| Coal is the fundamental energy in China. Accident caused by disaster resources is one of the major geological problems challenging coal mine safety. The construction of geology safety insurance system cannot be ignored. It is important to ascertain geological condition in a coal face or a mine area before mining, so as to secure a high output and high profit mine in safety production. Ground penetrating radar is a high-resolution and high-efficient non-destructive geophysical exploration technique. It is applied to a wide range of shallow geophysical investigation tasks including coal mine detection. In surface radar surveys, common-offset reflection profiling is usually preferred to determine subsurface structure. Resolution of a surface radar image decreases with depth quickly, which leads its detection distance restricted to 30 meters. Subsurface radar velocity characterization at the deeper parts of the section may not be accurate enough for reliable interpretations. Aqueous structures, which are of small scopes, call for relatively high resolution and recognition capability. In this study, velocity tomography of surface radar is carried out. With transmitting antenna and receiving antenna located on the two sides of detected profile respectively, surface radar velocity tomography can then be performed to determine the velocity distribution and identify potential harmful geology structures.This paper is on the background of Major National Scientific Apparatus Development Special Item(2012YQ030126) “Apparatus development and industry application for detection mine disaster resources”. The research is mainly focus on algorithms and experiments for ground penetrating radar tomography inversion, including data collection and preprocessing method, rectangular and complex observation area tomography inversion algorithms and parallel inversion algorithms. The proposed algorithms are verified through numerical simulation and physical model experiments. Tomography technique can greatly expand GPR’s application areas and provides effective technical support for coal mine safety issues, and has important practical significance for coal mine disasters prevention and safety production.Tomography is divided into two kinds: velocity tomography and attenuation tomography. Velocity tomography inverses velocity of the detected profile according to direct wave travel time, and attenuation tomography uses amplitude or frequency data. The detected medium by radar is mainly lossy dielectric medium in which electrical conductivity equals to zero, and dielectric constant is independent of attenuation. Rock formation in mine belongs to low lossy dielectric medium, so we use velocity tomography for detecting disaster resources in mine.In this paper, observation systems for rectangular, trapezoid and wedge detection area in coal mine are built. A new fast travel time collecting method, line-by-line method, is proposed based on point-by-point method, which improves detection efficiency. Its detection efficiency is higher than traditional points offset method and it can get more travel time data. A preprocessing method based on wavelet transform and cross correlation technique is also proposed. Wavelet transform is used for signal denoising and co-correlation method for extract travel time. The first inflection point in the waveform is chosen as extraction standard for the location of direct wave, and the first wave is the reference waveform.The first step of velocity tomography inversion is to discretize profile. The profile is discretized as a series of cells. For each transmitting point–receiving point pair, the length of each ray path segment that crosses a cell is computed. All the segment lengths are organized in a sparse matrix, called the coefficient matrix, which describes the geometry of the rays. Geology structure in coal mine is very complicated, including cavity, fault, collapse column, water and so on, which is usually consisted of inhomogeneous medium. There is error if straight ray tracing algorithm is used. Curve ray tracing algorithm is highly dependent on velocity of cells model and accuracy of initial velocity is very important. If we lack prior information of observation area, there is large difference between the initial velocity and real velocity, which could lead to great error of first ray tracing result. And in following iteration calculating, ray tracing error and velocity error will influence each other, which make us cannot get inversion result with enough accuracy.According to specialty of coal mine observation area and boundedness of straight ray tracing and curve ray tracing algorithms, an improved ray tracing algorithm is proposed in this paper, named coal mine non linear medium secondary subdivision dynamic ray tracing algorithm. This algorithm includes three steps: calculate initial velocity and inversion travel time through straight ray tracing; select abnormal rays and find start point of curve estimation through mathematical statistics methods; ray tracing in the following iteration through linear travel time interpolation. The first step provides relatively accurate initial velocity, and the second step could greatly reduce calculation amount in each iteration, then the shortage of present ray tracing algorithms is overcome.Because each ray crosses minority cells, there are lots of zero elements in the coefficient matrix. These zero elements waste lots of storage memory and calculating time. This paper designs sparse storage and solving method for tomography matrix. Number of nonzero elements in each line and the first nonzero element in each line are added to the traditional sparse matrix struct. The improved struct can greatly decrease both time and space complexity of tomography inversion algorithm and increase its efficiency. It is proved through three numerical simulation tests that the calculating time is significant decreased using sparse storage and solving method.There are various tomography inversion algorithms. In this paper, three commonly used algorithms, ART, SIRT and LSQR, are optimized through sparse storage. Besides, inversion termination conditions are improved. The commonly used travel time residual is changed to relative travel time residual and its change value and inversion result change value are added. The iteration calculation stops if inversion result or accuracy become stable. It is avoided that iteration continues until the max iteration time is reached.The three methods yield successful results with the numerical simulation using three models with representative disaster areas. The optimized algorithms are evaluated in respects of both precision and convergence rate. Also, cavities and water zone can be detected precisely in all algorithms. In particular, fault parallel to the survey line can be more accurately detected than fault perpendicular to the survey line. LSQR algorithm show better performance in resolving large sparse linear equations and obtain faster and stable convergence than ART and SIRT during the inversion processes. Though ART gives better results than SIRT, it cost more than two times calculating time. What deserves to be noticed is that there is a low velocity anomalous zone of strip structure in each ART result. In all methods, anomalous zone could lead to error in surrounding areas and these areas are in different shape. Due to multi solution of inversion, the inverted results may be unreliable even when iteration criteria is satisfied.Because the geology structure is complicated in mine, oblique roadway is also common. So we need consider proper algorithms for trapezoid and wedge detection area. The inversion equations become more ill-conditioned for each ray crosses less cells. In this paper, weighted sorted ART algorithm based on projection and partition is proposed. The optimized algorithm includes two parts. First, the oblique line is projected to the rectangular boundary and the profile is partitioned into three areas. Second, two weight coefficients are designed, ray angle and slowness difference, according to the influence by ray angle and disaster area.Numerical simulation proved that the proposed algorithm is feasible both in efficiency and accuracy when used in trapezoid area. The error of water zone model is higher than other models, which results in fake exceptions. In particular, fault parallel to the survey line can be more accurately detected than fault perpendicular to the survey line. The results of wedge models is quite consistent with models. This indicates that rays coverage has great an important influence on inversion accuracy. If travel time is collected with different direction survey line designed, inversion result can be improved. Another method is image fusion technique. In this paper, image fusion based on wavelet transform is employed. Contour map of inversion results inverted from data collected from two kinds of survey line design. Image fusion integrates exceptions from two original image and noise of background is eliminated. Experiments results shows that the better decomposition levels is from 8 to 12 and db4 and sym4 are more effective wavelet basis functions.In order to examine that whether the algorithms proposed is feasible for test data, several experiments, including rectangle and wedge, are taken using tomography GPR developed by State Key Laboratory of Coal Resources and Safe Mining in China University of Mining & Technology, Beijing. The roadway model is built at 1:1~1:3 in size. Travel time collected using two methods are nearly identical in rectangle experiments results, but there are obvious differences between them in wedge experiments results. So survey line length, sampling points number and detection resolution should be considered when choosing data collection method. The rectangle experiments results show that inversion results of ART and Damping LSQR are better than SIRT, which can reflect detection area structure accurately and comprehensively. The wedge experiments results show that the proposed optimized algorithm is also feasible for test data. The size, location and velocity of the cavity is accurately inverted.The data process for GPR tomography mainly consists two parts: ray tracing and inversion. Parallel strategy is designed for the two parts. Ray data can be uniformity reconstruction. ART distributes travel time error of each ray to the cells it crosses, so it suits for parallel procession. The cells are asymmetrical blocked, and each node is located at between cells. Parallel inversion for GPR tomography cannot be finished by one time Map Reduce parallel calculation framework. In this paper, associated Map Reduce parallel framework iterative composite applications are designed. The first time framework is used for ray tracing and others for ART iteration algorithm. The parallel calculation framework was tested using simulation data. The result showed that inversion efficiency was improved. |
PDF Full Text Request