Physical-data dual-drive bench blasting effect prediction method

Abstract

The invention discloses a physical-data dual-drive bench blasting effect prediction method in the technical field of blasting construction and intelligent mining, which comprises the following steps of: dynamically calibrating a fractal dimension field, and constructing a rock blasting classification model by combining blasting indexes such as fractal gradient, rock density, uniaxial compressive strength and drilling speed; performing three-dimensional reconstruction on the geometric morphology of the step before blasting by using an unmanned aerial vehicle high-precision modeling technology, and importing a finite element simulation model to predict a key blasting effect; collecting an actual blasting effect after blasting, and correcting parameters of the numerical simulation prediction model; and constructing a data set of actual lumpiness distribution of the muck pile, establishing a mapping relation between the rock detonability grade and the charge density, and optimizing the differential time sequence and the single-hole charge amount. According to the invention, through fractal dimension field dynamic mapping and real-time feedback control, accurate adaptation of blasting energy and a rock mass structure is realized.

Description

Technical Field

[0001] This invention relates to the fields of blasting construction and intelligent mining technology, and specifically discloses a physical-data dual-driven method for predicting the effect of bench blasting, which is applicable to the prediction of blasting effect and dynamic design optimization under complex geological conditions in open-pit mines. Background Technology

[0002] Today, clear development goals and technical requirements have been set for the construction of smart mines, emphasizing the establishment of a complete intelligent mining standard system by 2026. Blasting, as the starting point of mining operations, is crucial for achieving closed-loop control throughout the entire intelligent mining process. Its digitalization and intelligence levels directly affect the safety and production efficiency of the entire mine. In open-pit mining, the effect of bench blasting directly impacts engineering efficiency. Due to the wide area involved in bench blasting, the large scale of the resulting blast piles, and the numerous influencing factors (such as geological conditions, hole depth, charge quantity, and delay time), predicting blasting effects is challenging. Traditional blasting effect prediction methods generally only include the block size distribution of the blast pile, resulting in limited prediction content and low accuracy; therefore, the level of intelligence urgently needs development.

[0003] In recent years, data-driven deep learning methods have made significant progress in multi-factor prediction, applicable to complex scenarios such as predicting the effects of bench blasting. Intelligent sensing methods such as UAV aerial surveys and fiber optic sensing can provide rich evaluation data for bench blasting effects; however, due to the lack of characterization of the explosion physics process, problems exist regarding the unclear physical meaning of the data sources and poor model interpretability. Meanwhile, numerical simulation (such as the coupling of discrete element and finite element methods) can effectively reproduce bench blasting effects, achieving excellent results in physical reconstruction such as rock mass damage and fragmentation simulation. This can provide reliable structured training data for deep learning models, compensating for the shortcomings of purely data-driven models in terms of physical consistency. Some scholars have proposed that integrating the physical reconstruction capabilities of numerical simulation with the predictive capabilities of deep learning is an important direction for current research on intelligent blasting. The current bottlenecks lie in two aspects: first, how to rationally deploy advanced sensing equipment in mine blasting operations to obtain data that can guide numerical analysis; and second, how to achieve bidirectional coupling between multi-source sensing data and physical models to construct prediction methods with clear physical meaning and strong interpretability.

[0004] Therefore, this invention proposes a physical-data dual-driven method for predicting the effects of step blasting to solve the above problems. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a physical-data dual-driven method for predicting the effects of step blasting. By accurately modeling the rock mass structure and employing a simulation-measurement feedback mechanism, it breaks through traditional blasting effect prediction methods, thereby improving the accuracy and adaptability of blasting effect prediction.

[0006] To achieve the above objectives, the technical solution of the present invention is as follows: a physical-data dual-driven method for predicting the effect of step blasting, comprising:

[0007] S1 uses multi-source sensing technology to obtain the structural characteristics of the rock mass to be blasted: it obtains the three-dimensional contour, rock mass structure and mineral distribution information of the area to be blasted by using UAV oblique photography and multispectral imaging technology, and collects the spatial distribution and initial wave field data of the joint network by a distributed optical fiber sensing system to construct a three-dimensional joint structure model and lithology map.

[0008] S2 establishes a spatial distribution field that integrates rock mass structure and lithology: based on joint grid data, the box dimension algorithm is used to calculate the three-dimensional fractal dimension field of the rock mass joint network, and the multifractal spectrum analysis method is used to extract the spectral characteristics of rock minerals to help divide different lithological regions and dynamically calibrate the fractal dimension field;

[0009] S3 Multi-Source Parameter Fusion and Rock Blastability Grading Model Construction and Label Dataset Generation: This method fuses multi-source parameters of rock mass structure and constructs a rock blastability grading model. A genetic algorithm is used to train the mapping relationship between multi-source parameters and rock blastability levels. Label datasets for blast pile size and over- and under-excavation are constructed using UAV-based measured data from each blasting cycle for subsequent model training. Multi-source parameters include fractal gradient, rock density, uniaxial compressive strength, and drilling speed.

[0010] S4 Dual Data Source Sample Library Construction and SPH-FEM Numerical Model Construction: Based on the three-dimensional contour of the area to be detonated, a corresponding SPH-FEM numerical model is established, and the numerical prediction results are output; the numerical prediction results and the label dataset together form the dual data source sample library.

[0011] S5 blasting parameter fitting and LSTM-GAN hybrid model training: The LSTM-GAN hybrid model is used for deep learning modeling to obtain the correlation model; during the model training phase, the dual data source sample database is included in the training dataset of the correlation model to train the correlation model.

[0012] The correlation model includes a GAN module and an LSTM module; the GAN module generates the explosive block size distribution, and the LSTM module fits the optimal charge amount and differential delay scheme.

[0013] S6 Stress Wave Field Feedback and Real-time Numerical Model Correction: During the blasting implementation phase, based on the stress wave field data fed back by the fiber optic sensor, the coupling state between the explosive and the rock mass is analyzed through edge computing nodes, and the numerical response model is corrected in real time.

[0014] Furthermore, in step S5, the association model is trained through deep reinforcement learning.

[0015] Furthermore, the chromosome encoding in step S3 of the genetic algorithm includes:

[0016] Gradual gradient function of multi-source parameters of rock mass along borehole depth;

[0017] Explosiveness classification based on chaos optimization.

[0018] Furthermore, in step S5, the GAN module includes a generator network and a discriminator network, and the LSTM module includes an LSTM correction unit.

[0019] Generator network: It takes fractal dimension field, explosiveness classification and charge parameters as input, and outputs the predicted blasting block size and over- and under-excavation distribution;

[0020] Discriminator network: Compares the generator output with the measured data of the burst block size distribution to determine whether the generator output is true;

[0021] LSTM correction unit: dynamically adjusts the optimal charge amount and differential timing based on the loss value output by the discriminator.

[0022] Furthermore, the multifractal spectral analysis method in step S2 specifically includes:

[0023] A roughness coefficient distribution map of the joint surface is generated by reconstructing a 3D point cloud.

[0024] Calculate the fractal dimension volatility at different scales and screen the principal fractal dimension that characterizes the energy conduction path;

[0025] By combining multispectral mineral composition data, a mapping matrix between fractal dimension and dynamic tensile strength is established.

[0026] Furthermore, the real-time correction in step S6 includes: when the stress wave exceeds the preset threshold, correcting the rock dynamic strength and explosiveness level, regenerating the charge adjustment coefficient sequence for subsequent holes through the LSTM network, and completing parameter updates and command issuance before the next differential detonation.

[0027] Furthermore, it also includes: S7 automatic correction: when the deviation between the predicted results output by the associated model and the actual blasting effect exceeds 15%, the model calibration is automatically triggered to correct the inversion parameters input to the SPH–FEM numerical model and to perform incremental training on the associated model at the same time; the inversion parameters include the dynamic strength of the rock mass and the detonation pressure enhancement factor at the borehole interface.

[0028] Furthermore, in step S5, a variable diameter charge tube is prepared on-site using 3D printing technology. The gradient of its diameter change matches the topological solution output by the genetic algorithm, and an energy guiding groove is pre-placed on the tube wall to adapt to the orientation anisotropy of the fractal dimension field.

[0029] Furthermore, in step S5, the training data of the GAN module is fused with historical blasting cases. Through transfer learning, the fractal features under different geological conditions are mapped to a unified latent space, thereby achieving the generalization of the prediction model under the preset sample conditions.

[0030] Furthermore, the numerical prediction results include the outward expansion distance of the blast pile, the accumulation morphology, and the stress propagation characteristics; among them, the outward expansion distance is calculated by tracking the maximum horizontal displacement of particles in the blast pile rock mass, the accumulation morphology is obtained by reconstructing a three-dimensional volume model from the particle distribution at the final moment and extracting geometric parameters, and the stress propagation characteristics include the overpressure peak value and spatial distribution.

[0031] The three-dimensional contour of the area to be detonated was extracted from the orthophoto generated by UAV oblique photography and the point cloud model to extract the boundary of the rock mass to be detonated.

[0032] The above-mentioned approach has the following beneficial effects: 1. Compared with existing technologies, this invention significantly improves the prediction accuracy and scene adaptability of bench blasting effects through the deep integration of multi-source sensing technology and a physical-data dual-driven model. Specifically, by utilizing technologies such as UAV oblique photography, multispectral imaging, and distributed fiber optic sensing, the three-dimensional structure of the rock mass, joint network distribution, and lithological characteristics can be accurately acquired, constructing a multi-dimensional geological model that includes fractal dimension fields and lithological maps, fundamentally solving the problem of missing geological parameters in traditional methods. On this basis, the blasting physical process is reconstructed using the SPH-FEM numerical model, and multi-source data is processed by combining the LSTM-GAN hybrid model. This allows for the capture of physical mechanisms such as stress wave propagation and rock mass fracturing through numerical simulation, while also utilizing deep learning to mine nonlinear mapping relationships between data. For example, when the GAN module generates the block size distribution of the blast pile, it can integrate the physical constraints of numerical simulation with the statistical characteristics of measured data, while the LSTM module, trained with historical blasting data, can dynamically fit the optimal charge amount and delay scheme. In addition, the stress wave field feedback mechanism can correct the numerical model in real time. When the measured stress exceeds the threshold, it automatically adjusts the dynamic strength parameters of the rock mass, so that the prediction model can adapt to complex geological conditions. Compared with traditional methods, the prediction error of the blasting block size is reduced by more than 30%, and the over-excavation and under-excavation control accuracy is improved by 40%, effectively solving the pain points of traditional methods such as "single parameter prediction and poor adaptability".

[0033] 2. This invention achieves intelligent control of the entire process of bench blasting from design to implementation through a closed-loop mechanism of perception-simulation-prediction-feedback, significantly improving engineering efficiency and reducing safety risks. During the blasting implementation phase, fiber optic sensors provide real-time feedback of stress wave field data, and edge computing nodes can quickly analyze the coupling state between the explosive and the rock mass. When stress anomalies occur, the LSTM network adjusts the charge amount and delay time of subsequent boreholes in real time, avoiding blasting accidents caused by geological changes. For example, when the dynamic strength of the rock mass in a certain area is higher than expected, the system can automatically generate a charge adjustment coefficient sequence and update the parameters before the next detonation stage, ensuring controllable blast pile morphology. Simultaneously, the matching of 3D-printed variable-diameter charge tubes with the topological solution of the genetic algorithm can optimize energy distribution based on the anisotropy of the fractal dimension field, reducing energy waste and the risk of flyrock. Furthermore, the automatic correction mechanism triggers model calibration when the prediction deviation exceeds 15%. By inverting parameters such as rock mass dynamic strength and detonation pressure, incremental learning of the model is achieved, enabling continuous optimization of the system in long-term applications. In practical engineering, this method can increase blasting cycle efficiency by 25%, reduce explosive consumption by 15%, and reduce the wear and tear on subsequent excavation and loading equipment by precisely controlling over- and under-excavation. At the same time, it reduces safety hazards such as slope instability and promotes the construction of smart mines towards full-process automation.

[0034] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the physical-data dual-driven method for predicting the effect of step blasting according to the present invention. Detailed Implementation

[0036] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] The following detailed description illustrates the specific implementation method:

[0038] Example:

[0039] As attached Figure 1 As shown, the method for predicting the effect of step blasting driven by both physics and data includes the following steps:

[0040] S1 utilizes multi-source sensing technology to acquire the structural features of the rock mass to be blasted: It obtains the three-dimensional contour, rock mass structure, and mineral distribution information of the area to be blasted through UAV oblique photography and multispectral imaging technology. Combined with a distributed fiber optic sensing system, it collects the spatial distribution and initial wavefield data of the joint network to construct a three-dimensional joint structure model and lithological atlas. Specifically, the three-dimensional contour of the area to be blasted is extracted from the orthophoto image generated by UAV oblique photography and the point cloud model to determine the boundary of the rock mass to be blasted.

[0041] S2 establishes a spatial distribution field that integrates rock mass structure and lithology: Based on joint grid data, the box-count algorithm is used to calculate the three-dimensional fractal dimension field of the rock mass joint network, and the multifractal spectrum analysis method is used to extract the spectral characteristics of rock minerals to help divide different lithological regions and dynamically calibrate the fractal dimension field.

[0042] The multifractal spectrum analysis method specifically includes: generating a joint surface roughness coefficient distribution map through three-dimensional point cloud reconstruction; calculating the fractal dimension fluctuation rate at different scales and screening the principal fractal dimension that characterizes the energy conduction path; and establishing a mapping relationship matrix between fractal dimension and dynamic tensile strength by combining multispectral mineral composition data.

[0043] S3 Multi-Source Parameter Fusion and Rock Blastability Grading Model Construction and Label Dataset Generation: This method fuses multi-source parameters of the rock mass structure and constructs a rock blastability grading model. A genetic algorithm is used to train the mapping relationship between the multi-source parameters and the rock blastability level. Label datasets for blast pile size and over- and under-excavation are constructed using UAV-based measured data from each blasting cycle for subsequent model training. The multi-source parameters include fractal gradient, rock density, uniaxial compressive strength, and drilling speed. The chromosome encoding of the genetic algorithm includes: the gradual gradient function of the multi-source parameters of the rock mass along the borehole depth; and a blastability level classification based on chaos optimization.

[0044] S4 Dual Data Source Sample Library Construction and SPH-FEM Numerical Model Building: Based on the 3D contour of the area to be blasted, a corresponding SPH-FEM numerical model was established, outputting numerical prediction results. The numerical prediction results, together with the labeled dataset, constitute the dual data source sample library. The numerical prediction results include the outward expansion distance of the blast pile, the accumulation morphology, and stress propagation characteristics. The outward expansion distance is calculated by tracking the maximum horizontal displacement of particles in the blast pile rock mass; the accumulation morphology is obtained by reconstructing a 3D volume model from the particle distribution at the final moment and extracting geometric parameters; and the stress propagation characteristics include the overpressure peak value and spatial distribution.

[0045] S5 blasting parameter fitting and LSTM-GAN hybrid model training: The LSTM-GAN hybrid model is used for deep learning modeling to obtain the association model; during the model training stage, the dual data source sample database is included in the training dataset of the association model, and deep reinforcement learning is used to train the association model.

[0046] The correlation model includes a GAN module and an LSTM module. The GAN module generates the distribution of blast pile size. Specifically, the training data of the GAN module is integrated with historical blasting cases. Through transfer learning, the fractal features under different geological conditions are mapped to a unified latent space to achieve the generalization of the prediction model under the preset sample conditions. The LSTM module fits the optimal charge amount and differential delay scheme.

[0047] Specifically, the GAN module includes a generator network and a discriminator network, and the LSTM module includes an LSTM correction unit. The generator network receives fractal dimension field, explosiveness classification, and charge parameters as input, and outputs the predicted blasting block size and over- and under-exploitation distribution. The discriminator network compares the generator output with the measured data of blasting block size distribution to determine whether the generator output is true. The LSTM correction unit dynamically adjusts the optimal charge amount and time difference based on the loss value output by the discriminator.

[0048] Furthermore, since traditional charge tubes are standardized products with fixed diameters and cannot adapt to complex rock structures, variable-diameter charge tubes are manufactured on-site using 3D printing technology. The diameter variation gradient matches the topological solution output by the genetic algorithm, and pre-installed energy guiding grooves on the tube wall adapt to the anisotropy of the fractal dimension field. 3D printing technology can directly manufacture non-uniform variable-diameter charge tubes on demand based on the topological solution output by the genetic algorithm (i.e., the spatial distribution of parameters such as charge tube diameter, wall thickness, and guiding grooves), breaking through the geometric limitations of traditional processing techniques. Simultaneously, geological conditions at the blasting site may change in real time with excavation (e.g., newly exposed fissures). 3D printing supports rapid on-demand production, and combined with dynamic updates of the fractal dimension field (e.g., refreshed every 30 seconds), it achieves instant adaptation of the charge structure to geological conditions, avoiding the inventory and scheduling delays associated with traditional prefabricated tubes.

[0049] S6 Stress Wave Field Feedback and Real-time Numerical Model Correction: During the blasting phase, based on the stress wave field data fed back by fiber optic sensors, the coupling state between the explosive and the rock mass is analyzed through edge computing nodes, and the numerical response model is corrected in real time. Specifically, when the stress wave exceeds a preset threshold, the rock dynamic strength and explosiveness level are corrected, and the charge adjustment coefficient sequence for subsequent holes is regenerated through an LSTM network. Parameter updates and command issuance are completed before the next detonation stage.

[0050] S7 Automatic Correction: When the deviation between the predicted results output by the associated model and the actual blasting effect exceeds 15%, the model calibration is automatically triggered to correct the inversion parameters input to the SPH–FEM numerical model and to perform incremental training on the associated model. The inversion parameters include the dynamic strength of the rock mass and the detonation pressure boost factor at the borehole interface.

[0051] Taking an open-pit iron mine (the ore body is 1.2 km long, with an average excavation depth of 80 m, and the rock mass is mainly granite with quartzite interlayers) as an example, the differences in implementation process and effect between the traditional empirical blasting method and the physical-data dual-driven method are compared as follows:

[0052] Table 1 - Comparison of Implementation Steps

[0053]

[0054]

[0055] Table 2 - Comparison of Key Indicators and Benefit Analysis

[0056]

[0057] As shown in Tables 1 and 2, the physical-data dual-driven blasting method differs significantly from the traditional empirical blasting method in terms of implementation process and application effect. In terms of implementation steps, the traditional method relies on extensive methods such as borehole sampling and manual logging, taking 3 days for exploration and producing low data resolution. Parameter design is based on experience in classifying lithology, lacking a dynamic correction mechanism. In contrast, the dual-driven method achieves rapid 3D modeling within 2 hours through UAV oblique photography and distributed fiber optic sensing. It combines box-counting algorithms and genetic algorithms to quantify the fractal characteristics of the rock mass, constructs a multi-source parameter fusion blastability model, and achieves accurate prediction and real-time optimization of blasting effects through SPH-FEM numerical simulation and a hybrid LSTM-GAN model.

[0058] In terms of key indicators, the dual-drive method demonstrates significant advantages: geological exploration efficiency is increased by 36 times, the qualified rate of blasted block size is increased from <81.3% to >91.2%, the over- and under-excavation control accuracy is optimized from ±30cm to ±10cm, explosive consumption is reduced by 29%, peak blasting vibration is reduced by 33%, and model prediction error is reduced by more than 68.3%. Taking quartzite interlayer blasting as an example, traditional methods result in a large block rate of about 40% due to failure to identify lithological differences, requiring secondary crushing; while the dual-drive method dynamically calibrates the rock mass strength through fractal dimension field, optimizes charge parameters, and reduces the large block rate to 8%-15%, saving 12,000 yuan per blast.

[0059] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for predicting the effect of step blasting driven by both physics and data, characterized in that, include: S1 uses multi-source sensing technology to obtain the structural characteristics of the rock mass to be blasted: it obtains the three-dimensional contour, rock mass structure and mineral distribution information of the area to be blasted by using UAV oblique photography and multispectral imaging technology, and collects the spatial distribution and initial wave field data of the joint network by a distributed optical fiber sensing system to construct a three-dimensional joint structure model and lithology map. S2 establishes a spatial distribution field that integrates rock mass structure and lithology: based on joint grid data, the box dimension algorithm is used to calculate the three-dimensional fractal dimension field of the rock mass joint network, and the multifractal spectrum analysis method is used to extract the spectral characteristics of rock minerals to help divide different lithological regions and dynamically calibrate the fractal dimension field; S3 Multi-Source Parameter Fusion and Rock Blastability Grading Model Construction and Label Dataset Generation: This method fuses multi-source parameters of rock mass structure and constructs a rock blastability grading model. A genetic algorithm is used to train the mapping relationship between multi-source parameters and rock blastability levels. Label datasets for blast pile size and over- and under-excavation are constructed using UAV-based measured data from each blasting cycle for subsequent model training. Multi-source parameters include fractal gradient, rock density, uniaxial compressive strength, and drilling speed. S4 Dual Data Source Sample Library Construction and SPH-FEM Numerical Model Construction: Based on the three-dimensional contour of the area to be detonated, a corresponding SPH-FEM numerical model is established, and the numerical prediction results are output; the numerical prediction results and the label dataset together form the dual data source sample library. S5 blasting parameter fitting and LSTM-GAN hybrid model training: The LSTM-GAN hybrid model is used for deep learning modeling to obtain the correlation model; during the model training phase, the dual data source sample database is included in the training dataset of the correlation model to train the correlation model. The association model includes GAN and LSTM modules; The GAN module generates the explosive block size distribution, and the LSTM module fits the optimal charge amount and differential delay scheme. S6 Stress Wave Field Feedback and Real-time Numerical Model Correction: During the blasting implementation phase, based on the stress wave field data fed back by the fiber optic sensor, the coupling state between the explosive and the rock mass is analyzed through edge computing nodes, and the numerical response model is corrected in real time.

2. The method for predicting the effect of step blasting driven by both physical and data principles according to claim 1, characterized in that: In step S5, the association model is trained through deep reinforcement learning.

3. The method for predicting the effect of step blasting driven by both physical and data principles according to claim 1, characterized in that: The chromosome encoding in step S3 of the genetic algorithm includes: Gradual gradient function of multi-source parameters of rock mass along borehole depth; Explosiveness classification based on chaos optimization.

4. The method for predicting the effect of step blasting driven by both physical and data principles according to claim 1, characterized in that: In step S5, the GAN module includes a generator network and a discriminator network, and the LSTM module includes an LSTM correction unit. Generator network: It takes fractal dimension field, explosiveness classification and charge parameters as input, and outputs the predicted blasting block size and over- and under-excavation distribution; Discriminator network: Compares the generator output with the measured data of the burst block size distribution to determine whether the generator output is true; LSTM correction unit: dynamically adjusts the optimal charge amount and differential timing based on the loss value output by the discriminator.

5. The method for predicting the effect of step blasting driven by both physical and data principles according to claim 1, characterized in that: Step S2, multifractal spectral analysis, specifically includes: A roughness coefficient distribution map of the joint surface is generated by reconstructing a 3D point cloud. Calculate the fractal dimension volatility at different scales and screen the principal fractal dimension that characterizes the energy conduction path; By combining multispectral mineral composition data, a mapping matrix between fractal dimension and dynamic tensile strength is established.

6. The method for predicting the effect of step blasting driven by both physical and data principles according to claim 1, characterized in that: The real-time correction in step S6 includes: when the stress wave exceeds the preset threshold, correcting the rock dynamic strength and explosiveness level, regenerating the charge adjustment coefficient sequence for subsequent holes through the LSTM network, and completing parameter updates and command issuance before the next differential detonation.

7. The method for predicting the effect of step blasting driven by both physical and data principles according to claim 1, characterized in that: Also includes: S7 Automatic Correction: When the deviation between the predicted results output by the associated model and the actual blasting effect exceeds 15%, the model calibration is automatically triggered to correct the inversion parameters input to the SPH–FEM numerical model and to perform incremental training on the associated model. The inversion parameters include the dynamic strength of the rock mass and the detonation pressure boost factor at the borehole interface.

8. The method for predicting the effect of step blasting driven by both physical and data principles according to claim 6, characterized in that: In step S5, a variable diameter charge tube is prepared on-site using 3D printing technology. The gradient of its diameter change matches the topological solution output by the genetic algorithm. An energy guiding groove is pre-placed on the tube wall to adapt to the orientation anisotropy of the fractal dimension field.

9. The method for predicting the effect of step blasting driven by both physical and data principles according to claim 1, characterized in that: In step S5, the training data of the GAN module is fused with historical blasting cases. Through transfer learning, the fractal features under different geological conditions are mapped to a unified latent space, thereby achieving the generalization of the prediction model under the preset sample conditions.

10. The method for predicting the effect of step blasting driven by both physical and data principles according to claim 1, characterized in that: The numerical prediction results include the outward expansion distance, accumulation morphology, and stress propagation characteristics of the blast pile; among them, the outward expansion distance is calculated by tracking the maximum horizontal displacement of particles in the blast pile rock mass, the accumulation morphology is obtained by reconstructing a three-dimensional volume model from the particle distribution at the final moment and extracting geometric parameters, and the stress propagation characteristics include the overpressure peak value and spatial distribution. The three-dimensional contour of the area to be detonated was extracted from the orthophoto generated by UAV oblique photography and the point cloud model to extract the boundary of the rock mass to be detonated.

Images