Intelligent highway tunnel blasting one-shot design method

Abstract

The application discloses an intelligent one-shot one-design method for highway tunnel blasting excavation, and belongs to the technical field of tunnel construction process, which comprises the following steps: collecting the rock mass three-dimensional point cloud, joint image and physical and mechanical parameters of the tunnel face of a blasting round to form a single round exclusive rock mass data set; inputting the data set into an AI recognition model, extracting and identifying joint features, and outputting structured joint feature parameters; inputting the joint and rock mass mechanical parameters into an AI multi-objective optimization model to dynamically output core blasting parameters such as blast hole position, inclination and depth; checking the optimized parameters, generating a standardized design scheme after the qualified parameters, and synchronously feeding the parameters to the construction equipment, and re-optimizing the unqualified parameters; collecting the effect and process data after construction, feeding back the AI model parameters through the gradient descent algorithm, and realizing the closed-loop iteration of one-shot one-optimization; and the method solves the problems of insufficient dynamic response of joint rock mass and lack of feedback iteration mechanism in the existing laser scanning three-dimensional blasting design method.

Description

Technical Field

[0001] This invention relates to the field of tunnel construction technology, specifically to an intelligent method for one-shot-one-design blasting in highway tunnels. Background Technology

[0002] Drilling depth ≤ 5m, tilt angle adjustment range 0-90°, charge accuracy ≤ ±2%; the construction equipment control system includes instruction adaptation modules for different brands of hydraulic rock drills and automated charging equipment.

[0003] In highway tunnel excavation blasting, the distribution of joints in the rock mass at the tunnel face exhibits significant heterogeneity and dynamic variability, resulting in variations in excavation conditions for each blast. Existing laser scanning-based 3D blasting design methods often employ a "batch design" approach, where multiple blasts use the same design logic or parameter template, failing to provide flexible adaptation. Furthermore, low accuracy in joint identification and reliance on traditional algorithms for parameter optimization lead to insufficient adaptation of core parameters such as borehole location, dip angle, and charge quantity to real-time rock mass conditions, easily resulting in over- or under-excavation and excessive blasting vibration. In addition, existing methods lack a construction data feedback and iteration mechanism, making it impossible to optimize subsequent blast designs based on previous blasts, hindering continuous improvement in blasting accuracy. These are the main problems currently faced. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide an intelligent one-shot-one-design method for highway tunnel blasting, so as to solve the problems of insufficient dynamic response of jointed rock mass and lack of feedback iteration mechanism in the existing laser scanning three-dimensional blasting design method.

[0005] To address the above problems, the present invention provides the following technical solution:

[0006] A smart method for one-shot-one-design blasting of highway tunnels; it includes the following steps:

[0007] S1. Collect rock mass data for a single blast: Collect three-dimensional point cloud data, joint detail image data and rock physical and mechanical parameters of the tunnel face to be blasted, and form a dedicated rock mass dataset for a single blast.

[0008] S2. Intelligent identification of jointed rock mass features: Input the single-shot-specific rock mass dataset collected in step S1 into the preset AI identification model; extract and identify the joint strike, dip angle, spacing, aperture and distribution density through point cloud-image data fusion algorithm, and output structured joint feature parameters;

[0009] S3. AI-based multi-objective optimization of single-shot blasting parameters: Input the obtained joint feature parameters and rock mass physical and mechanical parameters into the AI ​​multi-objective optimization model; dynamically output the blasting hole location coordinates, inclination angle, borehole depth, single-hole charge amount and detonation sequence parameters for the current shot;

[0010] S4. Standardized blasting design output and verification: Verify the rationality of the single-blasting parameters optimized in step S3. If the verification is successful, generate a standardized design scheme including a hole layout diagram, a comparison table of hole parameters, and a detonation sequence flowchart, and synchronize it to the construction equipment control system. If the verification fails, return to step S3 for re-optimization.

[0011] S5. Construction Verification and Data Feedback Iteration: After completing the current blasting operation, blasting effect data and construction process data are collected and fed back to the AI ​​multi-objective optimization model through the gradient descent algorithm. The reward function weights and parameter optimization rules of the model are iteratively updated to provide optimization basis for the design of the next blast, realizing a closed-loop iteration of one blast, one optimization.

[0012] Preferably, in step S1, three-dimensional point cloud data and joint detail image data of the rock mass are acquired through a collaborative method of three-dimensional laser scanning and digital photogrammetry; wherein the scanning accuracy of the three-dimensional laser scanning unit is ≤±2mm, and the scanning range includes the current working face and an area 5m outside the excavation outline; the digital photogrammetry unit uses three industrial cameras to simultaneously capture images of the working face from the front, left 45°, and right 45° respectively; the physical and mechanical parameters of the rock mass are detected on-site using a portable rock mass strength tester; the acquired data are standardized and preprocessed to form a dedicated dataset for each blast, avoiding interference between data from different blasts.

[0013] Preferably, the AI ​​recognition model in step S2 adopts a deep learning architecture that integrates PointNet++ and CNN. The data processing flow of this recognition model is as follows: the three-dimensional point cloud data is voxelized to a size of 0.05m×0.05m×0.05m, and the geometric features of the rock mass are extracted; the Canny edge detection algorithm is used to extract the joint edge features of the image data; finally, the rock mass geometric features and joint edge features are fused through the attention mechanism to strengthen the feature weight of the joint area and suppress noise interference.

[0014] Preferably, the AI ​​multi-objective optimization model in step S3 is constructed based on a reinforcement learning algorithm: it uses the blasting hole location coordinates, inclination angle, drilling depth, single-hole charge amount, and detonation timing parameters as the action space, and the joint avoidance rate ≥98%, over-drilling / under-drilling amount ≤5cm, half-hole rate ≥85%, and blasting vibration velocity ≤10cm / s as the reward function; the parameter optimization rules are: the blasting hole location avoids dense joint areas by ≥0.3m, the inclination angle is ≥60° with the joint strike, the drilling depth is adapted to the joint layer thickness, the single-hole charge amount is adjusted according to the rock mass compressive strength gradient, and the detonation time difference is 50-200ms.

[0015] Preferably, the AI ​​multi-objective optimization model in step S3 has a built-in blasting parameter database and a transfer learning module: the database stores the historical best single-blast design parameters for different geological conditions and tunnel cross sections; the transfer learning module can transfer the basic model trained on historical data to the current engineering scenario.

[0016] Preferably, the equipment operating condition adaptation verification standard in step S4 is: drilling depth ≤ 5m, tilt angle adjustment range 0-90°, and charge accuracy ≤ ±2%.

[0017] Preferably, the blasting effect data in step S5 includes the over-excavation and under-excavation amount and half-hole ratio detected by three-dimensional laser scanning, and the blasting vibration velocity collected by distributed acceleration sensors; the construction process data includes the actual drilling position deviation, charge amount error, and detonation timing deviation. All collected data are quantified and labeled before being incorporated into the model iteration.

[0018] Preferably, both the AI ​​recognition model in step S2 and the AI ​​multi-objective optimization model in step S3 are lightweight edge models with ≤5G of model parameters. They are equipped with industrial-grade edge computing terminals that are dustproof, vibration-resistant, and have wide voltage power supply, enabling local data collection, local computation, and local distribution without network dependence. The edge terminal has a reserved 5G / Bluetooth module to support data cloud upload.

[0019] The beneficial effects of this invention are reflected in the following aspects:

[0020] 1. Achieving true one-blast-one-adaptation, ensuring that the parameters of each blast are highly matched with the real-time rock mass conditions: This invention proposes a data acquisition logic for a dedicated rock mass dataset for each blast, specifically designed for the real-time joint distribution, rock mechanics properties, and special working conditions of the face for each blast; combined with a multimodal joint recognition model integrating PointNet++ and CNN, it ensures that parameters such as borehole position and dip angle are highly matched with the current rock mass conditions of the blast, ultimately achieving borehole position positioning accuracy ≤ ±2cm, dip angle error ≤ ±0.3°, charge matching degree ≥95%, and over- or under-excavation amount stable at ≤5cm, solving the core problem of poor adaptability of existing technologies;

[0021] 2. The AI ​​multi-objective optimization model constructed in this invention incorporates four core indicators—joint avoidance rate, over- and under-drilling amount, half-hole rate, and blasting vibration velocity—into a unified reward function, achieving full parameter optimization of hole position, inclination angle, depth, charge amount, and detonation sequence. This reduces the optimization time for a single blast to ≤15 minutes, improving efficiency by more than 60% compared to traditional designs. Furthermore, this application combines transfer learning and small-sample augmentation algorithms, requiring only 30-80 sets of field data to complete model fine-tuning for new engineering scenarios, reducing the amount of adaptation data by 60%-75% compared to traditional designs, and significantly lowering the adaptation cost for new projects.

[0022] 3. This invention achieves the linkage and iteration of construction data and core parameters of AI model, feeding back quantitative data of blasting effect and construction process to the model and updating reward function weight and parameter optimization rules, rather than conventional surface parameter adjustment; through a closed loop of design, construction, detection, feedback and optimization, it enables the model to continuously learn the rock mass characteristics of the current project, achieve one-blast-one-optimization, and gradually improve the blasting quality with each blast, solving the defect of existing technology that cannot continuously improve blasting accuracy;

[0023] 4. This invention reduces blasting risks by avoiding hole positions and adjusting the charge amount, keeping the blasting vibration velocity stable at ≤10cm / s, and can be lowered to ≤8cm / s under special conditions to meet the safety requirements of nearby buildings. Engineering verification shows that this method can significantly improve the blasting quality qualification rate and design efficiency compared with traditional batch design methods, and is suitable for blasting excavation projects of highway tunnels with complex joints. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the intelligent one-shot-one-design method for highway tunnel blasting in this embodiment. Detailed Implementation

[0025] The present invention will be further described below with reference to the accompanying drawings and specific embodiments:

[0026] Example:

[0027] Reference Figure 1 This embodiment provides an intelligent one-shot-one-design method for highway tunnel blasting; it includes the following steps:

[0028] S1: Precise acquisition of rock mass data for a single shot;

[0029] For the tunnel face to be blasted, dedicated data collection will be conducted to avoid data confusion between different blasts and to provide basic data support for a one-blast-one-design approach. The collected data includes:

[0030] 1. Rock mass 3D point cloud data: The FAROFocus S150 high-precision 3D laser scanner is used. Its scanning unit has a scanning accuracy of ≤±2mm. The scanner is set up 5m behind the working face so that the scanning range covers the current working face and the area 5m outside the excavation outline to obtain the rock mass 3D point cloud data and capture the spatial distribution of the rock surface and shallow joints at the working face. Before scanning, the working face is cleaned to remove loose rocks and silt to avoid obscuring joint features.

[0031] 2. Joint detail image data: Three 24-megapixel industrial cameras were used to simultaneously capture images of the working face from the front, left 45°, and right 45° respectively. A ring-shaped supplementary lighting device was used during the shooting to avoid uneven lighting, supplement joint detail features, and improve the completeness of subsequent identification.

[0032] 3. Rock mass physical and mechanical parameters: Using a portable rock mass strength tester, 3-5 representative measuring points were selected at the working face to test the rock mass density, compressive strength, elastic modulus and other physical and mechanical parameters on site. The test error was ≤±3%, which provided a mechanical basis for optimizing parameters such as charge amount and drilling depth.

[0033] After data collection is completed, the collected point cloud, images and mechanical parameters need to be standardized, abnormal data should be removed, and integrated to form a single-shot-specific rock mass dataset to provide a unified data format for subsequent AI identification.

[0034] S2: Intelligent identification of jointed rock mass characteristics;

[0035] The single-shot-specific rock mass dataset formed in step S1 is input into the preset AI recognition model. The point cloud-image data fusion algorithm is used to extract and identify the joint strike, dip angle, spacing, aperture and distribution density, and output the structured joint feature parameters; thus solving the problem of low accuracy of the existing single recognition method.

[0036] The AI ​​recognition model adopts a deep learning architecture that integrates PointNet++ and CNN. The data processing flow of this recognition model is as follows:

[0037] The three-dimensional point cloud data is voxelized to a size of 0.05m×0.05m×0.05m to extract the geometric features of the rock mass surface. Then, the Canny edge detection algorithm is used to extract joint edge features from the image data. Finally, the two types of features are fused through an attention mechanism to strengthen the feature weight of the joint area, suppress noise interference, and accurately identify the joint direction, dip angle, spacing, opening and distribution density of the current blast face.

[0038] The structured joint feature parameters output in this step have a joint identification accuracy of ≥95% and a parameter measurement error of ≤±2° (tilt / direction) and ≤±5cm (spacing / opening).

[0039] S3: AI-based multi-target optimization of single-shot detonation parameters

[0040] The joint feature parameters output in step S2 and the rock mechanics parameters output in step S1 are input into the AI ​​multi-objective optimization model. The AI ​​multi-objective optimization model deployed locally at the edge dynamically outputs the blast hole location coordinates, inclination angle, borehole depth, single hole charge amount and detonation sequence parameters for the current blast, solving the problems of single optimization dimension and poor engineering adaptability in the existing optimization.

[0041] The AI ​​multi-objective optimization model is built on reinforcement learning algorithms and is equipped with an industrial-grade edge computing terminal that is dustproof, vibration-resistant, and has a wide voltage power supply. The model parameters are compressed to ≤5G, enabling local data collection, local calculation, and local distribution without network dependence. The edge terminal has a reserved 5G / Bluetooth module to support uploading data to the cloud database for long-term model training after construction is completed.

[0042] The AI ​​multi-objective optimization model uses the coordinates of the blasting hole position, tilt angle, depth, charge amount and detonation sequence as the action space, and the joint avoidance rate ≥98%, over-excavation and under-excavation amount ≤5cm, half hole rate ≥85%, and blasting vibration velocity ≤10cm / s as the multi-objective reward function, and sets dynamic weight coefficients, which can be flexibly adjusted according to the priority of engineering safety and efficiency.

[0043] The AI ​​multi-objective optimization model can output parameters through interactive iteration with the rock mechanics simulation environment. The specific parameter optimization rules are as follows:

[0044] ① Hole location optimization: Avoid areas with dense joints (≥0.3m) and arrange the holes at a foundation spacing of 1.0m × 0.8m to ensure drilling stability;

[0045] ② Inclination optimization: The angle with the joint direction is ≥60° to improve the blasting and fragmentation effect;

[0046] ③ Depth optimization: The thickness is adapted according to the joint layer, with an error of ≤±10cm, to avoid penetrating weak interlayers;

[0047] ④ Optimize the charge quantity: Adjust according to the rock mass compressive strength gradient. For every 5MPa increase in compressive strength, increase the charge quantity by 8-12%, with an error of ≤±2%. Reduce the charge quantity by 15-20% in special areas.

[0048] ⑤ Optimization of detonation sequence: Based on the basic logic of slotting holes, auxiliary holes, and peripheral holes, combined with the zoning planning of joint distribution, the time difference is calculated according to the rock mass wave velocity (50-200ms, error ≤±5ms) to reduce vibration superposition;

[0049] The AI ​​multi-objective optimization model incorporates a blasting parameter database and a transfer learning module. Its database stores historical optimal single-blast design parameters for different geological conditions such as granite, sandstone, and shale, as well as different tunnel cross sections, with a sample size of no less than 1,000 sets. The transfer learning module can transfer the basic model trained on historical data to the current engineering scenario. Only 50-80 sets of on-site single-blast data are needed to complete the model fine-tuning. The parameter prediction accuracy is ≥90%, and the time for single-blast parameter optimization is ≤15 minutes, which is much shorter than the interval between tunnel blasting construction procedures and does not affect the construction progress.

[0050] In this embodiment, both the AI ​​recognition model in step S2 and the AI ​​multi-objective optimization model in step S3 adopt a local deployment scheme at the edge, which is suitable for the on-site conditions of tunnel construction without network and in harsh environments.

[0051] S4: Standardized Blasting Design Scheme Output and Verification

[0052] The rationality of the single-shot blasting parameters optimized in step S3 is verified. After the verification is passed, a standardized design scheme including a hole layout diagram, a comparison table of parameters for each hole, and a detonation sequence flowchart is generated and synchronized to the construction equipment control system; this solves the problem of the existing design being disconnected from construction equipment and safety specifications.

[0053] The rationality check includes:

[0054] 1. Equipment operating condition adaptation verification: Built-in construction equipment parameter library to verify drilling depth ≤5m, tilt angle adjustment range 0-90°, and charge accuracy ≤±2%, ensuring that the parameters can be implemented.

[0055] 2. Blasting safety specification verification: Verify whether parameters such as charge amount, detonation time difference, and blasting vibration velocity meet the safety thresholds by referring to the tunnel blasting safety regulations;

[0056] After verification, the standardization generation unit generates a unique design scheme for the current blast, including a hole layout diagram (marking the coordinates and numbers of each hole), a parameter comparison table for each hole (listing the inclination angle, depth, and charge amount), and a detonation sequence flowchart (marking the detonation sequence and time difference). Then, through the data interaction unit, the scheme is synchronized to the control system of the hydraulic rock drill and the automated charging equipment according to the Profinet / Modbus TCP industrial communication protocol, avoiding errors from manual transcription and realizing direct connection between the design scheme and the construction equipment.

[0057] If the verification fails, the parameters will be fed back to step S3 for multi-objective optimization again until the equipment operating conditions and safety specifications are met.

[0058] S5: Construction Verification and Data Feedback Iteration

[0059] After completing the current blasting operation, quantitative data collection of blasting effects and processes is conducted. This data is then fed back to the AI ​​multi-objective optimization model via a gradient descent algorithm, iteratively updating the model's reward function weights and parameter optimization rules. This provides an optimization basis for the design of the next blast, achieving a closed-loop iteration of optimization per blast. To address the issue of insufficient depth in existing technologies, the data to be quantitatively collected includes:

[0060] 1. Blasting effect data: Three-dimensional laser scanning is used to detect the excavation outline and calculate the over-excavation and under-excavation amount and half-hole ratio; distributed accelerometers are used to collect blasting vibration velocity, and the maximum value is taken as the evaluation index;

[0061] 2. Construction process data: Record the actual borehole location deviation, charge quantity error, detonation timing deviation, etc., and quantify and label all data;

[0062] After the collected quantitative data is processed, it is fed back to the AI ​​multi-objective optimization model. The reward function weights and parameter optimization rules of the model are iteratively updated through the gradient descent algorithm, so that the model can continuously learn the rock mass characteristics and construction rules of the current project and provide a more accurate optimization basis for the design of the next blasting.

[0063] Continuous optimization: Repeat steps S1-S5 to ensure that the design of each shot is optimized based on the construction data of the previous shot, achieving a closed-loop iteration effect of optimization for each shot.

[0064] The design method described in this embodiment was applied to the blasting excavation of a Class IV jointed section of a highway tunnel in Chongqing. The tunnel has a horseshoe-shaped cross-section (12m wide and 8m high), with residential buildings located 30m away. The blasting vibration velocity threshold is ≤10cm / s. The rock mass is granite, and the joint distribution at the tunnel face varies with each blast. The specific implementation process of the design method described in this embodiment is as follows:

[0065] S1. For the working face of the first blast, a FAROFocus S150 scanner was set up to acquire point cloud data. Three industrial cameras simultaneously captured images from the front, left 45° and right 45° after ring lighting. A portable testing instrument detected the rock mass parameters as density 2.6 g / cm³ and compressive strength 35 MPa. After preprocessing, a dedicated dataset for the first blast was formed.

[0066] S2. Input the dataset into the AI ​​recognition model that integrates PointNet++ and CNN. Through the attention mechanism, the point cloud-image feature fusion is achieved, and the joint direction of the first blast face is identified as 30.5°, the dip angle is 64.8°, and the spacing is 0.9-1.1m. The recognition error is ≤±1.5° and ±3cm, and the joint recognition accuracy is 98%. The structured joint feature parameters are output.

[0067] S3. Input the joint characteristic parameters and rock mass mechanics parameters into the edge AI multi-objective optimization model. The model uses the joint avoidance rate ≥95%, over-excavation / under-excavation amount ≤5cm, half-hole rate ≥85%, and blasting vibration velocity ≤10cm / s as the reward function. The output parameters for the first shot are as follows: the holes are arranged in a 1.0m×0.8m pattern, avoiding the dense joint area by 0.4m; the inclination angle of the peripheral holes is 85° (angle with the joint is 64.5°), the depth is 4.5m, and the charge is 0.3kg / hole; the inclination angle of the auxiliary holes is 90°, the depth is 4.8m, and the charge is 0.6kg / hole; the inclination angle of the cut holes is 90°, the depth is 5.0m, and the charge is 1.1kg / hole; the detonation sequence is: the cut holes are detonated first, the auxiliary holes are detonated 100ms later, and the peripheral holes are detonated 200ms later; this optimization took 12 minutes.

[0068] S4. Verify that the parameters meet the equipment operating conditions (maximum drilling depth 5m, charge accuracy ±2%) and safety specifications; generate a dedicated design scheme for the first blast and synchronize it to the hydraulic rock drill and automated charge equipment control system according to the Modbus TCP protocol;

[0069] S5. After the first blast, quantitative data was collected: the over-drilling and under-drilling amount was 2cm, the half-hole rate was 87%, and the vibration velocity was 8.5cm / s; construction process data: drilling deviation ≤2cm, charge amount error ±1.5%; the data was fed back to the AI ​​multi-objective optimization model, and the model reward function weight and detonation time difference optimization rules were iteratively updated through gradient descent algorithm to complete the deep iteration of the model.

[0070] Design and construction of the second blast: Repeat steps S1-S5, based on the iterated AI model and combined with the newly collected data from the second blast face (joint strike 32°, dip angle 63°), optimize the output parameters of the second blast (peripheral hole dip angle 84°, charge amount 0.32kg / hole, detonation time difference adjusted to 110ms), optimization time is 11 minutes; after the construction of the second blast, quantitative data are detected and measured, the over-excavation and under-excavation amount is 1.8cm, the half-hole rate is 89%, and the vibration velocity is 8.2cm / s, the blasting effect is significantly improved compared with the first blast.

[0071] This embodiment, through continuous implementation and verification, shows that the design of each blast is accurately adapted to the joint conditions of the tunnel face at that time, and the blasting effect is gradually improved. The over-excavation and under-excavation are stable within 2cm, the half-hole rate is ≥87%, and the blasting vibration velocity is ≤9cm / s. Compared with the traditional batch design method, the design efficiency is improved by 65%, and the blasting quality qualification rate is improved by 30%. This proves that this method can effectively solve the dynamic adaptation problem of blasting in complex jointed rock mass tunnels, and ensure construction safety and progress.

[0072] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any person skilled in the art can make some modifications or equivalent substitutions to the above-disclosed technical content without departing from the scope of the technical solution of the present invention. All such modifications or equivalent substitutions should be covered within the protection scope of the present invention.

Claims

1. An intelligent method for one-shot-one-design blasting of highway tunnels, characterized in that: It includes the following steps: S1. Collect rock mass data for a single blast: Collect three-dimensional point cloud data, joint detail image data and rock physical and mechanical parameters of the tunnel face to be blasted, and form a dedicated rock mass dataset for a single blast. S2. Intelligent identification of jointed rock mass features: Input the single-shot-specific rock mass dataset collected in step S1 into the preset AI identification model; extract and identify the joint strike, dip angle, spacing, aperture and distribution density through point cloud-image data fusion algorithm, and output structured joint feature parameters; S3. AI-based multi-objective optimization of single-shot blasting parameters: Input the obtained joint feature parameters and rock mass physical and mechanical parameters into the AI ​​multi-objective optimization model; dynamically output the blasting hole location coordinates, inclination angle, borehole depth, single-hole charge amount and detonation sequence parameters for the current shot; S4. Standardized blasting design output and verification: Verify the rationality of the single-blasting parameters optimized in step S3. If the verification is successful, generate a standardized design scheme including a hole layout diagram, a comparison table of hole parameters, and a detonation sequence flowchart, and synchronize it to the construction equipment control system. If the verification fails, return to step S3 for re-optimization. S5. Construction Verification and Data Feedback Iteration: After completing the current blasting operation, blasting effect data and construction process data are collected and fed back to the AI ​​multi-objective optimization model through the gradient descent algorithm. The reward function weights and parameter optimization rules of the model are iteratively updated to provide optimization basis for the design of the next blast, realizing a closed-loop iteration of one blast, one optimization.

2. The intelligent one-shot-one-design method for highway tunnel blasting according to claim 1, characterized in that: In step S1, three-dimensional point cloud data and joint detail image data of the rock mass are acquired through a collaborative method of three-dimensional laser scanning and digital photogrammetry. The scanning accuracy of the three-dimensional laser scanning unit is ≤ ±2mm, and the scanning range includes the current working face and an area 5m outside the excavation outline. The digital photogrammetry unit uses three industrial cameras to simultaneously capture images of the working face from the front, left 45°, and right 45° respectively. The physical and mechanical parameters of the rock mass are detected on-site using a portable rock mass strength tester. After standardized preprocessing, the acquired data forms a dedicated dataset for each blast, avoiding interference between data from different blasts.

3. The intelligent one-shot-one-design method for highway tunnel blasting according to claim 1, characterized in that: The AI ​​recognition model in step S2 adopts a deep learning architecture that integrates PointNet++ and CNN. The data processing flow of this recognition model is as follows: the three-dimensional point cloud data is voxelized to a size of 0.05m×0.05m×0.05m, and the geometric features of the rock mass are extracted; the Canny edge detection algorithm is used to extract the joint edge features of the image data; finally, the rock mass geometric features and joint edge features are fused through the attention mechanism to strengthen the feature weight of the joint area and suppress noise interference.

4. The intelligent one-shot-one-design method for highway tunnel blasting according to claim 1, characterized in that: The AI ​​multi-objective optimization model in step S3 is constructed based on a reinforcement learning algorithm. It uses the blasting hole location coordinates, inclination angle, drilling depth, single-hole charge amount, and detonation timing parameters as the action space, and the joint avoidance rate ≥98%, over-drilling / under-drilling amount ≤5cm, half-hole rate ≥85%, and blasting vibration velocity ≤10cm / s as the reward function. The parameter optimization rules are: the blasting hole location avoids dense joint areas by ≥0.3m, the inclination angle is ≥60° with the joint strike, the drilling depth is adapted to the joint layer thickness, the single-hole charge amount is adjusted according to the rock mass compressive strength gradient, and the detonation time difference is 50-200ms.

5. The intelligent one-shot-one-design method for highway tunnel blasting according to claim 1, characterized in that: The AI ​​multi-objective optimization model in step S3 has a built-in blasting parameter database and transfer learning module: the database stores the historical best single-blast design parameters for different geological conditions and tunnel cross sections; The transfer learning module can transfer the basic model trained on historical data to the current engineering scenario.

6. The intelligent one-shot-one-design method for highway tunnel blasting according to claim 1, characterized in that: The equipment operating condition adaptation verification standards in step S4 are: drilling depth ≤ 5m, tilt angle adjustment range 0-90°, and charge accuracy ≤ ±2%.

7. The intelligent one-shot-one-design method for highway tunnel blasting according to claim 1, characterized in that: The blasting effect data in step S5 includes the over-excavation and under-excavation amount and half-hole ratio detected by three-dimensional laser scanning, and the blasting vibration velocity collected by distributed acceleration sensors; the construction process data includes the actual drilling position deviation, charge amount error, and detonation timing deviation. All collected data are quantified and labeled before being incorporated into the model iteration.

8. The intelligent one-shot-one-design method for highway tunnel blasting according to claim 1, characterized in that: Both the AI ​​recognition model in step S2 and the AI ​​multi-objective optimization model in step S3 are lightweight edge models with ≤5G of model parameters. They are equipped with industrial-grade edge computing terminals that are dustproof, vibration-resistant, and have wide voltage power supply, enabling local data collection, local computation, and local distribution without network dependence. The edge terminal has a reserved 5G / Bluetooth module to support data cloud upload.

Images