Blasting fragmentation prediction method suitable for complex terrain mine

Abstract

The invention discloses a blasting fragmentation prediction method suitable for a complex terrain mine, and the method is based on high-precision muck pile three-dimensional reconstruction, is combined with an image segmentation algorithm, and enables the calculated overall fragmentation error to be small through the correction of a spatial distribution model. Meanwhile, the overall lumpiness distribution obtained through the method is closer to actual lumpiness distribution, and through boulder rate comparison evaluation, a scientific basis can be provided for mine blasting effect evaluation and production management, blasting parameters can be optimized, and production efficiency can be improved.

Description

Technical Field

[0001] The invention relates to the technical field of blasting analysis, in particular to a blasting fragmentation prediction method suitable for mines with complex terrain. Background Art

[0002] In mining operations, due to the complexity of on-site geological structures and the dynamic changes in blasting parameters, blasting effects often show significant differences. Among them, the particle size distribution characteristics and spatial heterogeneity of blasting pile size are key technical indicators for evaluating the quality of blasting effects, and directly affect the operating efficiency and economic costs of subsequent shoveling, transportation and crushing processes. At present, traditional blasting pile size measurement mainly relies on manual measurement or screening methods, which have problems such as long time consumption, high labor intensity, high cost, and limited accuracy. It is difficult to widely promote its use in engineering blasting. Therefore, how to quickly count the blasting pile size and optimize the blasting parameters to reduce production costs according to actual on-site needs is one of the problems that need to be solved in actual engineering. Summary of the Invention

[0003] To solve the above problems, the present invention provides a blasting fragmentation prediction method applicable to mines with complex terrain, comprising the following steps: S1. Acquire images of the blast pile based on drone oblique photography technology; S2. Using the burst pile image obtained in step S1, construct a three-dimensional burst pile model; S3. Intercept local images of the implosion pile in the model to identify the surface fragmentation of the implosion pile, generate a fragmentation distribution curve, and output data such as average fragmentation and classification percentage; S4. Determine the average deviation rate between the overall and surface fragmentation of the implosion pile through experiments. Based on the deviation rate, correct the surface fragmentation of the local implosion pile in the model and predict the local overall fragmentation. Use manual screening to systematically verify the predicted results.

[0004] Based on the above solution, the specific method of step S4 is as follows: Five groups of blast pile samples were randomly selected within the same blasting area. Systematic sampling was performed on each blast pile to obtain five sets of overall blast pile average fragmentation data. Simultaneously, mobile phone photos were taken of these five groups of test blast piles to collect blast pile images, and blast pile models were established for each of them. Using S3 technology, different surface average fragmentation data were obtained for these five groups of blast piles, and the deviation rate between the overall and surface average fragmentation was determined. Based on the average deviation rate determined by the experiment, the surface fragmentation of the intercepted local explosion pile is corrected, and then the local corrected overall fragmentation is predicted, and the predicted results are systematically verified by manual screening method.

[0005] On the basis of the above scheme, the method for systematic sampling of each blast pile is as follows: systematic sampling of each blast pile: at the three longitudinal positions in front, middle and back of the blast pile, each position is divided into three horizontal measuring points at left, middle and right, and at the same time, rock samples are collected at the upper, middle and lower levels at each measuring point.

[0006] Based on the above scheme, the specific method of step S3 is as follows: select a local image of the blast pile for interception; import the intercepted blast pile image into ImageJ for preprocessing to highlight the rock block boundaries; then import the processed image into WipFrag, and extract the rock block boundaries through the edge detection algorithm; then use the watershed algorithm to segment the adjacent rock blocks; after the segmentation is completed, the software automatically calculates the equivalent particle size, aspect ratio and other parameters of each rock block; then convert the actual reference object pixel size in the calibration image into physical size to complete the scale conversion; finally, generate the block size distribution curve, and output the average block size and classification percentage.

[0007] Based on the above scheme, the method of step S2 includes: generating three-dimensional point cloud data from the explosive pile image based on motion recovery structure and multi-view stereo matching technology; gridding the three-dimensional point cloud data, constructing a triangular mesh and generating a three-dimensional explosive pile model; and evaluating the surveying and mapping accuracy of the three-dimensional explosive pile model through ground control point coordinate error analysis.

[0008] Based on the above scheme, the method of generating 3D point cloud is as follows: import the blast pile images collected by UAV into ContextCapture software for aerial triangulation, and then use aerial triangulation to improve the accuracy. However, before encryption, at least 3 control points need to be arranged on site before aerial survey and distributed in the entire aerial photography area. The longitude, latitude and elevation data of each control point are determined by GPS, and imported into ContextCapture software as ground image control points. The aerial film where each control point is located is accurately located in the image library. Each control point needs to be marked on no less than 4 different images. After completion, the aerial triangulation is used to calculate the control points. The calculation makes the coordinate system of the final model consistent with the geographic coordinate system required for engineering research; during the aerial triangulation analysis process, the software will automatically analyze and extract a large number of similar feature points in each aerial survey image, and then match each similar feature point, analyze the spatial position of the camera and the tilt angle of the camera when taking each aerial photo, and thus determine the spatial position relationship between each image; finally generate the camera pose and sparse point cloud; MVS determines the geometric relationship between images based on the generated camera pose, improves matching efficiency, and uses the sparse point cloud as the initial reference to constrain the depth estimation range, verify and optimize the reconstruction results, and the combination of the two finally generates a high-precision 3D point cloud.

[0009] On the basis of the above scheme, the specific method of gridding the three-dimensional point cloud data, constructing a triangular mesh and generating a three-dimensional explosion pile model is as follows: using the Delaunay triangulation algorithm on the high-precision point cloud, selecting three non-collinear points from the point cloud as the initial triangle vertices, adding the remaining points in sequence and adjusting the triangle connection relationship according to the Delaunay criterion, and finally forming an optimized TIN mesh; and based on this mesh, defining the topological structure of the model, performing splicing, stitching, denoising and smoothing processing to construct a three-dimensional model without texture information; and then extracting the high-pixel texture of the corresponding position from the image data according to the spatial position of the three-dimensional model for coloring, and finally outputting a three-dimensional model with realistic texture and bright colors.

[0010] The beneficial effects of the present invention are as follows: the method of the present invention is based on high-precision three-dimensional reconstruction of blast piles, combined with an image segmentation algorithm, and corrected through a spatial distribution model, so that the error in the inferred overall fragmentation is relatively small; at the same time, the overall fragmentation distribution obtained using the method of the present invention is closer to the actual fragmentation distribution, and through comparative evaluation of large-block rates, it can provide a scientific basis for mine blasting effect evaluation and production management, optimize blasting parameters, and improve production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0012] Figure 1 The original image of the local burst pile captured in step S1 of embodiment 2 of the present invention; Figure 2 The three-dimensional modeling diagram of the entire explosive pile generated in step S2 of Example 2 of the present invention; Figure 3 Renderings of the model explosion pile image before and after enhancement and noise filtering in step S3 of Example 2 of the present invention; Figure 4 This is the effect diagram after the edge detection algorithm is used to identify the rock contour in step S3 of Example 2 of the present invention; Figure 5 This is a diagram showing the effect of segmenting adjacent rock blocks using the watershed algorithm in step S3 of Example 2 of the present invention; Figure 6 The blockiness distribution curve obtained in step S3 of Example 2 of the present invention is generated; Figure 7 The particle size distribution diagrams for each particle size range before and after the screening method and correction in step S4 of Example 2 of the present invention are shown; Figure 8 This is the overall particle size distribution of the explosive pile after 5 correction tests in step S5 of Example 2 of the present invention. DETAILED DESCRIPTION

[0013] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0014] Example 1 The present invention provides a blasting fragmentation prediction method applicable to mines with complex terrain, comprising the following steps: S1. Acquire images of the blast pile based on drone oblique photography technology; S2. Using the burst pile image obtained in step S1, construct a three-dimensional burst pile model; S3. Intercept local images of the implosion pile in the model to identify the surface fragmentation of the implosion pile, generate a fragmentation distribution curve, and output data such as average fragmentation and classification percentage; S4. Determine the average deviation rate between the overall and surface fragmentation of the implosion pile through experiments. Based on the deviation rate, correct the surface fragmentation of the local implosion pile in the model and predict the local overall fragmentation. Use manual screening to systematically verify the predicted results.

[0015] Specifically, the specific method of step S4 is as follows: randomly select five groups of blast pile samples in the same blasting area, perform systematic sampling on each blast pile, and obtain five groups of overall blast pile average fragmentation data; synchronously, use a mobile phone to take photos of these five groups of experiments to collect blast pile images, and establish blast pile models for each of them; then use the S3 technology to obtain different surface average fragmentation data for these five groups of blast piles, and obtain the deviation rate between the overall and surface average fragmentation; Based on the average deviation rate determined by the experiment, the surface fragmentation of the intercepted local explosion pile is corrected, and then the local corrected overall fragmentation is predicted, and the predicted results are systematically verified by manual screening method.

[0016] Specifically, the method for systematic sampling of each blast pile is as follows: systematic sampling of each blast pile: at the three longitudinal positions of the front, middle and back of the blast pile, each position is divided into three horizontal measuring points of left, middle and right, and at the same time, rock samples are collected at the upper, middle and lower levels at each measuring point.

[0017] Specifically, the specific method of step S3 is as follows: select a local image of the blast pile for interception; import the intercepted blast pile image into ImageJ for preprocessing to highlight the rock block boundaries; then import the processed image into WipFrag, and extract the rock block boundaries through the edge detection algorithm; then use the watershed algorithm to segment the adjacent rock blocks; after the segmentation is completed, the software automatically calculates the equivalent particle size, aspect ratio and other parameters of each rock block; then convert the actual reference object pixel size in the calibration image into physical size to complete the scale conversion; finally, generate the block size distribution curve, and output the average block size and classification percentage.

[0018] Specifically, the method of step S2 includes: generating three-dimensional point cloud data from the explosive pile image based on motion recovery structure and multi-view stereo matching technology; gridding the three-dimensional point cloud data, constructing a triangular mesh and generating a three-dimensional explosive pile model; and evaluating the surveying and mapping accuracy of the three-dimensional explosive pile model through ground control point coordinate error analysis.

[0019] Specifically, the method for generating 3D point clouds is as follows: import the blast pile images collected by the drone into the ContextCapture software for aerial triangulation, and then use aerial triangulation encryption to improve the accuracy. However, before encryption, at least 3 control points must be arranged on site before the aerial survey and distributed throughout the aerial photography area. The longitude, latitude and elevation data of each control point are determined by GPS, and then imported into the ContextCapture software as ground image control points. The aerial film where each control point is located is accurately located in the image library, and each control point must be marked on no less than 4 different images. After completion, the aerial triangulation calculation is performed to make the final point cloud The coordinate system of the final model is unified with the geographic coordinate system required for engineering research; during the aerial triangulation analysis process, the software will automatically analyze and extract a large number of similar feature points in each aerial survey image, and then match each similar feature point, analyze the spatial position of the camera and the tilt angle of the camera when taking each aerial photo, and thus determine the spatial position relationship between each image; finally generate the camera pose and sparse point cloud; MVS determines the geometric relationship between images based on the generated camera pose, improves matching efficiency, and uses the sparse point cloud as an initial reference to constrain the depth estimation range, verify and optimize the reconstruction results, and the combination of the two finally generates a high-precision three-dimensional point cloud.

[0020] Specifically, the specific method of gridding the three-dimensional point cloud data, constructing a triangular mesh and generating a three-dimensional explosion pile model is as follows: using the Delaunay triangulation algorithm on the high-precision point cloud, selecting three non-collinear points from the point cloud as initial triangle vertices, adding the remaining points in sequence and adjusting the triangle connection relationship according to the Delaunay criterion, and finally forming an optimized TIN mesh; and based on this mesh, defining the topological structure of the model, performing splicing, stitching, denoising and smoothing processing to construct a three-dimensional model without texture information; and then extracting the high-pixel texture of the corresponding position from the image data according to the spatial position of the three-dimensional model for coloring, and finally outputting a three-dimensional model with realistic texture and bright color.

[0021] Example 2 Based on the method of Example 1, the present invention is described by taking the blasting of the Moergou open-pit quarry (located in the 167° direction of Yibin City, about 36.5 km away in a straight line, and mainly composed of limestone) as an example.

[0022] It should be understood that the specific implementation described herein is only used to explain the present invention and is not intended to limit the present invention. The method of this embodiment mainly includes the following contents: S1. Obtaining images of the blast pile based on drone oblique photography technology The UAV flight parameters were designed according to relevant specifications, and the explosion pile image data were collected in five different flight postures.

[0023] Specifically, the design relative altitude of the drone is the design elevation relative to the average ground elevation reference plane within the aerial photography area. According to the "Low-altitude Digital Aerial Photography Specification" (CH / Z3005-2010), the calculation formula for relative altitude is: Where, is the relative altitude, m; is the focal length of the drone camera, mm; is the ground resolution, cm; a is the pixel size, mm.

[0024] According to the national "3D Geographic Information Model Data Product Specifications", the 1:500 (field survey) scale mapping accuracy, GSD should be less than or equal to 5 cm. The maximum altitude of the 1:500 mapping scale can be obtained through the relative altitude formula.

[0025] In order to improve the modeling accuracy and data processing efficiency, it is necessary to increase the overlap of aerial survey images. The design heading overlap and lateral overlap rate can be set to 85%.

[0026] The calculation formulas for lateral overlap and heading overlap are as follows: Where c y is the lateral overlap, c x is the heading overlap, C y is the length of the overlapping parts of adjacent routes, C x is the length of the heading overlap, L x is the film length, L y is the image width.

[0027] Excessive speed of the drone will affect the flight stability and flight attitude of the drone, which may easily lead to poor clarity of aerial survey images and blurry images of ground objects. The maximum speed calculation formula is as follows: Where V max is the maximum speed, m / s; is the image motion compensation accuracy; GSD is the ground resolution, in cm; t is the camera exposure time, in seconds.

[0028] Based on experience, the image motion compensation accuracy requirement for high-resolution detailed inspection cameras or high-precision mapping cameras is better than 0.5%. To reduce the impact of motion on resolution, space cameras typically use a shorter exposure time, such as 1 / 200 s. The maximum speed is ultimately calculated based on actual site conditions and regulatory requirements.

[0029] Compared with vertical photogrammetry, oblique photography by drone can collect richer textures, reduce the loss of texture collection due to occlusion, and also make up for the shortcomings of orthophotometry. The drone is designed to take pictures in five postures. The first line camera shoots vertically downwards, and the second to fifth line cameras tilt at 45 degrees to shoot the front, back, left, and right directions of the explosive pile respectively. The original image of the local explosive pile is as follows: Figure 1 shown.

[0030] S2. Construct a 3D implosion model and evaluate the mapping accuracy Generating three-dimensional point cloud data from the explosion pile image based on structure from motion (SFM) and multi-view stereo matching (MVS) technology; Performing grid processing on the three-dimensional point cloud data, constructing a triangular grid and generating a three-dimensional explosion pile model; The surveying accuracy of the three-dimensional explosive pile model is evaluated by ground control point coordinate error analysis.

[0031] Specifically, each drone-collected aerial image data (S1) is imported into ContextCapture (10.20) for aerial triangulation. Aerial triangulation is then used to improve accuracy. However, prior to triangulation, at least three control points must be established on-site and distributed throughout the aerial survey area. GPS is used to determine the latitude, longitude, and elevation of each control point. These data are then imported into ContextCapture as ground control points. The aerial image where each control point is located is precisely located in the image library. Each control point must be marked on at least four different images. Once completed, aerial triangulation calculations are performed to align the coordinate system of the resulting model with the geographic coordinate system required for engineering research.

[0032] During the aerial triangulation (SFM) analysis process, the software automatically analyzes and extracts a large number of similar feature points from each aerial survey image. It then matches each similar feature point, analyzing the spatial position and tilt angle of the camera when each aerial photo was taken, thereby determining the spatial relationship between the images. However, close attention should be paid to the software's extraction and matching of image feature points. Abnormal data, such as insufficient feature point extraction (<500 points per image), low matching rate (<50 pairs of matching points in adjacent images), control point backprojection error >1.5 pixels, point cloud fragmentation, or coordinate system offset, will cause aerial triangulation failure and the inability to generate a correct dense point cloud for subsequent MVS. At this point, the cause must be analyzed and investigated in conjunction with the aerial triangulation report. For example, if feature points are insufficient, existing image data should be analyzed to locate defective areas and plan the shooting path or shooting point locations. Image overlap should be increased. Low matching rates require removing blurred and occluded images, or adjusting feature extraction parameters to use the SIFT algorithm. Large control point errors require re-measurement of control point coordinates, or correction of marker positions in the photo. Point cloud breaks and coordinate system offsets require manual addition of tie points and verification of control point coordinate system input. Finally, a camera pose and sparse point cloud are generated. MVS uses the generated camera pose to determine the geometric relationship between images, significantly improving matching efficiency. The sparse point cloud serves as an initial reference, constraining the depth estimation range and validating and optimizing the reconstruction results. The combination of the two ultimately generates a high-precision 3D point cloud.

[0033] These high-precision point clouds are triangulated using the Delaunay triangulation algorithm. Three non-collinear points are selected as the initial triangle vertices. Remaining points are then added sequentially, and the triangle connectivity is adjusted according to the Delaunay criterion, ultimately forming an optimized TIN mesh. Based on this mesh, the model's topology is defined, and stitching, stitching, denoising, and smoothing are performed to construct a texture-free 3D model. High-resolution textures at corresponding locations in the image data are then extracted based on the spatial position of the 3D model and colored. Finally, a 3D model with realistic textures and vibrant colors is output. After model construction is complete, the 3D model is inspected for integrity and texture quality. Any holes or blurred textures require additional image reconstruction or repair using mesh editing tools. Color differences and seams require adjustment of texture blending parameters or manual editing of the texture map.

[0034] Using the mean square error formula, the model's planar and elevation errors are calculated, and the overall model error accuracy is evaluated to determine if it meets the modeling accuracy requirements. The Viewer in ContextCapture software provides model measurement functions such as latitude, longitude, elevation, and distance. The mean square error is calculated by comparing the coordinates of the actual control points imported during aerotriangulation with the coordinates of the corresponding points measured in the 3D model. Multiple measurements are taken and the average is taken to calculate the model's planar and elevation errors, and the overall model error accuracy is evaluated as follows: Where m x is the mean error in the x direction, m; m y is the mean error in the y direction, m; m is the mean error in the plane, m; m z is the error value in elevation, m.

[0035] According to the national "3D Geographic Information Model Data Product Specification", the 1:500 (field survey) scale mapping accuracy requires that the plane error and elevation error are less than or equal to 0.3 and 0.5 m respectively. If the model does not meet the accuracy requirements, it needs to be re-modeled. The modeling example is Figure 2 shown.

[0036] S3. Intercept the local explosion pile image in the model to identify the surface fragmentation of the explosion pile The intercepted local explosion pile area of ​​the model must be greater than 15% of the total explosion pile area; Enhance and filter the model blast pile image to highlight the rock block boundary; Use edge detection algorithm to identify rock contours; Use the watershed algorithm to segment adjacent rock blocks; Identify the segmented rock blocks and calibrate the reference objects in the image to complete the conversion to actual size; Generate a block size distribution curve and output data such as average block size and classification percentage.

[0037] Specifically, the built blast pile 3D model is enlarged to an appropriate scale, the local blast pile model is adjusted to be parallel to the computer screen, the rock size ratios of the upper, lower, left and right sides of the image are ensured to be consistent, and the local image of the blast pile is selected (which must be greater than 15% of the total blast pile area) for interception; the intercepted blast pile image is imported into ImageJ for preprocessing. First, Gaussian filtering is used to reduce the noise of the image, which can retain the edge information while smoothing the noise to avoid subsequent edge detection distortion. Adjusting the contrast to enhance the grayscale difference between the rock and the background can improve the rock recognition rate. And enable the shadow removal function to suppress the local darkness or brightness caused by strong light exposure, and enhance the recognition of the prominent rock boundaries in the image. The effects before and after processing are as follows Figure 3 shown.

[0038] Import the processed image into WipFrag (4.0), use the edge detection algorithm (strong noise resistance and good edge continuity) to extract the rock boundary, and set the appropriate threshold to distinguish the real edge from the texture noise; the result is as follows Figure 4 shown.

[0039] Then, based on the edge detection results, the watershed algorithm is used to effectively separate the contacting rocks and avoid sticking rocks. The sensitivity parameter is adjusted to control the segmentation fineness, balancing the detection rate of small rocks and the risk of over-segmentation. The manual correction tool is used to handle abnormal areas and resolve areas that the algorithm misjudged. The results are as follows: Figure 5 shown.

[0040] After the segmentation is completed, the software automatically calculates the size of each rock block, and then converts the actual reference pixel size in the calibration image into physical size to complete the scale conversion. Finally, the output curves such as blockiness and graded percentage distribution are generated. The results are as follows Figure 6 shown.

[0041] S4. Based on the average deviation rate between the overall and surface fragmentation, the surface fragmentation of the local explosion pile (S3) is corrected, and then the local overall fragmentation is predicted, and the prediction result is verified.

[0042] The average deviation rate between the overall blast pile and the surface blast pile fragmentation is obtained through experiments; Based on the deviation rate, the surface fragmentation of the local implosion pile in the model is corrected (S3) to predict the local overall fragmentation; The predicted results were systematically verified using manual screening method.

[0043] Specifically, in order to evaluate the overall fragmentation distribution of the local blast pile in the model (since the internal ore rock of the blast pile is blocked by the surface ore rock, S3 can only obtain the fragmentation of the surface of the blast pile), a blasting test was selected for analysis, and five groups of blast pile samples were randomly selected in the same blasting area. Systematic sampling was performed on each blast pile: at the three longitudinal positions of the front, middle and back of the blast pile, each position was divided into three horizontal measuring points: left, middle and right. At the same time, rock samples were collected at the upper, middle and lower levels at each measuring point to ensure that the entire blast pile range was covered and the sampling bias was reduced. A total of 5 groups of overall blast pile average fragmentation data were obtained. Simultaneously, a mobile phone camera can be used to capture images of the five experimental burst piles using the S1 at five different shooting angles, and burst pile models can be established for each of them (the reason for re-modeling is that the burst pile images obtained after modeling are three-dimensional, and each block has a height difference, so the analyzed block size will be more accurate). Since the burst piles in the experiment are relatively small, based on mobile phone photography (human photography is at a low height), drones are no longer used, and direct modeling is sufficient. No control points are set and no accuracy evaluation is performed. Then, the S3 technology is used to obtain different surface average block size data for these five burst piles, and the deviation rate between the overall and surface average block size can be obtained, as listed in Table 1.

[0044] Table 1 Overall and surface average blockiness and deviation rate of the five groups of experiments Test number Overall average particle size / cm Average surface roughness / cm Deviation rate / % #1 15.82 13.81 12.71 #2 23.49 20.48 12.82 #3 29.53 25.67 13.07 #4 35.74 31.03 13.18 #5 41.85 36.18 13.55 As shown in Table 1, the average deviation rate of the five groups of tests is 13.07%. Based on the average deviation rate determined by the test, the surface fragmentation of the local explosive pile intercepted in the entire model was corrected (each particle size value increased by 13.07%), and then the local overall fragmentation (after correction) was predicted. The predicted results were systematically verified using the manual screening method. The data results are as follows: Figure 7 shown.

[0045] from Figure 8 It can be seen that the maximum deviation between the corrected overall particle size value and the sieving method particle size value is 3.06 cm, and the average deviation is 1.29 cm. The maximum deviation between the surface particle size value and the sieving method particle size value before correction is 12.98 cm, and the average deviation is 7.52 cm. The corrected overall fragmentation distribution is closer to the actual fragmentation distribution, indicating that the overall fragmentation calculated after correction is more accurate and can be used for subsequent fragmentation distribution statistical analysis.

[0046] S5. Use the large block rate as the evaluation index and optimize the blasting parameters by comparing with historical data.

[0047] Chunk size is defined by the on-site machinery capacity and primary crusher inlet dimensions.

[0048] Establish a blasting parameter-large block rate historical database, and optimize the blasting parameters by comparing the large block rates under different blasting parameters.

[0049] Specifically, the mining area adopts a step blasting scheme, mainly using mixed ammonium nitrate oil explosives, emulsion explosives as the detonating explosives, and a continuous charging structure. The on-site step height is 17~18 m, the slope angle is 75°, the drilling angle is 75 degrees, the hole depth is 18~19 m, and the over-depth is 1~2 m. The explosive consumption is 0.55 kg / m 3 The hole diameter was 90 mm, the minimum resistance line was 3-3.5 m, and the hole grid parameters were 4 m × 3.5 m. A digital electronic detonator millisecond differential hole-by-hole initiation network was used. The test was conducted by adjusting five different inter-hole delays: 15 ms, 17 ms, 20 ms, 24 ms, and 29 ms, respectively, with an inter-row delay of 55 ms.

[0050] The overall particle size distribution of the blast pile after the correction of these five blasting tests is statistically analyzed as follows: Figure 8 shown.

[0051] Considering that in the actual production process, the size of the blasted material must meet the allowable size of the coarse crusher inlet and the volume size of the loading (excavator) and transportation equipment (mining truck), the main determination methods are as follows: According to the requirements of excavator bucket capacity E: Requirements for the mining truck volume V: Where: a is the longest side dimension of the rock mass, m; E is the bucket capacity of the excavator, m 3 ; V is the bucket capacity of the mining card, m 3 .

[0052] The bulk ore size is ultimately defined based on the volumetric dimensions of the actual mechanical equipment on site and combined with the maximum ore size at the coarse crusher inlet.

[0053] Based on the excavator and truck models primarily used for mining and transportation at the on-site mine, we calculated using the volumetric dimension formula that the maximum side length of rock that can be loaded by the excavator is 85 cm, and the maximum side length that can be accommodated by the truck is 93 cm. To meet the allowable lumpiness at the on-site coarse crusher inlet, rock with a side length of 80 cm was ultimately defined as large. The large lump percentages for the five tests are listed in Table 2.

[0054] Table 2 Large block rate of ore and rock Serial number Large block rate (%) #1 4.8 #2 3.67 #3 2.77 #4 3.85 #5 5.91 In open-pit bench blasting operations, the large-block fraction of ore is typically required to be controlled within 5% to ensure smooth subsequent mining and processing. However, in this Test #5, the large-block fraction of the overall blast pile slightly exceeded the 5% standard. Analysis determined that this phenomenon was primarily due to excessive inter-hole delay settings, resulting in suboptimal blasting results. To this end, we have fully collected the relevant data from this test and incorporated it into our blasting database. In subsequent tests, we will adjust blasting parameters and conduct detailed comparative analysis with the data from this test. By systematically evaluating the relationship between different blasting parameters and the large-block fraction, we will achieve dynamic optimization of blasting parameters. This approach aims to strictly control the large-block fraction to a lower level, thereby effectively reducing the cost of secondary crushing operations. Ultimately, by continuously optimizing blasting processes, we will significantly improve the economic benefits of blasting operations and provide a strong guarantee for efficient and economical mining operations.

[0055] Each embodiment in this specification is described in a related manner. The same or similar parts between the embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences from other embodiments. The above description is only a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are included in the scope of protection of the present invention.

Claims

1. A method for predicting blasting fragmentation suitable for mines with complex terrain, characterized by: The following steps are involved: S1. Acquire images of the blast pile based on drone oblique photography technology; S2. Using the burst pile image obtained in step S1, construct a three-dimensional burst pile model; S3. Intercept local images of the implosion pile in the model to identify the surface fragmentation of the implosion pile, generate a fragmentation distribution curve, and output data such as average fragmentation and classification percentage; S4. Determine the average deviation rate between the overall and surface fragmentation of the implosion pile through experiments. Based on the deviation rate, correct the surface fragmentation of the local implosion pile in the model and predict the local overall fragmentation. Use manual screening to systematically verify the predicted results.

2. The blasting fragmentation prediction method applicable to complex terrain mines according to claim 1, characterized in that: The specific method of step S4 is as follows: Five groups of blast pile samples were randomly selected within the same blasting area. Systematic sampling was performed on each blast pile to obtain five sets of overall blast pile average fragmentation data. Simultaneously, mobile phone photos were taken of these five groups of test blast piles to collect blast pile images, and blast pile models were established for each of them. Using S3 technology, different surface average fragmentation data were obtained for these five groups of blast piles, and the deviation rate between the overall and surface average fragmentation was determined. Based on the average deviation rate determined by the experiment, the surface fragmentation of the intercepted local explosion pile is corrected, and then the local corrected overall fragmentation is predicted, and the predicted results are systematically verified by manual screening method.

3. The blasting fragmentation prediction method applicable to complex terrain mines according to claim 2, characterized in that: The method for systematic sampling of each blast pile is as follows: systematic sampling of each blast pile: at the three longitudinal positions of the front, middle and back of the blast pile, each position is divided into three horizontal measuring points of left, middle and right, and at the same time, rock samples are collected at the upper, middle and lower levels at each measuring point.

4. The blasting fragmentation prediction method applicable to complex terrain mines according to claim 1, characterized in that: The specific method of step S3 is as follows: select a local image of the blast pile for interception; import the intercepted blast pile image into ImageJ for preprocessing to highlight the boundaries of the rock blocks; then import the processed image into WipFrag and extract the rock block boundaries through the edge detection algorithm; then use the watershed algorithm to segment the adjacent rock blocks; after the segmentation is completed, the software automatically calculates the size of each rock block, and then converts the actual reference object pixel size in the calibration image into physical size to complete the scale conversion; finally, generate a curve of block size and grade percentage distribution.

5. The blasting fragmentation prediction method applicable to complex terrain mines according to claim 1, characterized in that: The method of step S2 includes: generating three-dimensional point cloud data from the implosion pile image based on motion recovery structure and multi-view stereo matching technology; gridding the three-dimensional point cloud data, constructing a triangular mesh and generating a three-dimensional implosion pile model; and evaluating the surveying and mapping accuracy of the three-dimensional implosion pile model through ground control point coordinate error analysis.

6. The blasting fragmentation prediction method applicable to complex terrain mines according to claim 5, characterized in that: The method for generating 3D point cloud is: Import the blast pile images collected by the drone into the ContextCapture software for aerial triangulation. On this basis, aerial triangulation is used to improve the accuracy. However, before encryption, at least three control points must be arranged on site before the aerial survey and distributed throughout the aerial photography area. The latitude, longitude and elevation data of each control point are determined by GPS and imported into the ContextCapture software as ground image control points. The aerial photographs where each control point is located are accurately located in the image library. Each control point must be marked on no less than four different images. After completion, the coordinate system of the final model is unified with the geographic coordinate system required for engineering research through aerial triangulation calculation. During the aerial triangulation analysis process, the software automatically analyzes and extracts a large number of similar feature points in each aerial survey image, then matches each similar feature point, analyzes the spatial position of the camera when each aerial photo was taken and the tilt angle of the camera, thereby determining the spatial position relationship between each image; finally, the camera pose and sparse point cloud are generated. MVS determines the geometric relationship between the generated camera poses and improves matching efficiency. The sparse point cloud is used as the initial reference to constrain the depth estimation range, verify and optimize the reconstruction results. The combination of the two ultimately generates a high-precision 3D point cloud.

7. The blasting fragmentation prediction method applicable to complex terrain mines according to claim 5, characterized in that: The specific method of gridding the three-dimensional point cloud data, constructing a triangular mesh and generating a three-dimensional explosion pile model is as follows: using the Delaunay triangulation algorithm on the high-precision point cloud, selecting three non-collinear points from the point cloud as initial triangle vertices, adding the remaining points in sequence and adjusting the triangle connection relationship according to the Delaunay criterion, and finally forming an optimized TIN mesh; and based on this mesh, defining the topological structure of the model, performing splicing, stitching, denoising and smoothing processing to construct a three-dimensional model without texture information; and then extracting high-pixel textures at corresponding positions from the image data according to the spatial position of the three-dimensional model for coloring, and finally outputting a three-dimensional model with realistic texture and vivid colors.

8. The blasting fragmentation prediction method applicable to complex terrain mines according to claim 4, characterized in that: When selecting a local image of the explosion pile for interception, the area of ​​the intercepted image should be larger than 15% of the explosion pile area.

Images