From a blasting perspective, an accurate mapping of the bench area to be blasted is critical for a precise blast pattern design and implementation. A detailed survey of the blast location and bench wall allows for quality control of drillhole alignment and pattern implementation. Moreover, it allows for calculation of the in situ volume of the rock mass in the blast location and a better understanding of the structural characteristics of the rock mass. The following section will describe the application of UAV technology for pre-blast monitoring.
3.1. Structural Mapping
One of the key advantages of digital photogrammetry is that once a 3D point cloud of the pit wall is generated, it is possible to both conduct virtual structural mapping and assess the blast induced damage of the pit wall on the same point cloud model. Structural mapping of the pit wall is important for both geomechanical stability analysis of pit slope and estimating the IBSD. The IBSD is defined by the intersection of naturally occurring joints within the rock mass [
27]. These joints are defined by their orientation, persistence, frequency and surface geometry [
28]. However, defining these characteristics can be challenging due to their three-dimensional nature and the field exposure of joints, which generally provides only two-dimensional information [
29]. Further to this limitation, exposures are often inaccessible for the manual mapping of joint characteristics. This is particularly true in the case of open pit mines, where once the benches are fully excavated, access to these benches becomes difficult or hazardous for personnel.
Through the use of UAV technology, it is possible to collect 2D images of the pit wall to generate a 3D model or an orthophoto for structural mapping and assessing the impact of blasting on the final wall geometry. By using virtual mapping, it is possible to accurately measure the orientation and trace length of joints exposed on the pit wall. The use of a UAV allows coverage of a large area, typically inaccessible for conventional structural mapping, such as window mapping, which involves having a person conduct mapping on or close to the rock face in a window with fixed height and width [
8]. It also generates a more extensive, high-resolution structural dataset. A statistical analysis of data can be used to generate a stochastic model of jointed rock mass based on a 3D discrete fracture network (DFN). A DFN is a stochastic model of the fractures within a rock mass, which can be used to estimate the IBSD required for the blast analysis [
30]. The accuracy and reliability of the DFN model is dependent on the quality of the mapping data collected [
31]. Therefore, the UAV’s ability to collect more data at a high resolution should produce more robust DFN models. Structural mappings were conducted at Mines A-D using the A1, B1, C1, and D1 UAV surveys, detailed in
Appendix A.
Using UAV and digital photogrammetry methods, the pit wall mapping time can be significantly reduced compared to manual mapping methods. At Mine D, a mapping-rate of approximately 44.6 m
/h was achieved using conventional window mapping in cells of 1.8 m high by 19.2 m long. On average, only 20 of these windows can be mapped during a shift by appropriately trained personnel. Through the use of a UAV, a mapping-rate of 490 m
/h was achieved, which is more than ten times faster than manual mapping. The images collected also provide a permanent record of the rock mass at a certain time, allowing site personnel to quantify any changes in the rock over time. It also becomes possible to collect and map data that were previously inaccessible, in addition to collecting data for a much larger area. The area used for mapping, the time required for virtual mapping, the estimated field mapping time based on the 44.6 m
/h rate and the virtual mapping method used for different mine locations are described in
Table 1. The methods and speed of mapping are described further below. Using UAV systems for data collection will not eliminate field observations, as the collected data need to be verified through ground sampling, and joint surface information cannot yet be accurately collected using UAV systems, which is required for rock mass rating systems [
8].
The joint mapping was done using the 3D point clouds directly in CloudCompare [
32], an open source software. Using this software, it was possible to measure the joint dip, dip direction and length of traces—all required to define joint sets and to estimate the IBSD of rock mass. The Compass tool by Thiele [
33] was used for manual and semi-automated structural mapping on the cloud using the default “Darkness” trace mode, while the Facets tool developed by Dewez [
34] was used for automated joint identification using default settings. Automated joint detection algorithms were run for all sites, with differing degrees of success. The automated option creates facets on the exposed joint surfaces and is useful where joint surfaces are clearly exposed on the pit wall. For highly irregular jointing or low-quality point cloud models, manual or semi-automated mapping should be used. Using manual virtual mapping can significantly increase the time required to complete the mapping of the wall. For example, for Mine B, the automated joint identification was significantly faster; 3D model creation, joint identification and analysis took only 5 h compared to 24 h for manual virtual mapping of the same wall.
The automated and manual mapping in Mine B had similar orientation results (see
Table 2). It should be noted that the trace lengths were only similar for Set 2, while Sets 1 and 3 had different mean trace lengths for the two mapping methods. The trace lengths were measured as the longest dimension of the facet when using the automatic mapping method. The difference in trace length was caused by facets splitting slightly offset joints in close spatial proximity into multiple components. In manual mapping, the same joints were mapped as a single joint. This resulted in longer trace lengths during manual mapping than automated mapping. The overall areal fracture intensity (P
) was similar for both manual and automated mapping, indicating that the total length of joints for each set was similar. However, the automated mapping still required manual post-processing to identify the joint sets and calculate the average trace lengths. This method was unable to detect the horizontal bedding, which was exposed as traces on the surface of the rock mass.
At Mine A, mapping was done with fully manual methods because of the lower quality of the point cloud. Using this approach involved manually tracing the length of the joint and visually fitting a plane in CloudCompare to estimate the dip and dip direction of the joint. To fit a new 3D plane in CloudCompare, the dip and dip direction were iteratively changed until the inserted plane visually fit the joint. This highly manual approach resulted in a very slow mapping rate. The flight plan at Mine A had a lower side overlap, which produced the lower quality point cloud during ODM’s dense cloud construction. To mitigate this at Mines B, C and D, a front and side overlap of 80% was used. Mine B had three well-defined and exposed joint sets, which made it suitable for automated joint identification, and therefore very fast for mapping. At Mines C and D, there were no clearly exposed joint surfaces, and the automated detection method could not be used. However, the semi-automated method could be used because the point clouds were sufficiently high quality; this method traces the selected joint and then estimates its length, dip and dip direction. This approach is faster than the manual method and field mapping. The difference in time requirement can be seen in
Table 1.
Figure 2 shows example windows of pit wall mapping for Mines A, B and C to illustrate the structural variability between sites, the joint trace and joint plane measurements captured and point cloud quality. With the joint trace length information and the area of the digital photogrammetric model, it is possible to calculate the areal fracture intensity (P
) of the mapped pit walls; a key input for the development of a DFN model [
35].
Figure 3 shows an example of the areal fracture intensity (P
in m/m
) of two benches mapped at Mine D.
Figure 3a shows the point cloud of the two benches that were mapped in CloudCompare using the Compass tool. The joint lengths were recorded and subsequently used to visualize the P
in CloudCompare. The major structure is highlighted by the dashed black lines. The “Estimate P
intensity” function with the Compass tool was used to visualize the P
by using the measured joint lengths and dividing them by the point cloud area. During virtual mapping, an attempt was made to exclusively map natural fractures (i.e., joints) for the estimate of P
. This ensures that the P
estimated represents in situ conditions. Other fractures on the wall include blast-induced fractures, which do not occur naturally and are only found in the rock mass local to blasting activity. The area of the bench was estimated in this study by cutting sections in 5 m intervals and measuring the heights over the length of the pit wall. The area of the bench can also be calculated using the 3D mesh, but in this study, we used 2D sections.
Figure 3b shows the P
values on the two mapped benches with the same major structure highlighted as in
Figure 3a. The figure demonstrates the spatial variation of P
on the pit walls, with the highest P
intensity on the middle of the lower bench around a local structure. Using this tool, it is possible to identify areas with high fracture intensities and areas of potential IBSD variation through visualization.
In this study, DFN models were developed for Mines A and B using data acquired from drone-based pit wall mapping. A DFN model stochastically describes the geometrical characteristics of rock mass fractures or, in general, discontinuities [
36]. The basic objective of DFN modelling is the generation of simulated fractures that accurately represent the salient characteristics of a population of fractures sampled in a particular rock mass. The fundamentals of discrete fracture network (DFN) modelling were explained by Dershowitz [
37]. DFN modelling has been used in a variety of mining applications, including dilution estimation [
38], caving fragmentation analysis [
39], estimation of rock mass strength [
40,
41], stability analysis of surface and underground excavations [
40,
42] and blast parameter assessment [
43,
44].
Joint sets are defined by their orientation, size, frequency and surface geometry. The availability of a high-quality 3D point cloud ensures the collection of accurate joint parameters, which leads to increased reliability of 3D DFN models [
45]. Once the joint sets were identified, it was possible to estimate a statistical distribution of their orientations and trace lengths. Furthermore, by using the total trace length of each set, the areal fracture intensity (P
) was estimated for the whole mapped area. In this study, the DFN generation was carried out using an iterative process in FracMan [
46], a discrete fracture network (DFN) modelling software, where an arbitrary volumetric fracture intensity (P
) (the area of fractures within a certain volume) was assigned to each joint set. The P
can be inferred from the P
since it cannot be directly measured in the field [
47]. A trace-plane or survey surface of mapped location was inserted into the DFN model, and the orientations and trace lengths of joints were recorded on the surface. The results were compared to the virtual structural mapping data measured using the UAV photogrammetric model. The P
of the individual sets were adjusted after each simulation until there was good agreement between the P
of the virtual structural mapping data and DFN model (
Table 3 and
Table 4), resulting in a model that was acceptable for use in IBSD estimation. A low P
percent error was considered as acceptable agreement for this study. The spatial variability within the rock mass for the individual joint sets was not considered; the rock mass was modelled using the average P
of each set. Similarly, the length of the joint sets was adjusted until the trace length of joint sets produced in the DFN matched the virtual structural mapping data (
Table 3 and
Table 4).
Figure 4 presents the 2D traces of the DFN models developed for Mine A and Mine B and the 2D trace-planes of the same pit walls that were virtually mapped.
Table 3 and
Table 4 show the distribution of DFN input data used for the joint sets and the resulting DFN mean trace lengths and P
; these results are also compared to the virtual structural mapping data for Mines A and B, respectively. The results demonstrate a good agreement between virtual mapping data and DFN models. It took over twenty DFN generations of Mine A to calibrate the model, compared to nine for that of Mine B. This difference in the number of DFN generations reflects the additional structural complexity of the rock mass at Mine A.
Figure 5 shows examples of stereonet agreement between field data and DFN model data for Mine A and Mine B. The bedding planes at Mine B visible in
Figure 4d were deterministically modelled in the DFN because they were horizontal with no variation in dip.