A Machine Learning-Based Fast-Forward Solver for Ground Penetrating Radar With Application to Full-Waveform Inversion

Abstract

The simulation, or forward modeling, of Ground Penetrating Radar (GPR) is becoming a more frequently used approach to facilitate interpretation of complex real GPR data, and as an essential component of full-waveform inversion (FWI). However, general full-wave 3D electromagnetic (EM) solvers, such as ones based on the Finite-Difference Time-Domain (FDTD) method, are still computationally demanding for simulating realistic GPR problems. We have developed a novel near realtime, forward modeling approach for GPR that is based on a machine learning (ML) architecture. The ML framework uses an innovative training method which combines a predictive principal component analysis technique, a detailed model of the GPR transducer, and a large dataset of modeled GPR responses from our FDTD simulation software. The ML-based forward solver is parameterized for a specific GPR application, but the framework can be applied to many different classes of GPR problems. To demonstrate the novelty and computational efficiency of our ML-based GPR forward solver, we used it to carry out FWI for a common infrastructure assessment application-determining the location and diameter of reinforcement bars in concrete. We tested our FWI with synthetic and real data, and found a good level of accuracy in determining the rebar location, size, and surrounding material properties from both datasets. The combination of the near real-time computation, which is orders of magnitude less than what is achievable by traditional full-wave 3D EM solvers, and the accuracy of our ML-based forward model is a significant step towards commercially-viable applications of FWI of GPR.

Finite difference time domain modelling as support to ground penetrating radar surveys of precast concrete units

“…Further improvement can be achieved by employing mode‐based interpretation methods like the full waveform inversion (FWI) and machine learning (ML) techniques that use FDTD models as essential components and allow for greater accuracy, quantitative information extraction and a degree of automation (Liu et al . ; Giannakis, Giannopoulos and Warren ). Computational requirements can be higher than the method employed for this study, although being FDTD, FWI and ML methods more and more popular, they are subjected to a constant and notable development which involves a sensible reduction of the computational costs (Warren et al .…”

A Machine Learning Scheme for Estimating the Diameter of Reinforcing Bars Using Ground Penetrating Radar

“…GPR has been established as a mainstream NDT tool in civil engineering [8], and it has been successfully applied for building inspection and for detecting reinforcing bars (rebars) in concrete structures [9]. Various approaches using GPR have been suggested for locating and characterising rebars [10], [11], and there are many commercial GPR systems that are custom-built for this purpose [12], [13]. Although GPR can reliably detect and locate rebars, assessing their quality and estimating their diameter is an ongoing area of research with, as of yet, no conclusive outcomes.…”

“…Due to that, additional NDT methods (e.g. eddy current [11], electromagnetic induction [14], [15]) need to be applied in the field to complement GPR, raising the overall computational and operational costs, and adding complexity to the acquisition.…”

“…E-mail: craig.warren@northumbria.ac.uk relationship between the diameter of the rebar and the received GPR signal [10], [16], [17]. However, these methods are based on simplified assumptions and they fail to provide a universal and reliable solution [11]. To tackle this, a detection algorithm based on full-waveform inversion (FWI) using shuffled complex evolution optimization has been suggested [18].…”

On the Introduction of Canny Operator in an Advanced Imaging Algorithm for Real-Time Detection of Hyperbolas in Ground-Penetrating Radar Data

“…Sparse reconstruction algorithms find application in the imaging of targets embedded in stratified dielectric media [38,39]. Moreover, fast and accurate forward solvers [40][41][42][43][44] are of high interest in this area of research, too, because they can be combined with imaging algorithms and full-wave inversion techniques [45,46].…”