Application of laser scanning for rock mass characterization and discrete fracture network generation in an underground limestone mine
According to the Mine Health and Safety Administration (MSHA), between 2006 and 2016, the underground stone mining industry had the highest fatality rate in 4 out of 10 years, compared to any other kind of mining , . Additionally, during that same time, 40% of the fatalities were due to ground control issues, such as roof and rib collapses and pillar bursts. The National Institute for Occupational Safety and Health (NIOSH) developed guidelines for designing underground stone mines . However, these guidelines do not apply to all underground stone mines.
Monsalve, Baggett, Bishop, and Ripepi present a case study of an underground limestone mine experiencing a structurally controlled mode of failure, and analyze different methods to study that type of instability . They conclude that the integration of terrestrial laser scanning (TLS) with discrete element modelling (DEM) can be used to prevent rock falls in underground excavations to enhance worker safety. However, an adequate rock mass characterization and structural mapping must be performed in order to generate reliable models that allow the engineer to have a better understanding of the rock mass behavior.
The final product of a laser scan is a point cloud that represents the surfaces of the objects scanned, in this case the rock surface in an underground environment. The density of the point cloud depends on a series of parameters from both the laser scanner operational conditions and the specifications of the project, such as the resolution and the quality of the laser scan, the section of the tunnel, the distance between scanning stations, and the purpose of the scans. The lead engineer defines these parameters. One purpose of the scan can be to characterize the rock mass and statistically analyze the discontinuities that compose it for further discontinuous modeling. In these instances, additional data processing and a detailed analysis should be performed on the point cloud to extract the parameters that define a discrete fracture network (DFN) for each discontinuity set. A DFN is a geometric representation in three dimensions of a geological structure defined by statistical information of the structure characteristics measured in the field, such as orientation, density, and size .
This paper describes the methodology used in a case study mine to perform a structural mapping on a rock mass in order to obtain the input data needed to create a DFN for DEM. Different software packages were used to perform these analyses:
(1) SCENE was used to import the laser scans and generate a project point cloud .
(2) I-Site studio was used to perform the structural mapping and to estimate the trace length and spacing of the different fractures .
(3) Dips was used to characterize and group the discontinuity sets .
(4) 3DEC™ was used to generate the DFN .
A specific area was defined in the case study mine to focus the present research. The mine research team and personnel selected this area because of the presence of significant ground control issues, mainly defined as structurally controlled instability failure. This was evidenced by: (1) jointing pattern where at least four joint sets were well defined; observed wide-joint spacing, generally ranging from 0.6 to 2 m; amount of fallen blocks observed on the floor with cubical and tabular shapes; joint surfaces defined as mostly closed, flat and smooth, with a JRC ranging from 2 to 4, completely dry and fresh; and other geological structures, such as faults and contacts, that could generate a rock fall in the absence of the required support.
This failure mechanism is enhanced by the multiple karst formations present in the mine, which, during excavation, tends to generate rock blocks up to 4 m3 that pose a high risk for workers, equipment, and the overall mining plan .
Fig. 1 shows a plan view of the area of interest. This illustrates a karst structure that crosses both tunnels and has a variable aperture reaching up to 0.5 m. Fig. 1a is a view from the tunnel to the karst void, while Fig. 1b presents a pile of material that fell down from the karst structure. Water and mud have also fallen out of the karst void. For these reasons, laser scans were carried out close to this area in order to map the structural conditions and to perform further numerical modeling to understand and prevent rock falls in this section of the mine.
The following methodology describes in detail the procedures used by the authors to perform the laser scans in the study area, to import and process the information obtained from those scans, and to create the respective DFNs resulting from the virtual structural mapping. This methodology was designed based on lessons learned from experiences presented in previous works and practices using the laser scan and different software packages for these procedures , , . Additionally, documents, such as the laser scan user manual and user guides on planning and performing the laser scans provided by the laser scan provider company, were used to complement this methodology , .
Definition of operational conditions
A laser scanner is a surveying apparatus that produces a massive point cloud (millions of points per scan), indicating the positions of the scanned objects. This equipment sends an infrared laser beam into the center of a rotating mirror, which deflects the laser beam onto a vertical rotation into the environment being scanned (Fig. 2a). Scattered light from surrounding objects is then reflected back into the scanner. The equipment is able to identify two waves: the one sent out by the equipment and the one reflected by the object. The phase shift between both waves is used to calculate the distance of the measured object . Then, by using angle encoders to measure the mirror rotation and horizontal rotation of the scanner, the x, y, z coordinates of each point are calculated, resulting in a point cloud that is stored in a removable SD memory card . Fig. 2b presents the vertical and horizontal rotation of the laser scan. The laser scan unit used during this study was a FARO® Focus3D, acquired by the Mining and Minerals Engineering Department of Virginia Tech in 2011.
The two main variables to be defined during the scan are the resolution and the quality. These two variables affect the time of the scan and the density of the point cloud . The size of the excavation to be scanned, the distance of the scanner from the tunnel face, and the pacing between stations also affect the results. According to Fekete, the optimal positioning of a laser scan from the face is between 0.5 and 1 times the diameter of the excavation . In this particular case, the diameter of the excavation is 13 m, and 1 times the diameter was used as the distance between the scanner and the face.
Once the distance from the face to the scanner was defined, the adequate resolution and quality were selected. Since these parameters vary from site to site, a set of scans were performed in order to define which combination of resolution and quality values provided, in a reasonable scan time, a scan with an acceptable point cloud density to perform an adequate structural mapping of the rock mass.
Table 1 shows the different operational conditions tested in the mine, the real time that it took to perform each scan, and the point cloud size and the point cloud density. Twelve scans were performed, processed, and converted into point clouds, which were imported to I-site studio for analysis. The total point count for each cloud was measured. To measure the average point cloud density, three 5 m × 5 m mapping windows were generated, one on the right wall, another one on the left wall, and the last one on the roof. The amount of points contained within each mapping window was measured and recorded. The point cloud density was calculated by dividing the number of points by the area of each mapping window (25 m2). Then, for each scan, the average point cloud density was calculated by averaging the point cloud density on each mapping window. The scans performed with the resolutions of 710.7 million of points and 177.7 million of points were discarded since the amount of data obtained exceeded the requirements and significantly increased the scan time.
For selecting the best operational conditions, the real scan time and the average point cloud density were taken into account. These values were normalized and averaged. The best operational conditions for this case were a resolution of 1/4 and a quality of 1x, which required a scan time of 5 min and 12 s and yielded a point cloud density of 11 points/cm2. The obtained point cloud density is considered acceptable for structural mapping, bearing in mind that previous work has performed structural mapping on a LiDAR extracted point cloud with either a density of 4 points/cm2 or with a density of 16 points/cm2 , .
Laser scanning has been used for multiple applications in both the tunneling and mining industry, such as support evaluation, scaling assessments, leakage mapping, analysis of structurally controlled overbreak, structural mapping, roughness evaluation, and deformation analysis . Fekete determined that the best practice for the laser scanner distance from the face is from 0.5 to 1 diameter . In addition, they defined the optimal separation between stations is 1 diameter of the excavation to ensure the maximum coverage of the area with minimum overlap between scans and data redundancy .
Using these recommendations, the stations were positioned 13 m apart. Fig. 3 shows the estimated locations of the different stations in which the laser scanner was set. The survey was performed around the pillar between crosscuts 15 and 16.
The laser scanner used does not have an integrated global positioning system. This means the scans are not georeferenced unless there is a reference point with known coordinates. The laser scanner used does have an internal compass and inclinometer that allows orientation of each scan with respect to magnetic north .
Without a georeferencing tool, the only way to integrate different scans into a single model is to identify common reference points, or objects, present between scans. FARO suggests the use of checkboards or spherical references placed in strategic places that can be detected from the two stations that will be referenced together. In this particular case, 21.5 cm diameter inflatable balls were used as spherical targets between scans. Table 2 shows the recommended distance from the target to the laser as a function of the target size and laser scan operational conditions. Considering the recommended target distances, spacing between stations, and reference target sizes, four references were used in order to reference the current station with the two other stations (one forward and one behind). As the station moves, the references that were behind it will be leapfrogged forward as the scans advance in order to reference all the scans performed in the area together. This procedure is depicted in Fig. 4.
Data importing and processing
Once the operational conditions and the scanning procedure were defined, the laser scans were performed. Nine laser scanning stations were set around the pillar between crosscuts 15 and 16. In order to download the laser scans and convert them into point clouds, the software SCENE was used. Before generating the final compiled point cloud, the scans were spatially referenced with each other using the reference points. This process is defined as “registration.”
When a laser scan is performed, the scanner sets the position of the mirror as the coordinates x = 0, y = 0 and z = 0. If a reference point is observed from two different places, a spatial relation between those stations can be made . SCENE allows the user to open two stations at the same time and select the common points between them. Fig. 5 presents the spatial registration process at Stations 015 and 016. There are two reference spheres on each scan.
After referencing all the scans, SCENE evaluates the distance error between points and the overlap between scan stations. This software suggests that errors with a value of less than 8 mm and overlap values greater than 25% provide acceptable results on the registration process . When the errors are greater than 20 mm and the overlap between stations is less than 10%, the results from the laser scan are not acceptable; therefore, the scanning process must be repeated by using more reference objects or by reducing the distance between stations. Values between this ranges may still be accepted, considering a less precision on the laser scan results.
Table 3 presents the results obtained from the registration of the nine laser scans, which obtained acceptable values for both the errors (maximum point error 6.7 mm) and the overlap (minimum overlap 25.2%). According to these results, the recommendations made by Fekete regarding the laser scan station locations are acceptable in this particular case. Finally, in order to proceed with the analysis, the point cloud was saved as a laser scan file .
Structural data processing
I-Site studio is a point cloud processing software that allows users to edit and process laser scans. This software contains a set of geotechnical analysis tools that assist engineers during the structural mapping process. This software allows for greater and more representative data regarding the structural information of the rock mass, which can be used to generate DFNs. The point cloud generated from the laser scans was imported into I-Site studio to perform a structural mapping along the scanned tunnels. The generated file contained 331,230,871 points, or 14.7 gigabytes, of spatial data. Due to computational limitations, the point cloud was divided into 10 smaller sections, and each section was divided into right wall, left wall, and roof. This allowed researchers to perform the structural mapping with a useful amount of information that prevented the system from crashing.
Fig. 6 shows the left wall of the third section generated from the overall point cloud, where it is possible to observe the geological structures. Some of these structures exhibit similar orientations and relatively large exposed areas, while others are not easy to observe due to the sight angle. Additionally, the bottom half of Fig. 6 shows the mapped discontinuities output from I-Site studio. For the mapping process, the points that belong to the same discontinuity plane have to be manually selected. Using the geotechnical tool “query dip and strike,” an average plane is generated that accounts for the coordinates of each selected point. This process was repeated along the whole model section by section. Fig. 7 shows a plan view of the final point cloud with the mapped discontinuities. In total, 874 structures were mapped in the area.
Once all the relevant structural features observed in the 3D model were mapped, the structural data was exported into an Excel file. Structural features extracted from a point cloud into I-Site studio contain x, y, and z coordinates, orientation information (dip, dip direction, and strike, and size information (trace length and area). This information was imported into the software dips, which was used to analyze and classify the structures into joint sets. The defined joint sets can be filtered and reimported into I-Site studio as individual joint sets, where geotechnical tools can calculate the spacing between discontinuities from the same family.
3DEC is a DEM software that can use DFNs to generate geological structures in a rock mass. These DFNs are a set of discrete, planar, disk-shaped, and finite size fractures, which are used to cut through the blocks that constitute the model. These sets of disks are defined based on statistical information on the characteristics of the measured fractures in the field, such as orientation, size, and density . All of these characteristics can be obtained from the information mapped from the laser scans.
The orientation is defined by the dip (steepest declination) and dip direction (measured clockwise from true north) of a plane. This defines the spatial position with respect to the true north and a horizontal plane . Orientation parameters are defined as circular data. Due to this, the most adequate distribution to represent this data is the Fisher distribution, which is used to represent 3D orientation vectors . This distribution depends on the factor k that can be estimated form datasets greater than 30 poles. The software dips is able to calculate each k coefficient, mean dip, and dip direction for each joint set.
While the actual size of a structure is difficult to measure, it is regularly estimated as the trace length of the discontinuity, which is the length of the exposed structure. In this case, the trace length was considered the maximum length of the mapped plane. This information provided a statistical distribution for the trace length of each joint set.
Fracture density is a measure of the frequency of discontinuities belonging to the same family. This parameter could be measured as the number of fractures per unit length (P10), the length of fractures per unit area (P21), or the area of fractures per unit volume (P32) . When a DFN is generated in 3DEC, this parameter serves as a stopping condition . For instance, when P10 is used as the density parameter, a sampling line is defined, and 3DEC will randomly create discs based on the predefined size and orientation distributions until the number of fractures intersecting that scan line meet the value set as P10. For this work, the P10 estimate was based on the spacing of the discontinuities calculated using I-Site studio, considering the definition of discontinuity frequency on a scan line as the reciprocal of the mean spacing . In this case, spacing values less than 50 cm were not considered when calculating the fracture density because their respective fracture densities were not comparable with those observed in the field.
Results and discussion
The methodology described in this work allowed the authors to obtain a 3D virtual structural mapping of the area of interest. This information is important to help visualize and understand the structural setting for an area of interest and to identify possible blocks that may form and generate rock falls. Fig. 8 summarizes the virtual structural mapping and the statistical analysis for the trace length and the lineal fracture density of each identified structural set. Four discontinuity sets were defined from the mapped discontinuities:
(1) Set 1 is almost perpendicular to the tunnel orientation and presents a sub-vertical dip
(2) Sets 2 and 3 are oblique joints with a steep dip
(3) Set 4 corresponds to the bedding planes and contacts between rock units, which are almost parallel to the tunnel orientation and its mean dip of 29°.
Table 4 shows the statistical summary for the orientation, size, and density of each individual discontinuity set. This information was used to generate a DFN in 3DEC for each discontinuity set. Fig. 9 shows each individual DFN and the intersection of all four DFNs. It is also possible to observe a stereographic analysis of the generated disks, which can be compared with the stereographic analysis obtained from the virtual mapping. The major challenge identified during DFN generation was density definition. If the lineal density is considered, the value is only considered along a single sample line. Therefore, if the control volume in which the fracture network is being generated is significantly larger than the sizes of the disks, a large number of disks will be generated before the fracture density condition is met.
It is important to note that the DFNs obtained in this work that resulted from a virtual mapping on a laser-scanned area do not represent the final structural setting of the simulated rock mass. 3DEC uses these discs to cut through the blocks that conform the model, ultimately obtaining a set of blocks with fractures that were generated based on the DFNs. Because of this, it is important to compare the final structural setting of the simulated rock mass with conditions observed in the field. If these differ, the model should be calibrated to obtain the best representation of the actual rock mass. Additionally, it is worth mentioning that the DFN models result from statistical distributions obtained from measured values; therefore, each time the model is run, it generates a different DFN producing different outputs. In order to obtain a significant result for the model, a stochastic analysis must be performed to address significant conclusions from the simulations.
Since the methodology presented in the present study was developed to obtain information relevant for generating DFNs only, no mechanical properties of discontinuities on the rock mass have been measured yet. Thus, performing rock mechanics laboratory tests to measure the joint strengths is necessary. In addition, fieldwork aimed at measuring the wall strength and the roughness of each joint set will be performed. In addition, intact rock samples will also be collected to evaluate the mechanical properties of the footwall, orebody, and hanging wall. These parameters will be used as inputs for the numerical models and compared with observed conditions in the mine.
This paper describes the methodology used to perform virtual structural mapping in an underground mine from information obtained from a set of laser scans in an area that presents a structurally controlled failure mechanism. The following conclusions are derived from this work:
(1) Planning the scanning project is fundamental in order to save time during the scanning process, reduce errors on the resulting point cloud, and obtain the necessary information required for the specific need.
(2) While there is no existing guideline for performing an underground laser-scanning project, the experiences from other authors are valuable in defining the conditions of the scanning. For this study, following the recommendations of Fekete and Diedrichs allowed us to obtain a maximum point error of 6.7 mm and a minimum overlap between scans of 25.2%, both acceptable values according to studies by researchers , .
(3) It is important that geologists or engineers performing the structural mapping from the project point cloud have spent adequate time in the field in order to avoid any misleading interpretation and to ensure the results concur with those observed in the field.
(4) Information extracted from the laser scans provides sufficient information to perform a statistical analysis of the variables required to generate a DFN for each identified structural set. However, further numerical models are required to define the effectiveness of this methodology in accurately representing the rock mass structure.
(5) In order to perform this kind of analysis, the integration of four different software packages is required. Therefore, it is important to define an adequate workflow, which identifies the inputs required for each program and the outputs to be obtained. It is also important for the engineer to understand the limitations of each software and how these limitations can affect the results of the models.
(6) Finally, the criteria of the engineer, based on their experience in the field and observed ground conditions, are fundamental for deriving a conclusion from the results of the analyses.
Application of laser scanning for rock mass characterization and discrete fracture network generation in an underground limestone mine
Juan J.Monsalve, Jon Baggett, Richard Bishop, Nino Ripepi
International Journal of Mining Science and Technology
Volume 29, Issue 1, January 2019, Pages 131-137