Projection of soil properties in banks

Abstract

The present disclosure describes a method for predicting and projecting soil properties in banks in which drilling according to a blasting plan has not yet commenced. For this purpose, the method uses information from ad-hoc test drilling and information from sensor data from drilling adjacent banks, vertically and laterally. The information is subsequently processed using pattern recognition and projection techniques, such as neural networks or other machine learning techniques, and hardness is projected for the bank in which drilling according to a blasting plan has not yet commenced.

Description

PROJECTION OF LAND PROPERTIES IN BANKS FIELD OF INVENTION

[0001] The present invention relates to the mining and construction industries. In particular, the present invention relates to a method for predicting and projecting soil and rock properties in a bench that has not yet been drilled. STATE OF THE ART

[0002] Currently, in civil construction and open-pit mining, masses of soil material are extracted, either to create space for civil works or to extract mineral resources from the excavated material. The extraction of these masses is typically carried out in stages, progressing through horizontal layers that are subdivided into smaller units called benches, which in turn form stepped structures.

[0003] Various techniques are used to extract the masses of soil material that form the benches, including bench blasting. Bench blasting involves drilling a series of vertical holes or boreholes into the upper surface of the bench in patterns known as a blast plan. These holes are then filled with explosives, and the material is fragmented and broken up by detonating the explosives.

[0004] The design of blasting plans in mining relies heavily on geological information about soil properties, which is the primary parameter to consider when deciding how to carry out a blast. This geological information is mainly obtained from spot borehole tests and geostatistical models that interpolate the data measured between different boreholes to estimate soil properties across the bench. These soil properties can include material hardness, lithology, volatility, indications of water presence, high temperatures, presence of blast-reactive material, material density, percentage of any valuable material to be extracted, and so on.

[0005] However, borehole testing is expensive and scarce, so it may be necessary to plan the blasting of a material deposit that has not been tested or that only has one or two boreholes for analysis. In such situations, limited information leads to inaccurate predictions and subsequent inefficient blasting plans.

[0006] Other methods used to project soil properties include using sensor data to estimate hardness or other geological parameters, such as seismic measurements and measurements taken on a drilling rig during drilling operations, such as when drilling boreholes, pits, or blast holes in a blasting plan. These technologies allow for inferences of geological parameters based on sensor data, improving the amount of information available for projecting soil properties. A bench blasting plan typically involves drilling a number of blast holes, ranging from a few dozen to several thousand, in some cases up to 5,000 blast holes per bench.However, this additional information is only obtained after executing a drilling plan and after the drilling has already been carried out, which means that sometimes with the additional information, the conclusion is reached that the blasting plan executed is inefficient or inadequate.

[0007] Thus, current technologies and methods only allow for marking decisions to be made for the design of a bench in which a blasting drilling plan is already being executed or which has already been drilled, including the charge or gunpowder factor, the allocation of the explosive material and the detonation sequence.

[0008] Among the known methods is publication W02024050637A1, which describes a drilling and blasting method and system with fragmentation modeling. The publication explains that fragmentation is measured using sensors installed on the excavator shovels and these measurements are correlated with blasting parameters. However, it does not allow for design decisions to be made regarding future benches that have not yet been worked.

[0009] There is also publication W02022016207A1 which describes a process for improving blast design that includes the steps of: acquiring geological data on a blast site from multiple available data sources; extracting one or more explosivity labels from the geological data; mapping the explosivity labels in two or three dimensions; designing a blast hole pattern based on this mapping; designing the placement of the explosive within each blast hole based on the explosivity label for that blast hole; and loading each blast hole in the blast hole pattern with the explosive. designed for that blast hole; and detonate the explosive. However, it does not allow design decisions to be made regarding future benches that have not been worked on, it only operates with respect to mining benches from which data is directly extracted from processes already executed on them.

[0010] Therefore, new techniques are required to project soil properties more accurately in benches where a blasting plan has not yet begun, in order to improve the blasting pattern and adjust both the drilling pattern and the spacing between holes based on the soil properties, in order to improve the final fragmentation and reduce inefficiencies in the process. SUMMARY DESCRIPTION OF THE INVENTION

[0011] This disclosure describes a process for predicting and projecting soil properties in benches where blasting has not yet begun. This is achieved by using information from spot sounding tests and sensor data from drilling operations in adjacent benches, both vertically and laterally. This information is then processed using pattern recognition and projection techniques, such as convolutional neural networks (CNNs) or other machine learning techniques, to project the hardness of the bench where blasting has not yet commenced. DESCRIPTION OF THE FIGURES

[0012] Figure 1A shows a schematic representation of a first banking exploitation sign

[0013] Figure 1B shows a schematic representation of a second bank exploitation sign

[0014] Figure 1C shows a schematic representation of a third banking exploitation sign.

[0015] Figure 2A shows a schematic representation of a plan view of a bank where its volatility has been projected based on prior art techniques.

[0016] Figure 2B shows a schematic representation of a plan view of the bank in Figure 2A where its volatility has been projected based on the technology of the present application.

[0017] Figure 3 shows a schematic representation of a plan view of a bench where its soil hardness has been projected based on prior art techniques.

[0018] Figure 4 shows a schematic representation of a blasting plan based on the soil hardness of Figure 3.

[0019] Figure 5 shows a schematic representation of a plan view of the bench in Figure 3 where its soil hardness has been projected based on the technology of the present application.

[0020] Figure 6 shows a schematic representation of a blasting plan based on the soil hardness of Figure 5.

[0021] Figure 7 shows a schematic representation of the main stages of the technology of the present application.

[0022] Figure 8 shows a schematic representation of the main stages of the process for segmenting at least one adjacent bank according to the technology of the present application.

[0023] Figure 9 shows a schematic representation of the main stages of the process for determining soil properties for an adjacent bank according to the technology of the present application.

[0024] Figure 10 shows a schematic representation of the main stages of the process for predicting soil properties for a bank under study according to the technology of the present application. DETAILED DESCRIPTION OF THE INVENTION

[0025] The present invention aims to obtain a better description of soil properties such as rock mass hardness and volatility, from borehole drilling data and some geotechnical field measurements.

[0026] Classical geostatistical block models in the medium term are taken as a reference, to obtain a measure of the gaps in the description of hardness, in order to evaluate the performance of our data-based proposal.

[0027] The technology can be broken down into two stages; the first aims to model the behavior of soil properties, such as hardness or volatility, based on the measurement of the hole while drilling data and the truth of the ground, in order to obtain an inference of soil properties, such as hardness or volatility, based on drilling data, using semi-supervised machine learning techniques.

[0028] Furthermore, the present technology addresses the problems of missing data, using supervised learning algorithms, filling in missing or incomplete data gaps with a prediction of the properties based on the behavior of the filled gaps.

[0029] The second part of the technology is related to predicting soil properties, such as hardness or volatility, of a subsequent blasting polygon, using only the position of the rock mass and employing a supervised learning algorithm trained with data from previous drilling.

[0030] Given the above procedure, the resolution of the hardness prediction depends on the true ground data provided by the on-site measurements, and can deliver, for example, a hardness range, as a label, or a numerical prediction, based on a magnitude that could be MPa, with a certain probability.

[0031] The results are then interpolated to obtain a continuous estimate of the hardness in the exploitation zones. This is used by blast designers to improve their design decisions, allowing us to obtain better fragmentation, more safety, and cost savings throughout the drilling and blasting process.

[0032] In the first stage, which involves measuring and labeling drilling data, the process aims to represent and characterize the influence of the drilled holes. Measurements that can be taken in the field during drilling include torque, penetration rate, air pressure, RPM, and pulldown, among others. This information is often referred to as MWD (Measures While Drilling). However, it is also possible to insert probes into drilled holes to measure soil properties using sensors on the probe.

[0033] These data are assumed to be uniform in a geometry around each borehole that forms a finite spatial domain, or they are considered uniform in a subsection of the finite spatial domain for cases where it is desired to evaluate the change in soil properties with depth. The geometry used can be the projection of a hexagon or the projection of another geometric solid, such that the The set of geometric bodies allows tessellation of the plane or at least substantially covering the plane while avoiding overlaps.

[0034] In some cases, data from a borehole may be missing or deemed statistically invalid. In these cases, the process allows for interpolating measurements to complete the missing information.

[0035] In this stage, a soil property scale is constructed, including rock mass hardness and volatility, and each finite spatial domain associated with each borehole is classified. On the diagram, the scale ranges from 1 to 5 (very soft, soft, medium, hard, very hard). Different scales can be used depending on the desired granularity level.

[0036] To assess soil hardness, the UCS (Ultra-Stepping Compound) value from spot borehole tests is used as input data. This value allows the hardness scale labels to be correlated with a field measurement, which is then considered in blast design. The set of spot borehole tests is measured using a sample dataset, which is often less than 5% of the total data volume.

[0037] Next, the MWD parameters are correlated with the truth on the ground to determine their weight and obtain calibration metrics.

[0038] In the second stage, the properties of the soil, such as hardness or volatility, of an adjacent bank that has not been drilled are estimated.

[0039] In this procedure, a model is trained to predict the hardness of adjacent benches based on the labels and their position. The hardness distributions of the lower benches were estimated in the diagram below. Future hardness zones can be predicted using machine learning systems, such as a convolutional neural network (CNN) classifier.

[0040] The predictive model of this technology is applicable to various soil properties, and in particular to soil hardness. In one mode, the predictive model is trained using the (x,y,z) positions of historical boreholes, and the hardness label of these boreholes, obtained in the previous step, allows for predicting the hardness level for a new (x,y,z) position. Specifically, for the design use case For blasting, a discrete number of points are taken within a polygon or finite space of interest, their respective hardness is predicted, and the predictions are extrapolated within the area of ​​interest using proximity criteria. These criteria can be, for example, inverse square of distance, nearest neighbors, etc.

[0041] The above allows hardness to be assigned to the different zones in a generalized way, so that the conditions by zone can be visualized, and can be used to design the parameters of the blast in question.

[0042] Once the hardness distribution of the bench under study has been obtained, design constraints may include the drilling pattern (burden and spacing), borehole length and diameter, quantity and type of explosive, loading factor, and detonation sequence. These decisions allow for the optimization of blasting results.

[0043] The invention is able to integrate the models into an interface that allows users to obtain hardness information and predictions by simply loading a blast polygon design file (usually a *.dxf file).

[0044] When neural networks are used for pattern recognition, the network is prepared by obtaining baseline data that is filtered to select the most consistent or stable data within the training data, forming filtered data. This filtered data is then divided into training data and test data. The training data is used to train neural networks, such as convolutional neural networks. The training is then tested with the test data in a supervised or semi-supervised training process until a success rate considered acceptable according to a reliability criterion is achieved.The result of this process is a pre-trained neural network; however, pre-trained neural networks can undergo an adjustment process with additional data from the specific analysis site where it will be used. With the additional data, a retraining is performed that allows the results obtained to be better adjusted to the particular parameters of each area.

[0045] Figure 1A shows a schematic representation of a first bank exploitation sequence, illustrating 200 adjacent banks of this first exploitation sequence.

[0046] Figure 1B shows a schematic representation of a second bank operating system whose process is carried out after the exploitation of the banks illustrated in Figure 1 A and where adjacent banks 200 of this second series of exploitation are illustrated.

[0047] Figure 1C shows a schematic representation of a third series of bench exploitation whose process is carried out after the exploitation of the benches illustrated in Figure 1B. In this figure, a bench under study 100 is illustrated, for which the present technology allows its soil properties to be projected based on borehole drilling information and based on the soil property information of one or more of the adjacent benches 200 illustrated in Figures 1A, 1B and 1C.

[0048] Figure 2A shows a schematic plan view of a deposit where its volatility has been projected based on prior art techniques. The projection established that the deposit consisted mainly of high-volatility material 304 and a low-volatility zone 302.

[0049] Figure 2B shows a schematic plan view of the deposit in Figure 2A, where its volatility has been projected based on the technology of this application. The projection established that the deposit consisted mainly of medium-volatility material 303 and high-volatility material 304, with some small areas of very low-volatility material 301, low-volatility material 302, and very high-volatility material 305. Comparing Figure 2A with Figure 2B, it can be seen that the volatility projection in the prior art underestimated the deposit's volatility by assigning a low-volatility value of 302 to a large area of ​​the deposit.

[0050] Figure 3 shows a schematic representation of a plan view of a bench where its soil hardness has been projected based on prior art techniques, illustrating areas of very soft soil 401 with compressive strength between 10 and 25 MPa, areas of soft soil 402 with compressive strength between 25 and 50 MPa, and areas of medium soil 403 with compressive strength between 50 and 100 MPa.

[0051] Figure 4 shows a schematic representation of a blasting plan based on the soil hardness of Figure 3 where a blast drilling plan was projected with a burden x spacing pattern of type 1 1 ,5 x 11 ,5 meters, with a total of 278 blast drillings 500.

[0052] Figure 5 shows a schematic representation of a plan view of the bench in Figure 3 where its soil hardness has been projected based on the technology of the present application, illustrating areas of very soft soil 401 with compressive strength between 10 and 25 MPa, areas of soft soil 402 with compressive strength between 25 and 50 MPa, areas of medium soil 403 with compressive strength between 50 and 100 MPa, areas of hard soil 404 with compressive strength between 100 and 150 MPa and areas of very hard soil 405 with compressive strength between 150 and 250 MPa.

[0053] Figure 6 shows a schematic representation of a blasting plan based on the soil hardness shown in Figure 5. This plan projects a blast hole pattern of 11 x 13 meters, with a total of 256 blast holes. Compared to the blasting pattern in Figure 4, the new pattern illustrated in Figure 6 requires 8% fewer holes. In this particular case, this optimization avoids drilling 22 blast holes and the consequent drilling of 264 linear meters of rock.

[0054] In one embodiment, the present technology refers to a method for predicting soil properties in a bank under study comprising the steps of: segmenting at least one adjacent bank (600); determining soil properties for each adjacent bank (700); and predicting soil properties for the bank under study (800) from the soil property values ​​of the adjacent banks and the relative position of each adjacent bank with respect to the bank under study.

[0055] The step of segmenting at least one adjacent bench (600) involves the steps of defining a geometric drilling space (601) around drilling holes in the adjacent bench, defining a geometric drilling space (602) around blasting holes in the adjacent bench, and defining projected geometric spaces (603) in those locations of the adjacent bench that have not been covered by geometric drilling spaces or geometric drilling spaces.

[0056] The stage of determining soil properties for each adjacent bench (700) involves the steps of selecting at least one soil property (701), assigning a soil property value to each borehole geometric space (702) based on borehole information associated with each borehole, and assigning a soil property value to each borehole geometric space (703) based on information measured in the drilled blast holes; and assign a soil property value to each projected geometric space (704) based on a projection made from the soil property values ​​of the borehole geometric spaces and the drilling geometric spaces.

[0057] The stage of predicting soil properties for the bank under study (800) involves the steps of segmenting the bank under study (801) by defining geometric spaces under study, defining the relative position of each adjacent bank (802) with respect to the bank under study, and predicting and assigning a soil property value for each geometric space under study (803) based on a projection made from the soil property values ​​of the geometric spaces of the adjacent banks and the relative position of each adjacent bank with respect to the bank under study.

[0058] In some versions of this technology, the value of each soil property corresponds to a range on a scale defined for that soil property. Assigning a range instead of a specific soil property value facilitates the grouping of areas with similar values, simplifying the interpretation of input variables and reducing the analytical resources required for value projections. Working with ranges also allows for approximating values ​​by area, so that if an area has several values ​​all within the same range, the soil property value can be approximated to an average value within that range.

[0059] In some versions of this technology, the step of assigning a soil property value to each borehole geometry space (703) also includes identifying boreholes with incomplete information and then predicting and assigning a value to the incomplete information based on a projection made from the soil property values ​​of the borehole geometry spaces and the borehole geometry spaces with complete information. This allows the use of information from boreholes where data only exists partially or where some of it was discarded because it fell outside filtering parameters.

[0060] In some forms of the present technology, the step of assigning a soil property value to each projected geometric space (704) is carried out by pattern recognition and projection techniques.

[0061] In some forms of this technology, pattern recognition and projection techniques are machine learning techniques. using deep neural networks (DNNs), including convolutional neural networks (CNNs). Deep neural networks (DNNs) have densely connected layers with nodes for learning global features. Convolutional neural networks (CNNs), which are a particular type of deep neural network (DNN), include layers specialized in spatial hierarchies, making them particularly efficient for tasks that require matrix representation of data, such as image processing.

[0062] In some forms of the present technology, the step of assigning a soil property value to each projected geometric space (704) is carried out by geostatistical interpolation methods such as Gaussian process regression techniques (kriging).

[0063] In some versions of this technology, the geometric spaces of the sounding, the geometric spaces of the drilling, the projected geometric spaces, and the geometric spaces under study are vertically subdivided into subsections, where the step of predicting and assigning a soil property value for each geometric space under study (803) predicts and assigns a soil property value for each subsection. This allows the incorporation of records of soil property variations at different depths within the geometric space and their weighting according to their proximity to the bench under study.Thus, if for example in an adjacent bank that is located above the bank under study, the soil property values ​​of the lower subsections of the geometric spaces of the adjacent bank positioned above the bank under study will have a greater weight in the projection of soil properties compared to the soil property values ​​of the upper subsections of the geometric spaces of the adjacent bank positioned above the bank under study.

[0064] In some forms of this technology, the step of predicting and assigning a soil property value for each geometric space under study also includes assigning a soil property value to the geometric spaces under study that contain a borehole based on associated borehole information.

[0065] In some forms of this technology, the information measured in the drilled blast holes is measured while drilling (MMP) information obtained from sensors on a drilling machine and selected from torque, penetration rate, air pressure, RPM, pulldown, and vibrations.

[0066] In some forms of this technology, the information measured in the blast holes drilled is obtained by means of probes or analysis of the material extracted from the drilling.

[0067] In some forms of this technology, based on the prediction of properties in the study bench, design parameters are selected from burden, spacing, loading factor, borehole diameter and length, type and quantity of explosive, and detonation sequence. APPLICATION EXAMPLES

[0068] The present technology was applied in an open pit mine and allowed modification of the drilling pattern (burden x spacing), allowing a reduction in the meters of drilled wells and adjusting the distribution of the drillings to more efficient patterns.

[0069] For example, in a study bench, the soil property of hardness was selected using prior art techniques based on hardness information from historical sounding tests. The result of this analysis projected the hardness illustrated in Figure 3, and from this, a blast drilling plan was projected with a burden x spacing pattern of the type 11.5 x 11.5 meters, with a total of 278 blast holes illustrated in Figure 4.

[0070] Subsequently, the hardness of the bench was projected using data measured while drilling (MMP) from adjacent benches that had been drilled prior to the bench under study. This data was processed using the current technology, supplementing the information from the adjacent benches and employing machine learning techniques. The result of this new analysis projected the hardness illustrated in Figure 5, and from this, a blasting plan was projected with a 1 x 13 meter pattern, with a total of 256 blast holes, illustrated in Figure 6.

[0071] In the example described above, this technology allowed for the prediction and projection of soil and rock properties in a bench under study that had not yet been drilled, specifically projecting the hardness in order to optimize the blasting plan. As a result, the new blasting pattern required 8% fewer holes. In this particular case, this optimization avoided drilling 264 linear meters of rock. REFERENCE NUMBERS

[0072] Below is a list of the elements of the invention illustrated in the figures and their respective reference numbers. 100 Bank under study 200 Adjacent Bank 301 Very low volatility 302 Low volatility 303 Average volatility 304 High Volatility 305 Very high volatility 401 Very soft soil 402 Soft soil 403 Average floor 404 Hard floor 405 Very hard ground 500 Blasting Drilling 600 Segment at least one adjacent bank 601 Define a geometric sounding space 602 Define a geometric drilling space 603 Define projected geometric spaces 700 Determine soil properties for each adjacent bank 701 Select at least one land property 702 Assign a soil property value to each geometric sounding space 703 Assign a soil property value to each geometric drilling space 800 Predict soil properties for the bank under study 801 Segment the bank under study 802 Define the relative position of each adjacent bank 803 Predict and assign a soil property value for each geometric space under study

[0073] Finally, it should be noted that various specific parameters of the invention, such as dimensions, choice of materials, and specific aspects of the preferred configurations described above, may vary or be modified according to operating requirements. Consequently, the configurations The specific variations and / or modifications described above are not intended to be limiting, and such variations and / or modifications are within the spirit and scope of the invention.

Claims

CLAIMS 1. A method for predicting soil properties in a bench under study comprising the steps of: segmenting at least one adjacent bench (600) which for each segmented adjacent bench involves the steps of: defining a borehole geometric space (601) around boreholes in the adjacent bench; defining a blast hole geometric space (602) around blast holes in the adjacent bench; defining projected geometric spaces (603) in those locations of the adjacent bench that have not been covered by borehole geometric spaces or blast hole geometric spaces; determining soil properties for each segmented adjacent bench (700) which for each segmented adjacent bench involves the steps of: selecting at least one soil property (701); assigning a soil property value to each borehole geometric space (702) based on borehole information associated with each borehole;assign a soil property value to each drill hole space (703) based on information measured in the drilled blast holes; assign a soil property value to each projected space space (704) based on a projection made from the soil property values ​​of the borehole spaces and the drill hole spaces; predict soil properties for the bench under study (800) involving the steps of: segmenting the bench under study (801) by defining study spaces; defining the relative position of each adjacent bench (802) with respect to the bench under study; predicting and assigning a soil property value to each study space (803) based on a projection made from the soil property values ​​of the adjacent bench spaces and the relative position of each adjacent bench with respect to the bench under study.

2. The method according to claim 1, wherein the at least one soil property is selected from hardness, density, volatility, conductivity, resistivity, chemical characterization, electromagnetism, porosity, temperature, degree of deviation in drilling, presence of water, presence of reactive materials to explosive materials, grade of a valuable mineral.

3. The method according to claim 1, wherein the value of each soil property corresponds to a range of a scale defined for said soil property.

4. The method according to claim 1, wherein the step of assigning a soil property value in each borehole geometric space (703) further includes determining boreholes with incomplete information and then predicting and assigning a value to the incomplete information based on a projection made from the soil property values ​​of the borehole geometric spaces and the borehole geometric spaces with complete information.

5. The method according to claim 1, wherein the step of assigning a soil property value to each projected geometric space (704) is performed using pattern recognition and projection techniques.

6. The method according to claim 5, wherein the pattern recognition and projection techniques are machine learning techniques using neural networks selected from convolutional neural networks (CNNs) and deep neural networks (DNNs).

7. The method according to claim 1, wherein the step of assigning a soil property value to each projected geometric space (704) is performed by geostatistical interpolation methods such as Gaussian process regression techniques (kriging).

8. The method according to claim 1, wherein the geometric spaces of sounding, the geometric spaces of drilling, the projected geometric spaces and the geometric spaces under study are vertically subdivided into subsections, wherein the step of predicting and assigning a soil property value for each geometric space under study (803) predicts and assigns a soil property value for each subsection.

9. The step of predicting and assigning a soil property value for each geometric space under study also includes assigning a soil property value to the geometric spaces under study that contain a borehole based on associated borehole information.

10. The method according to claim 1, wherein the information measured in the drilled blast holes is measured-while-drilling (MMP) information obtained from sensors on a drilling machine and selected from torque, penetration rate, air pressure, RPM, pulldown, and vibrations.

1. The method according to claim 1, wherein the information measured in the drilled blast holes is obtained by means of probes or analysis of the material extracted from the borehole.

12. The method according to claim 1, wherein, based on the prediction of properties in the study bench, design parameters are selected from burden, spacing, loading factor, borehole diameter and length, type and quantity of explosive, and detonation sequence.

Images