Final Report:
A Multi-Sensor Array Inductive Scanner for Rapid Imaging of Reinforced and Pre-Stressed Concrete
Grant reference: GR/M87368/01
To the General Engineering Programme, EPSRC
Submitted June 2003
Principal investigator:
Dr P Gaydecki, Department of
Instrumentation and Analytical Science
Co-applicant:
Professor FM Burdekin, Manchester Centre
for Civil & Construction Engineering
Department of Instrumentation and Analytical Science
UMIST
PO Box 88
Manchester M60 1QD
United Kingdom
Tel:
Fax:
E-mail:
[UK-44] (0) 161 200 4906
[UK-44] (0) 161 200 4911
patrick.gaydecki@umist.ac.uk
www.dias.umist.ac.uk
1.
Introduction
This project, which commenced in February 2000 concerned the research and development of a rapid, multisensor array inductive scanner to generate dimensionally accurate images of steel reinforcing components
embedded within reinforced and pre-stressed concrete, for the purposes of condition assessment and
structural integrity monitoring. For its successful realisation, it required expertise in several fields, including
electromagnetic field theory, civil engineering, digital signal processing, software coding, analogue and
digital electronic design. The research team has been engaged with inductive scan imaging work since the
early 1990’s and, through journal publications and presentations at national and international level, has
established a reputation in this area. Final reports for previously funded programmes in this area
(GR/H37860 and GR/K91217) have been highly rated. The team, shown in Figure 13, presently comprises the
following members.
Name
Patrick Gaydecki
F Michael Burdekin
Bosco Fernandes
Graham Miller
Post
Reader
Professor (retired)
Research associate
(EPSRC)
PhD student (EPSRC)
Sung Quek
PhD student
Muhammad Zaid
PhD student
Activity
Principal investigator
Co-applicant
Coil modelling; sensor design; corrosion detection; finite
element analysis; mechanical design
Sensor instrumentation; real-time DSP system design;
corrosion detection
Image layer separation; depth / diameter profiling; 3D
visualisation; control and acquisition software
Image interpolation; neural network profiling and corrosion
estimation
The project sought to address some of the remaining fundamental problems associated with the technology, in
order for the method to be adopted by industry as a reliable testing and monitoring modality. These problems
are summarised as follows:
1.
Sensor and system miniaturisation. Early sensors employed detection coils of 40 mm, which yielded
good depth penetration but poor spatial resolution. For multi-sensor array designs, smaller coil diameters
were required that retained their sensitivity and stability. In addition, it was necessary for the electronic
signal post-processing systems to be small, robust and reliable.
2.
The detection range of the sensors was to be extended to at least 100 mm. Most reinforcing bar of
importance lies within 75 mm of the concrete surface, but margins were desirable for unusual structures
or structures coated with other protection layers.
3.
Image de-blurring, 3D rendition and quantitative information extraction. As has been documented many
times in the past, images produced by inductive scanning are blurred, due to the point spread function
(PSF) of the sensor; this problem becomes more severe as the distance between the sensor and the
steel reinforcing increases. Image processing and analysis routines were to be developed that could not
only enhance the raw data, but extract accurate dimensional information relating to the rebar depth and
diameter. Clearly, this information is crucial to site inspection engineers. Rebar layer separation and 3D
visualisation was also significant to aid structural integrity assessment and coring or drilling operations.
4.
Corrosion detection. In common with many other nondestructive testing techniques, inductive sensors
developed prior to the award of this grant could not reliably distinguish between good and corroded steel
bar, especially with low to moderate levels of corrosion. Again, for field inspection purposes, this was an
important objective.
5.
Scan time reduction. The laboratory-based, single-sensor systems typically required 2 hours to generate
a high-resolution image, which would be intolerable under field conditions. Methodologies were to be
investigated to reduce this to approximately five minutes.
6.
Fabrication of a portable inductive imaging system suitable for on-site testing, forming the basis of a precommercial instrument.
Objectives 1 to 5 have been completely achieved. In some cases the performance of the system has exceeded
the expectations of the team. Objective 6 is on-going and nearly complete. At the time of writing, the inductive
scanning system research at UMIST, funded under this award, is being developed as a commercial instrument
by Elcometer Instruments Ltd, one of the collaborating companies named in the original application. As is
frequently the case in research, the manner in which some of the problems were solved differed from the
solutions proposed in the Case for Support; we highlight below where such changes in research approach were
adopted. Respecting the raw data, the new system provides very much higher signal-to-noise ratios and
cleaner images than previous designs. Using small 15 mm diameter sensor coils, it is possible to detect steel
plate to a depth of 300 mm, a 16 mm steel reinforcing bar to a depth of 150 mm, and a 15 mm steel ball
bearing to a depth of 100 mm. Data provided by the sensor enable a three-dimensional image of the bar
mesh to be constructed and visualised. With no prior knowledge of the bar location or dimension, the scan
depth and bar diameters can be calculated to an accuracy of 5% in a single scan. Notably, the system is
1
capable of detecting even small amounts of corrosion product, something that is currently beyond the
capabilities of ultrasonic, X-ray or microwave sensing systems. At the present time, the sensor is capable of
detecting and imaging a 1 mm thick layer of corrosion on a 15 mm diameter steel bar, located 40 mm below
the surface concrete.
Key advances made in this research will be detailed in Section 3; for the purposes of clarity, mathematical
details have been omitted, which may in any case be found in the journal and conference publications listed
in Appendix 1.
2.
Theoretical considerations
Inductive sensors exploit the principle that the impedance of a coil carrying an alternating current changes
when a metal target is positioned within the flux of the radiating magnetic field. This occurs because eddy
currents are set up in the metal target, which in turn affect the electrical properties of the excitation coil. The
phenomenon can be explained mathematically with recourse to the laws of Ampère, Faraday and Lenz.
If the target is mainly conductive but of low permeability (such as copper, where the relative permeability, µr
= 1 and the conductivity, σ = 5.7 × 107 Sm-1), then the eddy currents generate a reflected magnetic field that
itself generates a secondary current in the coil, opposing the original sensing current. Effectively, the
resistance of the coil increases. Furthermore, the Q of the coil drops since energy is absorbed by the target.
The Q of a coil within a tuned system is the ratio of the input energy necessary to maintain oscillation,
divided by the energy lost (due to resistance). In other words, Q describes the degree of damping of a
resonant system; the higher the Q, the lower the damping.
Conversely, for purely permeable targets such as ferrite or corrosion product, no eddy currents flow since
there are no conductive paths. However, the local increase in permeability increases the effective inductance
of the coil. If the coil forms part of a tuned oscillator, then its resonant frequency falls, because this is
inversely proportional to the root of the inductance.
For targets that are both permeable and conductive, the modified electrical properties of the coil respecting
Q and L are determined by a combination of the factors given above, and therefore depend on the absolute
values of permeability and conductivity of the target in question.
It is clear that high-precision and repeatable measurements of these two parameters are fundamental to this
work; detection of both are required for the imaging of steel and corrosion. However, for small targets, or
targets at a considerable distance from the coil, the changes in Q and L are minute (typically, for the sensors
described below, one part in 100,000 for a 16 mm diameter steel bar at a distance of 150 mm). For this
reason, considerable effort was expended in the design and fabrication of sensitive and stable sensor
systems.
3.
Key advances
The most significant progress over the last three years relates to sensor design, digital signal processing
(DSP) electronics, image reconstruction, information extraction and speed of operation. It is important to
stress that each of these areas is critical when considering the performance of an effective field instrument.
3.1 Sensor design
For the inductive coil array, the original intention was for each coil to be etched as a multi-layer configuration
on a PCB substrate. Each layer would describe a flat, spiral structure, sometimes known as a pancake coil. It
was argued that the use of PCB technology would make possible the fabrication of a linear array of such
coils, reducing the size of each element and obviating problems associated with repeatability and
robustness. Furthermore, finite element analysis using the electromagnetic modelling package Opera 3D
suggested that the induced EMF in the receivers would be sufficiently strong to enable practical usage of this
basic design. A wide range of coils was manufactured in this manner, two of which are shown in Figures 1a
and 1b. In practice however, it was found that for square, 25 mm2 coils, even with densely packed, fine
tracks 100 microns in width, it was not possible to achieve inductances greater than a few micro henries.
When combined with the sensor electronics, these coils could not reliably image steel bars at distances
greater than 10 mm. Clearly, this was unacceptable. Larger inductances could be achieved by increasing the
area of the coils, but this would compromise the miniaturisation process required by the final design. A
decision was therefore made to optimise small diameter, multi-turn coils wound on conventional hollow ferrite
cores.
The Q-sensor. This sensor was primarily intended for the visualisation of the parent steel. The front end
comprised a standard Colpitts-type LC oscillator, with the coil acting as both transmitter and receiver. The
coil was wound around a 15 mm diameter hollow ferrite, giving an inductance of 60 µH. The selected
2
operating frequency, given the capacitors, was 40 kHz. The relatively heavy gauge of the wire, combined
with the chosen excitation frequency, ensured a high Q factor under null conditions, together with a
pronounced skin effect in the target material. This resulted in the surface of the bar being imaged, facilitating
the visualisation of corrosion (see below). To detect changes in Q, the electronics were designed to produce
a signal whose amplitude was proportional to that of the excitation oscillation. The signal from the oscillator
was fed to a precision active full-wave rectifier, whose output was in turn sent to an offset (common-mode
signal) subtraction circuit, configured using a close-tolerance Wheatstone bridge arrangement. The nulled
signal was now input to a gain-selectable amplifier and finally to an analogue filter bank combining a lowpass and a 50Hz (mains) notch filter. Photographs of the sensor are shown in Figures 1c and 1d. Without
signal post-conditioning, this sensor yields an imaging range of 50 mm for a 16 mm diameter steel bar. It is
the most stable, sensitive and noise-free sensor designed by the team to date. Further details and
justification of its design are provided in references [5, 9, 15].
The heterodyne sensor. This H-sensor was primarily intended for the visualisation of corrosion, and
detected the change in inductance by measuring the shift in the resonant frequency of the oscillator. It
consisted of exactly the same front end coil and oscillator circuit as the Q-sensor. In order to detect the shift
in frequency, the output from the sensor was compared to a fixed frequency reference signal generated by a
crystal oscillator. The comparison was performed using a phase-sensitive detector and frequency-to-voltage
converter [7, 9, 17, 20].
The comparatively small diameter of the coils, combined with the small area of the oscillator circuitry, implied
that it would be straightforward to construct a widely-spaced array of such sensors. Continued work in this
area led the development of a combined Q-H design, further reducing the size [22]. In this revised circuit, a
single coil and oscillator fed both a Q-detection and frequency sensitive circuit. It is shown in Figures 1e and
1f.
3.2 Real-time DSP system
In the original proposal it was stated that a real-time DSP system would be developed as part of the sensor
instrumentation package for the purposes of image storage, signal-to-noise ratio enhancement and to aid in
the extraction of quantitative information. Two such systems have been designed and employed, based on
the Motorola DSP56002 DSP device and the more recent and powerful DSP56309. The Q-detection sensor
gave good results for bars located less than 50 mm below the concrete surface, but suffered from increasing
problems of noise beyond this range. This was both due to the small size of the search coil and the limited
resolution of the first analogue-to-digital converter (ADC) used (12-bit). The first DSP system, shown in
Figures 2a and 2b, comprised a DSP56002 DSP device operating at 30 million instructions per second
(MIPS), a 16-bit sigma-delta ADC sampling at 24 kHz, 32 k words of 24-bit local fast memory for image
storage and a serial port for communicating with the host PC. This system was programmed to perform realtime averaging of signals taken from the sensor. Because of its speed, it was possible to take hundreds of
reading per sample point without incurring a time penalty in the scanning process. In addition, the system
applied high levels of very pure digital gain to the sensor input, relaxing the requirement for the analogue
amplifier, minimising the noise contribution still further [9, 15, 20].
The later generation system, shown in Figures 2c and 2d, was similar in concept but employed the
DP56309, operating at 100 MIPS. This instrument also included a digital-to-analogue converter (DAC),
whose output was fed back to the Q-sensor’s nulling circuit. This represented a significant stage in sensor
development since it was no longer necessary to manually correct for sensor drift prior to a scan. The
software could now instantaneously calculate the offset voltage and feedback the required nulling signal.
The DSP systems extended considerably the range of the sensors. Figures 3a and 3b for example, show
profile scans obtained by the Q-sensor for a 16 mm bar located at a depth of 60 mm below the concrete.
Figure 3a was obtained under conditions of no averaging, and Figure 3b was obtained using 600 averages
per sample in real time. Both scans required the same time to perform (10 seconds), determined by the
speed of the X-Y motor drive assembly. DSP has not only improved the detection capability of the sensor, it
has also extended the range over which the bars can be accurately located, sized and ultimately rendered in
three dimensions. This is because the reconstruction algorithms operate using quantitative relationships of
the signal profile (see below). The more uniform and regular the profile is, the more reliable the derived
statistics.
3.3 Layer separation, information extraction, image interpolation and 3D visualisation
In the original proposal it was asserted that this phase of the work would be central to the success of the
project, and therefore most research effort would be expended in this area. Quoting from the Case for
Support, we stated:
3
“Image generation, reconstruction, recognition and evaluation. This phase of the research will be the
most critical. The unprocessed images obtained by mapping the outputs of the receivers as grey-level
values will be blurred, as discussed previously. Processing algorithms will be coded not only to sharpen
these images, but also to extract dimensional and condition information in respect of the steel bars”.
The problems associated with these images are severe indeed. They are blurred, and the blurring worsens
with depth. Moreover, the various bar mesh layers, located at different depths, simply appear as lighter or
darker structures in the composite 2D image (much like a conventional radiograph). It was proposed in the
Case to enhance the images either by developing a deconvolution algorithm based on a Biot-Savart model,
or by exploiting the spatial response of a single receiver. In the event, the latter route was adopted, and this
has proven highly successful. The software allows the operator to take a raw, 2D image of a rebar mesh,
separate the layers, obtain very accurate depth and diameter information, and finally to reconstruct a virtual
3D image of the bar arrangement within the scan volume, which can be rotated at will. The algorithms are
fast, robust and as far as the team is aware, scientifically novel in this application [4, 6, 11-13, 16].
The key to the success of the technique is founded on four particular properties of the curve obtained from a
transverse scan across a steel bar: first, the relationship between the curve width at half-height and the peak
height is unique to a given bar at a given depth; second, both the curve and the depth profile may be
accurately represented using non-linear polynomial models. Third, polynomial interpolation allows the
separation of multiple layers. Finally, for parallel bars separated by more than five diameters, the curves may
be treated as linearly separable.
Using a single sensor, the motorised bench-mounted X-Y scanner originally required two hours to obtain a
410 × 410 resolution image with pixels of 1.1 mm2 area. Careful analysis of the spatial frequency response of
the sensor suggested, however, that more widely spaced samples could be taken and Fourier-based
interpolation applied to reconstruct a high resolution image, without loss of information. The Fourier-based
method employs zero padding in the spatial frequency domain and inverse Fourier transformation to obtain
higher resolution data in the time/spatial domain. The reconstruction is highly satisfactory, with insignificant
deviation from the results obtained for a conventional high resolution scan. Furthermore, the Fourier
algorithm is efficient, straightforward to code and yields an ideal band-limited interpolation. Experiments
show that this new methodology is faster than the traditional scanning protocol by at least a factor of ten.
There is a limit on how widely spaced the sample points can be, before information is lost. This limit is
encapsulated within the Shannon Sampling Theorem, which stipulates that a signal comprising frequencies
extending to f must be sampled with a frequency of at least 2f. If this condition is violated, aliasing will occur.
This situation cannot be remedied (information retrieved) by any interpolation method. In order to define the
maximum spatial sampling interval therefore, it is necessary to establish the spatial frequency response of
the sensor. We have described in various publications the method that we used to ascertain this; suffice it so
say here that it is possible, using the sensors developed, to take scan readings every 20 mm, rather than
every 1.1 mm, and still retain all relevant information [10, 18, 21, 23]. Clearly, the time needed to scan is
further reduced if multiple sensors are employed. In the commercial product, a widely spaced linear area will
be combined with image interpolation. In this case, it will require perhaps a few tens of seconds to scan a 0.5
m2 area of concrete.
Therefore, to generate a virtual 3D bar mesh from the raw image, the method proceeds as follows.
1.
2.
3.
4.
5.
6.
Obtain a low-resolution image of the bars and perform Fourier-based interpolation for high-resolution
reconstruction.
Represent the image as a series of vertical and horizontal signal vectors, and apply polynomial
interpolation to separate the upper and lower bar layers.
Fit Pearson VII curves to the separated bar profiles, and use these to extract the full width at height,
peak intensity and peak position.
Using a Bleasdale power-law regression model, calculate the depth of the bar.
Using the results from (3) and (2), calculate the bar depth.
Combine the reconstructed image with heterodyne data to map areas of corrosion.
This procedure enables a three-dimensional image of the bar mesh to be constructed and visualised, with no
prior knowledge of depth or diameter; this is a unique feature of the method. It allows estimates to be made
of a bar’s depth and diameter to an accuracy of 5%.
3.5 Laboratory scanner system
The laboratory-based scanner system included the bench-mounted, motorized X-Y scanner assembly
described in previous publications [1, 3-4], together with the modified sensors, processing systems and
software. The configuration is shown in Figure 4.
4
3.6 Portable scanner design
Elcometer Ltd is primarily responsible for producing multi-element sensors based on the UMIST design, and
for constructing the field-worthy mechanical scanner. Originally, it was intended for the scanner to be
motorised. However, progress in the areas of sensor development and image interpolation suggested a
manually operated, linear-scanning array would be faster, more robust, reliable and cheaper. At the present
time, we continue to collaborate closely with Elcometer on this phase of the work. Figure 5a shows a
photograph of the current scanner prototype. It comprises a rigid plastic arm moving with the aid of gear
wheels located at either end, which allow it to travel within a frame without slippage. In addition, the wheels
contain shaft encoders that transmit the lateral position of the arm to the controlling notebook computer.
Within the arm is mounted an array of sensors, one of which, together with its analogue electronics, is shown
in Figure 5b.The output from each sensor is multiplexed to a central acquisition unit, converted into digital
form and finally transmitted to the computer. The new digital processor board designed by Elcometer is
shown in Figure 5c. This phase of the work is ongoing and will require another 9 months to complete.
4.0 Results
4.1 Unprocessed images
Figure 6a shows a 2D image of several 16 mm diameter steel reinforcing bars, cast within a concrete test
block, obtained with this latest Q-detection sensor. The block was 450 x 450 x 150 mm, with the four
horizontal bars located 20 mm below the surface, and the three vertical bars located 36 mm below the
surface. The sensor was scanned at a height of 5 mm above the concrete, ie giving a sensor to upper bar
distance of 25 mm. This test specimen has been used many times in the past to evaluate the imaging
capabilities of various inductive-scan sensor designs, but never has an image been obtained with such good
contrast and such a high signal-to-noise ratio, especially of the lower bars. The previous range limit of this Qdetection device was 50 mm; using real-time averaging it is possible to extend this considerably. Figure 6b
for example, shows an image obtained using 200 averages per sample point, of three horizontal and three
vertical 16 mm diameter bars, with the upper bars positioned 60mm below the sensor (ie, the vertical bars
were located 76 mm below the device). Despite blurring, the upper bars are readily visible, whilst those at
the lower depth are also still discernable.
4.2 Layer separation, dimensional analysis and 3D reconstruction
As discussed above, image reconstruction and analysis is based upon curvilinear models of the line scan
vectors and depth response. Figures 7a and 7b depict how the greyscale image may be represented as a
series of vectors, each of which is than approximated by a Pearson VII model. Figure 7c shows the
apparatus being used to obtain a series of line scans of a single bar over a range of distances. These are
than mapped as a 2D depth response image, as shown by Figure 7d. These data are used to construct a
curvilinear estimation of the depth response, based on a Bleasdale model.
Figure 8 depicts the separation and spatial quantification of an image of a double layer, 3 × 3 mesh of bars,
shown in Figure 8a. Polynomial estimation is used to isolate the upper (8b) and lower (8c) layers. The profile
signals are fed to the models, which produce measurements as shown in the table that accompanies this
figure. The numbers in brackets denote the actual bar diameter (10 mm) and depths of the upper (30 mm)
and lower (40 mm) layer. As the figures show, the models produce very accurate results, the repeatability
having been substantiated by a wide range of bar diameters and depths [12, 16]. These measurements may
now be used to construct a 2D grid of the bar positions, as shown in Figure 8c. Alternatively, by employing
all the information, a virtual 3D reconstruction of the raw image may be synthesised, as illustrated in Figure
9. The technique is therefore highly effective at removing the effects of blur in the original image. This is
confirmed in Figure 10, which shows an original double layer, 2 × 2 mesh of bars (a), the separated upper
and lower layers (b and c), and the final reconstructed plan view image (d). Of particular significance is the
ability of the method to recover the lower layer, whose bars appear as faint structures in the original image,
not only because they are further from the sensor, but because the image will normalise according to the
feature with the highest intensity.
4.3 Corrosion detection and visualisation
Figure 11 depicts images of corroded steel bars produced by the heterodyning sensor. The 20 mm diameter
bar shown in Figure 11a was corroded using an automated, salt-water immersion/drying process until the
corrosion layer was approximately 2 mm thick. The bar was then butt-joined with another of the same
material and original dimension, using a threaded stud screwed into the tapped ends of each bar. Figure 11b
depicts the resulting image, obtained with a scan height of 10 mm. The corroded area is clearly visible, as is
a dark band denoting the region of the butt-join. The dark band is a consequence of the interruption of eddy
currents at the interface, and suggests that this sensor might also be useful in detecting vertically orientated
cracks.
5
In a further test, five sections of a 20 mm diameter hollow steel bar were corroded from one to five weeks in
the corrosion tank. They were then threaded onto a plastic bolt (Figure 11c) and scanned at a height of 10
mm; the image is shown in Figure 11d. Here, the dark bands represent the interfaces between the sections,
with the light bands produced by the steel. To maximise contrast, the sections were located on the plastic
bolt not in order of corrosion severity. For example, the second section was least corroded, and appears as
the brightest band. The most significant finding from this test was that the sensor was capable of responding
to small amounts of corrosion. Figure 11e represents a repeat scan of Figure 11a, but now conducted under
a 30 mm thick ceramic plate. As expected, this makes no difference to the quality of the signal. This test was
repeated under various moisture conditions, without detrimental effect.
4.4 Image interpolation and reconstruction
Figure 12 confirms how Fourier-based interpolation may be used to significantly reduce the scan time
required. The stipulation remains, however, that the acquisition process must not violate the Shannon
Sampling Theorem. In Figure 12a is shown an original high-resolution image, with scan steps of 1.1 mm that
required 2 hours to acquire. Figure 12b shows a low resolution image, with vertical scan steps of 20 mm,
which required 10 minutes. Figure 12c depicts a Fourier interpolation of (b). To confirm the validity of the
method, the reconstructed data were fed to the layer separation and dimensional analysis models; the depth
and diameter statistics were in all cases as precise as those yielded by the original data [18].
5.
Ongoing work and collaboration with Elcometer Ltd and CAPCIS Ltd
Elcometer, with whom the UMIST team continues to work closely, is producing the final commercial system.
The various software modules developed during the course of the research programme are being combined
into a single, stand alone package suitable for use by non-expert personnel. In addition, the team continues
to research new algorithms based on neural networks for faster curve profiling and corrosion estimation. It is
envisaged that CAPCIS Ltd, the other industrial partner (a corrosion consultancy), will conduct field-trials of
the product and also purchase it under a preferential licensing arrangement. The financial and legal
arrangements are under the management of UMIST Ventures Ltd. There is very considerable interest in the
system both from industry and academia, and strong evidence that it will be commercially successful.
6.
Financial statement
Overall, the budget for this project was under-spent by a small amount - £587. Certain areas were marginally
overspent – travel for example – since group members made two presentations to QNDE in the USA, one of
the premier international conferences for nondestructive testing. In general, we trust that the outcomes of the
work represent good value for money and a worthwhile investment of the public purse.
7.
Publications and dissemination of information
The results of this research programme have been widely published in peer-reviewed journals and in the
proceedings of international and national conferences [1-23]. They are listed in Appendix 1. EPSRC is
always acknowledged. We have also demonstrated the system to many academic and industrial visitors to
UMIST.
8.
Future work
The team will maintain vigorous research activity in this area. In addition, it is also strongly involved with the
development of real-time DSP systems. On the basis of the DSP work, Motorola Europe is sponsoring the
establishment of an advanced real-time DSP laboratory within the group. By combining the sensor
technology with new generation DSP devices, we believe it will be possible to produce a solid-state scanner
capable of producing live, video data of steel reinforcing bar. This will form the subject of our next proposal to
EPSRC.
9.
Summary and conclusion
During this research programme, new inductive sensors have been developed based on Q-detection and
heterodyning principles to image steel rebar and corrosion product. Using small 15 mm diameter sensor
coils, it is possible to detect steel plate to a depth of 300 mm, a 16 mm steel reinforcing bar to a depth of 150
mm, and a 15 mm steel ball bearing to a depth of 100 mm. Data provided by the sensor enable a virtual 3D
image of the bar mesh to be constructed and visualised. With no prior knowledge of the bar location or
dimension, the scan depth and bar diameters can be calculated to an accuracy of 5%. Signal to noise ratios
have been significantly enhanced by the construction of dedicated real-time DSP systems, which also
perform automatic sensor stabilisation and image storage. A large suite of software routines has been
developed for signal acquisition, image interpolation, reconstruction, dimensional measurement, corrosion
mapping and virtual 3D display. The entire system is now being developed as a commercial, manuallyoperated multi-element scanner that will produce images of steel reinforcing within 2 minutes. The research
has been widely published and the team continues to investigate novel research avenues in this field of
nondestructive testing.
6
Appendix 1: Publications under this award
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
Fernandes BT, Silva I and Gaydecki PA 2000 Vector Extraction from Digital Images of Steel Bars Produced by an
Inductive Scanning System using a Differential Gradient Method combined with a Modified Hough Transform NDT
& E International 33 69-75
Gaydecki PA Silva I, Fernandes BT and Yu ZZ 2000 A portable inductive scanning system for imaging steel
reinforcing bars embedded within concrete Sensors and Actuators A: Physical 84 25-32.
Fernandes B, Gaydecki P, Sung Q, Miller G and Burdekin FM 2000 A Multi-Sensor Array Inductive Scanner for
High-Speed Imaging of Reinforcing and Pre-Stressing Steel in Concrete Proc. Conf. BInstNDT 2000, pp217-222.
Quek S, Miller G, Fernandes B, Gaydecki P and Burdekin FM 2001 Imaging and Defect Detection in Steel
Reinforcing Bars And Cables using a Multi-Sensor Array Inductive Scanner Conf. Proc. Structural faults & repair
2001 4-6 July, London UK 1-8
Gaydecki P, Fernandes B, Sung Q, Miller G and Burdekin FM 2001 A Multi-Sensor Array Inductive Scanner for
Rapid Imaging of Reinforced and Pre-Stressed Concrete Review of Progress in Quantitative Nondestructive
Evaluation 20B 1171-1178
S Quek, P Gaydecki, G Miller and B Fernandes 2001 Polynomial-Based Layer Separation applied to Images of
Reinforcing Bar Mesh embedded in Concrete, obtained using an Inductive Scanning System Proc. Conf.
BInstNDT 2001, 19-24
G Miller, B Fernandes, S Quek and P Gaydecki 2001 Visualization of Damage and Corrosion to Reinforcing Bars
in Concrete using a New Solid-State Inductive Scanning Sensor Proc. Conf. BInstNDT 2001, 3-8
John G, Gaydecki P, Fernandes B and Silva I 2000 Development of inductive imaging of corrosion damaged
reinforced and prestressed concrete structures Conf. Proc. Corrosion 2000, NACE Int. Paper no. 00293.
G Miller, P Gaydecki, S Quek, B Fernandes and M A M Zaid 2002 A combined Q and heterodyne sensor
incorporating real-time DSP for concrete reinforcement imaging and corrosion detection Proc. Conf BInstNDT
2002 351-357
M A M Zaid, S Quek, P Gaydecki, G Miller and B Fernandes 2002 High-resolution image generation of steel
reinforcing bars using Fourier-domain interpolation of a sparsely populated data set Conf BInstNDT 2002 345-350
S Quek, P Gaydecki, M A M Zaid, G Miller and B Fernandes 2002 Three-dimensional visualization of steel
reinforcing bars using Pearson and Bleasdale models applied to depth response images taken from an inductive
sensor Proc. Conf BInstNDT 2002 339-344
Quek S, Gaydecki P, Fernandes B and Miller G 2002 Multiple layer separation and visualisation of inductively
scanned images of reinforcing bars in concrete using a polynomial-based separation algorithm NDT & E
International 35 233-240
S Quek, B Fernandes, P Gaydecki and G Miller 2002 Multiple Layer Visualization of Inductively scanned images
of Reinforcing Bar Mesh using a Polynomial-Based Separation Algorithm Review of Progress in Quantitative
Nondestructive Evaluation, 21B 1227-1232
B Fernandes, G Miller, P Gaydecki and S Quek 2002 Damage and Corrosion Visualization of Reinforcing Bars
Embedded in Concrete using a New Solid-State Inductive Scanning Sensor Review of Progress in Quantitative
Nondestructive Evaluation, 21B 1233-1238
Gaydecki P, Quek S, Miller G, Fernandes B T and Zaid M A M 2002 Design and evaluation of an inductive Qdetection sensor incorporating DSP for imaging of steel reinforcing bars in concrete Measurement Science and
Technology 13 1327-1335
Quek S, Gaydecki P, Zaid M A M, Miller G and Fernandes B 2003 Three-dimensional image rendering of steel
reinforcing bars using Curvilinear models applied to orthogonal line scans taken by an inductive sensor NDT & E
International 36 7-18
Miller G, Gaydecki P, Quek S, Fernandes B and Zaid M A M 2003 Detection and imaging of surface corrosion on
steel reinforcing bars using a phase-sensitive inductive sensor intended for use with concrete NDT & E
International 36 19-26
Zaid M A M, Gaydecki P, Quek S, Miller G and Fernandes B 2003 Image reconstruction of steel reinforcing bars
in concrete using Fourier-domain interpolation applied to a sparsely populated data set Journal of Nondestructive
Evaluation (accepted)
G Miller, P Gaydecki, S Quek, B Fernandes and M A M Zaid 2003 A combined sensor incorporating real-time
DSP for the imaging of concrete reinforcement and corrosion visualisation Proc. Sensors and their Applications
XII, Limerick, Ireland (accepted)
G Miller, P Gaydecki , S Quek, M Zaid and B Fernandes 2003 A sensor for imaging steel in reinforced concrete
structures and visualisation of surface corrosion, incorporating real-time DSP Proc. Conf BInstNDT 2003
(submitted)
M Zaid, P Gaydecki, S Quek, G Miller and B Fernandes 2003 Extraction of dimensional information of steel
reinforcing bars in concrete using artificial neural networks trained on data from an inductive sensor Proc. Conf
BInstNDT 2003 (submitted)
Miller G, Gaydecki P, Quek S, Fernandes B and Zaid M A M 2003 A combined Q-type and heterodyne inductive
sensor for imaging and corrosion detection of steel bars embedded in concrete Sensors and Actuators A:
Physical (in preparation)
Zaid M A M, Gaydecki P, Miller G and Quek S 2003 Neural network systems in the quantification of corrosion
imaged by an inductive sensor NDT & E International (in preparation)
Appendix 2: Figures
(a)
(b)
(d)
(c)
(e)
(f)
Figure 1. Evolution of sensor design. (a) and (b): Low and high density multi-layer PCB etched coils. Each
coil has sides of 25 mm. (c) and (d): The Q-sensor. (e) and (f): The combined Q and heterodyne sensor,
capable of imaging both steel and corrosion. This design is now being used in the multi-element array.
Appendix 2: Figures
(a)
(b)
(c)
(d)
Figure 2. The two real-time DSP systems developed for digitizing, processing and storing the inductive
sensor signals. (a) and (b): The 1st generation system, based on a DSP56002 architecture. (c) and (d): The
2nd generation system, based on a DSP56309 architecture. Significantly, the 2nd generation system also
incorporates digital feedback control for sensor stabilisation.
Appendix 2: Figures
0.05
Sensor voltage
0.04
0.03
0.02
0.01
0
0
50
100
150
200
250
300
250
300
Distance, mm
(a)
0.05
Sensor voltage
0.04
0.03
0.02
0.01
0
0
50
100
150
200
Distance, mm
(b)
Figure 3. Transverse line scan of a single 16 mm diameter steel reinforcing bar, taken at a distance of 60
mm. (a) No averaging. (b) 600 averages per sample point in real-time, using the DSP system.
Appendix 2: Figures
Processor control
Digitized processed signal
Digitized processed signal
Software command
Control and
acquisition software
Real-time
DSP system
Analysis software
Analogue sensor signal
Image data
Scan control information
Motorized X-Y Scanner with inductive sensor
Figure 4. Principal components of the current laboratory-based inductive scanning system. Sensor signal
paths are shown in blue.
Appendix 2: Figures
(a)
(b)
(c)
Figure 5. The portable scanning system, now being developed as a commercial product by Elcometer Ltd.
(a) The manually operated scanner assembly, containing an array of sensors. (b) One pre-production sensor
and analogue electronics. (c) The digital circuitry.
Appendix 2: Figures
(a)
(b)
Figure 6. Images produced by the Q-sensor. (a). 450 × 450 mm image of a concrete test block containing
four 16 mm diameter horizontal bars, 25 mm below the surface, and three vertical bars 41 mm below the
surface. (b). 387 × 387 mm image of three 16mm diameter steel bars at 60 mm depth, and three vertical bars
at 76 mm depth.
Appendix 2: Figures
(a)
(b)
(c)
(d)
Figure 7. Modelling the sensor signal as a series of polynomials; (a) raw image; (b) image as a series of
horizontal and vertical vectors; (c) obtaining the line scan response at different depths; (d) multiple line scan
responses as a composite vector image.
Appendix 2: Figures
(a)
(b)
(c)
Y Scanned Dimension (mm)
350
300
Calculated Items
Top Layer
Bottom Layer
250
diameter (10)
10.66
10.24
200
diameter (10)
9.34
9.58
150
diameter (10)
10.43
10.44
100
Depth (30 and 40)
30.87
40.37
50
Depth (30 and 40)
30.19
40.95
Depth (30 and 40)
31.20
41.09
0
0
100
200
300
X Scanned Dimension (mm)
(d)
Figure 8: (a) Original image using 10 mm diameter bars. (b) Separated upper layer. (c) Separated lower
layer. (d) Extracted bar locations. Table shows calculated mean dimensional information.
Appendix 2: Figures
Figure 9. Virtual 3D reconstruction of raw image shown in Figure 8a. The scan volume is depicted by the
black region. All scalings are now correct respecting diameters, relative bar placements and position from the
upper surface.
Appendix 2: Figures
(a) Blurred image of bar mesh with 16 mm
diameter bars scanned at a depth of 30
mm.
(b) Separated top layer.
(c) Separated bottom layer.
(d) Top view of the reconstructed bar mesh, with
accurate diameter information.
Figure 10. Four images depicting the raw data, separated layers and reconstructed bar mesh, using a model
analysis of the bar profiles.
Appendix 2: Figures
(a)
(b)
(c)
(d)
(e)
Figure 11. Images of corrosion produced by the heterodyne sensor. (a) Photograph of a 20 mm corroded
steel bar butt-joined to an uncorroded section. (b) Sensor image of Figure 11a. Note presence of dark band
towards the centre representing the interruption of eddy currents at the interface. (c) Five steel bar sections
corroded from one to five weeks (not in order). The least corroded section is second from the left. (d) Sensor
image of Figure 11c. Brightest band corresponds to least corroded section. Intermittent dark bands represent
bar interfaces. (e) Sensor image of a bar shown in Figure 11a, placed under a 30 mm thick ceramic tile. Both
the corroded region and the interfacial region are clearly visible.
Appendix 2: Figures
(a)
(b)
(c)
Figure 12. Results from Fourier interpolation. (a) Original high resolution 410 × 410 image, using scan steps
of 1.1 mm. (b) Low-resolution image using vertical scan steps of 20 mm. (c) High resolution image
reconstructed by applying Fourier interpolation to image (b).
Appendix 2: Figures
Figure 13. The present research team. From left to right: Muhammad Zaid, Graham Miller, Sung Quek,
Bosco Fernandes and Patrick Gaydecki