Blasting effect monitoring peripheral hole blasting optimization method and system based on deep learning

Abstract

The invention relates to the technical field of blasting engineering and deep learning, in particular to a blasting effect monitoring peripheral hole blasting optimization method and system based on deep learning. In order to solve the problems that an intelligent and automatic optimization mechanism is lacked in blasting parameter design and a monitoring means is lagged, a blasting defect rate prediction model is constructed based on an SO-DBN model, and geological related parameters, peripheral hole related parameters and outer ring auxiliary hole related parameters of an existing blasting scheme are input into the blasting defect rate prediction model; predicting the residual mark rate after blasting; when the predicted residual mark rate meets a preset standard, a blasting scheme is output; a Faster Net-YOLOv8 model is used to carry out identification and information extraction on the residual mark rate image after blasting; according to the identified image, calculating the actual residual mark rate, and supplementing the relevant parameters with the actual residual mark rate error being less than 5% into the initial database;

Description

Technical Field

[0001] The present invention relates to the field of blasting engineering and deep learning technology, and in particular to a method and system for optimizing blasting effects monitoring and peripheral eye blasting based on deep learning. Background Art

[0002] With the large-scale advancement of projects such as tunneling and mining, blasting technology has been widely used in civil engineering and the mining industry. Traditional blasting operations rely on experience and theoretical calculations to design blasting parameters. However, due to factors such as complex geological conditions, difficulty integrating multi-source data, and differences in blasting areas, precise control of blasting results is difficult, and post-blasting residue rates and over-excavation are prominent issues. Irrational blasting parameter design not only reduces blasting efficiency but also may lead to safety hazards and economic losses. The rapid development of data science and artificial intelligence technologies, especially the application of deep learning models in image recognition and data mining, has brought new opportunities for the intelligent and automated optimization of traditional blasting technology.

[0003] Currently, many blasting plans rely on limited field test data and expert experience, failing to effectively integrate complex geological and blasting data for global optimization. Furthermore, existing blasting effect monitoring often relies on manual calculations or simple image recognition algorithms, which are inefficient and lack real-time feedback on blasting results, making it difficult to quickly adjust blasting plans. Therefore, developing an intelligent blasting effect monitoring system based on deep learning would not only improve blasting accuracy but also reduce post-blasting scarring and overexcavation by optimizing blasting parameters in real time. This is crucial for improving the safety and cost-effectiveness of blasting operations.

[0004] Under the current technical background, there are the following major shortcomings:

[0005] Disadvantage 1:

[0006] Blasting parameter design relies on experience and lacks intelligent and automated optimization mechanisms: In traditional blasting operations, blasting parameter design relies heavily on engineers' experience and simple theoretical calculations, failing to fully utilize modern intelligent technologies for system optimization. Due to the complexity of geological conditions and the variability of blasting areas, manually designed blasting plans often fail to accurately match actual scenarios, resulting in significant deviations in blasting results. Common problems include high residual scar rates and significant overexcavation, which directly impact project quality and safety.

[0007] Traditional solutions struggle to adapt to complex and changing geological environments, with high adjustment costs and long cycles, making it difficult to ensure continuous optimization and efficient execution of blasting operations. Furthermore, the multi-source data involved in the blasting process (such as geological parameters, parameters of peripheral holes, and outer ring auxiliary holes) is not fully integrated and analyzed, making it difficult to achieve automated, precise blasting plan design and optimization.

[0008] Disadvantage 2:

[0009] Lagged monitoring methods prevent real-time feedback on blasting results and adjustments to plans: Existing blasting monitoring methods primarily rely on manual measurement and traditional image analysis techniques. These methods are inefficient and cannot provide fast and accurate feedback on blasting results. Post-blasting residual rates and overexcavation typically require manual inspection and calculation, resulting in long feedback cycles and difficulty implementing real-time adjustments at the blasting site. This delayed feedback mechanism often means that when blasting results are unsatisfactory, adjustments are often too late, resulting in safety risks for subsequent blasting.

[0010] In addition, although remote sensing technology and multi-source data collection have made great progress, the existing system is difficult to fully integrate data from remote sensing, geology and other aspects for effective monitoring, resulting in insufficient accuracy and response speed in blasting plan optimization, which limits the comprehensive application of intelligent monitoring and automated optimization. Summary of the Invention

[0011] In response to the shortcomings pointed out in the above background technology, this application proposes the following solutions:

[0012] Solution 1:

[0013] This paper proposes a deep learning-based intelligent blasting parameter optimization method and system. To address the problem of blasting parameter design relying on experience and lacking intelligence, this application proposes a deep learning-based blasting optimization system that can fully utilize multi-source geological data and blasting parameters to automatically design and optimize blasting plans. By introducing a deep belief network (DBN) and a snake optimization algorithm (SO), the system can effectively analyze complex geological conditions and generate and adjust blasting parameters in real time.

[0014] The system no longer relies on traditional empirical judgment. Instead, it uses extensive historical data and model training to achieve intelligent parameter design, ensuring that blasting plans remain adaptable to complex and changing geological environments. Furthermore, the system dynamically optimizes blasting plans by continuously learning from updated data, reducing residual damage and overexcavation, thereby improving blasting efficiency and safety.

[0015] Solution 2:

[0016] To address the lag and untimely feedback issues of existing monitoring methods, this proposed blasting effect feedback and optimization mechanism, based on real-time monitoring, employs advanced image recognition technology (such as FasterNet-YOLOv8) to monitor the post-blasting residual rate in real time and provide rapid feedback on the blasting effect. The system integrates a multi-source data fusion module capable of instantly processing multi-dimensional data from remote sensing, on-site imagery, geology, and other sources, automatically calculating the post-blasting residual rate and overexcavation, and comparing these with the model's predicted results.

[0017] If the system detects unsatisfactory blasting results, it automatically adjusts the parameters of the peripheral holes and outer ring auxiliary holes to optimize subsequent blasting plans. This closed-loop optimization mechanism ensures real-time adjustments to blasting operations, enhances the automation and intelligence of the entire process, and minimizes safety risks during blasting operations.

[0018] Based on the above technical ideas, the technical solution adopted by the present invention is:

[0019] One object of the present invention is to provide a method for optimizing blasting effects monitoring and peripheral eye blasting based on deep learning, comprising:

[0020] S1. Construct a blasting scar rate prediction model based on the SO-DBN model;

[0021] S2. Inputting geological parameters, peripheral hole parameters, and outer ring auxiliary hole parameters of the existing blasting plan into a blasting scar rate prediction model to obtain a predicted scar rate;

[0022] If the predicted residual rate meets the preset standard, the blasting plan is output;

[0023] If not, adjust the geological parameters, peripheral hole parameters and outer ring auxiliary hole parameters, and continue to input them into the blasting scar rate prediction model until the predicted scar rate meets the preset standard and output the blasting plan;

[0024] S3. Using the output blasting plan, blast the tunnel face on site to obtain a post-blasting scar rate image;

[0025] S4. Build an optimized FasterNet-YOLOv8 model;

[0026] S5. Use the optimized FasterNet-YOLOv8 model to identify and extract information from the scar rate image after blasting to obtain the actual scar rate of the tunnel face;

[0027] S6. performing error analysis between the obtained actual scar rate of the tunnel face and the predicted scar rate;

[0028] If the error is less than 5%, the actual scar rate and the geological parameters of the blasting plan, the parameters of the peripheral holes, and the parameters of the outer ring auxiliary holes are added to the initial database to obtain an optimized blasting database, realizing fully automated blasting plan optimization and automatic calculation of the scar rate, and improving the generalization ability of the blasting scar rate prediction model;

[0029] If the error is greater than or equal to 5%, the relevant data of the blasting plan will be discarded.

[0030] To further define the above technical solution, in step S1, the SO-DBN model is constructed by the following steps:

[0031] S101: Use the snake optimization algorithm to optimize the deep belief network model and obtain the optimized parameters;

[0032] S102: Input the optimized parameters into the deep belief network model, train the restricted Boltzmann machine layer by layer, and perform supervised training on the neural network through back propagation to obtain the SO-DBN model.

[0033] To further define the above technical solution, S101 includes the following steps:

[0034] S1011: In the exploration phase of the snake optimization algorithm, the candidate solution of the deep belief network model is updated to obtain the updated candidate solution, and the initial hyperparameters of the deep belief network model are determined through extensive search;

[0035] S1012: During the development phase of the snake optimization algorithm, the updated candidate solution is optimized, and the initial hyperparameters are globally and locally optimized based on the optimized candidate solution to obtain optimized parameters.

[0036] To further limit the above technical solution, in step S2, the preset standard is that the hard rock scar rate is ≥80%; the medium hard rock scar rate is ≥50%; and the soft rock scar rate is ≥20%.

[0037] To further define the above technical solution, step S4 includes the following steps:

[0038] S401: Based on the residual structure FasterNet-Block of the FasterNet network, the bottleneck residual structure in the C2f structure of the FasterNet-YOLOv8 model is replaced with the residual structure FasterNet-Block;

[0039] Among them, FasterNet-Block consists of a partial convolution and two convolutional neural networks.

[0040] To further limit the above technical solution, step S5 includes the following steps:

[0041] S501: Select one of the residues, convert the size of the residue rate image after the explosion into the display size, and calculate the conversion scale coefficient k;

[0042] S502: Input the post-blasting scar rate image into the optimized FasterNet-YOLOv8 model, use the detection frame in the FasterNet-YOLOv8 model to identify and extract information from the post-blasting scar rate image, and obtain the coordinates of the upper left corner and lower right corner of the scar detection frame;

[0043] S503: Calculate the pixel length of the residue according to the coordinates of the upper left corner and the lower right corner of the residue detection frame;

[0044] S504: Convert the pixel length of the scar to a display size according to the scale factor k to obtain the actual length of the scar. If the actual length of the scar is greater than 70% of the designed depth of the blasthole, it is considered a visible scar.

[0045] S505: Repeat the above steps S501-S504 to calculate the actual length of the remaining scars in sequence, and count the number of all visible scars in the scar rate image. That is, the scar rate of the tunnel face is obtained by the ratio of the number of visible scars to the number of blastholes.

[0046] To further limit the above technical solution, in step S6, the initial database includes:

[0047] Peripheral eye related parameters, outer ring auxiliary eye related parameters, geological related parameters and post-blasting scar rate.

[0048] Another object of the present invention is to provide a blasting effect monitoring and peripheral eye blasting optimization system based on deep learning, which specifically includes:

[0049] The scar rate prediction module is used to build a blasting scar rate prediction model based on the SO-DBN model, and output the predicted scar rate according to the geological related parameters of the existing blasting plan, the related parameters of the peripheral holes and the related parameters of the outer circle auxiliary holes;

[0050] The parameter adjustment module is used to dynamically adjust the geological parameters, the parameters related to the peripheral holes, and the parameters related to the outer ring auxiliary holes. If the predicted scar rate meets the preset standard, the blasting plan is output; if the predicted scar rate does not meet the preset standard, the parameters are updated and re-input into the scar rate prediction module until the predicted scar rate meets the preset standard, and the blasting plan is output;

[0051] The output module is used to execute and output a blasting plan that meets the preset standards, perform on-site blasting on the tunnel face, and generate a post-blasting scar rate image;

[0052] The image processing module is used to identify and extract information from the scar rate image after blasting based on the optimized FasterNet-YOLOv8 model to obtain the actual scar rate of the tunnel face;

[0053] The error analysis module is used to perform error analysis on the actual scar rate of the tunnel face and the predicted scar rate, and to judge the optimization effect based on the error results.

[0054] Furthermore, it also includes:

[0055] The database module is used to store the peripheral eye parameters, outer circle auxiliary eye parameters and their corresponding scar rate data of the initial blasting plan.

[0056] Furthermore, it also includes:

[0057] The optimization module is used to add the actual scar rate and the geological parameters of the blasting plan, the peripheral eye parameters and the outer circle auxiliary eye parameters to the database module when the error is less than 5%, expand and optimize the database module, and thus enhance the generalization ability of the blasting scar rate prediction model.

[0058] Compared with the prior art, the technical innovation and random classification method introduced in this invention bring the following significant effects:

[0059] 1. Realizes intelligent and automated optimization of blasting parameters: By introducing deep learning technology, the present invention can fully integrate multi-source geological and blasting data, automatically generate the optimal blasting parameter design scheme, and reduce dependence on engineers' experience. By combining the deep belief network (DBN) with the snake optimization algorithm (SO), the system can accurately predict the blasting effect in a complex and changeable geological environment, automatically optimize the blasting parameters, and reduce the incidence of residual rate and over-excavation. This intelligent and automated design scheme not only improves the accuracy of blasting operations, but also greatly improves work efficiency and construction safety. In addition, the system can dynamically learn and adjust the blasting scheme, so that each blasting operation is constantly optimized, solving the problem that traditional methods cannot adapt to complex environments.

[0060] 2. Provides a real-time blasting effect monitoring and feedback mechanism: The present invention adopts advanced image recognition technology (FasterNet-YOLOv8) and multi-source data fusion to achieve real-time monitoring and feedback of blasting effects, overcoming the feedback lag problem of existing technologies. Through real-time analysis of the residual rate after blasting, the system can quickly detect blasting effects that do not meet expectations, and optimize subsequent blasting operations by automatically adjusting the parameters of the peripheral eyes and outer ring auxiliary eyes. This real-time feedback and automatic adjustment mechanism ensures the continuity and safety of the blasting process, greatly improves operating efficiency, and reduces the time delay of human intervention. Compared with the traditional manual monitoring method that requires manual measurement of length to calculate the residual rate, the use of image recognition and calculation can eliminate the manual measurement process, further improving the reliability and safety of the blasting effect.

[0061] 3. Improved blasting effect prediction and model generalization capabilities: By combining multi-source data and deep learning algorithms, the present invention not only achieves accurate prediction of blasting effects, but also continuously updates and optimizes the database, thereby improving the generalization capabilities of the model. Through the learning and training of a large amount of historical blasting data by the deep learning model, the system can adapt to various complex geological conditions and significantly improve the prediction accuracy of blasting effects. In addition, with the automatic collection and updating of each blasting data, the model can self-adjust and optimize according to the new data to ensure that the prediction model is always in the optimal state. This dynamic learning capability enables the system to quickly adapt to different types of blasting operations, enhances the robustness and wide applicability of the blasting plan, and thus further improves blasting efficiency and construction safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0063] Figure 1 This is a flow chart of the peripheral eye blasting optimization method for blasting effect monitoring based on deep learning of the present invention.

[0064] Figure 2 This is the DBN network structure diagram of the blasting effect monitoring peripheral eye blasting optimization method based on deep learning.

[0065] Figure 3 FasterNet-YOLOv8 model diagram for the deep learning-based blasting effect monitoring peripheral eye blasting optimization method.

[0066] Figure 4This is the optimized C2f structure diagram of the peripheral eye blasting optimization method for blasting effect monitoring based on deep learning.

[0067] Figure 5 Diagram of the image-based scar rate calculation method for the peripheral eye blasting optimization method for blasting effect monitoring based on deep learning.

[0068] Figure 6a This is the model fitting effect of the tunnel maximum linear overbreak prediction model training process.

[0069] Figure 6b This is the model fitting effect of the tunnel maximum linear overbreak prediction model prediction process.

[0070] Figure 6c It is the error loss diagram of the SO-DBN model.

[0071] Figure 7 The figure is a scene blasting diagram in an engineering example in which the present invention is applied.

[0072] Figure 8 The figure is an on-site blasting analysis diagram in an engineering example in which the present invention is applied.

[0073] Figure 9 The figure is a tunnel face contour diagram in an engineering example in which the present invention is applied.

[0074] Figure 10 The figure is a peripheral blast hole trace diagram in an engineering example of applying the present invention. DETAILED DESCRIPTION

[0075] This application proposes a method for optimizing blasting around holes using deep learning-based blasting effect monitoring. This method is primarily designed to address the challenges of traditional blasting operations, which rely on experience, have difficulty accurately controlling blasting effects, and cannot optimize blasting parameters in real time. This application provides an effective and innovative method based on multi-source data fusion and deep learning models. Through automated blasting parameter optimization and real-time monitoring, it achieves the effects of improving blasting accuracy, reducing residual scar rates and over-excavation, and significantly improving the efficiency and safety of blasting operations. This method is effectively applicable to blasting operations in complex geological environments.

[0076] Deep learning and intelligent optimization algorithms are used to address the issues of blasting effect monitoring and peripheral eye blasting optimization. First, an initial database is constructed based on geological parameters, peripheral eye parameters, outer ring auxiliary eye parameters, and post-blasting scar rate. Next, a blasting scar rate prediction model is established using a deep belief network (DBN) model optimized using the snake optimization algorithm (SO). By inputting the peripheral eye and outer ring auxiliary eye parameters of the existing blasting plan, the model predicts the post-blasting scar rate. If the predicted result does not meet the preset standards, the system automatically adjusts the parameters and re-predicts. Subsequently, the FasterNet-YOLOv8 model is used to identify the post-blasting scar image, calculate the actual scar rate, and perform an error analysis compared to the predicted value. If the error is less than the preset value, the database is updated and the model is optimized to ensure the accuracy and reliability of subsequent blasting plans.

[0077] Furthermore, the proposed deep learning-based blasting effect monitoring and peripheral eye blasting optimization method addresses the challenges of traditional blasting operations, such as reliance on experience for parameter design, delayed monitoring, and untimely feedback. This application provides a comprehensive intelligent blasting optimization method for complex geological environments. A suitable system for this method, built using the SO-DBN and FasterNet-YOLOv8 models, avoids issues such as high blasting scar rates, severe overexcavation, and the inability to adjust blasting plans in real time, often caused by insufficient manual experience or improper data processing.

[0078] Below in conjunction with specific embodiment, further set forth the present invention.Should be understood that these embodiments are only used to illustrate the present invention and are not used in limiting the scope of the present invention.In addition, should be understood that after reading the content taught by the present invention, those skilled in the art will make various changes or modifications to the present invention, and these equivalent forms fall equally within the scope limited by the appended claims of the application.

[0079] Example 1

[0080] like Figure 1-10 As shown, this embodiment provides a method and system for optimizing blasting effects and peripheral eye blasting based on deep learning.

[0081] Among them, Figure 1 As shown, specifically including:

[0082] S1. Construct a blasting scar rate prediction model based on the SO-DBN model;

[0083] S2. Inputting geological parameters, peripheral hole parameters, and outer ring auxiliary hole parameters of the existing blasting plan into a blasting scar rate prediction model to obtain a predicted scar rate;

[0084] If the predicted residual rate meets the preset standard, the blasting plan is output;

[0085] If not, adjust the geological parameters, peripheral hole parameters and outer ring auxiliary hole parameters, and continue to input them into the blasting scar rate prediction model until the predicted scar rate meets the preset standard and output the blasting plan;

[0086] S3. Using the output blasting plan, blast the tunnel face on site to obtain a post-blasting scar rate image;

[0087] S4. Build an optimized FasterNet-YOLOv8 model;

[0088] S5. Use the optimized FasterNet-YOLOv8 model to identify and extract information from the scar rate image after blasting to obtain the actual scar rate of the tunnel face;

[0089] S6. performing error analysis between the obtained actual scar rate of the tunnel face and the predicted scar rate;

[0090] If the error is less than 5%, the actual scar rate and the geological parameters of the blasting plan, the parameters of the peripheral holes, and the parameters of the outer ring auxiliary holes are added to the initial database to obtain an optimized blasting database, realizing fully automated blasting plan optimization and automatic calculation of the scar rate, and improving the generalization ability of the blasting scar rate prediction model;

[0091] If the error is greater than or equal to 5%, the relevant data of the blasting plan will be discarded.

[0092] In an embodiment of the present invention, in step S1, the SO-DBN model is constructed by the following steps:

[0093] S101: Use the snake optimization algorithm to optimize the deep belief network model and obtain the optimized parameters;

[0094] S102: Input the optimized parameters into the deep belief network model, train the restricted Boltzmann machine layer by layer, and perform supervised training on the neural network through back propagation to improve the precision and accuracy of the deep belief network model to obtain the SO-DBN model.

[0095] S101 includes the following steps:

[0096] S1011: In the exploration phase of the snake optimization algorithm, the candidate solution of the deep belief network model is updated to obtain the updated candidate solution, and the initial hyperparameters of the deep belief network model are determined through extensive search;

[0097] S1012: During the development phase of the snake optimization algorithm, the updated candidate solution is optimized, and the initial hyperparameters are globally and locally optimized based on the optimized candidate solution to obtain optimized parameters.

[0098] In the embodiment of the present invention, in step S2, the preset standard is that the scar rate of hard rock is ≥80%; the scar rate of medium-hard rock is ≥50%; and the scar rate of soft rock is ≥20%.

[0099] Specifically, the SO-DBN model uses the Snake Optimization (SO) algorithm to optimize the DBN model. SO is a meta-heuristic swarm optimization algorithm proposed by Hashim et al. in 2022. As the name suggests, it is inspired by the foraging and reproduction behavior patterns of snakes in nature. The authors abstracted this from a mathematical perspective and proposed a simple and efficient hyperparameter optimization algorithm. The specific calculation principle is as follows:

[0100] (1) Exploration phase (Q < 0.25, Q = c1 × exp(tT / T)) The position update formula is shown in equations (1) and (2).

[0101] X i,m (t+1)=X rand,m (t)±c2×A m ×((X max -X min )×rand+X min ) (1)

[0102]

[0103] Where: c1 = 0.5, t - current iteration, T - maximum number of iterations, X i,m —Position of the ith male, X rand,m —Position of a random male, A m —The predatory ability of individual male snakes, The fitness value, f i m — The fitness value, rand—a random number in [0,1], X max — Upper limit, X min —lower limit, c2—0.05, Q—food quantity.

[0104] (2) Development stage (Q>0.25) "Searching for food" (Temp>0.6, Temp=exp(-t / T)) as shown in formula (3).

[0105] X i,j (t+1)=X food ±c3×Temp×rand(X food -X i,j (t)) (3)

[0106] Where: c3=2, X i,j —Individual position, Xfood —Global optimal individual position, Temp—temperature.

[0107] (3) The movement formulas for the “combat mode” (rand<0.6) in the development stage are shown in equations (4) and (5).

[0108] X i,m (t+1)=X i,m (t)±c3×FM×rand×(Q×X best,f -X i,m (t)) (4)

[0109]

[0110] Where: FM—male combat capability, X best,f —The optimal position for females, —Fitness value of the female's optimal position, c3-2, f i —Fitness value of individual i.

[0111] (4) The movement formula of the “mating mode” (rand>0.6) in the development stage is shown in Equations (6) and (7).

[0112] X i,m (t+1)=X i,m (t)±c3×M m ×rand×(Q×X i,f -X i,m (t)) (6)

[0113] M m =exp(-f i f / f i m ) (7)

[0114] Where: M m —Male mating ability, X i,f —position of the ith female, f i f 、f i m —Search agents for females and males, if larger than the initial state, are replaced;

[0115] Deep Belief Network (DBN) is a neural network model that can be trained both supervised and unsupervised, and can solve high-dimensional nonlinear function problems. DBN consists of an unsupervised lower layer of several RBMs stacked together and a supervised upper layer composed of a BP neural network. During training, the output of the previous RBM is the input of the next RBM, and training is done layer by layer to the supervised upper layer. Through reverse supervision training of the neural network, parameter fine-tuning is performed to improve the precision and accuracy of the model. The DBN network structure is as follows: Figure 2 shown.

[0116] In an embodiment of the present invention, step S4 includes the following steps:

[0117] S501: Based on the residual structure FasterNet-Block of the FasterNet network, the bottleneck residual structure in the C2f structure of the FasterNet-YOLOv8 model is replaced with the residual structure FasterNet-Block;

[0118] Among them, FasterNet-Block mainly consists of a partial convolution PConv and two convolutional neural networks Conv (such as Figure 3-4 ).

[0119] Specifically, the residual structure is a short-circuit structure (Shortcut) proposed to solve the model training difficulties caused by network stacking and the large training errors of deep network models. By referring to the residual structure FasterNet-Block in the FasterNet network structure, it is introduced into the C2f structure of the YOLOv8 network. The optimized C2f module is as follows: Figure 4 shown.

[0120] Depend on Figure 4 As can be seen, the FasterNet-Block architecture replaces the bottleneck residual structure in C2f. FasterNet-Block primarily consists of a PConv (partial convolution) and two Conv (convolutional neural networks). PConv (partial convolution) only extracts regular features from a portion of the input. Compared to Conv networks, it focuses more on the center position, thus reducing memory usage and improving speed.

[0121] In an embodiment of the present invention, step S5 includes:

[0122] S501: Select one of the residues, convert the size of the residue rate image after the blasting to the display size, and calculate the conversion scale coefficient K;

[0123] Image size conversion such as Figure 5As shown, based on the known upper step height and the pixel coordinates of the height in the image, the image and display size are converted. The proportional coefficient calculation formula is:

[0124] Where H is the actual step height; (X2, Y2) is the coordinate of the top of the step, (X1, Y1) is the coordinate of the lower right corner of the image, and K is the conversion coefficient.

[0125] S502: Input the post-blasting scar rate image into the optimized FasterNet-YOLOv8 model, use the detection frame in the FasterNet-YOLOv8 model to identify and extract information from the post-blasting scar rate image, obtain the coordinates of the upper left corner and lower right corner of the scar detection frame, and calculate the pixel points.

[0126] S503: Calculate the pixel length of the residue according to the coordinates of the upper left corner and the lower right corner of the residue detection frame;

[0127]

[0128] Among them, (x3, y3) is the coordinate of the upper left corner of the detection box, (x4, y4) is the coordinate of the lower right corner of the detection box, and l is the pixel length of the residue.

[0129] S504: Convert the pixel length of the scar to a display size according to the scale factor K to obtain the actual length of the scar. If the actual length of the scar is greater than 70% of the designed depth of the blasthole, it is considered a visible scar.

[0130] The actual length of the scar is calculated as follows: L = kl;

[0131] Where L is the actual length of the scar.

[0132] S505: Repeat the above steps S501-S504 to calculate the actual length of the remaining scars in sequence, and count the number of all visible scars in the scar rate image. That is, the scar rate of the tunnel face is obtained by the ratio of the number of visible scars to the number of blastholes.

[0133] In an embodiment of the present invention, in step S6, the initial database includes: peripheral eye related parameters, outer ring auxiliary eye related parameters, geological related parameters and post-blasting scar rate.

[0134] Example 2

[0135] like Figure 1-10As shown, in this embodiment, a deep learning-based blasting effect monitoring and peripheral eye blasting optimization system is built based on the SO-DBN model and the FasterNet-YOLOv8 model to achieve the effect of accurately predicting the residual rate after blasting and automatically optimizing the blasting parameters, so as to achieve the purpose of improving blasting accuracy, reducing the residual rate and over-excavation phenomenon, and improving the safety and efficiency of blasting operations.

[0136] Specifically, the optimization system includes:

[0137] The scar rate prediction module is used to build a blasting scar rate prediction model based on the SO-DBN model, and output the predicted scar rate according to the geological related parameters of the existing blasting plan, the related parameters of the peripheral holes and the related parameters of the outer circle auxiliary holes;

[0138] The parameter adjustment module is used to dynamically adjust the geological parameters, the parameters related to the peripheral holes, and the parameters related to the outer ring auxiliary holes. If the predicted scar rate meets the preset standard, the blasting plan is output; if the predicted scar rate does not meet the preset standard, the parameters are updated and re-input into the scar rate prediction module until the predicted scar rate meets the preset standard, and the blasting plan is output;

[0139] The output module is used to execute and output a blasting plan that meets the preset standards, perform on-site blasting on the tunnel face, and generate a post-blasting scar rate image;

[0140] The image processing module is used to identify and extract information from the scar rate image after blasting based on the optimized FasterNet-YOLOv8 model to obtain the actual scar rate of the tunnel face;

[0141] The error analysis module is used to perform error analysis on the actual scar rate of the tunnel face and the predicted scar rate, and to judge the optimization effect based on the error results.

[0142] Specifically, the optimization system also includes:

[0143] The database module is used to store the peripheral eye parameters, outer circle auxiliary eye parameters and their corresponding scar rate data of the initial blasting plan.

[0144] Specifically, the optimization system also includes:

[0145] The optimization module is used to add the actual scar rate and the geological parameters of the blasting plan, the peripheral eye parameters and the outer circle auxiliary eye parameters to the database module when the error is less than 5%, expand and optimize the database module, and thus enhance the generalization ability of the blasting scar rate prediction model.

[0146] Examples of engineering examples:

[0147] 1. Initial database:

[0148] In summary, the integrity coefficient A1, uniaxial compressive strength A2 (MPa), structural surface inclination A3, structural surface position A4, peripheral eye charge A5 (kg), peripheral eye hole spacing A6 (cm), outer ring auxiliary eye hole spacing A7 (cm), outer ring auxiliary eye charge A8 (kg), and smooth blasting layer thickness A9 (cm) are selected as the input parameters of the tunnel smooth blasting design parameter optimization model, as shown in Table 1 below.

[0149] Table 1 Some databases

[0150]

[0151]

[0152] 2. Model training results

[0153] Figure 6 shows the regression test results of the training set, test set, and model error loss of the tunnel maximum linear overbreak prediction model. Figure 6a 、 Figure 6b The two figures show the scatter plot relationship between the predicted value and the measured value. From the analysis, we can see that the correlation coefficients of the maximum linear over-mining in the training set and the test set are 0.9887 and 0.9843 respectively, which are strong linear correlations. Figure 6c This shows the error loss of the SO-DBN model. Analysis shows that after 800 iterations, the error loss stabilizes and approaches 0. The above analysis shows that the model has a strong correlation with the sample dataset, indicating that the established maximum linear overbreak model for tunnels is sufficiently reliable.

[0154] 3. The blasting parameter output and prediction are shown in Table 2:

[0155] Table 2 Blasting parameter output and prediction

[0156]

[0157] 4. On-site blasting effect.

[0158] Through Figures 7 to 10Analysis shows that after multiple blasting operations, the face contour line is highly flat, and the traces of the surrounding blastholes remain intact and clear, indicating that the range of surrounding rock disturbance is well controlled and the missing parts are significantly reduced. This shows that the selection and implementation of blasting design parameters have achieved an optimization effect, avoiding the occurrence of large-scale over-excavation or under-excavation. This feature not only ensures a significant reduction in the amount of material used in the subsequent shotcrete process, directly reducing construction costs, but also greatly improves process efficiency. In addition, the overall flatness of the face provides ideal construction conditions for subsequent support and lining operations, reducing the workload of additional adjustments, thereby improving the overall efficiency of the project construction. The above results further verify the effectiveness and importance of precision blasting technology in tunnel construction, and provide a strong guarantee for subsequent construction processes.

[0159] 5. Calculation of residual rate: Based on the model recognition results, the number of residual eyes is calculated in sequence according to the formula, and the residual rate is 88% with an error of 3%.

[0160] 6. Expand the scar rate and design parameters to the original database, and the optimization scheme is shown in Table 3:

[0161] Table 3 Residue rate and design parameter optimization scheme

[0162]

[0163] The above content is a further detailed description of the present invention in conjunction with specific preferred embodiments, which is convenient for those skilled in the art to understand and apply the present invention. It cannot be determined that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art of the present invention, several simple deductions or substitutions can be made without departing from the concept of the present invention, without having to go through creative work. Therefore, based on the disclosure of the present invention, simple improvements made to the present invention by those skilled in the art should be within the scope of protection of the present invention.

Claims

1. A method for optimizing blasting effects and peripheral eye blasting based on deep learning, characterized by: include: S1. Construct a blasting scar rate prediction model based on the SO-DBN model; S2. Inputting geological parameters, peripheral hole parameters, and outer ring auxiliary hole parameters of the existing blasting plan into a blasting scar rate prediction model to obtain a predicted scar rate; If the predicted scar rate meets the preset standard, the blasting plan is output; If not, adjust the geological parameters, peripheral hole parameters and outer ring auxiliary hole parameters, and continue to input them into the blasting scar rate prediction model until the predicted scar rate meets the preset standard and output the blasting plan; S3. Using the output blasting plan, blast the tunnel face on site to obtain a post-blasting scar rate image; S4. Build an optimized FasterNet-YOLOv8 model; S5. Use the optimized FasterNet-YOLOv8 model to identify and extract information from the scar rate image after blasting to obtain the actual scar rate of the tunnel face; S6. performing error analysis between the obtained actual scar rate of the tunnel face and the predicted scar rate; If the error is less than 5%, the actual scar rate and the geological parameters of the blasting plan, the parameters of the peripheral holes, and the parameters of the outer ring auxiliary holes are added to the initial database to obtain an optimized blasting database, realizing fully automated blasting plan optimization and automatic calculation of the scar rate, and improving the generalization ability of the blasting scar rate prediction model; If the error is greater than or equal to 5%, the relevant data of the blasting plan will be discarded.

2. The method for optimizing blasting effect monitoring and peripheral eye blasting based on deep learning according to claim 1 is characterized in that: In step S1, the SO-DBN model is constructed through the following steps: S101: Use the snake optimization algorithm to optimize the deep belief network model and obtain the optimized parameters; S102: Input the optimized parameters into the deep belief network model, train the restricted Boltzmann machine layer by layer, and perform supervised training on the neural network through back propagation to obtain the SO-DBN model.

3. The method for optimizing blasting effect monitoring and peripheral eye blasting based on deep learning according to claim 2 is characterized in that: S101 includes the following steps: S1011: In the exploration phase of the snake optimization algorithm, the candidate solution of the deep belief network model is updated to obtain the updated candidate solution, and the initial hyperparameters of the deep belief network model are determined through extensive search; S1012: During the development phase of the snake optimization algorithm, the updated candidate solution is optimized, and the initial hyperparameters are globally and locally optimized based on the optimized candidate solution to obtain optimized parameters.

4. The method for optimizing blasting effect monitoring and peripheral eye blasting based on deep learning according to claim 1 is characterized in that: In step S2, the preset standard is that the scar rate of hard rock is ≥80%; the scar rate of medium-hard rock is ≥50%; and the scar rate of soft rock is ≥20%.

5. The method for optimizing blasting effect monitoring and peripheral eye blasting based on deep learning according to claim 1 is characterized in that: Step S4 includes the following steps: S401: Based on the residual structure FasterNet-Block of the FasterNet network, the bottleneck residual structure in the C2f structure of the FasterNet-YOLOv8 model is replaced with the residual structure FasterNet-Block; Among them, FasterNet-Block consists of a partial convolution and two convolutional neural networks.

6. The method for optimizing blasting effect monitoring and peripheral eye blasting based on deep learning according to claim 1 is characterized in that: Step S5 includes the following steps: S501: Select one of the residues, convert the size of the residue rate image after the explosion into the display size, and calculate the conversion scale coefficient k; S502: Input the post-blasting scar rate image into the optimized FasterNet-YOLOv8 model, use the detection frame in the FasterNet-YOLOv8 model to identify and extract information from the post-blasting scar rate image, and obtain the coordinates of the upper left corner and lower right corner of the scar detection frame; S503: Calculate the pixel length of the residue according to the coordinates of the upper left corner and the lower right corner of the residue detection frame; S504: Convert the pixel length of the scar to a display size according to the scale factor k to obtain the actual length of the scar. If the actual length of the scar is greater than 70% of the designed depth of the blasthole, it is considered a visible scar. S505: Repeat the above steps S501-S504 to calculate the actual length of the remaining scars in sequence, and count the number of all visible scars in the scar rate image. That is, the scar rate of the tunnel face is obtained by the ratio of the number of visible scars to the number of blastholes.

7. The method for optimizing blasting effect monitoring and peripheral eye blasting based on deep learning according to claim 1 is characterized in that: In step S6, the initial database includes: Peripheral eye related parameters, outer ring auxiliary eye related parameters, geological related parameters and post-blasting scar rate.

8. A deep learning-based blasting effect monitoring and peripheral eye blasting optimization system, characterized by: The method according to any one of claims 1 to 7 is implemented, specifically comprising: The scar rate prediction module is used to build a blasting scar rate prediction model based on the SO-DBN model, and output the predicted scar rate according to the geological related parameters of the existing blasting plan, the related parameters of the peripheral holes and the related parameters of the outer circle auxiliary holes; The parameter adjustment module is used to dynamically adjust the geological parameters, the parameters related to the peripheral holes, and the parameters related to the outer ring auxiliary holes. If the predicted scar rate meets the preset standard, the blasting plan is output; if the predicted scar rate does not meet the preset standard, the parameters are updated and re-input into the scar rate prediction module until the predicted scar rate meets the preset standard, and the blasting plan is output; The output module is used to execute and output a blasting plan that meets the preset standards, perform on-site blasting on the tunnel face, and generate a post-blasting scar rate image; The image processing module is used to identify and extract information from the scar rate image after blasting based on the optimized FasterNet-YOLOv8 model to obtain the actual scar rate of the tunnel face; The error analysis module is used to perform error analysis on the actual scar rate of the tunnel face and the predicted scar rate, and to judge the optimization effect based on the error results.

9. The blasting effect monitoring and peripheral eye blasting optimization system based on deep learning according to claim 8 is characterized in that: Also includes: The database module is used to store the peripheral eye parameters, outer circle auxiliary eye parameters and their corresponding scar rate data of the initial blasting plan.

10. The blasting effect monitoring and peripheral eye blasting optimization system based on deep learning according to claim 8 is characterized in that: Also includes: The optimization module is used to add the actual scar rate and the geological parameters of the blasting plan, the peripheral eye parameters and the outer circle auxiliary eye parameters to the database module when the error is less than 5%, expand and optimize the database module, and thus enhance the generalization ability of the blasting scar rate prediction model.

Images