitudes are in phase, as expected for reflecting objects, then there will be constructive interference among them and enhancement of such objects in ( ) f x, y . The opposite takes place for the background noise, since noise components in different signals are generally not in phase [2], [16]- [18]. Therefore, SNR and resolution are enhanced [2], [3], [19].
Nevertheless, from a practical standpoint, image quality might be hindered by the SAFT's assumptions. The pulse velocity c is not constant across the image plane since concrete is a dispersive medium [3]; hence, objects might be shown with wrong shape or position in ( ) f x, y [20]. The implications of attenuation are disregarded; hence, the brightness of objects in ( ) f x, y might not be proportional to their elastic properties [13]. Consequently, objects might not be shown as expected and their recognition might be hindered despite the SNR and resolution enhancements.
Moreover, every signal component is assumed a one-time reflected wave. Therefore, artifacts might arise from phenomena such as multiple reflections [9] and surface waves [13], [21] that make up the signals but are disregarded. Further artifacts might arise from mode conversion [3] and acoustic shadowing [4]. Acoustic shadowing occurs when objects of interest behind reflecting objects (e.g. steel bars) are not shown in ( ) f x, y because not enough energy was transmitted across these objects [20].

SOFTWARE DEVELOPMENT
The ( ) P x, y should be proportional to inspection system's sampling frequency [22]. The size of our ( ) for each ( ) i g t should be known with precision. The data acquisition device used in this work was an array with fixed elements and known operation mode; therefore, i T and i R were easily determined from the device's characteristics. As the transducers are on the structure's surface, it was assumed that T R y y 0 = = . The value of c should be assumed or estimated with testing or signal processing. The c value estimated by the data acquisition device from the arrival time of surface waves was used in this work [14]. The values of * t determined from Equation 2 are real but the values of t are integer multiples of the sampling period. The floor function was selected to determine the sample of ( ) i g t corresponding to * t . The signals were not rectified, to preserve the phase information used in Equation 3 [23], and were not pre-processed either.
The SAFT assumes in Equation 2 that the entire concrete volume was insonified by every i T [17]. Since this is not true, the windowing function ( ) i w x, y in Equation 4 [12] was used to consider the volume portion that was insonified by each i T . Essentially, Additionally, the upper envelope of each column of ( ) f x, y was determined with the Hilbert transform and taken as its new values [12]. ( ) f x, y was then transformed into a grayscale image, outputted, and displayed with a color scheme where pixel brightness is proportional to color warmness. Though the signals were not processed prior to mapping and ( ) f x, y was not processed after transformation with performance-improving techniques, modules for this could be easily added to the software.

Data acquisition
The JPCP test section (TS) and the concrete slab (CS) imitating doweled regions of JPCPs in Figure 2 were both imaged in this work. TS comprised 250 mm-thick concrete slabs, was built in 1999, and, according to visual inspections [24], was in sound condition at data acquisition. Its as-built thickness was measured with coring right after data acquisition. The CS was 190mm thick (in design and after construction) and had three 25 mm steel bars inserted 54, 74, and 94 mm deep. It was built in laboratory in 2019; dowel positioning was ensured by tying the dowels to trusses and the trusses to the forms. The TS' and CS' concrete mixes are presented in Table 1.  Data was acquired on six points on the TS in 2018 and nine points on the CS in 2019. The TS' points were on plain regions. The CS' points were three over each dowel's axis. The number of points was deemed sufficient since this work aimed to validate the software for research purposes and the scenarios of interest were restricted to plain and doweled regions of sound JPCPs. A larger and more diverse set of points should be used to validate similar software for practical purposes or for a large set of scenarios. A practical obstacle for increasing the number and diversity of points for any purpose is the fact that a structure's as-built characteristics must be known with precision for validation.
The data acquisition process was the same for every point. The area where the data acquisition device would be placed was swept free of debris and a point was marked with a construction pencil. The device was positioned centered on the point and data acquisition was carried out. A coupling layer or surface preparation was not required. The acquired data was labelled, stored in the device, and eventually transferred to a personal computer for processing.
ACSYS' A1040 MIRA was the data acquisition device. It comprises a 4-by-12 array of transducers operated as a 1-by-12 one. Its transducers are 30 mm apart and transmit 50 kHz shear-horizontal waves. Its operation mode is illustrated in Figure 1. Each transducer, except the rightmost one, is operated in transmission ( T ) once, while the transducers to its right are in reception 5/11 mode ( R ). The transducer T is switched from left to right to sweep the concrete volume beneath the device. A signal ( ) i g t is acquired for each T -R pair. I = 66 signals are acquired on each point then. The device also estimates the pulse velocity c in the inspected concrete volume with the time-of-flight of surface waves [14].

Validation procedure
The TS' and CS' data were inputted to the herein developed software and the SAFT UI software accompanying the data acquisition device. The latter was assumed to result in standard images since it was developed for practical purposes. Its images are hereafter called standard images, while the images obtained with the software being validated are called evaluated images.
The herein developed software was evaluated by comparing the quality of its images to the standard ones. The images were analyzed from a practical standpoint. A good-quality image was assumed to be one where the backwall and dowel were recognizable and positioned correctly. An object was assumed recognizable if it was shown with the expected shape and brightness and around their expected positions. Artifacts were expected and considered in the analysis. The TS' and CS' thicknesses were measured from the images by taking the vertical position of the backwall's brightest pixel as the thickness estimate [11], [14] and considered in the analysis.
The differences in image quality were used to infer differences between the software; namely, to enumerate potential issues with the evaluated software and signal and image processing techniques potentially used by the standard software but not by the evaluated one. The information presented above and in the literature was considered for this purpose.

RESULTS AND DISCUSSION
The TS' standard and evaluated images in Figure 3 both show a backwall and multiple-reflection artifacts. The backwalls are the 200-250 mm deep objects. The backwall was recognizable in both images because it was a bright and wide object at depths roughly the concrete element's as-built thickness (240.50 mm). Its brightness is due to the slabbase course interface's large reflection coefficient, and its width is due to the interface being sensed by every T -R pair. The standard image's backwall seems larger than the evaluated image's one because the former is presumably saturated. The multiple-reflection artifacts are the 425-475 mm deep objects ( Figure 3). As the SAFT does not recognize multiple reflections, the waves bouncing between backwall and surface (multiple reflections) are computed as reflections at ever deeper backwalls and represented as mirror images of the backwall. The mirror images are not carbon copies of the backwall due to attenuation.
The CS' standard images in Figures 4-6 show a dowel, a backwall, and multiple-reflection artifacts, and its evaluated images shown a dowel, a backwall, surface-waves artifacts and multiple-reflection artifacts. The backwall was recognizable in every image. The multiple-reflection artifacts comprise mirror images of the backwall and dowel.
The dowel is bright, round, and at the expected position in the evaluated images and bright and at the expected position, but not round, in the standard ones (Figures 4-6). The dowels were nevertheless recognizable in every image. The shape information on the mirror images in the standard images were used to recognize the dowels. Artifacts, if well understood, might therefore provide some information on the inspected structure.   The backwalls should be shown brighter than the dowels since the slab-base course interface's reflection coefficient is larger than the slab-dowel's coefficient. Attenuation prevented this in the evaluation images in Figures 4-6. The T -backwall path is larger than the T -dowel path. Hence, the waves striking the backwall have smaller intensity, and the waves reflected at the backwall might have smaller intensity despite the backwall's larger reflection coefficient. In addition, the waves reflected at the backwall are attenuated to a greater degree on their way to R . As a result of this and the fact that attenuation has not been compensated in the evaluated software, the dowels are shown brighter than the backwall in the evaluated images. Figures 4-6 due to saturation and presumably attenuation compensation. The multiple-reflection artifacts in the standard images are brighter and wider in Figures 3-6 for the same reasons.

The dowels and backwalls have similar brightness in the standard images in
The surface-waves artifacts are shown above the dowels in the evaluated images in Figures 4-6. As the SAFT does not recognize surface waves, it computes the waves travelling on the concrete structure's surface from T to R as reflected waves in the inspected volume's interior, resulting in those artifacts. The surface waves are unavoidable. The brightness distribution in Figure 3 occluded them. The standard software has presumably filtered them out as the artifacts are not presented in the standard images in Figures 4-6.
The TS and CS were roughly 220 mm and 170 mm thick respectively according to Figures 3-6. As the same thickness measure was obtained with the standard and evaluated images, the evaluated software does not result in gross quantitative errors for the inspected scenarios. The thickness estimates' inaccuracy was due to the c estimates' inaccuracy. The c estimate's accuracy was disregarded for this work.
The standard and evaluated images obtained for the other points are presented in Figures 7-11. The observations above are therefore true for all standard and evaluated images obtained int this work. Mode-conversion artifacts were not observed in this work because shear-horizontal waves are not mode converted. Acoustic shadowing was not observed in this work either, despite being reported for similar scenarios [4].

CONCLUSIONS
A SAFT UI software for JPCP NDE was developed and validated in this work. The SAFT's fundamentals were reviewed and every step required to implement the SAFT was discussed. The validation was carried out by comparing the JPCP images obtained with the herein developed software to the ones obtained with a software assumed to yield standard images. The backwalls and dowels (steel bars) were recognizable in both evaluated and standard images. The characteristic of these objects and artifacts in both images were discussed and considered in the validation. The thickness estimated from the evaluated and standard images were the same. Therefore, from a practical standpoint, the quality of the images obtained with the herein developed software and the standard one was essentially the same and no gross quantitative errors were observed in the evaluated images. The differences between the evaluated and standard images were presumably due to further signal and image processing steps in the standard software and not due to errors in the evaluated software. The herein developed software was therefore validated for JPCP NDE, though restricted to the scenarios represented by the inspected structures. The software can nevertheless be used for testing and development of solutions to tackle SAFT UI's issues.