Numerical modelling of high-frequency ground-penetrating radar antennas

摘要

Ground-Penetrating Radar (GPR) is a non-destructive electromagnetic investigativeudtool used in many applications across the fields of engineering andudgeophysics. The propagation of electromagnetic waves in lossy materials isudcomplex and over the past 20 years, the computational modelling of GPR hasuddeveloped to improve our understanding of this phenomenon.udThis research focuses on the development of accurate numerical models ofudwidely-used, high-frequency commercial GPR antennas. High-frequency, highresolutionudGPR antennas are mainly used in civil engineering for the evaluationudof structural features in concrete i. e., the location of rebars, conduits, voidsudand cracking. These types of target are typically located close to the surfaceudand their responses can be coupled with the direct wave of the antenna. Mostudnumerical simulations of GPR only include a simple excitation model, such as anudinfinitesimal dipole, which does not represent the actual antenna. By omittingudthe real antenna from the model, simulations cannot accurately replicate theudamplitudes and waveshapes of real GPR responses.udNumerical models of a 1.5 GHz Geophysical Survey Systems, Inc. (GSSI) antennaudand a 1.2 GHz MALÅ GeoScience (MALÅ) antenna have been developed.udThe geometry of antennas is often complex with many fine features that must beudcaptured in the numerical models. To visualise this level of detail in 3d, softwareudwas developed to link Paraview—an open source visualisation application whichuduses the Visualisation Toolkit (VTK)—with GprMax3D—electromagnetic simulationudsoftware based on the Finite-Difference Time-Domain (FDTD) method.udCertain component values from the real antennas that were required for theudmodels could not be readily determined due to commercial sensitivity. Valuesudfor these unknown parameters were found by implementing an optimisationudtechnique known as Taguchi’s method. The metric used to initially assessudthe accuracy of the antenna models was a cross-corellation of the crosstalkudresponses from the models with the crosstalk responses measured from the realudantennas. A 98 % match between modelled and real crosstalk responses wasudachieved.udFurther validation of the antenna models was undertaken using a series ofudlaboratory experiments where oil-in-water (O/W) emulsions were created toudsimulate the electrical properties of real materials. The emulsions providedudhomogeneous liquids with controllable permittivity and conductivity and enableduddifferent types of targets, typically encountered with GPR, to be tested. The laboratory setup was replicated in simulations which included the antennaudmodels and an excellent agreement was shown between the measured andudmodelled data. The models reproduced both the amplitude and waveshape ofudthe real responses whilst B-scans showed that the models were also accuratelyudcapturing effects, such as masking, present in the real data. It was shownudthat to achieve this accuracy, the real permittivity and conductivity profiles ofudmaterials must be correctly modelled.udThe validated antenna models were then used to investigate the radiationuddynamics of GPR antennas. It was found that the shape and directivity ofudtheoretically predicted far-field radiation patterns differ significantly from realudantenna patterns. Being able to understand and visualise in 3d the antennaudpatterns of real GPR antennas, over realistic materials containing typicaludtargets, is extremely important for antenna design and also from a practicaluduser perspective.