Identification of the oversized rock pieces in hydraulic hammer breaking process based on the scanner data geometry features analysis

Due to the harsh conditions in the mining environment, it is difficult to fully automate some of the tasks carried out by mining machines and devices. As part of the technological process of extracting copper ore from Polish mines, the material is broken with a hammer to fit through the screen of a 40×40 cm square shape slot, after which it is loaded and conveyed for further processing. Usually, most of the material remaining on the screen cannot be considered oversized, additionally smaller particles can form blockages or rest on larger blocks. In this case, the crushing is not necessary, and only its movement may be sufficient. The focus of the presented research is on the identification of the particles that would be considered oversized, as it is a crucial task in the breaking hammer automation process. The authors propose a method based on the analysis of initially extracted 3D data copper ore blocks obtained from laser scanning at one of the transfer stations in KGHM Polska Miedź S.A., Polkowice-Sieroszowice mine. The difficulties of measurement are multiplied by the necessity of performing the scan from a single point of view, which was forced by safety concerns. Segmented but incomplete 3D data are processed, which results in finding the geometric features that prevent the material from going through the screen in an automatic manner.


Introduction
Hydraulic Breaker Hammers (HBHs) are commonly used in various technological processes, such as mining, to disintegrate or reduce the size of materials.They are also widely used in civil engineering, including building and road construction, demolition [1], and the processing of construction and demolition waste (CDW) for the desired granulation [2].In mining, HBHs are used to fragment oversized blocks after blasting [3] and before crushing in the stone refining process [4].Furthermore, they are used for scaling applications in underground excavations that remove loose and unstable rocks [5], and to prevent damage to technical infrastructure, such as belt conveyors, caused by large blocks falling at transfer points where material is classified as coarse and fine [6].
KGHM Polska Miedź S.A. mines employ hydraulic hammers to efficiently break up large rocks at unloading points.This material comes from the process of blasting where it is fragmented to pieces of varying size [7].Uniformity of the rock particle distribution is based on a set of blasting and material parameters (e.g.burden to blasthole diameter ratio, spacing to burden ratio, charge factor, stiffness ratio) [8].Some of the pieces can be considered to large for further processing IOP Publishing doi:10.1088/1755-1315/1295/1/012003 2 and need to be broken beforehand.Copper ore is transported by mining loaders and hauling trucks and then loaded onto conveyor belts.A 40 x 40 cm screen window is used to determine the largest size of blocks that can be safely loaded onto the belt conveyor by gravity.Solids larger than this require initial fragmentation with a hammer.The material that hits the screen has varying grain classes.Sometimes, the material is bulky enough to flow through the screen without requiring fragmentation.In some cases, the size of the oversized blocks can exceed the required size imposed by the dimensions of the screen.Depending on the result of the blasting process, the sizes of the rock fragments can be very different, sometimes even several times larger than required.
LiDAR (Light Detection and Ranging) technology enables the acquisition of information about an object's geometry at high resolution.On the basis of the measurement of the distance traveled by the laser beam, the position of the object in the 3D space is determined.The data provided are in the form of a point cloud.LiDAR systems, because of their speed and noninvasiveness, are successfully used in demanding and hazardous mining environments.They are used, among other things, to study the geometry of underground tunnels [9][10][11], deformation monitoring [12,13], detection of damage to the components of the mining infrastructure [14], Simultanous Localization and Mapping implementations [15,16], or calculation of the ore volume [17].
When it comes to underground measurements at mining plants, it is essential to consider various limitations.These measurements must not interfere with the plant's operation, resulting in downtime.However, some environmental and technical factors can impact the accuracy of the measurements, including high temperature, humidity, dust, poor lighting, limited space, a lack of measuring stations, and the potential for random events or equipment and personnel failures.Therefore, it is crucial to use appropriate apparatus and measuring devices that can effectively operate under such conditions.
Based on the preliminary measurements taken while using the hydraulic hammer, it appears that laser scanning could be a viable option.However, it is important to note that in the case of presented measurements only one perspective was available.Taking measurements from multiple perspectives was not feasible during the hydraulic hammer breaking operation due to limited space and safety constraints.The surface of the rock block that needs to be recorded is often covered by solid fragments in the foreground, often preventing proper identification.The authors attempted to resolve the problem by using laser scanning data from a singular perspective to detect the excessively large copper ore blocks.

Experiment description
The laser scanning measurements were performed in an underground mine during the hydraulic hammer rock breaking operation performed on the screen.The measurement scene is shown in Figure 1.Data collection was performed from a single scanner station using a Livox sensor.It is a LiDAR scanner with a linear cloud pattern that captures 240,000 points/s (first or dual return).The scanner field of view is 81.7 x 25.1 degrees, and the maximum scanning range is 260 m.The standard deviation when measuring distances is less than 2 cm at a distance of 20 m, and less than 0.05 • when measuring angles [18]. 3

Point cloud data preprocessing
Processing the raw data in the form of a point cloud acquired from the LiDAR sensor involved three steps.The first step was scan extraction, which separated 20 scans from each measurement of a given object, every 0.2 seconds of measurement.The final point cloud was the combined scans from 4 seconds, which ensured sufficient density of the point cloud.The second stage included filtering the point cloud using the intensity parameter.Points in a specific intensity range, below the set of values that show the highest abundance in the point cloud, were identified as measurement noise and removed.Outlier points in the cloud were also excluded using the Statistical Outlier Removal (SOR) filter.It operates under assumption that the distance between given point and its neighbours has normal (gaussian) distribution, which allows to identify outliers based on a set of parameters (number of nearest neighbours and number of standard deviations from the mean) [19].
The final stage of the initial extraction was to identify the oversized rock particles and extract them as separate point clouds.

Geometry analysis
The main goal of the presented geometry analysis is to identify oversized rock pieces.In the case of the hammer-breaking process during which the data was collected, oversized rocks can be defined as pieces that would not fit through a screen window of a square shape with 40x40 centimeter dimensions.
A single geometry feature, such as the distance between two points, cannot be used as a definitive criterion to determine the material as oversized.Since the blocks can be rotated, which would make them fit through the screen, at least two mutually dependent sets of two points are necessary.The flow chart of the algorithm for geometry analysis can be seen in Figure 2.  The minimum condition, which can also be treated as an initial step, is to find two points with the largest Euclidean distance within the obtained data set.Euclidean distance is a direct extension of the Pythagorean theorem applied to the 3D space.Its formula can be seen in Eq. 1.
where the x 1 , y 1 , z 1 are coordinates of a point P 1 and x 2 , y 2 , z 2 coordinates of a point P 2 , respectively.
If the maximum distance found is smaller than the defining feature of the screen window shape, the rock is classified as not oversized.In the case of this experiment, the minimum distance required is based on the diagonal length of the square.
If the first requirement is met, the found points are used to define the axis to which a perpendicular plane is created.For that, a direction vector of a line is necessary and it can be based on the points found in previous steps.The equation for the direction vector of a said line can be seen in Eq. 2, where the coordinates come from the points P 1 and P 2 which are the selected furthest points inside the segmented point cloud.
Every plane can be defined by the general form as in Eq. 3.
where A, B and C coefficients are the components of the normal vector perpendicular to the plane itself [20].In the vector form, the plane equation can be written as in Eq. 4.
where ⃗ n is a normal vector of a plane, which in this case is equal to the direction vector of a line defined in Eq. 2 (so ⃗ n = ⃗ l), ⃗ r is a position vector of a general point on the plane and ⃗ r 0 is a position vector of a selected point on the plane, which in case of the presented algorithm is a point (either P 1 or P 2 ) of intersection between the plane and a perpendicular line In summary, we create a plane perpendicular to the line defined by two points (P 1 and P 2 ) with the point of intersection being one of them.The exact intersection point between the plane and the axis is not important, as the distance of points from the plane is not mapped (a depth map is not created).Due to that, the intersection point was taken based on one of the line-defining points (P 1 ).The equation of such a plane can be written as in Eq. 5 and the coefficients of that plane can be described by the equations presented in Eq. 6.
In the next step, every point from the scanning data is projected onto the created plane.This is done by defining the normal vector between the point and plane, perpendicular to the plane (so it can be defined as the closest distance from a point to the plane) and then finding the point of intersection.The new coordinates P ′ = (x ′ , y ′ , z ′ ) of any point in the cloud P = (x, y, z) can be described as in Eq. 7, which is based on the orthogonal projection [21].
Once more, the largest possible distance between two points is found, this time over the data processed to a 2D space.If the minimum distance requirement is not met, the object is classified as not oversized.Analogically, if the maximum distance found is larger than the required one, the object is classified as oversized.

Results
The processing of the raw point cloud obtained from one measurement station was performed according to the steps described in Subsection 3.1.The results of the data before and after processing are presented in Figure 3. Initially extracted point cloud of the given rock block can be subjected to further processing.First, the coordinates of the furthest points have been found and the distance between them calculated.The calculated distance was equal to 0.66 m, which is larger than the side (0.40 m) of the square that defines the screen window or its diagonal (0.56 m).Therefore, this rock is initially classified as potentially oversized and needs further processing.Scheme of a measurement station used in the experiment can be seen in Figure 4.  To find the second pair of points that allow for the final classification, a plane perpendicular to the line obtained in the previous step is created.The intersection point between the plane and the line is defined by one of the line endpoints.The visualization of this step is presented in Figure 5. Furthermore, every point in the point cloud was projected onto the created plane as can be seen in Figure 6a.Then, the furthest points in the newly created point space were found.The distance between them was equal to 0.47 m, which is still larger than the side of the square described previously, which allows one to classify the measured rock block as oversized.Visualization of the distance found can be seen in Figure 6b.

Conclusions
The authors presented a methodology for the classification of oversized rocks using data obtained from a single laser scan station.The methodology was based on the calculation of the geometry features: two interconnected lines that define the shape as impossible to fit through the window on the screen.Segmentation, the initial part of the algorithm, was done by manual selection of points after additional preprocessing ensuring removal of the unwanted points.This means that for the full automatization of the method, the segmentation has to be resolved in the future work.Most promising approaches involve use of the machine learning, either by segmentation performed directly on the obtained point cloud, or indirectly by segmenting the RGB and applying the segmentation to the point cloud afterwards.
In the case of the site where the measurements were made, said windows had a square shape with a side of 40 cm size.The example provided for the methodology was correctly classified as oversized due to geometric features in the form of two perpendicular lines.Each line was formed from the two farthest points in the processed cloud.Presented methodology can be easily adapted to different shapes and in most of the cases it won't require any modifications other than changing the line length values.

Figure 1 :
Figure 1: Measurement station in an underground mine.

Figure 2 :
Figure 2: Flowchart of the classification procedure.

Figure 3 :
Figure 3: On the left -raw point cloud in form of a whole hammer operation scene.In the left-down corner a screen with a hammer hanging above it can be seen.Colors are based on the surface reflectivity.On the right -fragment extracted from the raw point cloud, representing potential oversized rock particle.

Figure 4 :
Figure 4: Visualization of found two furthest points in the scan with a line drawn between them: a) front perspective, b) side perspective

Figure 5 :
Figure 5: Created plane perpendicular to the line defined by two found points

Figure 6 :
Figure 6: a) Visualization of the point projection process.Original points represented by black dots and projected points represented by red dots, b) Effect of the projection with a line drawn between two new furthest points