sensors Article

Detection and Inspection of Steel Bars in Reinforced Concrete Structures Using Active Infrared Thermography with Microwave Excitation and Eddy Current Sensors Barbara Szymanik 1, *, Paweł Karol Frankowski 1, *, Tomasz Chady 1, * and Cyril Robinson Azariah John Chelliah 2 1 2

*

Szczecin Department of Electrical and Computer Engineering, West Pomeranian University of Technology, ul. Sikorskigo 37, Szczecin 70-313, Poland Department of Nanosciences and Technology, School of Science and Humanities, Karunya University, Coimbatore, Tamilnadu 641114, India; [email protected] Correspondence: [email protected] (B.S.); [email protected] (P.K.F.); [email protected] (T.C.); Tel.: +48-91-449-4134 (B.S.); Fax: +48-91-449-4859 (B.S.)

Academic Editor: Piervincenzo Rizzo Received: 24 December 2015; Accepted: 8 February 2016; Published: 16 February 2016

Abstract: The purpose of this paper is to present a multi-sensor approach to the detection and inspection of steel bars in reinforced concrete structures. In connection with our past experience related to non-destructive testing of different materials, we propose using two potentially effective methods: active infrared thermography with microwave excitation and the eddy current technique. In this article active infrared thermography with microwave excitation is analyzed both by numerical modeling and experiments. This method, based on thermal imaging, due to its characteriatics should be considered as a preliminary method for the assessment of relatively shallowly located steel bar reinforcements. The eddy current technique, on the other hand, allows for more detailed evaluation and detection of deeply located rebars. In this paper a series of measurement results, together with the initial identification of certain features of steel reinforcement bars will be presented. Keywords: infrared thermography; microwave heating; eddy current testing; multi frequency eddy current technique; concrete testing; rebar detection

1. Introduction Reinforced concrete has been an universally dominant construction material for over a century, although structures made of this material are often exposed to many types of damage and deterioration due to different causes and exposure conditions. Therefore, in most countries relatively frequent inspections are required due to building code requirements. Usually, such tests should be conducted after the concrete has hardened and without damaging the structure. For this reason nondestructive testing methods (NDTs) are commonly used for this evaluation [1–3]. There are several aspects in reinforced concrete testing, which are important from the practical point of view: ‚

Assessing the dimensions of structural elements and locating damage and defects (such as voids, cracks and inclusions). Here the most popular NDT methods are: ultrasonic sensors and ground penetrating radar [4–12]. Both methods are fast and reliable, but the results obtained are not easy to interpret. Active and passive thermography can also be used to inspect the inner structure of concrete, but due to some limitations, in practice these techniques are used to detect defects or plaster damages near the surface [12,13].

Sensors 2016, 16, 234; doi:10.3390/s16020234

www.mdpi.com/journal/sensors

Sensors 2016, 16, 234

‚

‚

2 of 16

Reinforcement location and corrosion assessment. Here the natural choices are electro- magnetic methods, such as radiography (a very efficient method, but hard to use in practice and dangerous for the operator) [14–17], eddy current sensors (a promising method, that allows not only the detection of the reinforcement, but also identification) [18–20] and impedance tomography, which is, in contrast to previously mentioned methods, a contact technique. Dampness assessment. Here the most commonly used methods are impedance tomography [21–23], thermography and some chemical techniques.

In this article the authors propose a new concept of two-step concrete structures analysis, using two different non destructive techniques: active thermography and eddy current sensors. The active thermography method is proposed to be used as a preliminary technique, which may be useful in initial rebar detection, whereas the eddy current technique is used for structure assessment and damage identification. In the case of active thermography, an external energy source should be used to induce a thermal contrast inside the inspected specimen. In the case of concrete structures conventional heating methods (like halogen lamps, flash lamps, infrared radiators, hot air and so on) may usually be not effective due to the large size of the inspected structures. In this study we propose using microwave heating [24–27]. In the literature microwave heating of concrete structures is used mainly for acceleration of curing, cement decontamination and drilling or melting the concrete [28]. Here we propose the microwaves as the energy source in infrared thermography (this concept is very rare in the literature, and the authors are familiar with only one publication concerning this subject [24]). The main advantage of the microwave heating is its volumetric character. A certain volume of the specimen is heated at once, which means that this technique is very fast. On the other hand, in the case of this non-conventional energy source, the heating ratio is dependent not only on the thermal properties of the material, but also on some electrical properties. Also an interesting phenomenon is observed during the interaction of the microwaves with metal—The heating is strongly limited due to a very low penetration depth value. This phenomenon will be demonstrated in both numerical models and experiments. To the best of the authors' knowledge the numerical modeling of reinforcement detection in concrete structures using microwave heating has not been previously described in the literature. After heating, a thermovision camera is used to observe the inspected specimen surface. The inner material flaws and inclusions will cause the non-uniform heating of the sample, and this will be visible as hotter or cooler places. Therefore steel bars should be easily detected using this method. Due to the large scale of the inspected structures and the characteristics of concrete heating (which will be further discussed), this method will be here used as a preliminary study technique. The second method proposed here, the multi-frequency eddy current technique, may be used as a second inspection step as it allows not only for the detection of the rebars, but also identification of some features of these structures. Previous researches prove that the single-frequency eddy current method (SFM) can be successfully used to evaluate many structural parameters. The identification may involve: rebar diameter (D), physical properties of rebar (represented by rebar class), and rebar location, including thickness of the concrete cover (h). Therefore, the eddy current evaluation is one of the most promising NDT methods. Nevertheless, the same tests prove that evaluating three parameters at once is a problem, especially, when the concrete cover thickness is high. In this paper, the aim is to present the multi-frequency method (MMFM) which can be used as an alternative to the standard assessment [29–32]. However, MMFM can also be utilized complementarily to the SFM. During the investigation, it is very important to correctly select the frequency and transducer size, especially the distance between the transducer columns. Small transducers have good spatial resolution, while bigger transducers are more sensitive. Therefore, small transducers work correctly only when distance between the transducer and a ferromagnetic material is low, so in this case the frequency used should be high. The larger transducers provide better results greater distances, however their spatial resolution and repeatability are lower. The MMFM allow testing over 50 different frequencies during the one measurement and easily selecting the best one. Also, if transducer size is

Sensors 2016, 16, 234

3 of 16

too small or too large for a particular sample, the test will show this up. In this paper we present the new method of frequency selection in MMFM, and we also present suitable statistical data mining methods that are used to obtain signals in the multi-frequency method allowing for very efficient structure identification. 2. Active Infrared Thermography with Microwave Excitation The mechanisms of microwave heating of a variety of materials are well known and have been extensively described in the literature [4,9]. Obviously, this heating is very closely correlated with the incident electric field frequency—for lower microwave frequencies the ionic conductivity is the most important cause of the material heating, whereas for higher frequencies the energy absorption is primarily due to effects connected with the polarity of the materials' molecules. In our study the incident microwave frequency is fixed at a 2.45 GHz value—The most common frequency for industrial and domestic purposes. As mentioned previously, in this case the heating mechanism is primarily due to the existence of dipole molecules which tend to re-orientate in the presence of a microwave field. The loss mechanism is due to inability of the polarization to follow the fast field alternations. The rate of microwave heating is dependent not only on the thermal properties of the material, but also on some of its electric properties. An important factor is the value of the dielectric loss tangent tanδ = ε''/ε' where ε’ indicates the real part of complex dielectric permittivity, and ε'' is its imaginary part. The dielectric loss tangent is a function of the frequency, and describes the ability of a material to absorb the microwave energy: the higher its value, the more energy can be absorbed by the material. In case of metals, and other materials with very large electric conductivity values, the crucial factor in case of microwave heating is penetration depth, described by: d δ“

1 πµσ f

(1)

where f is the electric field frequency, µ is the magnetic permeability of the material and σ is the material conductivity. In our case, for steel rebar the penetration depth is equal to about 4 µm, which in practice may be treated as negligible. This basically means that metal (even if not treated as a perfect conductor, with resistivity equal to zero) can be considered as a microwave-reflecting material. 2.1. Numerical Modelling of Microwave Heating The numerical modelling was performed using commercial software-COMSOL Multiphysics ver. 4.4 (COMSOL, Inc., Burlington, MA, USA). This software is a powerful tool for solving partial differential equations using Finite Element Method (FEM). The main advantage of this tool relates to its ability to solve multiphysics problems—i.e., ones where several physical phenomena are connected. Microwave heating is an example of a problem where one has to take into account more than one physical module. The created model was based on several governing equations—Electromagnetic wave equation, electromagnetic power dissipation equation derived from Poynting theorem and the heat transfer equation. Therefore two COMSOL modules were used: “Electromagnetic waves”, solved in the frequency domain and “Heat transfer in solids”, solved in the time domain. These modules were coupled by using the domain condition “Heat source” defined as the density of total electromagnetic power dissipated in the material. Figure 1a presents the model's geometry. The position and diameter of the rebar was parameterized as shown in Figure 1b. Four rebar dimension values were analysed (0.8, 1, 1.2 and 1.4 cm) and nine rebar positions below the concrete surface (from 5 cm to 1 cm in 0.5 cm steps).

Sensors 2016, 16, 234 Sensors 2016, 16, 234

4 of 16 4 of 16

Figure 1. 1. (a) (b) problem problem Figure (a) The The geometry geometry used used for for 3D 3D numerical numerical modelling modelling in in COMSOL; COMSOL; (b) parameterization: the its position position below below the the concrete concrete surface surface parameterization: the chosen chosen diameters diameters of of steel steel rebar rebar and and its are shown. shown. are

The wave excitation was carried out using the COMSOL port boundary condition, with the The wave excitation was carried out using the COMSOL port boundary condition, with the power power fixed at 1000 W. The time of heating was set to 60 s, and after that the natural cooling stage fixed at 1000 W. The time of heating was set to 60 s, and after that the natural cooling stage of the of the system was simulated. Total time of simulation was set to 360 s. The simulated material system was simulated. Total time of simulation was set to 360 s. The simulated material properties of properties of concrete, air and steel are collected in Table 1. concrete, air and steel are collected in Table 1. Table 1. Basic Basic thermal thermal and and dielectric dielectric (at (at 2.45 2.45 GHz) GHz) properties properties of of materials materials used used in in simulations. simulations. Table 1.

Material

Material

Concrete

Concrete

Steel

Air Steel

Loss Tangent Loss Tangent tan δ = ε'' / ε' tan    ' ' /  ' 0.36/4.5 Here simulated as 0.36/4.5 conductivity Here as σ =simulated 8.41 ˆ 106 S/m conductivity

Density Density [kg/m3 ] [kg/m3] 2400

Thermal Thermal Conductivity Conductivity [W/(m¨ K)] [W/(m·K)] 0.8

Heat Capacity Capacity atat Heat Constant Pressure Constant Pressure [J/(kg¨ K)] [J/(kg·K)] 750

2400

0.8

750

1.29 7850

0.022 66

1010 490

7850

66

490

 = 8.41 × 106 S/m As previously mentioned, was parameterized Air - the simulation1.29 0.022 in terms of the rebars' 1010 diameter and the position of the rebar below the concrete surface. As a result we performed 36 simulations, one for each As case. previously mentioned, the simulation was parameterized in terms of the rebars' diameter To present result,the the concrete temperature distribution waswe plotted for one located and the positionanofexemplary the rebar below surface. As a result performed 36slice, simulations, in of the geometry, in the yz plane (see Figure 2). The temperature was plotted for the onethe formiddle each case. chosen rebar's diameter and position = 10distribution mm and position = 40 mm the concrete To present an exemplary result,(i.e., the diameter temperature was plotted forbelow one slice, located surface). The first plot (Figures 2a and 3a) shows the temperature distribution after the heating process in the middle of the geometry, in the yz plane (see Figure 2). The temperature was plotted for the (the firstrebar's 60 s of diameter the simulation). It is clearly that= the is not heated microwaves. chosen and position (i.e., visible diameter 10 steel mm rebar and position = 40 by mm below the Heating itself is not uniform, the heated volume is dependent on the waveguide flange shape (here the we concrete surface). The first plot (Figures 2a and 3a) shows the temperature distribution after simply an open distance between the waveguide the sample (here, 10 cm). heatinguse process (thewaveguide) first 60 s ofand the the simulation). It is clearly visible that and the steel rebar is not heated Obviously the temperature distribution is also disturbed by the presence of the steel rebar inside the by microwaves. Heating itself is not uniform, the heated volume is dependent on the waveguide concrete structure. It is clear that the resulting images should be properly analyzed to obtain an image flange shape (here we simply use an open waveguide) and the distance between the waveguide and of detected rebar. thethe sample (here, 10 cm). Obviously the temperature distribution is also disturbed by the presence of the steel rebar inside the concrete structure. It is clear that the resulting images should be properly analyzed to obtain an image of the detected rebar.

Sensors 2016, 16, 234

5 of 16

Sensors 2016, 16, 234 Sensors 2016, 16, 234

5 of 16 5 of 16

Sensors 2016, 16, 234

5 of 16

Figure 2. Exemplary results of numerical modeling (rebar diameter = 10 mm and position = 40 mm below

Figure 2. Exemplary ofnumerical numerical modeling (rebar diameter = and 10 mm and= position = 40 mm Figure 2. Exemplaryresults results of modeling (rebar diameter = 10 mm position 40 mm below the concrete surface). The comparison between the temperature distributions after the heating phase below concrete surface). The comparison between the temperature distributions after the heating thethe concrete surface). The comparison between the temperature distributions after the heating phase (60 s) and cooling phase (360 s) is shown. Figure 2. Exemplary results of numerical modeling (rebar diameter = 10 mm and position = 40 mm below (60(60 s) and cooling phase (360 s) is shown. phase s) and cooling phase (360 s) is shown. the concrete surface). The comparison between the temperature distributions after the heating phase (60 s) and cooling phase (360 s) is shown.

(a) (a)

(b) (b)

Figure 3. Zoomed region of interest. The comparison between the temperature distributions after:

d = 14 dmm dmm = 12 dmm dmm = 10 dmm dmm = 8 mm =d14 mm =d12 mm =d10 mm d =d8=mm = 14 = 12 = 10 8 mm

Figure 3. Zoomed region of interest. The comparison between thethe temperature distributions after: after: Figure region The comparison temperature distributions (a)3. theZoomed heating phase (60ofs) interest. and(a) (b) cooling phase (360 s).between(b) (a) the heating phase (60 s) and (b) cooling phase (360 s). (a) the heating phase (60 s) and (b) cooling phase (360 s). Figure 3. Zoomed region of interest. The comparison between the temperature distributions after: It the canheating be clearly seen in Figures 2 and 3, that(360 in most cases detection of the rebar can not be done (a) phase (60 s) and (b) cooling phase s). It can be clearly seen in Figures 2 and 3, that in most cases detection of the rebar can not be done directly (based on the raw data). Although the rebar itself is not heated and the temperature It can be clearly in Figures 2 and 3 the thatrebar in most of the can not be directly (based on seen the raw data). Although itselfcases is notdetection heated and the rebar temperature distribution, observed at in theFigures concrete surface, hasintomost be disturbed by its presence, thecan effect It can be clearly seen 2 and 3, that cases detection of the rebar notisbehidden done observed at the concrete surface, has tothe be rebar disturbed byis itsnot presence, the effect istemperature hidden donedistribution, directly (based on the raw data). Although itself heated and the because(based the concrete is heated quite effectively microwaves too. and The the rawtemperature data for an directly on the itself raw data). Although the rebar by itself is not heated because the concrete itself is heated quite effectively by microwaves The raw forisan distribution, observed thes)concrete surface, has is toshown be disturbed by4. itstoo. presence, thedata effect hidden exemplary time stepat(200 for all configurations in Figure distribution, observed at the concrete surface, has to be disturbed by its presence, the effect is hidden exemplary time step (200 s) for all configurations is shown in Figure 4. because the concrete itself is heated quite effectively by microwaves too. The raw data for an exemplary because the concrete itself is heated quite effectively by microwaves too. The raw data for an Depth = 5 cm Depth = 4.5 cm Depth = 4 cm Depth = 3.5 cm Depth = 3 cm Depth = 2.5 cm Depth = 2 cm Depth = 1.5 cm Depth = 1 cm timeexemplary step (200 time s) Depth forstep all configurations is shown inisFigure 4. s) for= 4all shown in Figure Depth = 5 cm = 4.5(200 cm Depth cmconfigurations Depth = 3.5 cm Depth = 3 cm Depth = 2.5 cm4.Depth = 2 cm Depth = 1.5 cm Depth = 1 cm Depth = 5 cm Depth = 4.5 cm Depth = 4 cm Depth = 3.5 cm Depth = 3 cm Depth = 2.5 cm Depth = 2 cm Depth = 1.5 cm Depth = 1 cm

Figure 4. The temperature distribution at the concrete surface obtained for all rebar diameters (d) Figure 4. The temperature distribution at the concrete surface obtained for all rebar diameters (d) and depths. and depths. Figure 4. The temperaturedistribution distribution at obtained for all diameters (d) Figure 4. The temperature at the theconcrete concretesurface surface obtained forrebar all rebar diameters (d) and depths.

and depths.

Sensors 2016, 16, 234

6 of 16

Sensors 2016, 16,visible 234 16 It is clearly that for all rebar diameter cases the object can be detected only if it6isof located shallowly beneath the concrete surface (i.e., 1 cm, 1.5 cm and 2 cm). In case of numerical modeling it It 2016, is clearly Sensors 16, 234visible that for all rebar diameter cases the object can be detected only if it is located 6 of 16 is possible to model the case with solid concrete structure, without the steel bar inside (in Figure 5 shallowly beneath the concrete surface (i.e., 1 cm, 1.5 cm and 2 cm). In case of numerical modeling it one may see temperature distribution at the concrete for this model), which may be5 used It is the clearly visible all rebar diameter cases thesurface object can be detected only if(in it isFigure located is possible to model thethat casefor with solid concrete structure, without the steel bar inside to remove the effect of concrete heating, by simple subtraction. The results of such operation shallowly beneath the concrete surface (i.e., cm, 1.5 cm surface and 2 cm). of numerical modeling one may see the temperature distribution at 1the concrete for In thiscase model), which may be useditfor an exemplary timethe (200 s) are shown in Figure 6.structure, The results are interpret, the rebar is is remove possible tostep model the case with solid concrete without theeasy steel bar inside (inasFigure 5 to effect of concrete heating, by simple subtraction. Thenot results ofto such operation for an one may see temperature at the concrete surface for model), which be used onceexemplary indicated asthe astep hotter gray)inplace and once as a are cooler (darker gray) spot inrebar the middle time (200(lighter s) aredistribution shown Figure 6. The results notthis easy to interpret, asmay the is toimage. remove the effect of concrete heating, by simple subtraction. The results of such operation for an indicated as a hotter (lighter gray) place and once as a cooler (darker gray) spot in the middle of of theonce This phenomena may be explained after the insightful analysis of the heating/cooling exemplary time step (200 s) are shown in Figure 6. The results are not easy to interpret, as the rebar the image. This phenomena may be explained after the insightful analysis of the heating/cooling process of the modeled system. When the rebar is located deeper beneath the concrete surface itislimits once indicated as hotter (lighterWhen gray)In place and once as a cooler (darker gray) spot in the surface middle of process of the modeled system. the rebar is located deeper beneath the concrete it will the volume heated bya the microwaves. such a situation, the temperature just above the rebar the image. This phenomena may be explained after the insightful analysis of the heating/cooling limits the volume heated by the microwaves. In such a situation, the temperature just above the be much higher, than the temperature at the same spot in the model without the rebar. The same process modeled When the rebar deeper beneath concrete surface rebar willofbethe much higher,system. than the temperature at is thelocated same spot in the model the without the rebar. Theit phenomenon occurs, obviously, in the case when the rebar is located shallowly, but the cooling process limits the volume heated by the microwaves. In such a situation, the temperature just above the same phenomenon occurs, obviously, in the case when the rebar is located shallowly, but the cooling for small volumes is fast enough, so the rebar is visible as the cooler spot. rebar will be much higher, than the temperature at the same spot in the model without the rebar. The process for small volumes is fast enough, so the rebar is visible as the cooler spot.

same phenomenon occurs, obviously, in the case when the rebar is located shallowly, but the cooling process for small volumes is fast enough, so the rebar is visible as the cooler spot.

(a)

(b)

Figure 5. The temperature distribution at the concrete surface obtained in a model without the rebar.

Figure 5. The temperature distribution in a model without the rebar. (a) at the concrete surface obtained (b) (a) observed after the heating phase (60 s), (b) after the cooling phase (360 s) (a) observed after the heating phase (60 s), (b) after the cooling phase (360 s)

d = 14 mm d = 12 mm d = 10 mm d = 8 mm d = 14 mm d = 12 mm d = 10 mm d = 8 mm

Figure 5. The temperature distribution at the concrete surface obtained in a model without the rebar. 3.5 cm Depth = 3 cm = 2.5 cm Depth =phase 2 cm (360 Depth Depth = 1 cm (a) observedDepth after=the heating phase (60 s),Depth (b) after the cooling s) = 1.5 cm Depth = 3.5 cm

Depth = 3 cm

Depth = 2.5 cm

Depth = 2 cm

Depth = 1.5 cm

Depth = 1 cm

Figure 6. The result of subtracting the temperature profile obtained in a model without the rebar from the original data. Results for all the rebar diameters and chosen depths. Figure 6. result The result of subtracting temperature profile obtainedinina amodel modelwithout withoutthe therebar rebar from Figure 6. The of subtracting the the temperature profile obtained Obviously, in reality, one for does notrebar havediameters the access to thedepths. case of solid concrete, so the from the original data. Results all the and chosen

the original data. Results for all the rebar diameters and chosen depths. previously shown operation has to be replaced by some numerical estimation of the trend. The idea Obviously, in reality, doesmask. not have thetheaccess to the of solid from concrete, so the is to use median filtering withone a large Finally, obtained datacase is subtracted the original Obviously, inone reality, one does not the by access the case of the solid concrete, so the previously previously operation has to behave replaced someto numerical estimation of the trend. The idea image. Thisshown is of the methods used in background removal from thermal images [33]. This is to useismedian filtering a large Finally, thethe obtained data isofsubtracted from theidea original shown operation has be with replaced bymask. some numerical estimation theanalysis trend. The process shown intoFigure 7. In Figure 8 one can see result of such an for one caseisofto use image. This iswith one the used incases, background removal the thermal images [33]. rebar diameter (14 mm, formethods the remaining the results are similar). It can befrom clearly seen thatThis the median filtering aoflarge mask. Finally, the obtained data isfrom subtracted the original image. process is shown in Figure 7. In Figure 8 one can see the result of such an analysis for one case of This is one of the methods used in background removal from the thermal images [33]. This process is rebar diameter (14 mm, for the remaining cases, the results are similar). It can be clearly seen that the

shown in Figure 7. In Figure 8 one can see the result of such an analysis for one case of rebar diameter

Sensors 2016, 16, 234

7 of 16

Sensors 2016, 16, 234 16 (14 mm, for the remaining cases, the results are similar). It can be clearly seen that the rebar7 ofdetection Sensors 2016, 16, 234 7 of 16 is limited and possible only in the case when it is located shallowly beneath the concrete surface (i.e., rebar Sensorsdetection 2016, 16, 234is limited and possible only in the case when it is located shallowly beneath7 the of 16 1 cm, 1.5 cm and 2 cm). concrete surface (i.e., 1 cm, 1.5 cmpossible and 2 cm). rebar detection is limited and only in the case when it is located shallowly beneath the

rebar detection limited possible only in the case when it is located shallowly beneath the concrete surfaceis(i.e., 1 cm,and 1.5 cm and 2 cm). concrete surface (i.e., 1 cm, 1.5 cm and 2 cm).

(b) (c) (a) (b) (c) (a) Figure 7. Visualization of the background removal method based on median filtering, shown for the Figure 7. Visualization of the background removal(b) method based on median filtering, (c)shown for the (a) case of rebar with 14 mm locatedremoval at 2 cm method depth. (a) original data, (b) data after median Figure 7. Visualization of diameter the background based on median filtering, shown for the case of rebar with 14 mm diameter located at 2 cm depth. (a) original data, (b) data after median filtering with large results of located subtraction case of7. rebar withmask, 14 mm diameter at 2 (a,b). cm depth.based (a) original data,filtering, (b) datashown after median Figure Visualization of(c) the background removal method on median for the filtering with large mask, (c) results of subtraction (a,b). filtering with with large 14 mask, results of subtraction (a,b). case of rebar mm(c) diameter located at 2 cm depth. (a) original data, (b) data after median filtering with large mask, (c) results of subtraction (a,b).

(a) (b) (c) (d) (a) (b) (c) (d) Figure 8. The result of background removal for the chosen case. Rebar diameter—14 mm, (a) (b) (c) (d) (a) depth—2.5 cm,result (b) depth—2 cm, (c) depth—1.5 Figure 8. The of background removal cm, for (d) thedepth—1 chosen cm. case. Rebar diameter—14 mm, Figure 8. The result of background removal for the chosen case. Rebar diameter—14 mm, (a) depth—2.5 cm, (a) depth—2.5 cm, (b) depth—2 cm, (c) depth—1.5 cm, (d) depth—1 cm. Rebar diameter—14 mm, Figure 8. The result of background removal for the chosen case. (b) Experimental depth—2 cm,Methods (c) depth—1.5 cm, (d) depth—1 cm. 2.2. and Results (a) depth—2.5 cm, (b) depth—2 cm, (c) depth—1.5 cm, (d) depth—1 cm.

2.2. Experimental Methods and Results The numerical modeling has shown that the detection of rebar is possible using active 2.2. Experimental Methods and Results 2.2. Experimental Methods and Results thermography with microwave To that validate previous of experiments The numerical modeling excitation. has shown the the detection of results, rebar isa series possible using active have also been conducted. The numerical modeling shown that the detection detection rebar isseries possible using The numerical modelinghas has shown To that the ofofrebar is apossible activeactive thermography with microwave excitation. validate the previous results, ofusing experiments The experimental setup is excitation. composed microwave heating device, an A325 thermovision thermography with microwave excitation. of validate the results, a series of experiments have also been conducted. thermography with microwave Toavalidate theprevious previous results, a series of experiments (FLIR Systems, Inc., OR,ofUSA) and the tested with an have also been conducted. experimental setupWilsonville, is composed a microwave heatingsample. device,Aanmagnetron A325 thermovision havecamera also The been conducted. output power of 1 kW, operating on 2.45 GHz, connected directly to rectangular waveguide, is usedan The experimental setup is composed of a microwave heating device, an A325 thermovision camera (FLIR Systems, Inc., Wilsonville, OR, USA) and the tested sample. A magnetron with The experimental setup is composed of a microwave heating device, an A325 thermovision camera ascamera a highpower power open waveguide's aperture is pointed directlywith at the (FLIR Systems, Inc., generator. Wilsonville, OR, USA) and the tested A magnetron an output ofmicrowave 1 kW, operating on 2.45The GHz, connected directly to sample. rectangular waveguide, is used (FLIR Systems, Inc., Wilsonville, OR, USA) and the tested sample. A magnetron with an output power sample surface. For safety purposes accuracy theaperture setupdirectly is located in output power of 1microwave kW, operating onand 2.45experimental GHz, connected directly toexperimental rectangular waveguide, isat used as a high power generator. The open waveguide's is pointed the of 1 kW, operating on 2.45 GHz, connected directly to rectangular waveguide, is used as a high power an anechoic chamber. as a high power microwave generator. open waveguide's aperture is pointed directly at the sample surface. For safety purposes and The experimental accuracy the experimental setup is located in microwave generator. open waveguide's apertureaccuracy is pointed at the sample surface. sample surface. For The safety purposes and experimental the directly experimental setup is located in For an anechoic chamber. safetyanpurposes and experimental accuracy the experimental setup is located in an anechoic chamber. anechoic chamber.

Figure 9. (a) photo of the experimental setup, (b) samples: S1—rod diameter 8 mm, S2—rod diameter 10Figure mm, and S3—rod diameter 12 mm. setup, (b) samples: S1—rod diameter 8 mm, S2—rod diameter 9. (a) photo of the experimental 10 mm,9.and S3—rod 12 mm. setup, (b) samples: S1—rod diameter 8 mm, S2—rod diameter Figure (a) photo of diameter the experimental

Figure 9. (a) photo of the experimental setup, (b) samples: S1—rod diameter 8 mm, S2—rod diameter 10 mm, and S3—rod diameter 12 mm. 10 mm, and S3—rod diameter 12 mm.

Sensors 2016, 2016, 16, 16, 234 234 Sensors

of16 16 88of

Figure 9 presents the described setup. Three samples, with different steel rebar diameters Sensors 2016, 16, 234 8 of 16 (8, 10 Figure presents theThe described Three different rebar diameters and 12 mm)9were tested. positionsetup. of the rod wassamples, set to 1.5with cm below the steel concrete surface. The Figure 9 presents the described setup. Three samples, with different steel rebar diameters (8, 10 (8, 10 and 12 mm) were tested. The position of the rod was set to 1.5 cm below the concrete surface. time of heating was set to 60 s. For safety reasons, the sample was not observed during the heating and 12ofmm) were was tested. The position of safety the rodreasons, was set tothe 1.5 sample cm below thenot concrete surface. The the The time set 60 s. For observed during phase. Afterheating the heating the to sample's surface was observed using awas thermovision camera for an time of heating was set to 60 s. For safety reasons, the sample was not observed during the heating heating phase. After the heating the sample's wasInobserved thermovision camera additional 5 min, during the natural coolingsurface process. total, forusing eacha sample sequences of phase. After the heating the sample's surface was observed using a thermovision camera for an for an additional 5 min, during the natural cooling process. In total, for each sample sequences of 300 additional thermograms were recorded (1 thermogram perIn second). 10 presents 5 min, during the natural cooling process. total, for Figure each sample sequencesthe of raw 300 thermograms were recorded (1 thermogram per second). Figure 10 presents the raw thermograms thermograms acquired for each sample, just after the heating (1st thermogram in the sequence) 300 thermograms were recorded (1 thermogram per second). Figure 10 presents the raw and acquired eachacquired sample, just after the heating in the sequence) and at the end of at thethermograms endfor of cooling process (300th thermogram from the sequence). It can beinnoticed that results for each sample, just after(1st thethermogram heating (1st thermogram the sequence) and cooling process (300th thermogram from to thethose sequence). It can beIt noticed that the results obtained obtained experimentally are quite similar from modeling. at the end of cooling process (300th thermogram from thenumerical sequence). can be noticed that the results experimentally are quite similar to those from numerical modeling. obtained experimentally are quite similar to those from numerical modeling.

(a) (a)

(a')(a')

(b) (b)

(b') (b')

(c) (c)

(c') (c')

Figure 10. Thermograms obtainedfor for each each sample: sample: S1 heating phase, (a') after the the Figure 10. Thermograms obtained S1(a) (a)after afterthethe heating phase, (a') after Figure 10. Thermograms obtained for each sample: S1 (a) after the heating phase, (a') after the cooling cooling phase, S2 (b) after the heating phase, (b') after the cooling phase, S3 (c) after the heating cooling phase, S2 (b) after the heating phase, (b') after the cooling phase, S3 (c) after the heating phase, S2 (b) after the heating phase, (b') after the cooling phase, S3 (c) after the heating phase, (c') after phase, (c') after cooling phase. phase, (c') after thethe cooling phase. the cooling phase.

In the case of the experimental data the previously shown technique of background removal In the case of the experimental data the previously shown technique of background removal was also used. results are shown in the Figure 11. It can shown be noticed that this of simple image processing In the case of The the experimental data previously technique removal was was also used. The results are shown in Figure 11. It can be noticed that thisbackground simple image processing technique enables detection of the rebars in the case of sample S3. Also forsimple sampleimage S2 theprocessing rebar also used. The results are shown in Figure 11. It can be noticed that this technique enables detection theberebars in that the case of sample S3. be Also for sample S2 the position is visible. It can of also noticed, the best result can obtained after 100 s of rebar technique enables detection of thebe rebars in the that case of sample S3. Also forbe sample S2 theafter rebar100 position position is visible. It can also noticed, the best result can obtained observation. For sample S1 the rebar detection was not successful. This result shows the method s of is visible. It can also be noticed, that the best result can be obtained after 100 s of observation. For observation. For sample rebar waswas notshown successful. This result shows limitations, related not S1 onlythe with rebardetection depth (which in numerical modeling), but the also method its sample S1 the rebar detection was not successful. This result shows the method limitations, related limitations, related not only with rebar depth (which was shown in numerical modeling), but alsonot its diameter.

only with rebar depth (which was shown in numerical modeling), but also its diameter. diameter.

(a)

(a)

(a')

(b)

(c)

(b)

(c)

(b')

(c')

(d)

(d)

(d')

Figure 11. Cont.

(a')

(b') Figure 11. Cont. Figure 11. Cont.

(c')

(d')

Sensors 2016, 16, 234

9 of 16

Sensors 2016, 16, 234

9 of 16

(a'')

(b'')

(c'')

(d'')

Figure Figure 11. 11. The The results results of of background background removal removal obtained obtained for for all all samples. samples. S1 S1 (a) (a) 1st 1st thermogram thermogram in in the the sequence, 200 s, (d) after 300300 s, S2 1st thermogram in theinsequence, (b') after sequence, (b) (b)after after100 100s,s,(c)(c)after after 200 s, (d) after s, (a') S2 (a') 1st thermogram the sequence, (b') 100 (c') s, after (d')s,after S3 (a'') thermogram in thein sequence, (b'') after (c'')s,after afters, 100 (c') 200 afters,200 (d') 300 afters,300 s, S31st (a'') 1st thermogram the sequence, (b'') 100 afters,100 (c'') 200 s, (d'') after 300 s. after 200 s, (d'') after 300 s.

3. Eddy 3. EddyCurrent CurrentTechnique Technique The second second technique technique discussed discussed here here may may be be used used both both to to detect detect and and evaluate evaluate rebars. rebars. The eddy The current method method uses uses the the phenomena phenomena of of electromagnetic electromagnetic induction induction to to detect detect inner inner flaws flaws or or inclusions inclusions current in the the examined examined materials. materials. This method is a natural natural choice choice in the the case case of of conducting conducting materials materials and and in the possibility possibility of of using using it it in in rebar rebar detection detection has has been been discussed discussed earlier earlier [34–37]. [34–37]. Here we we present present the the the multi-frequency technique and we propose a method that may be used for the problem of multi-frequency technique and we propose a method that may be used for the problem of identification identification of chosen the rebar the basis of obtained signals.should This method should of chosen parameters of parameters the rebar onof the basis of on obtained signals. This method be considered becomplementary considered as complementary thermovision. combination both methods may create the as to thermovision.toThe combinationThe of both methodsof may create the full, multisensor full, multisensor system for rebar detection and identification. system for rebar detection and identification. 3.1. 3.1. Single Single Frequency Frequency Methods Methods The waswas presented in previous work [34,35]. It consists an E-shaped The system systemutilized utilized presented in previous work [34,35]. It of consists of andifferential E-shaped transducer 12b) and four an XYZ an scanner, an excitation subsystem, a dataa differential (Figure transducer (Figure 12b) subsystems: and four subsystems: XYZ scanner, an excitation subsystem, acquisition subsystem and a controlling computer. The exciting coils of the eddy current transducer data acquisition subsystem and a controlling computer. The exciting coils of the eddy current are poweredare by powered two independent function synthesizers powerthrough amplifiers. Theamplifiers. frequenciesThe of transducer by two independent functionthrough synthesizers power the both excitation sinusoidal signals are equal. The alternating currents in the two excitation frequencies of the both excitation sinusoidal signals are equal. The alternating currents in thecoils two have the same amplitude and opposite phase in the case of differential mode or the same phase in excitation coils have the same amplitude and opposite phase in the case of differential mode or the the case of absolute mode. In the case of theIndifferential excitation coils create equalcoils but same phase in the case of absolute mode. the case oftransducer, the differential transducer, excitation opposite directed that cancel eachfluxes other out. in equilibrium a voltage create equal butmagnetic oppositefluxes directed magnetic that Therefore, cancel each other out. state Therefore, in measured at the pick-up coil is small and constant. Variations of the electrical conductivity or magnetic equilibrium state a voltage measured at the pick-up coil is small and constant. Variations of the permeability of the tested object, orpermeability the presenceofofthe flaws, cause changes the eddy flow electrical conductivity or magnetic tested object, or theinpresence ofcurrent flaws, cause and corresponding changes in the phase and amplitude of the measured voltage. An example of the changes in the eddy current flow and corresponding changes in the phase and amplitude the signal achieved during movement of the transducer above the specimen with a single rebar is shown measured voltage. An example of the signal achieved during movement of the transducer above the in Figure 12c,d. measurements were taken 12c,d. in steps ofmeasurements 0.61 mm for different thicknesses of 0.61 the specimen with a The single rebar is shown in Figure The were taken in steps of concrete covers and different samples (the rebars having different diameters D = {8,10,12,14} mm, and mm for different thicknesses of the concrete covers and different samples (the rebars having belonging to three classes). For the SFM of attributes used For in the shape different diameters D = {8,10,12,14} mm, two and kinds belonging to three are classes). theanalysis: SFM twothe kinds of factors d and maximal value of the Umax Values the shape factors d are independent attributes arethe used in the analysis: thewaveform shape factors d .and the of maximal value of the waveform Umax. of the waveform maximal describe of only shape ofmaximal a normalized (e.g., d30the is Values of the shape factorsvalue d are and independent thethe waveform valuewaveform and describe only ashape difference between Xmax and X30(e.g., , where X30 representsbetween position X ofmax the transducer when the signal of a normalized waveform d30 is a difference and X30, where X30 represents amplitude at the pick-up coil was equal U = 30%¨ U ). At the beginning almost 100 attributes position of the transducer when the signal 30 amplitude at the pick-up coil was equal U30 = 30%·U max). max were However, classification when the problemindimensionality increases, At theconsidered. beginning almost 100inattributes were tasks considered. However, classification tasks when the the volume the space increases verythe fastvolume and the become sparse. In the order to obtain a problemofdimensionality increases, ofavailable the spacedata increases very fast and available data reliable result, the amount of data needed to support the result often grows exponentially with the become sparse. In order to obtain a reliable result, the amount of data needed to support the result dimensionality. The curse of dimensionality can be avoided by of reducing the number often grows exponentially with the dimensionality. The curse dimensionality can of bedimensions. avoided by reducing the number of dimensions. Figure 13 presents this phenomena achieved in the case of results obtained for very simple model including Kullback–Leibler divergence attribute filter and Naive Bayes classifier.

Sensors 2016, 16, 234

10 of 16

Figure 13 presents this phenomena achieved in the case of results obtained for very simple model Sensors 2016, 16, 234 10 of 16 including Kullback–Leibler divergence attribute filter and Naive Bayes classifier.

Figure 12. 12. (a) Tested Tested specimen; specimen; (b) (b) Cross-section Cross-section of of the the transducer; transducer; (c) (c) Example Example of of the the signal signal received received Figure at the pick-up coil of the differential transducer; (d) Example of the signal received at the pick-up coil at the pick-up coil of the differential transducer; (d) Example of the signal received at the pick-up coil of the absolute transducer. of the absolute transducer.

Figure 13. 13.Simple Simple classification model correctness of classification vs.ofnumber ofParameters: attributes. Figure classification model correctness of classification vs. number attributes. Parameters: (a) concrete cover thickness, (b) rebar diameter, (c) rebar class. (a) concrete cover thickness, (b) rebar diameter, (c) rebar class.

The simple model presented can be used to determine the complexity for every particular The simple model presented can be used to determine the complexity for every particular classification task obtained areare notnot satisfactory. A final model was classification task (Figure (Figure 13). 13).However, However,the theresults results obtained satisfactory. A final model built based on the rough set theory [7,9]. This method allows one to presents relations between was built based on the rough set theory [7,9]. This method allows one to presents relations between received voltage waveform and tested sample in the form of “if/then” rules. Moreover, the method received voltage waveform and tested sample in the form of “if/then” rules. Moreover, the method also allows allowsthe theselection selectionofofanan optimal attributes During a relatively transducer also optimal attributes set.set. During the the teststests a relatively largelarge transducer was was used (T20 ). As one can see in Table 2, the transducer achieves its best performance for a concrete 20 used (T20 ). As one can see in Table 2, the transducer achieves its best performance for a concrete cover cover thickness h to from 15 to mm.distances For these only is the classification thickness range h range from 15 35 mm. For35these notdistances only is thenot classification correctness the correctness the highest, but also the attribute standard deviation is the lowest (statistics highest, but also the attribute standard deviation is the lowest (statistics were calculated based onwere the calculated based on the database including 110 instances for every value of the concrete cover database including 110 instances for every value of the concrete cover thickness h). Studies showed the thickness h).ofStudies showed the importance the especially correctly selected transducer size,the especially the importance the correctly selected transducerofsize, the distance m between transducer distance m between the transducer columns (between the pick-up coil and the excitation coil). In this columns (between the pick-up coil and the excitation coil). In this research three different transducers research three different transducers were used: T25 where m = 25 mm, T20 20 where m = 20 mm and T55 were used: T25 where m = 25 mm, T20 where m 25 = 20 mm and T5 where m = 5 mm. The smaller where m = 5have mm.good The smaller transducerswhile havebigger good spatial resolution, while biggerThis transducers are transducers spatial resolution, transducers are more sensitive. means that more sensitive. This means that the small transducers work correctly only when the thickness of the the small transducers work correctly only when the thickness of the concrete cover (i.e., the distance concrete the cover (i.e., the distance between the ferromagnetic bar and thetransducer transducer) is small. The between ferromagnetic bar and the transducer) is small. The larger provides better larger transducer provides better results for a bigger thickness h, however its spatial resolution and repeatability are low. Moreover, large size transducers provide relatively good results for a big interval of h values. Small transducers should be used when the distance between the pick-up coil and the sample is small and when the interval of distances h is small. Results obtained for the three different transducers are presented in Table 3. Both kinds of the transducer operation modes were

Sensors 2016, 16, 234

11 of 16

results for a bigger thickness h, however its spatial resolution and repeatability are low. Moreover, large size transducers provide relatively good results for a big interval of h values. Small transducers should be used when the distance between the pick-up coil and the sample is small and when the interval of distances h is small. Results obtained for the three different transducers are presented in Table 3. Both kinds of the transducer operation modes were evaluated. In many cases the differential eddy current transducers are more effective. Small magnitudes of voltage on the pick-up coil result in very high sensitivity, therefore these kinds of transducers are perfect for crack detection and evaluation. In this case, the quality of results collected by the SFM for both types of connection is comparable. However the MMFM is more effective when the transducer coils are connected in other ways. Therefore, results presented in the next section were collected using absolute connection. Table 2. Correctness of the classification and standard deviation of selected attributes. h [mm]

5

10

15

20

25

30

35

40

45

50

D Class σd80 σd10

75% 78% 0.530 5.986

81% 83% 0.480 4.259

91% 92% 0.295 0.707

96% 100% 0.257 0.450

92% 87% 0.295 0.450

93% 94% 0.295 0.502

84% 95% 0.257 0.518

75% 79% 0.498 1.068

78% 78% 0.561 1.179

66% 74% 1.114 1.841

Table 3. Transducer parameters and statistics. Transducer size

T5

T20

T25

Optimal range of h [mm] Correctness of D classification in the optimal range [%] Correctness of D classification for h = 40 to 50 mm [%]

0–25 94–98 58–68

15–35 84–96 66–78

15–35 91–93 72–78

3.2. Massive Multi-Frequency and Spectrogram Method The MMFM [18,19] is based on a few simple physical phenomena. The first one is an eddy current phenomenon. The other is the fact that low frequencies have a large skin depth and hence give clear signals from support structures that are located away from the coil. Because of the different depths of penetration for different frequencies, the relationship between signals for different samples (h) changes with frequency. Consequently, it is possible to combine the signals from many different frequencies. It was proved [29–32] that the MMFM system allows changes of the reinforcement structure to be observed through changes in the corresponding spectrograms. The frequencies for which the amplitude of components display the largest changes depend on changes in the tested structure [32]. This can be used not only for measuring the thickness of conductive coatings on conductive base metals, detection and identification of flaws in conducting materials, or differentiating between flaws in various layers of built-up structures, but also in the task of evaluating the distance between a transducer and ferromagnetic elements [29,30] or identification of ferromagnetic shapes or material. An additional advantage of the utilization of numerous frequency components in the excitation is that every single frequency consists of independent information, therefore, the impact of noise can be reduced. Moreover the presented method is much faster than Swept Frequency Methods. The scheme of the system [31] is presented in Figure 14a. Exciting coils are excited simultaneously by the multi-frequency signals given as: y ptq “

n ÿ

Ui ¨ sin pvi ¨ t ` ϕi q

i“1

The transducer is moved over the sample. The voltage waveform measured during the movement at the pick-up coil is decomposed by a FFT algorithm. Normally, when a sample is uniform, the FFT products can easily be set equal. However, to receive equal output signals the magnitude of the

Sensors 2016, 16, 234

12 of 16

excitation signal components must be precisely selected, as presented in Figure 15. In order to build the multi-frequency excitation signal 45 components having frequencies from 0.5 kHz to 200 kHz were used (Figure 15). The frequency set was selected with inconstant steps, that is, the size of the steps changed logarithmically to show the phenomena properly. The basic concept of MMFM is based on the spectrogram Sensors 2016, 16, peak 234 [29] (highest value in spectrogram) parameters (Smax , f max ) (Figure 14b). When 12 ofthe 16 distance between Sensors 2016, 16, 234 the transducer and rebar becomes larger, then the spectrogram peak f max is shifted 12 of 16 spectrogram peak fmax is shifted frequencies. An example of input a frequency spectrum of towards higher frequencies. An towards example higher of a frequency spectrum of typical and output signals typical input and output signals is presented in Figure 15. spectrogram peak fmax is is presented in Figure 15.shifted towards higher frequencies. An example of a frequency spectrum of

typical input and output signals is presented in Figure 15.

(a)

(b)

(a)

(b)

Figure 14.14. TheThe Massive Multi-Frequency Method (a) (a) system; (b)(b) basic concept of identification. Figure Massive Multi-Frequency Method system; basic concept of identification. Figure 14. The Massive Multi-Frequency Method (a) system; (b) basic concept of identification.

Figure 15. The frequency spectral of signal magnitude fed to: (a) exciting coils; (b) pick-up coil. The one periodThe of time signal fed to: (c) exciting coils; (d) pick-up coil.exciting coils; (b) pick-up coil. The Figure Figure 15. 15. The frequency frequency spectral spectral of of signal signal magnitude magnitude fed fed to: to: (a) (a) exciting coils; (b) pick-up coil. The one one period period of of time time signal signal fed fed to: to: (c) (c) exciting exciting coils; coils; (d) (d) pick-up pick-up coil. coil.

3.3. Results and Identification

3.3. and The MMFM can be used to select the best frequency for SFM inspection [19]. From the 3.3. Results Results and Identification Identification investigator’s point of be view thetomost interesting the frequency where changes [19]. in the received The used select the best best isfrequency frequency for SFM SFM inspection The MMFM MMFM can can be used to select the for inspection [19]. From From the the signal are the largest. Therefore, in most cases the best testing frequency is f max , the frequency where investigator’s point of view the most interesting is the frequency where changes in the received investigator’s point of view the most interesting is the frequency where changes in the received the peak valuelargest. Smax is observed. in most cases the best testing frequency is fmax, the frequency where signal signal are are the the largest. Therefore, Therefore, in most cases the best testing frequency is f max , the frequency where The value method can also be used to select the best among available transducers. Tables 2 and 3 the peak S isisobserved. the peak value Smax observed. max showThe thatmethod each transducer a range of distances from particular sample, for which it works can also behas used select best among available transducers. Tables 3 The method can also be used toto select thethe best among available transducers. Tables 2 and23and show properly. In thetransducer SFM measurements to of detect this range interval it sample, was necessary to build the show that each has a range distances from particular for which it works that each transducer has a range of distances from particular sample, for which it works properly. corresponding and calculate to thedetect statistics. the testsit110 measurements each properly. thedatabase SFM measurements this During range interval necessary to for build theh In the SFMInmeasurements to detect this range interval it was necessarywas to build the corresponding were carried out. A much faster way to get the same result is to use the MMFM and create the corresponding database calculate the statistics. During the tests 110 for each h database and calculate theand statistics. During the tests 110 measurements for measurements each h were carried out. A corresponding spectrograms (Figure 16). The the largest changes are observed for theand spectrograms were carried out. A much faster way to get same result is to use the MMFM create the much faster way to get the same result is to use the MMFM and create the corresponding spectrograms achieved for h values between(Figure 7 mm 16). to 30The mm.largest The differences between spectrograms obtained for corresponding spectrograms changes are observed spectrograms (Figure 16). The largest changes are observed for the spectrograms achievedfor for the h values between intervals of h between 2 mm to 7 mm and 30 mm to 40 mm are also significant. For higher valuesfor of achieved formm. h values 7 mm to 30 mm. The differences between spectrograms obtained 7 mm to 30 The between differences between spectrograms obtained for intervals of h between 2 mm concrete cover thickness the changes are small, therefore identification can be difficult. The statistical intervals of h 30 between mm toare 7 mm 30 mm toFor 40 higher mm arevalues also significant. higher valuesthe of to 7 mm and mm to240 mm alsoand significant. of concreteFor cover thickness results presented in Tables 2 and 3 could be expanded to a larger depth interval by increasing the concrete cover thickness the changes are small, therefore can be difficult. Thein statistical changes are small, therefore identification can be difficult.identification The statistical results presented Tables 2 numberpresented of excitation frequencies. transducers can be utilized also for low distances. results in Tables 2 and Therefore 3 could belarge expanded to a larger depth interval by increasing the This is one of the biggest advantages MMFMlarge has over the SFM, because one large transducer can be number of excitation frequencies. Therefore transducers can be utilized also for low distances. usedisover whole range. This onethe of the biggest advantages MMFM has over the SFM, because one large transducer can be identification used The overbasic the whole range. process is based on the fmax. However, studies show that the changes of fmax are only for theprocess 5–15 mm range. be conducted wider Theobserved basic identification is based onThe the identification fmax. However,can studies show thatfor themuch changes of

Sensors 2016, 16, 234

13 of 16

and 3 could be expanded to a larger depth interval by increasing the number of excitation frequencies. Therefore large transducers can be utilized also for low distances. This is one of the biggest advantages Sensors 2016, 16, 234 13 of 16 MMFM has over the SFM, because one large transducer can be used over the whole range.

Figure 16. 16. Spectrograms Spectrograms and and spectrograms spectrograms max. max. value value profiles profiles created created for for 45 45 different different values Figure values of of hh ..

The basic identification process is based on of thedistance f max . However, show that the changes of During this investigations the assessment betweenstudies the ferromagnetic material and f max are observed only for the mm range. The identification canlike be conducted transducer was considered the5–15 simplest task (Figure 13). However with SFM, for themuch set ofwider most intervals of h. To was make this possible, profiles ofrules maximal spectrograms are used. These profiles useful attributes selected and association were value inferred [35,36]. Rough set theory was used show local f max values for this purpose [38,39].(a component with the highest magnitude) for every single transducer position (Figure below the spectrograms). For16, low h identification can be carried out based only on three parameters: global fmax, and two assessment of position distancewhen between ferromagnetic material and local During fmax. Thethis firstinvestigations one is a local fthe max for transducer one the of the excitation coils is directly transducer was considered the simplest task (Figure 13). However likebetween with SFM, the set ofand mostrebar useful above the rebar. This attribute is very significant when the distance transducer is attributes selected association rules were theory for this small. Thewas other one is and a local fmax taken when theinferred rebar is[35,36]. placed Rough over 40set mm fromwas the used pick-up coil (when the impact of the eddy current is small). This one is significant when the distance between purpose [38,39]. transducer and rebar is higher than 30 mm.out Based only on these three parameters it is possible totwo get For low h identification can be carried based only on three parameters: global f max , and about 95% correct identifications (iffor wetransducer know theposition rebar diameter class). However, the most local f max . The first one is a local f max when oneand of the excitation coils is directly universal useful surface area under line. This attribute allows to above the and rebar. Thisattribute attributeisisthe very significant when the the profile distance between transducer andone rebar expand the identification over the the whole range. Even over if the40 other of the is small.very The effectively other one is a local f max taken when rebar is placed mmparameters from the pick-up structure is still very and differences between h and coil (whenare theunknown, impact of the the classification eddy current error is small). This onesmall is significant when the distancereal between predicted are smaller 2 mm. transducerh and rebar isthan higher than 30 mm. Based only on these three parameters it is possible to The identification of the rebar diameter is even more difficult. Attributes from the previous case get about 95% correct identifications (if we know the rebar diameter and class). However, the most are the most However to surface improvearea the under correctness of classification it is necessary toone addto a universal andsignificant. useful attribute is the the profile line. This attribute allows few more attributes. The number of attributes and correctness of classification in this case depends on transducer size and the method used for identification of a rebar class and ranges between 92%–98%. The last case is rebar class identification. The class classification is relatively difficult when conducted based only on the profile surface area, and local/global fmax parameters. However, correctness of classification still can be around 90%.

Sensors 2016, 16, 234

14 of 16

expand very effectively the identification over the whole range. Even if the other parameters of the structure are unknown, the classification error is still very small and differences between real h and predicted h are smaller than 2 mm. The identification of the rebar diameter is even more difficult. Attributes from the previous case are the most significant. However to improve the correctness of classification it is necessary to add a few more attributes. The number of attributes and correctness of classification in this case depends on transducer size and the method used for identification of a rebar class and ranges between 92%–98%. The last case is rebar class identification. The class classification is relatively difficult when conducted based only on the profile surface area, and local/global f max parameters. However, correctness of classification still can be around 90%. Previous investigations have shown that the rebar class is strongly correlated with the signal magnitude [34]. Therefore, to get better results it is reasonable to add Smax to the database or the maximal value of all frequency waveforms. 4. Conclusions Thermography with microwave excitation allows for the rough detection of rebar reinforcements. This method can be used on large areas (using for example antenna arrays) which may simultaneously be observed using a thermovision camera. This is the main advantage of this technique. According to the measurement and numerical simulation results, changes in the rebars’ diameter (in the range between 8 and 14 mm) has a very moderate influence on their detectability. Unfortunately the method has also some strong limitations. Detection of rebars becomes difficult when the rebars are located deeper than 20 mm under the surface of the concrete. This issue can be overcome by utilizing signal processing techniques like principal component thermography. Another solution is to use an additional different technique. In this paper the authors proposed the successful multi-frequency eddy current method as a complementary diagnostic tool. The massive multi-frequency method (MMFM) allows for fast, easy and accurate identification of the reinforced concrete structure parameters. Although the MMFM requires building more complex hardware and software than the SFM system, it has many advantages. Large numbers of frequency components in the excitation signal allow the full testing of a structure with high accuracy, a far bigger than it is possible with SFM. The method can also be used to expand the operating range of the transducer and the identification process is easier than in the case of the single frequency method. The method can therefore be used as a substitute or a complementary technique to the standard single-frequency identification method. In summary, the multi-frequency method is more effective, but also more complex than the single-frequency one. Acknowledgments: This work is supported by Foundation for Polish Science and cosponsored by UE (Program Operacyjny Innowacyjna Gospodarka, Priorytet 1, Działanie 1.2), under grant VENTURES/2013-11/5. Author Contributions: Barbara Szymanik—conception of utilizing the active thermography method with microwave excitation in rebar detection. Conduction of experiments and numerical modelling in the field of thermography. Data analysis. Writing the article. Paweł Frankowski—Leading the research project. Overall conception of this paper. Conduction of experiments in the field of eddy currents. MMFM system implementation and development. Sample preparation (both methods). Data analysis and identification. Writing the article. Tomasz Chady—conception and development of the massive multi frequency method (MMFM) for eddy current testing, conception and development of the hardware system for MMFM ECT, conception of the identification based on the spectrogram analysis, conception and manufacturing of the multiple size eddy current transducers. Cyril R. A. John Chelliah—assistance in conduction of experiments and numerical modelling in the field of thermography. Conflicts of Interest: The authors declare no conflict of interest.

References 1.

Masoumi, F.; Akgül, F.; Mehrabzadeh, A. Condition Assessment of Reinforced Concrete Bridges by Combined Nondestructive Test Techniques. IACSIT Int. J. Eng. Technol. 2013, 5, 708–711. [CrossRef]

Sensors 2016, 16, 234

2. 3. 4. 5. 6. 7. 8. 9.

10. 11. 12. 13.

14.

15. 16.

17.

18. 19. 20.

21.

22.

23.

15 of 16

Hoła, J.; Schabowicz, K. State-of-the-art non-destructive methods for diagnostic testing of building structures-anticipated development trends. Arch. Civil Mech. Eng. 2010, 10, 5–18. [CrossRef] Maierhofer, C.; Reinhardt, H.W.; Dobmann, G. Non-Destructive Evaluation of Reinforced Concrete Structures; Woodhead Publishing CRC Press: Cambridge, UK, 2010. Sakata, Y.; Ohtsu, M. Crack evaluation in concrete members based on ultrasonic spectroscopy. ACI Mater. J. 1995, 92, 686–698. Schickert, M. Ultrasonic NDE of concrete. In Proceedings of the 2002 IEEE Ultrasonics Symposium, Munich, Germany, 8–11 October 2003; pp. 739–748. Lorenzi, A.; Tisbierek, F.T.; Silva, L.C.P. Ultrasonic Pulse Velocity Analysis in Concrete Specimens. In Proceedings of the IV Conferencia Panamericana de END, Buenos Aires, Argentina, 22–26 October 2007. Clayton, D.A. Nondestructive Evaluation of Thick Concrete Structures. In Proceedings of the International Symposium Non-Destructive Testing in Civil Engineering (NDT-CE), Berlin, Germany, 15–17 September 2015. Al-Qadi, L.; Lahour, S. Ground Penetrating Radar: State of the Practice for Pavement Assessment. J. Mater. Eval. 2004, 42, 759–763. Morcous, G.; Erdogmus, E. Use of Ground Penetrating Radar for Construction Quality Assurance of Concrete Pavement; NDOR Project Number P307, FINAL REPORT; University of Nebraska-Lincoln: Lincoln, NE, USA, 2009. Maierhofer, C. Nondestructive Evaluation of Concrete Infrastructure with Ground Penetrating Radar. ASCE J. Mater. Civil Eng. 2003, 15, 287–297. [CrossRef] Hasan, M.I.; Yazdani, N. Ground penetrating radar utilization in exploring inadequate concrete covers in a new bridge deck. Case Stud. Construct. Mater. 2014, 1, 104–114. [CrossRef] Clark, M.; McCann, D.; Forde, M. Application of infrared thermography to the non-destructive testing of concrete and masonry bridges. NDT&E Int. 2003, 36, 265–275. : Maierhofer, C.; Arndt, R.; Rollig, M.; Rieck, C.; Walther, A.; Scheel, H.; Hillemeier, B. Application of impulse-thermography for non-destructive assessment of concrete structures. Cem. Concr. Compos. 2006, 28, 393–401. [CrossRef] Martz, H.; Roberson, G.; Skeate, M.; Schneberk, D.; Azevedo, S. Computerized tomography studies of concrete samples. Nucl. Instrum. Methods Phys. Res. Sect. B Beam Interact. Mater. Atoms 1991, 58, 216–226. [CrossRef] Martz, H.; Scheberk, D.; Roberson, G.; Monteiro, P. Computerized tomography analysis of reinforced concrete. ACI Mater. J. 1993, 90, 259–264. Paetsch, O.; Baum, D.; Ehrig, K.; Meinel, D.; Prohaska, S. Automated 3D Crack Detection for Analyzing Damage Processes in Concrete with Computed Tomography. In Proceedings of the ICT Conference Wels 2012, Wels, Austria, 19–21 September 2012; pp. 321–330. Paetsch, O.; Baum, D.; Prohaska, S.; Ehrig, K.; Meinel, D.; Ebell, G. 3D Corrosion Detection in Time-dependent CT Images of Concrete. In Proceedings of the Digital Industrial Radiology and Computed Tomography (DIR 2015), Ghent, Belgium, 22–25 June 2015. Chady, T.; Enokizono, M. Flaw Reconstruction using a Multi-Frequency Method and an Artificial Intelligence. JSAEM Stud. Appl. Electromagn. Mech. 1999, 8, 203–208. Chady, T.; Enokizono, M. Multi-frequency exciting and spectrogram based ECT method. J. Magn. Magn. Mater. 2000, 215–216, 700–703. [CrossRef] Chady, T.; Łopato, P. Flaws Identification Using Eddy Current Differential Transducer and Artificial Neural Networks. In Review of Quantitative Nondestructive Evaluation; Thompson, D.O., Chimenti, D.E., Eds.; American Institute of Physics, AIP CP 820: Melville, NY, USA, 2005; Volume 25, pp. 783–790. Berowski, P.; Filipowicz, S.F.; Sikora, J.; Wójtowicz, S. Determining Location of Moisture Area of the Wall by 3D Electrical Impedance Tomography. In Proceedings of the 4th World Congress on Industrial Process Tomography, Aizu, Japan, 2–5 September 2005. Seppänen, A.; Hallaji, M.; Pour-Ghaz, M. Electrical impedance tomography-based sensing skin for detection of damage in concrete. In Proceedings of the 11th European Conference on Non-Destructive Testing (ECNDT 2014), Prague, Czech Republic, 6–10 October 2014. Karhunen, K.; Seppänen, A.; Lehikoinen, A.; Monteiro, P.J.M.; Kaipio, J.P. Electrical resistance tomography imaging of concrete. Cem. Concr. Res. 2010, 40, 137–145. [CrossRef]

Sensors 2016, 16, 234

24.

25. 26.

27. 28. 29. 30.

31.

32. 33.

34. 35.

36. 37. 38. 39.

16 of 16

Brachelet, F.; Keo, S.; Defer, D.; Breaban, F. Detection of reinforcement bars in concrete slabs by infrared thermography and microwaves excitation. In Proceedings of the QIRT 2014 Civil Engineering & Buildings, Bordeaux, France, 7–11 July 2014. Metaxas, A.C.; Meredith, R.J. Industrial Microwave Heating; Peter Peregrinus Ltd.: London, UK, 1983. Szymanik, B. Zastosowanie Aktywnej Termografii Podczerwonej ze Wzbudzeniem Mikrofalowym do Wykrywania Niemetalicznych min Ladowych. ˛ Ph.D. Thesis, West Pomeranian University of Technology, Szczecin, Poland, 2013. (In Polish). Meredith, R.J. Engineers’ Handbook of Industrial Microwave Heating; Short Run Press: London, UK, 1998. Makulan, N.; Rattanadechob, P.; Agrawal, D.K. Applications of microwave energy in cement and concrete—A review. Renew. Sustain. Energy Rev. 2014, 37, 715–733. [CrossRef] Chady, T.; Enokizono, M.; Sikora, R.; Takeuchi, K.; Kinoshita, T. Eddy Current Testing of Concrete Structures. Int. J. Appl. Electromagn. Mech. 2002, 15, 33–37. Nagata, S.; Chady, T.; Shidouji, M.; Enokizono, M. Development of Practical MFES System for Concrete Materials. In Electromagnetic Nondestructive Evaluation VI; Kojima, F., Takagi, T., Udpa, S.S., Pávó, J., Eds.; IOS Press: Amsterdam, The Netherlands, 2002; pp. 104–107. Chady, T.; Gratkowski, S.; Nagata, S.; Sikora, R.; Wójtowicz, S. Eddy Current Inspection of Reinforcement Bars in Concrete Structures. In Electromagnetic Nondestructive Evaluation (IX); Udpa, S.S., Nicola Bowler, N., Eds.; IOS Press: Amsterdam, The Netherlands, 2005; pp. 203–210. Chady, T.; Sikora, R. Optimization of Eddy-Current Sensor for Multifrequency Systems. IEEE Trans. Mag. 2003, 39, 1313–1316. [CrossRef] Szymanik, B.; Unnikrishnakurup, S.; Balasubramaniam, K. Background Removal Methods in Thermographic Non Destructive Testing of Composite Materials. In Proceeding of the NDE 2014, Pune, India, 4–6 December 2014. Chady, T.; Frankowski, P. Electromagnetic Evaluation of Reinforced Concrete Structure. Rev. Prog. Quant. Nondestruct. Eval. 2013, 32, 1355–1362. Frankowski, P.; Chady, T.; Sikora, R. Identification of Rebars in a Reinforced Mash Using Eddy Current Method and Association Rule Learning. In Proceedings of the Far East NDT, Zhuhai, China, 22–24 June 2015; pp. 202–207. Frankowski, P.K.; Chady, T.; Sikora, R. Knowledge Extraction Algorithms Dedicated for Identification of Steel Bars in Reinforced Concrete Structures. Rev. Prog. Quant. Nondestruct. Eval. 2014, 40, 822–827. Frankowski, P.K. Knowledge Extraction From the Eddy Current Measurement Data, (pl). Inform. Autom. ´ Pomiary Gospod. Ochr. Srodowiska 2014, 4.2, 822–827. Pawlak, Z. Rough Sets: Theoretical Aspects of Reasoning About Data; Kluwer Academic Publishing: Dordrecht, The Netherlands, 1991. Shearer, C. The CRISP-DM model: The new blueprint for data mining. J. Data Warehous. 2000, 5, 13–22. © 2016 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons by Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).