Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
1System Description This chapter was prepared by 3d-Radar and describes how the 3d-Radar system was improved for the field evaluation in Task 7. Figure 1.1 shows the improved system assembled behind a pickup truck before the uncontrolled field evalua- tion in Florida. The following sections briefly describe each component of the system in detail. GeoScope The 3d-Radar GeoScope emits a step frequency continuous waveform, which is a sine wave with constant amplitude and stepwise frequency variation. The waveform is speci- fied by a start frequency, a stop frequency, a frequency step, and a dwell time (amount of time spent transmitting a sine wave at a given frequency). A laptop is necessary to set up the GeoScope and visualize the data stream, via an Ethernet connection, while preprocessing and data storage take place within the GeoScope. Data are displayed in the three dimensions (inline and crossline B-scans; C-scan) directly during the acquisition in navigable panels in real time. Antenna Array An array model B3231 was used for the acquisition. The system makes it possible to collect up to 31 parallel B-scans simultaneously, with a minimum spacing of 10 cm (approxi- mately 4 in.). The array has an overall width of 3.2 m (approx- imately 10.5 ft) and is connected to the GeoScope through a single cable. Vehicle Mount A special tower mount has been designed to quickly shift between data acquisition and system transportation. The mount consists of a cylinder fixed to the hitch receiver of the vehicle, connected on top to a rotatable cylinder. The array is attached to a frame that can be raised and lowered along the main tower by an electrical hinge. During data acquisition, the array is lowered and perpendicular to the vehicle, cov- ering most of a traffic lane. During transportation between survey sites, the array can be lifted and turned 90°, thus mini- mizing possible traffic safety issues. System Settings To determine the speed of the survey, the time required to perform a complete array scan, which is sending and receiving through all the active channels along with the inline sampling, needs to be calculated. 3d-Radar provides an Excel-based utility to assist with these calculations to define the maximum surveying speed during the acquisition. As summarized in Table 1.1, three different system configurations were tested, allowing for different maximum vehicle speeds. The objective was to implement configurations to match the purposes of a routine control (high speed), a more careful evaluation (medium speed), and an in-depth assessment of the road substructure (slow speed). For this phase of the testing, it was preferred to keep the same frequency characteristics of the signal in terms of bandwidth and sampling (frequency step), while decimating the spatial sampling moving from a slow-speed to a high-speed acquisi- tion. The high-speed settings meet the minimum requirement of 1 sample per foot. The bandwidth (BW) allows a resolution: 1 1.36 ns BW = When a relative permittivity value of 5 as average for asphalts is used, the spatial resolution on the shallow subsurface of a road structure is approximately 2.39 cm (0.94 in.). C h a p t e r 1 3d-Radar Report
2real-time processing and postprocessing The real-time processing sequence consists of ⢠Inverse Fast Fourier Transform (including frequency filter); ⢠Background subtraction; ⢠Time windowing; and ⢠Range gain compensation. These operations are performed on all active channels, with user-selectable parameters, and allow for a full 3-dimensional visualization of the data cube directly during the acquisi- tion phase on the laptop that controls the GeoScope ground- penetrating radar (GPR) device. The steps listed above are the basic steps necessary to pro- vide a classical time domain visualization of the radar data. Because the GeoScope is a step-frequency GPR, it is natural to have data recorded in frequency domain, because what is measured is very similar to a frequency response of the subsurface. Postprocessing through 3d-Radar Examiner touches the same processing steps as does the real time, plus some other tasks that can be fine-tuned to improve visualization or pro- duce an output for specific applications (i.e., the delamina- tion detection algorithm designed for this project). Given the wide bandwidth that can be recorded with the system and the variety of scenarios that are faced in GPR survey work, data stored in frequency domain represents a versatile, raw starting point to build a solid, fine-tuned processing sequence to produce the time domain synthetic section that will be the basis for interpretation and analysis. Nonetheless, the user has an either-or option for storing the acquired data in time domain, exactly as captured during real-time processing. If confidence in the processing settings is high enough, this indicates a good way to save time in post- processing. For this set of data captures, frequency domain data were collected in order to provide greater options dur- ing postprocessing. Delamination Detection algorithm The delamination detection algorithm was prototyped on the basis of data collected at the National Center for Asphalt Technology (NCAT) facility during 2009 and 2010. The algorithm was built to account for the fact that delamination can occur at a relatively wide range of depths and show a variety of amplitude characteristics in the recorded data. The approach we chose is based on isolating a subvolume where delamination is most likely to happen and performing an energy-based study of frequency inter- vals on the subvolume. The advantage is that considering frequencies, every sample will carry information about the whole depth range to be analyzed, while sorting the energetic Table 1.1. System Configuration for Three Speeds Slow Speed Medium Speed High Speed Frequency range 200 MHzâ3 GHz 200 MHzâ3 GHz 200 MHzâ3 GHz Frequency step 2.5 MHz 2.5 MHz 2.5 MHz Dwell time 1 µs 0.6 µs 0.5 µs Inline sampling 10 cm (~4 in.) 20 cm (~8 in.) 30 cm (~12 in.) Crossline sampling 10 cm (~4 in.) 20 cm (~8 in.) 30 cm (~12 in.) B-scans per swath 31 16 11 Maximum speed ~3.5 mph ~18 mph ~45 mph Figure 1.1. Antenna array and tower mount in position for data collection.
3 values takes care of the varying amplitudes of signatures due to delamination. Schematically, the time window of interest is extracted from every trace and converted to frequency domain, where the spectrum is divided into frequency intervals, or âbins.â The algorithm computes the energy contained in every bin and then sorts the obtained values. The value that will be finally extracted is relative to only one bin, which is selected by the user (Figure 1.2). The user then defines a threshold value for Es and the mini- mum size for an anomaly of interest to produce the final out- put (Figure 1.3). Difficulties encountered, Future Implementation, and Next Steps The settings for data collection could potentially be modi- fied to meet the needs of each department of transporta- tion in terms of traffic flow and safety. Feedback from the operators during some of the field work indicated that some- times slower speeds are safer because traffic flow is more controllable. Additionally, studying the data can allow fine-tuning of the waveform characteristics for the specific application, which can change the maximum possible speed without data loss. The analysis of the data collected has also indicated that slow-speed measurements are affected by some sort of move- ment of the antenna, particularly visible close to the surface. Suspicions were that this behavior is due to the resonant fre- quency of the vehicle suspension; that is, the antenna vibrates together with the vehicle and produces small undulations in the data. Crosschecking typical values for resonance fre- quency of the suspensions with the variations in the data shows this to be the case. Future implementations will need to address this issue, directly in the hardware if possible, by either modifying the antenna mount, using an optional antenna trailer, or possibly in postprocessing software. The size of the generated files is another issue. This issue is present when storing files, transferring files, and postpro- cessing. Modification of the Examiner software package to support optional routines for semiautomatic partitioning to allow fast processing and access of subsections is being imple- mented. Within future releases of Examiner, it will be possible Figure 1.2. Schematic representation of the delamination detection algorithm: operations on the trace in position x = j, y = i. Es = user defined. Figure 1.3. The red rectangles in the picture are the final output of the algorithm, a result of a statistical analysis based on parameters input by the user.
4to divide the data into smaller portions, an operation now performed manually by the user. The delamination detection algorithm has given reason- ably good results for some of the data collected and lacked consistency for others. As the algorithm was developed on the basis of experience at the NCAT facility, where human- introduced simulated delamination provided solid and clean characteristics, this should be regarded as a fairly good first result. Further study of the data collected in real situations, combined with ground truth, will refine the signature given by a delamination both qualitatively and quantitatively. The next step for the detection algorithm is to increase its robustness based on the type of anomaly to be detected. Additionally, the algorithm will attempt to factor in characteristics of the road substructure where variation in depth of targeted layers and eventual lateral discontinuities are present. Finally, the current version of the algorithm requires multiple inputs from the user, which means that the data analyst should have a background in GPR data processing. A further refinement that includes a detailed and targeted analysis of the data and ground truths collected during this project should minimize the input required in future revisions, making it a more user-friendly tool for nonspecialist operators.
