Rock stratum blast hole depth and spacing optimization method based on deep learning

Abstract

This invention provides a deep learning-based method for optimizing borehole depth and spacing in rock blasting. The method includes: acquiring historical blasting datasets, rock geological survey data, and real-time drilling response data of the drilling rig during drilling operations in the target blasting area; performing spatiotemporal alignment and cleaning on the above data to construct a multi-source heterogeneous blasting feature database; constructing a deep learning-based rock charge coupling effect analysis model, which includes a drilling feature extraction module, a medium impedance matching correction module, and a blasting parameter optimization decision module; and training the model using the multi-source heterogeneous blasting feature database. This invention inversely generates three-dimensional fracture distribution characteristics, calculates the multi-medium acoustic impedance matching coefficients between the PE pipe, the borehole filling medium, and the borehole wall rock, and generates an effective energy transfer efficiency matrix. Finally, the blasting parameter optimization decision module outputs the target borehole depth and target borehole spacing values ​​that maximize the half-hole ratio after blasting.

Description

[0001] This invention relates to the field of open-pit mining and digital construction technology, and in particular to a method for optimizing the depth and spacing of blasting holes in rock strata based on deep learning. Background Technology

[0002] In bench blasting of fractured rock strata in open-pit mines, PE pipes are widely used in conjunction with explosives for decoupled charging to obtain a smooth slope and strictly control the damage to the rock mass caused by blasting vibrations. Hole depth and spacing, as core geometric parameters determining the spatial distribution of blasting energy, directly determine the success or failure and economic benefits of the blasting operation. Existing blasting parameter designs mainly rely on the experience and analogy of engineers or on general empirical formulas from macroscopic geological survey reports. This static design model typically assumes that the rock mass properties within the same blasting area are homogeneous or simply gradually changing. However, in actual mining geological environments, the degree of fracture development, hardness coefficient, and structural plane orientation of the rock mass exhibit strong spatial heterogeneity and randomness. When drilling rigs operate at different hole positions, a single and fixed hole network parameter is difficult to adapt to the actual physical and mechanical state of the rock mass at the hole position. Surface survey data alone often cannot accurately capture hidden weak interlayers or localized hard rock zones in deep rock strata, leading to excessive dissipation or concentrated damage of explosive energy during transmission.

[0003] Furthermore, in special cases where PE pipes are used for uncoupled explosive loading, the blast shock wave propagates from the explosive to the rock through multiple transmissions and reflections via the pipe wall and annular gap. If the acoustic impedance matching relationship between the explosive structure and the rock mass is ignored, simply adjusting parameters based on the rock strength coefficient often fails to accurately assess energy transfer efficiency. Existing design methods lack refined quantitative analysis of this cross-medium coupling effect, making it impossible to find the optimal balance between protecting the borehole wall integrity and ensuring sufficient rock fragmentation in the construction parameters. This easily leads to engineering quality problems such as residual material at the bottom, over-excavation, insufficient half-hole ratio, or excessively high large-block ratio, increasing the difficulty of subsequent loading and transportation and slope maintenance costs. Summary of the Invention

[0004] The purpose of this invention is to provide a method for optimizing the depth and spacing of blasting holes in rock strata based on deep learning, so as to solve the problems pointed out in the background art.

[0005] This invention provides a deep learning-based method for optimizing the depth and spacing of blasting holes in rock strata. The method is applied to bench blasting operations in fractured rock strata of open-pit mines. The operations utilize drilling rigs for drilling and PE pipes for loading explosives. The method includes the following steps: Acquire historical blasting datasets, rock geological survey data, and real-time drilling response data of the drilling rig during drilling operations for the target blasting area; Spatiotemporal alignment and cleaning were performed on the historical blasting dataset, the rock mass geological survey data, and the real-time drilling response data to construct a multi-source heterogeneous blasting feature database. A deep learning-based rock mass charging coupling effect analysis model is constructed; the rock mass charging coupling effect analysis model is equipped with a drilling feature extraction module, a medium impedance matching correction module, and a blasting parameter optimization decision module; The rock mass charge coupling effect analysis model was trained using the multi-source heterogeneous blasting feature database; The method further includes: The drilling feature extraction module generates a three-dimensional fracture distribution feature vector that reflects the roughness of the borehole wall and the degree of fracture development based on the real-time drilling response data. The medium impedance matching correction module calculates the multi-medium acoustic impedance matching coefficient between the PE pipe, the filling medium inside the hole, and the rock on the hole wall based on the preset physical parameters of the PE pipe. The dielectric impedance matching correction module uses the multi-medium acoustic impedance matching coefficient to weight and correct the three-dimensional crack distribution feature vector to generate an effective energy transfer efficiency matrix. The blasting parameter optimization decision module maps the effective energy transfer efficiency matrix to the blasting parameter space and outputs the target hole depth and target hole spacing values ​​that maximize the half-hole ratio after blasting.

[0006] Optionally, obtaining the real-time drilling response data specifically includes: The sensor array installed on the drilling rig synchronously collects the rotational torque of the drill rod, axial drilling pressure, instantaneous drilling speed, and slag discharge air pressure at a preset sampling frequency. Using Teale's mechanical energy equation, the collected slewing torque, axial drilling pressure, and instantaneous drilling speed are converted into a sequence of mechanical work required to break a unit volume of rock. The mechanical energy sequence is denoised and smoothed to obtain a drilling mechanical specific energy curve that can characterize the change of rock hardness along the hole depth direction.

[0007] Optionally, when constructing the multi-source heterogeneous blasting feature database, the following data alignment operation is performed: The coordinates and elevation data of each borehole were obtained using the Global Navigation Satellite System. Establish a three-dimensional spatial coordinate system for the blasting area; The rock strata occurrence information in the rock mass geological survey data, the depth information in the real-time drilling response data, and the blasting effect information in the historical blasting dataset are uniformly mapped to the three-dimensional spatial coordinate system to form structured tensor data with spatial topological relationships.

[0008] Optionally, the method further includes an engineering safety verification step before outputting the target hole depth value and the target hole distance value: Using Sadovsky's empirical formula, based on the single-hole charge amount corresponding to the currently calculated target hole spacing value, the maximum surface vibration velocity induced by blasting is predicted. Determine whether the maximum ground vibration velocity exceeds the safety allowable standard of the protected building or structure; If the safety allowable standard is exceeded, a constraint penalty mechanism is triggered to forcibly reduce the target hole spacing value or adjust the micro-delay detonation time until the safety verification requirements are met.

[0009] Optionally, the method further includes a digital construction instruction generation step: The target hole depth and target hole spacing values ​​are converted into a visualized electronic hole pattern; Mark the geographical coordinates, design depth, and corresponding recommended length of PE pipe for each hole location on the electronic hole layout map; The electronic hole layout diagram is transmitted to the on-board terminal screen of the drilling rig via a wireless communication network to assist operators in precise hole layout and positioning.

[0010] Optionally, the specific logic for the dielectric impedance matching correction module to calculate the multi-dielectric acoustic impedance matching coefficient is as follows: Obtain the density and longitudinal wave velocity of the PE pipe material, and calculate the characteristic acoustic impedance of the PE pipe; Based on the three-dimensional fracture distribution feature vector, the dynamic acoustic impedance of the pore wall rock is estimated; Based on the ratio of the characteristic acoustic impedance of the PE pipe to the dynamic acoustic impedance of the rock in the borehole wall, and combined with the properties of the gap filler between the PE pipe and the borehole wall, the energy transmittance is calculated using the stress wave transmission theory formula. The energy transmittance is normalized and used as the multi-medium acoustic impedance matching coefficient, which is used to characterize the effective proportion of explosion energy passing through the PE pipe charge structure and entering the rock mass.

[0011] Optionally, the rock mass charge coupling effect analysis model adopts a two-stream attention network architecture: The drilling feature extraction module includes a long short-term memory network branch, which is used to extract the temporal dependency features of the real-time drilling response data in the depth direction. The medium impedance matching correction module includes a graph convolutional neural network branch, which is used to aggregate spatial geological correlation features between adjacent boreholes. The dual-stream attention network architecture introduces an attention mechanism in the feature fusion layer, dynamically adjusting the weight values ​​of each feature channel based on the multi-medium acoustic impedance matching coefficient.

[0012] Optionally, the training process of the rock mass charging coupling effect analysis model adopts a composite loss function that includes a half-porosity constraint; The composite loss function consists of a hole mesh parameter prediction error term, a half-porosity prediction error term, and a rock blockiness uniformity penalty term. During training, when the model predicts a half-porosity lower than a preset quality threshold, the gradient weight of the half-porosity prediction error term in the composite loss function is automatically increased, forcing the model parameters to converge in a direction that is conducive to protecting the integrity of the pore walls.

[0013] Optionally, the method further includes a closed-loop feedback iteration step based on 3D scanning: After the blasting operation is completed, point cloud data of the blasted section is obtained using a 3D laser scanner; The actual blasting half-hole ratio and rock fragmentation size distribution are calculated using a point cloud segmentation algorithm. Calculate the residual between the actual blasting half-hole ratio and the model prediction value; The residual is fed back as a reward signal for reinforcement learning to the rock mass charge coupling effect analysis model, and the weights of the blasting parameter optimization decision module are updated online incrementally.

[0014] Optionally, the blasting parameter optimization decision module is also used to fine-tune parameters for special geological conditions: When the three-dimensional fracture distribution feature vector indicates the presence of a weak interlayer at the bottom of the hole, an over-depth margin is automatically added to the target hole depth value to overcome the absorption effect of the weak interlayer on the explosion energy. When the three-dimensional fracture distribution feature vector indicates that the fractures in the orifice area are extremely developed, an automatic instruction to adjust the packing length is issued, and the target hole spacing value is reduced accordingly to reduce the risk of blasting gas prematurely escaping from the orifice and causing a blasting accident.

[0015] The present invention has achieved the following beneficial effects: This invention constructs a deep learning-based rock mass charge coupling effect analysis model, incorporating real-time drilling response data into the blasting parameter design process. It utilizes a drilling feature extraction module to accurately invert the three-dimensional fracture distribution within the rock mass, transforming blasting design from macroscopic geological inference to in-situ measured perception, thus achieving precise response to rock mass inhomogeneity. This invention integrates physical mechanisms into the deep learning network, calculating the energy transfer efficiency between the PE pipe charge structure and the borehole wall rock through a medium impedance matching correction module. This efficiency is then used to weight and correct fracture distribution characteristics, forcing the model to consider both rock fragility and energy transfer effectiveness in decision-making, overcoming the technical limitations of traditional methods in quantifying and evaluating multi-medium coupling effects. Furthermore, this invention comprehensively considers the goal of maximizing half-pore ratio and engineering safety verification mechanisms. It can automatically optimize the target borehole depth and spacing based on local lithological changes, and perform targeted parameter fine-tuning when encountering weak interlayers or fractured zones. This effectively improves the uniformity of slope half-pore ratio and rock fragmentation size while ensuring the vibration safety of protected structures, thereby reducing the overall cost of mining.

[0016] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0017] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating the method for optimizing the depth and spacing of rock blasting holes based on deep learning in an embodiment of the present invention. Figure 2 This is a schematic diagram of the composition structure of the rock blasting hole depth and spacing optimization system based on deep learning, according to an embodiment of the present invention. Detailed Implementation

[0019] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0020] This invention provides a method for optimizing the depth and spacing of blasting holes in rock strata based on deep learning. The method includes the following steps: Acquire historical blasting datasets, rock geological survey data, and real-time drilling response data of the drilling rig during drilling operations for the target blasting area; Spatiotemporal alignment and cleaning were performed on the historical blasting dataset, the rock mass geological survey data, and the real-time drilling response data to construct a multi-source heterogeneous blasting feature database. A deep learning-based rock mass charging coupling effect analysis model is constructed; the rock mass charging coupling effect analysis model is equipped with a drilling feature extraction module, a medium impedance matching correction module, and a blasting parameter optimization decision module; The rock mass charge coupling effect analysis model was trained using the multi-source heterogeneous blasting feature database.

[0021] The method of the present invention is as follows Figure 1 As shown, this method is applied to bench blasting in fractured rock strata of open-pit mines. The construction utilizes drilling rigs for drilling operations and employs PE pipes in conjunction with explosives for loading operations. The project implementation relies on a digital construction system for open-pit mines, such as... Figure 2 As shown, the system includes: a front-end perception and execution layer, a middle data transmission layer, and a back-end intelligent decision-making layer. A heavy-duty tracked hydraulic down-the-hole (DTH) drill rig serves as the data acquisition terminal and construction execution mechanism; drilling-while-monitoring sensor arrays are configured on key moving parts of the drill rig. High-precision pressure transmitters are installed in the hydraulic cylinder pipelines of the drill frame's propulsion beam for real-time monitoring of propulsion and return oil pressures; an industrial-grade laser ranging module is installed on the side of the slide. Furthermore, vortex flow meters and piezoresistive pressure sensors are installed on the main air outlet pipeline of the air compressor and the air supply connector of the drill pipe, respectively. A dual-antenna carrier phase differential global navigation satellite system receiver is installed on the top of the drill rig mast. An industrial-grade edge computing terminal is mounted on the drill rig vehicle. A ground server cluster is used to carry out the training and inference tasks of the rock mass charge coupling effect analysis model.

[0022] The multi-source heterogeneous blasting feature database includes: drilling response data, geological survey data, and historical blasting data.

[0023] In the vehicle-mounted edge computing terminal, the received slewing torque T, axial drill pressure F, drill pipe rotation speed N, and instantaneous drilling speed u are subjected to Kalman filtering. The improved Teale equation is used to calculate the mechanical efficiency while drilling (MSE) point by point. The specific formula is: MSE equals the mechanical efficiency coefficient multiplied by (axial drill pressure divided by the borehole cross-sectional area A, plus twice pi multiplied by the slewing torque multiplied by the rotation speed, then divided by the product of the borehole cross-sectional area and the instantaneous drilling speed). The mechanical efficiency coefficient is a correction factor dynamically calibrated based on the hydraulic system characteristic curve of the drilling rig and the real-time oil temperature, typically ranging from 0.85 to 0.95. Simultaneously, the coefficient of variation of MSE in the depth direction and the energy entropy are calculated to capture subtle bedding changes and damage zones within the rock.

[0024] Since drilling data is a one-dimensional time series based on borehole trajectories, while geological data and blasting design are based on three-dimensional volume data, there is a dimensional mismatch between the two. This embodiment uses voxel mapping technology to solve this problem. First, a local three-dimensional Cartesian coordinate system is established for the blasting area. Based on GNSS coordinates and inclinometer data, the drilling MSE curve of each borehole is discretized using spatial analytical geometry algorithms, mapping it to a series of attributed coordinate points in three-dimensional space. Second, using Kriging interpolation or inverse distance weighting, the discrete rock mass geological survey data is mapped to a voxel grid in the same coordinate system, so that each spatial grid is assigned macroscopic geological attributes. Finally, historical blasting effect data is back-projected back to the original borehole location. For example, the half-porosity of a borehole location measured by post-blast scanning is used as a label value for the drilling data sequence of that borehole location. After the above processing, a structured tensor data containing spatial topological relationships is formed. This database not only stores the longitudinal physical characteristics of individual boreholes, but also records the spatial adjacency relationships between boreholes through a graph structure. The specific construction process is as follows: Set the resolution of the 3D voxel mesh in the blasting area to be [value missing]. ( Using the borehole coordinates as the center, a spatial neighborhood is extracted, and the drilling response features are mapped to voxels using three-dimensional ordinary kriging interpolation, generating a dimension of... The input tensor.

[0025] in, Batch Size; The number of voxel grids along the borehole depth direction (e.g., corresponding to a borehole depth of 15 meters). ); and These represent the number of neighboring grid cells extracted in the horizontal direction (in this embodiment, we take...). That is, covering an area of ​​3 meters around the hole). The number of characteristic channels includes four characteristic dimensions: mechanical energy during drilling, slewing torque, slag discharge air pressure, and rock wave impedance.

[0026] The rock mass charge coupling effect analysis model includes: a feature extraction while drilling module, a medium impedance matching correction module, and a blasting parameter optimization decision module. The feature extraction while drilling module uses a bidirectional long short-term memory network (Bi-LSTM) as the backbone network for feature extraction.

[0027] In blasting operations using PE pipes for decoupled explosive charges, the explosive energy does not act directly on the rock, but undergoes a complex cross-medium transfer process: explosive → PE pipe → annular gap (air / water / rock powder) → rock on the borehole wall. According to the theory of elastic wave propagation, when a stress wave is incident perpendicularly at the interface of two different media, its transmission coefficient T and reflection coefficient R depend on the acoustic impedance of the two media. , For density, The impedance matching degree is crucial. If there is a severe impedance mismatch (e.g., between a PE pipe and air), most of the energy will be reflected back into the pipe, resulting in extremely low work efficiency. This is a major cause of uneven foundation and rock fragmentation in traditional blasting. Therefore, the dielectric impedance matching correction module performs the following logical operation: it reads the density of the PE pipe material from a preset material parameter library. and longitudinal wave velocity Calculate the characteristic acoustic impedance of PE pipe Using the feature vectors output by the drilling feature extraction module, the dynamic acoustic impedance of the borehole wall rock distribution along the depth direction is estimated through a pre-trained multilayer perceptron (MLP) subnetwork. Based on the slag discharge air pressure and conductivity in the drilling data, it was determined that the annular gap between the PE pipe and the borehole wall was due to air coupling. ≈0), water coupling ( ≈1.5×10 6 (kg / m²s) or rock powder filling. Based on the one-dimensional stress wave multilayer medium transmission formula, the total energy transmittance from explosive to rock is calculated. For example, for an explosive-PE pipe-water-rock system, the specific calculation steps are as follows: Step A: Gap medium determination. Set the slag discharge air pressure threshold. .

[0028] If real-time wind pressure If the gap medium is determined to be air, its acoustic impedance is... ; If real-time wind pressure The interstitial medium is determined to be either drill cuttings or water, and its acoustic impedance is... Take 30% of the acoustic impedance of the rock in the borehole wall.

[0029] Step B: Energy transmittance calculation. Based on the one-dimensional plane wave perpendicular incidence theory, define any adjacent medium With medium Energy transmission coefficient at the interface for:

[0030] In the formula, and medium respectively and medium The characteristic acoustic impedance.

[0031] Then, the total energy transmittance from the explosive to the rock wall of the borehole is calculated. :

[0032] In the formula, The transmittance coefficient at the interface between the explosive and the PE pipe; The transmission coefficient is the interface between the PE pipe and the annular gap. is the transmission coefficient at the interface between the annular gap and the rock wall of the borehole.

[0033] Step C: Normalization correction. This is to address air coupling issues. To address the problem of vanishing gradients caused by extremely small numerical values, logarithmic normalization is used to generate the final multi-medium acoustic impedance matching coefficients. :

[0034] In the formula, The theoretical maximum transmittance corresponds to the fully coupled operating condition (i.e., assuming...). (Calculated value at time) This represents the theoretical minimum transmittance, corresponding to the air-decoupled condition (i.e.) (Calculated value at time) Represents the natural logarithm operation.

[0035] Effective energy transfer efficiency matrix The generation process uses the calculated matching coefficient λ(z) as weights to apply to the three-dimensional fracture distribution feature vector. Perform element-wise weighting. The physical significance of this operation is that it guides the neural network to detect even well-developed rock fissures in a certain area. High values ​​theoretically make it prone to breakage, but if the impedance matching at that point is extremely poor ( If the value is low (energy cannot be transferred), then this characteristic contributes little to the blasting.

[0036] This invention employs a dual-stream network architecture, specifically a Bi-LSTM branch and a Graph Convolutional Neural Network (GCN). The output features of the Bi-LSTM branch and the GCN are concatenated in a fusion layer and then weighted again using a channel attention mechanism before being input into the subsequent burst parameter optimization decision module. The concatenation formula is as follows:

[0037] In the formula, For the first The node feature matrix of the layer; For the first The node feature matrix of the layer; For the first The learnable weight matrix of the layer; It is a non-linear activation function (Leaky ReLU is used in this embodiment); For the adjacency matrix with added self-loops, where It is the identity matrix; This is the original adjacency matrix, and its elements are... Defined as: if the hole With Kong Euclidean distance Rice, then ,otherwise ; for The degree matrix, whose diagonal elements .

[0038] The rock blasting parameter optimization decision module uses a supervised learning regression network based on the DDPG architecture. The supervised learning regression network includes a policy subnetwork and a value subnetwork; the policy subnetwork uses the Swish activation function (…). The model's output layer is connected to a hyperbolic tangent activation function, constraining the output value within the standardized interval [-1, 1]. The value sub-network uses a parallel input structure, receiving geological state features and action parameters from the policy sub-network, respectively. The inputs are concatenated in an intermediate layer after passing through independent feature extraction layers to obtain a scalar Q-value, which guides the training of the policy sub-network model. A composite loss function is used to guide the training of the value sub-network model. for:

[0039] in, For the error term of the mesh parameter prediction ( The weighting coefficients are used to characterize the strength of the fitting constraint between the model output and historical expert experience data; For the half-porosity prediction error term ( The weight coefficients of the model are used to characterize the gradient penalty weights generated when the predicted half-porosity is lower than the preset quality threshold, forcing the model parameters to converge in a direction that is conducive to protecting the integrity of the hole wall. The penalty term for the uniformity of rock block size ( The weighting coefficients are used to characterize the physical constraint weights for situations where rocks are over-crushed or have an excessively high proportion of large pieces.

[0040] The first part is the error term for the prediction of mesh parameters ( This is the basic supervised learning loss, used to constrain the model output from deviating from the distribution range of historical best cases. This embodiment uses the Huber Loss function to calculate the difference between the model output value and historical expert data. The reason for using Huber Loss is that historical mining blasting data inevitably contains outliers caused by human error or recording mistakes. Huber Loss is insensitive to outliers, which enhances the robustness of model training and prevents the model from being biased by erroneous historical data.

[0041] The second part is the error term for semi-porosity prediction ( Based on the mapping relationship from geological features and blasting parameters to half-porosity, a surrogate model is configured. When training the main model, if the parameters generated by the main model are input into the surrogate model and the predicted half-porosity is lower than a preset quality threshold (e.g., 90%), a surrogate model is configured. The value of the item will grow exponentially.

[0042] To achieve gradient backpropagation for the aforementioned physical constraints, this embodiment pre-constructs and trains a differentiable semi-porosity prediction surrogate model (HCF-Proxy Net). The input to this surrogate model is the aforementioned geological feature tensor. and the action of the hole mesh parameters output by the strategy network The output is the predicted half-pore rate. Before training the main model, HCF-ProxyNet is trained under supervised conditions using historical data until convergence, and its parameters are frozen.

[0043] Half-porosity prediction error term The specific calculation formula is as follows:

[0044] In the formula, The preset target threshold for semi-porosity is (in this embodiment, it is set to 0.90, i.e., 90%). For the predicted output function of a pre-trained and parameter-frozen differentiable surrogate model; This is the current input geological state feature tensor; The action vector (i.e., target hole depth and target hole distance) output by the policy network. For a linear rectified function, when Time output Otherwise, output 0. The physical meaning of this formula is: a penalty gradient is generated only when the predicted half-aperture is lower than the target value.

[0045] The third part is the penalty for uniformity of rock size ( To balance the semi-porosity and the crushing effect.

[0046] For model training: Phase 1: ... Set to 1.0. and Set to 0. Pre-train the model using historically standardized blasting data, allowing it to learn to mimic the basic perforation logic of human experts and quickly converge to a reasonable parameter space; Second stage: Gradually reduce... Up to 0.1, while increasing and Up to version 1.0.

[0047] After the model output, a series of engineering safety verification steps were implemented: First, based on the target hole spacing (S) and target hole depth (H) output by the model, combined with the linear charge density of the PE pipe (kg / m), the maximum charge Q per hole was calculated. Next, the straight-line distance R from the blasting center to the nearest protected object (such as a residential building, high-voltage power line tower, or substation) was automatically calculated using the GIS system. Finally, the Sadovsky empirical formula was used for vibration prediction.

[0048] Here, K (site coefficient) and The attenuation index is a dynamic value obtained by the system through least squares regression fitting based on the measured vibration data of the most recent 20 blasts in the area.

[0049] The predicted vibration velocity V will be compared with the safety allowable standards for this type of building as specified in the "Safety Regulations for Blasting" (GB 6722). ) for comparison.

[0050] like The parameters passed the verification and were marked as compliant, then proceed to the next step.

[0051] like The system immediately triggers the constraint penalty mechanism. The execution logic of this mechanism is as follows: First-level adjustment: Adjust the differential detonation time. Second-level adjustment: If adjusting the differential detonation time is ineffective, forcibly reduce the target hole spacing value. The system gradually reduces the hole spacing in 0.2-meter increments. Reducing the hole spacing means reducing the load area per hole, thus allowing a reduction in the charge Q per hole while maintaining constant unit consumption, thereby reducing vibration at the source. The iterative process is repeated until the safety requirements are met.

[0052] Furthermore, using a coordinate transformation algorithm, the optimized target hole depth and target hole spacing are mapped back to the mine's digital elevation model, generating an electronic hole layout map containing three-dimensional spatial information. The electronic hole layout map indicates that each hole node is encapsulated with a unique ID, absolute geographic coordinates (N, E, Z), design azimuth, design inclination, and recommended PE pipe length.

[0053] Furthermore, after each blasting operation is completed, a long-range 3D laser scanner (LiDAR) installed on the opposite side of the slope is used to acquire high-density point cloud data of the blasted section.

[0054] Image segmentation based on the watershed algorithm is performed on the point cloud of the blast pile surface. The three-dimensional size distribution of rock block size is statistically analyzed, and the large block ratio (K80) and fine ore ratio are calculated.

[0055] The measured semi-porosity and block size indices are compared with the predicted values ​​before the model blasting to calculate the residuals.

[0056] Rewards are given based on the calculation results; reward function The definition is as follows:

[0057] In the formula, The actual semi-porosity (range 0 to 1) is measured by post-explosion scanning. The rock fragmentation uniformity index is obtained by fitting the Rosin-Rammler distribution. For the foundation indicator variable: when the height of the remaining foundation after blasting exceeds 0.3 meters, ,otherwise ; As weighting coefficients, they are set to [values] in this embodiment. ,in Setting a larger value aims to severely punish root remnants.

[0058] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A deep learning-based method for optimizing the depth and spacing of blasting holes in rock strata, the method being applied to bench blasting in open-pit mines with well-developed fractures, wherein the blasting operation utilizes a drilling rig for drilling and PE pipes for loading explosives; the method includes the following steps: Acquire historical blasting datasets, rock geological survey data, and real-time drilling response data of the drilling rig during drilling operations for the target blasting area; Spatiotemporal alignment and cleaning were performed on the historical blasting dataset, the rock mass geological survey data, and the real-time drilling response data to construct a multi-source heterogeneous blasting feature database. A deep learning-based rock mass charging coupling effect analysis model is constructed; the rock mass charging coupling effect analysis model is equipped with a drilling feature extraction module, a medium impedance matching correction module, and a blasting parameter optimization decision module; The rock mass charge coupling effect analysis model was trained using the multi-source heterogeneous blasting feature database; The method is characterized in that it further includes: The drilling feature extraction module generates a three-dimensional fracture distribution feature vector that reflects the roughness of the borehole wall and the degree of fracture development based on the real-time drilling response data. The medium impedance matching correction module calculates the multi-medium acoustic impedance matching coefficient between the PE pipe, the filling medium inside the hole, and the rock on the hole wall based on the preset physical parameters of the PE pipe. The dielectric impedance matching correction module uses the multi-medium acoustic impedance matching coefficient to weight and correct the three-dimensional crack distribution feature vector to generate an effective energy transfer efficiency matrix. The blasting parameter optimization decision module maps the effective energy transfer efficiency matrix to the blasting parameter space and outputs the target hole depth and target hole spacing values ​​that maximize the half-hole ratio after blasting.

2. The deep learning-based rock stratum blast hole depth and spacing optimization method according to claim 1, characterized in that, The acquisition of the real-time drilling response data specifically includes: The sensor array installed on the drilling rig synchronously collects the rotational torque of the drill rod, axial drilling pressure, instantaneous drilling speed, and slag discharge air pressure at a preset sampling frequency. Using Teale's mechanical energy equation, the collected slewing torque, axial drilling pressure, and instantaneous drilling speed are converted into a sequence of mechanical work required to break a unit volume of rock. The mechanical energy sequence is denoised and smoothed to obtain a drilling mechanical specific energy curve that can characterize the change of rock hardness along the hole depth direction.

3. The method for optimizing the depth and spacing of rock blasting holes based on deep learning according to claim 1, characterized in that, When constructing the multi-source heterogeneous blasting feature database, the following data alignment operation is performed: The coordinates and elevation data of each borehole were obtained using the Global Navigation Satellite System; Establish a three-dimensional spatial coordinate system for the blasting area; The rock strata occurrence information in the rock mass geological survey data, the depth information in the real-time drilling response data, and the blasting effect information in the historical blasting dataset are uniformly mapped to the three-dimensional spatial coordinate system to form structured tensor data with spatial topological relationships.

4. The method for optimizing the depth and spacing of rock blasting holes based on deep learning according to claim 1, characterized in that, Before outputting the target hole depth value and the target hole distance value, the method also includes an engineering safety verification step: Using Sadovsky's empirical formula, based on the single-hole charge amount corresponding to the currently calculated target hole spacing value, the maximum surface vibration velocity induced by blasting is predicted. Determine whether the maximum ground vibration velocity exceeds the safety allowable standard of the protected building or structure; If the safety allowable standard is exceeded, a constraint penalty mechanism is triggered to forcibly reduce the target hole spacing value or adjust the micro-delay detonation time until the safety verification requirements are met.

5. The method for optimizing the depth and spacing of rock blasting holes based on deep learning according to claim 1, characterized in that, The method also includes a digital construction instruction generation step: The target hole depth and target hole spacing values ​​are converted into a visualized electronic hole pattern; Mark the geographical coordinates, design depth, and corresponding recommended length of PE pipe for each hole location on the electronic hole layout map; The electronic hole layout diagram is transmitted to the on-board terminal screen of the drilling rig via a wireless communication network to assist operators in precise hole layout and positioning.

6. The method for optimizing the depth and spacing of rock blasting holes based on deep learning according to claim 1, characterized in that, The specific logic for the dielectric impedance matching correction module to calculate the multi-dielectric acoustic impedance matching coefficient is as follows: Obtain the density and longitudinal wave velocity of the PE pipe material, and calculate the characteristic acoustic impedance of the PE pipe; Based on the three-dimensional fracture distribution feature vector, the dynamic acoustic impedance of the pore wall rock is estimated; Based on the ratio of the characteristic acoustic impedance of the PE pipe to the dynamic acoustic impedance of the rock in the borehole wall, and combined with the properties of the gap filler between the PE pipe and the borehole wall, the energy transmittance is calculated using the stress wave transmission theory formula. The energy transmittance is normalized and used as the multi-medium acoustic impedance matching coefficient, which is used to characterize the effective proportion of explosion energy passing through the PE pipe charge structure and entering the rock mass.

7. The method for optimizing the depth and spacing of rock blasting holes based on deep learning according to claim 1, characterized in that, The rock mass charge coupling effect analysis model adopts a dual-stream attention network architecture: The drilling feature extraction module includes a long short-term memory network branch, which is used to extract the temporal dependency features of the real-time drilling response data in the depth direction. The medium impedance matching correction module includes a graph convolutional neural network branch, which is used to aggregate spatial geological correlation features between adjacent boreholes. The dual-stream attention network architecture introduces an attention mechanism in the feature fusion layer, dynamically adjusting the weight values ​​of each feature channel based on the multi-medium acoustic impedance matching coefficient.

8. The method for optimizing the depth and spacing of rock blasting holes based on deep learning according to claim 1, characterized in that, The training process of the rock mass charge coupling effect analysis model adopted a composite loss function that includes a half-porosity constraint. The composite loss function consists of a hole mesh parameter prediction error term, a half-porosity prediction error term, and a rock blockiness uniformity penalty term. During training, when the model predicts a half-porosity lower than a preset quality threshold, the gradient weight of the half-porosity prediction error term in the composite loss function is automatically increased, forcing the model parameters to converge in a direction that is conducive to protecting the integrity of the pore walls.

9. The method for optimizing the depth and spacing of rock blasting holes based on deep learning according to claim 1, characterized in that, The method also includes a closed-loop feedback iteration step based on 3D scanning: After the blasting operation is completed, point cloud data of the blasted section is obtained using a 3D laser scanner; The actual blasting half-hole ratio and rock fragmentation size distribution are calculated using a point cloud segmentation algorithm. Calculate the residual between the actual blasting half-hole ratio and the model prediction value; The residual is fed back as a reward signal for reinforcement learning to the rock mass charge coupling effect analysis model, and the weights of the blasting parameter optimization decision module are updated online incrementally.

10. The method for optimizing the depth and spacing of rock blasting holes based on deep learning according to claim 1, characterized in that, The blasting parameter optimization decision module is also used for fine-tuning parameters for special geological conditions: When the three-dimensional fracture distribution feature vector indicates the presence of a weak interlayer at the bottom of the hole, an over-depth margin is automatically added to the target hole depth value to overcome the absorption effect of the weak interlayer on the explosion energy. When the three-dimensional fracture distribution feature vector indicates that the fractures in the orifice area are extremely developed, an automatic instruction to adjust the packing length is issued, and the target hole spacing value is reduced accordingly to reduce the risk of blasting gas prematurely escaping from the orifice and causing a blasting accident.

Images