Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Construction and Building

Construction and Building Materials 19 (2005) 755762

MATERIALS
www.elsevier.com/locate/conbuildmat

Modelling ground penetrating radar by GprMax


A. Giannopoulos
School of Engineering and Electronics, Institute for Infrastructure and Environment, University of Edinburgh, Alexander Graham Bell Building, Kings Buildings, Edinburgh EH9 3JN, UK Available online 2 August 2005

Abstract This paper deals with the fundamentals of ground penetrating radar (GPR) operation and presents a software tool that can be used to model GPR responses from arbitrarily complex targets. This software tool called GprMax is available free of charge for both academic and commercial use and has been successfully employed in situations, where a deeper understanding of the operation and detection mechanism of GPR was required. Examples from both 2D and 3D models are presented which demonstrate the use of GprMax. The tool can be downloaded from www.gprmax.org or by contacting the author. 2005 Elsevier Ltd. All rights reserved.
Keywords: Ground penetrating radar; Modelling; Non-destructive testing; GprMax; Finite-dierence time-domain

1. Introduction Ground penetrating radar is customarily used for the non-destructive testing (NDT) of structures and transport systems [1,2]. The use of GPR for NDT is just one of the many dierent areas, where radar is being applied as an important tool assisting the engineer or scientist in their eorts to determine the presence or absence as well as the kind of key underground features [3]. The main advantages of GPR are: its fast data acquisition capability, its high resolving ability and the fact that it responds equally well to metallic and nonmetallic targets. Its main drawback is the complex nature of its data and the diculty that the GPR user faces when trying to interpret them. Interpretation of the data acquired using a GPR is often a complicated and error prone procedure mainly due to the complexity of the GPR signals and the variety of factors that inuence and determine them. In order to successfully interpret GPR data a considerable amount of expertise is required by the systems user or the data analyst. This experience
E-mail address: A.Giannopoulos@ed.ac.uk. 0950-0618/$ - see front matter 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.conbuildmat.2005.06.007

is usually obtained either by expensive experimentation or by referring to prior knowledge often obtained after costly mistakes in the eld. Due to the frequent complexity of GPR data their interpretation is usually limited to dening general areas of interest or just locating anomalies instead of accurately determining the type and size and exact position of targets. The inability of interpreting GPR data quantitatively lies partly on the lack of understanding of how they have been created (i.e., understanding of the complex interactions between radars electromagnetic elds and the targets) as well as on the lack of advanced processing tools that can process GPR data beyond the simple ltering procedures employed by most GPR users today. In order to improve our ability to understand and get more out of GPR data more sophisticated data analysis tools as well as better understanding of the radars detection mechanism are required. Modelling of GPR responses either analytically or numerically plays a central role in advancing our understanding of GPR as well as providing the means for testing new data processing techniques and

756

A. Giannopoulos / Construction and Building Materials 19 (2005) 755762

interpretation software. The availability of a free GPR modelling tool gives both researchers and practitioners the opportunity to numerically experiment with GPR on their computers without incurring a substantial cost by creating expensive physical models at least at an initial stage of a project. Simulating what-if scenarios can save money and time as well as provide data to support project proposals that could employ GPR but need some preliminary evidence in order to convince more sceptic project managers of GPRs suitability to solve a given problem. Most importantly however, a freely available and well documented modelling tool avoids the syndrome of re-inventing the wheel that plagues many new research eorts that need GPR modelling facilities. Successful modelling attempts of ground penetrating radar have been reported by many authors [46]. Most of the proposed approaches are based on the nitedierence time-domain (FDTD) method. The main reasons for such widespread use of the FDTD method are: its ease of implementation in a computer programme at least at a simple introductory level and its good scalability when compared with other popular electromagnetic modelling methods such the nite-element and integral techniques [7]. The main drawbacks of the FDTD technique are: the need to discretize the volume of the problem space which could lead to excessive computer memory requirements and the staircase representation of curved interfaces. Details of the development of the FDTD method can be found in [8] which provides an excellent introduction to the technique.

rE

oB ; ot oD J c J s; rH ot r B 0; r D qv ;

1 2 3 4

where t is time (seconds) and qv is the volume electric charge density (coulombs/cubic meter). In Maxwells equations, the eld vectors are assumed to be single-valued, bounded, continuous functions of position and time. Further, the macroscopic eects of charged particles contained in a given medium on the electromagnetic elds interacting with it are described by the constitutive relations which are [10]: D  E; B l H ; J c r E; 5 6 7

2. Numerical modelling of GPR The analytical solution of the GPRs forward problem is a dicult and if realistic problems are to be addressed an almost impossible task for even experienced scientists and engineers. The analytical diculties in solving the GPR problem arise from the lack of closed form solutions to general electromagnetic dipole sources over a half space category of problems. The formulations of such problems lead to the well known Sommereld type of integrals [9] which are dicult to be accurately evaluated even numerically. Most importantly though the need to model with as much detail and realism possible complex GPR targets as well as include in our advanced models the GPR transducers precludes the use of purely analytical techniques. All electromagnetic phenomena, on a macroscopic scale, are described by the well known Maxwells equations. These are rst order partial dierential equations that express mathematically the relations between the fundamental electromagnetic eld quantities and their dependence on their sources [10].

where * denotes convolution and the tensors , l and r are the permittivity (farads/meter), permeability (henries/meter) and conductivity (siemens/meter) of the medium, respectively, and are usually referred to as the constitutive parameters. They could, in general, be functions of position, direction, the strength of the applied eld and time. These relations couple the ux quantities with the electric and magnetic eld intensities. In order to simulate the GPR response from a particular target or set of targets the above equations have to be solved subject to the geometry of the problem and its initial conditions. The nature of the GPR forward problem classies it as an initial value open boundary problem. That means that in order to obtain a solution one has to dene an initial condition (i.e., the excitation of the GPR transmitting antenna) and allow for the resulting elds to propagate through space reaching a zero value at innity because there is usually no specic boundary limiting the problems geometry where the electromagnetic elds can take a predetermined value. The rst part (initiation of a source) is relatively easy to accommodate if the exact details of the GPR antenna are not taken into account. However, the second part (i.e., the truncation of the model) can not as easily be tackled by using a nite computational space. The FDTD approach to the numerical solution of Maxwells equations is to discretize both the space and time continua. Thus, the spatial and temporal discretization steps play a very signicant role since the smaller they are the closer the FDTD model is to a real representation of the problem. However, the values of the discretization steps always have to be nite, since computers have a limited amount of storage and nite processing speed. Hence, the FDTD model represents a discretized version of the real problem and of limited

A. Giannopoulos / Construction and Building Materials 19 (2005) 755762

757

size. The building block of this discretized FDTD grid is the Yee cell named after Kane Yee who pioneered the FDTD method [11]. The 3D Yee cell is illustrated in Fig. 1. The 2D FDTD cell is easily obtained as a simplication of the 3D Yee cell. By assigning appropriate

(i,j+1,k-1) Ez (i,j+1,k) Ex Hy

Ex Ez

(i+1,j+1,k-1)

Ey Hx

Hz

Ey Hx Ey (i+1,j,k-1)
y

Ey

Hz

Ez

Hy

Ex Ez

(i,j,k)

Ex

(i+1,j,k)
z

Fig. 1. The Yee cell.

constitutive parameters to the locations of the electromagnetic eld components complex shaped targets can be easily included in the models. However, objects with curved boundaries are represented using a staircase approximation. The numerical solution is obtained directly in the time domain by using a discretized version of Maxwells curl equations that are applied in each FDTD cell. Since these equations are discretized in both space and time the solution is obtained in an iterative fashion. In each iteration the electromagnetic elds advance (propagate) in the FDTD grid and each iteration corresponds to an elapsed simulated time of one Dt. Hence by specifying the number of iterations one can instruct the FDTD solver to simulate the elds for a given time window. The price one has to pay for obtaining a solution directly in the time domain using an explicit numerical method like FDTD is that the values of the temporal discretization step Dt and the spatial discretization Dx, Dy and Dz cannot be assigned independently of each other. FDTD is a conditionally stable numerical process. The stability condition known as the CFL condition after the initials of Courant, Freidrichs and Lewy [8] is 1 Dt 6 q ; 1 1 1 c Dx2 Dy 2 Dz2 8

Boundary of computer model

ABC

Air Earth

Tx

targets

Fig. 2. FDTD view of the models space.

where c is the speed of light. Hence t is bounded by the values of Dx, Dy and Dz. The stability condition for the 2D case is easily obtained by letting Dz ! 1. One of the most challenging issues in modelling open boundary problems like GPR is the truncation of the computational domain at a nite distance from the sources and targets where the values of the electromagnetic elds can not be calculated directly by the numerical method. Therefore, approximate conditions known as absorbing boundary conditions (ABCs) are applied at a sucient distance from the source to truncate and therefore limit the computational space. The role of these ABCs is to absorb any waves impinging on them, hence simulating an unbounded space. The computational space (i.e., the model) limited by these ABCs

Table 1 Some GprMax2D commands Command #domain: #dx_dy: #time_window: #medium: #box: #cylinder: #triangle: #tx: #rx: #scan: Function Controls the physical size of the model Denes the discretization steps Denes the simulated time window for the GPR trace Introduces the electrical properties of dierent media in the model Introduce a rectangle of specic properties into the models space Like the box: but introduces a cylinder into the model Like the box: but introduces a triangular patch. Species the details of a transmitter (Tx) Species the details of a receiver (Rx) Can be used to automatically generate B-Scans

758

A. Giannopoulos / Construction and Building Materials 19 (2005) 755762

should contain all important features of the problems geometry such as all sources and output points and important targets. Fig. 2 illustrates the basic dierence between a real problems space and the actual FDTD modelled space. In Fig. 2 it has been assumed that the half-space, where the target is situated is of innite extent. As a result of this assumption the only reected waves will be the ones originating from the target. In cases, where the host medium cannot be assumed to be of innite extent (e.g., a nite concrete slab) the assumption can still be made as far as the slabs actual size is large enough
400

that any reected waves that will originate at its termination will not aect the solution for the required time window. In general, any objects that span the size of the computational domain (i.e., model) are assumed to extend to innity. The only reections that will originate from their termination at the truncation boundaries of the model are due to imperfections of the ABCs and in general are of very small amplitude compared with the reections from target(s) inside the model. All other boundary conditions which apply at interfaces between dierent media in the FDTD model are automatically enforced.

Simulated GPR Trace

300

Rebar reflection

200 Rebar multiple reflection 100 Volts/m

100

200

300 Direct arrival 400 0 1 2 3 4 5 nanoseconds 6 7 8

Fig. 3. Simulated GPR trace from a rebar in concrete.

Fig. 4. Geometry of a GprMax2D concrete slab with rebars model.

A. Giannopoulos / Construction and Building Materials 19 (2005) 755762

759

3. GprMax GprMax2D and GprMax3D are two computer programmes that implement the FDTD method for GPR modelling in 2D and 3D, respectively. Some of their key features are: an easy to use command interface, the ability to model dispersive materials, the modelling of complex shaped targets as well as the simulation of unbounded space using powerful absorbing boundary conditions. GprMax3D allows the simulation of GPR antennas and even the introduction of their feeding transmission lines into the model. GprMax2D is mainly used for GPR signature simulation whereas GprMax3D is used for more detail and realistic simulations especially when comparisons with real GPR data are important. Both GprMax2D and 3D programmes use a simple ASCII (text) le to dene the models parameters. In this le special commands are used which instruct the software to perform specic functions that are required by the type of the model the user wants to create. Some of the commands of GprMax2D are shown in Table 1. The software package includes a comprehensive Users Manual in which details of the functionality of the programmes can be found.

As an illustration of the simplicity of creating models for GprMax2D the following input le creates a GPR model that can simulate a single GPR trace (A-scan) from a rebar embedded in a concrete half-space: #domain:0.5 0.5 #dx_dy:0.005 0.005 #time_window:8.0e-9 #medium:6.0 0.0 0.0 0.01 1.0 0.0 concrete #box:0.0 0.0 0.5 0.4 concrete #cylinder:0.25 0.3 0.02 pec #tx:0.20 0.42 1.0 1500e6 ricker trace.out b #rx:0.30 0.42 The result of this simulation-a GPR trace over a rebar can be seen in Fig. 3.

4. GprMax2D modelling example The exibility of GprMax2D allows the modelling of complex what-if scenarios. In the following a simple example of modelling rebars in concrete is presented. The geometry of the problem consists of a 2 m wide

Fig. 5. Simulated GPR scans from the 2D concrete slab model using dierent antenna centre frequencies.

760

A. Giannopoulos / Construction and Building Materials 19 (2005) 755762

slab, where rebars of 20 mm diameter are located at an average depth of 150 mm from the slabs surface. Although, the horizontal distance between each rebar is xed at 100 mm their cover depth is randomly varied between 4 mm from the average cover depth of 150 mm. An illustration of the models geometry is presented in Fig. 4. The electrical properties of concrete were assumed to be r = 6 and r = 0.01 S/m. In Fig. 5 a comparison is made of simulated GPR scans obtained from the model when dierent centre frequencies for the modelled GPR transducers were used. As it was expected, it evident that the resolution of the 900 MHz antenna is less than the 1.5 GHz one since the 1.5 GHz response resolves the rebars more clearly. Fig. 6 illustrates three simulated scans from the same model when dierent values for the relative permittivity

of concrete are used. In all these scans the GPR antenna is simulated using a 1.5 GHz ricker pulse. The values for the relative permittivity of concrete that were used for this comparison are: r = 6, r = 12 and r = 18. As it was expected the resolution of the modelled GPR data improves as the propagating wavelength becomes smaller since the velocity of propagation decreases with increasing values of relative permittivity. Further, it can be seen as well that although the targets positions have not been altered in the model their responses appear to register at a later time in the simulated GPR scans. The above example illustrates in a simple way that the model predicts key features of GPR responses reasonably well especially when one takes into account the fact that GprMax2D provides only a solution based on a 2D approximation of a real 3D problem.

Fig. 6. Simulated GPR scans from the 2D concrete slab model for dierent values of relative permittivity of concrete.

A. Giannopoulos / Construction and Building Materials 19 (2005) 755762

761

5. GprMax3D modelling example Modelling GPR in 3D using GprMax3D requires substantially more computer resources than simple 2D modelling. However, the continuous increase of the amount of computer power that is readily available to the average GPR user an eect of the rapid advancements in computer technology makes 3D GPR modelling a clear and better alternative to simple 2D simulations. Currently the main use of GprMax3D is to model as realistically as possible GPR antennas and their eld distribution when they are in close proximity with targets. GprMax3D is especially useful in studying

numerically computed radiation patterns of GPR antennas which demonstrate signicant dierences from customarily used in GPR analysis theoretical eld patterns that are calculated assuming innitesimal Hertzian dipoles as GPR sources and they are based in far-eld approximations. In this paper, a simple 3D example is presented. This example is from a simulation of a simple bowtie antenna located 10 mm above a concrete slab. The antenna is unshielded and has a total length of 150 mm and a are angle of 60. A twin-pair transmission line with impedance of 200 Ohms is used to feed the bow-tie. The excitation pulse is a Gaussian of 900 MHz centre frequency.

Fig. 7. Snapshots of three principal planes through a bow-tie antenna (units are in FDTD cells).

Fig. 8. Snapshot of the plane containing the antenna structure (units are in FDTD cells).

762

A. Giannopoulos / Construction and Building Materials 19 (2005) 755762

In Figs. 7 and 8 snapshot images obtained from the models space are taken after 3 ns had passed from the initiation of the source pulse in the feeding transmission line. The snapshot images presented here are of the eld component which has the same orientation with the long axis of the bow-tie antenna. It is interesting to note how the model can be used to provide information that leads to the visualization of the complex radar elds that exist close to the antenna. Fig. 7 includes all three principal planes of the model in an attempt to demonstrate how this simple GPR antenna radiates in 3D space. Fig. 8 presents the snapshot image from the plane that contains the antenna structure itself. It should be noted here that construction of 3D models using GprMax3D follows the same simple procedure as for the case of 2D models employing similar commands in a simple ASCII le. The only signicant dierence is obviously the fact that in GprMax3D objects need to be specied as 3D entities and not as simple 2D shapes.

Acknowledgement The author will like to acknowledge the support of the Building Research Establishment(BRE), UK for providing nancial support for the development of the rst version of GrpMax.

References
[1] Bungey JH, Millard SG, Shaw MR. Radar assessment of posttensioned concrete. Proceedings of the 7th international conference on structural faults and repair-97, Edinburgh, 810 July 1997, vol. 1. Edinburgh: Engineering Technics Press; 1997. p. 3319. [2] Forde MC, McCavitt N. Impulse radar testing of structures. Proc Instn Civ Engrs Structs and Bldgs 1993;99:969. [3] Daniels DJ. Subsurface penetrating radar. London: The Institution of Electrical Engineers; 1996. [4] Bergmann T, Robertsson JO, Holliger K. Finite-dierence modelling of electromagnetic wave propagation in dispersive and attenuating media. Geophysics 1998;63:85667. [5] Giannopoulos A. The investigation of transmission-line matrix and nite-dierence time-domain methods for the forward problem of ground probing radar, DPhil Thesis, University of York, Department of Electronics, York, UK; 1997. [6] Bourgeois JM, Smith GS. A fully three-dimensional simulation of a ground-penetrating radar: FDTD theory compared with experiment. IEEE T Antenn Propag 1996;34:3644. [7] Millard SG, Shaw MR, Giannopoulos A, Soutsos MN. Modelling of subsurface pulsed radar for nondestructive testing of structures. ASCE J Mater Civil Eng 1998;10:18896. [8] Taove A. Computational electrodynamics: the nite-dierence time-domain method. Artech House; 1995. [9] King RWP, Owens M, Wu TT. Lateral electromagnetic waves. Springer-Verlag; 1992. [10] Balanis CA. Advanced engineering electromagnetics. Wiley; 1989. [11] Yee KS. Numerical solution of initial boundary value problems involving Maxwells equations in isotropic media. IEEE T Antenn Propag 1966;14:4958.

6. Conclusions Numerical modelling of GPR is very useful in enhancing our understanding of the GPRs detection mechanism. The GprMax suite of programmes allows the simulation of realistic scenarios encountered in everyday use of GPR. As computer power is constantly increasing GPR modelling will become an important tool in training new GPR users as well as improving data interpretation of complex GPR sections. The introduction of more sophisticated features into the modelling programmes is a continuing eort especially for the realistic modelling of actual GPR transducers.

You might also like