Shield tunneling machine intelligent tunneling regulation and control system and method based on muck digital imaging and knowledge graph

Abstract

The invention discloses a shield tunneling machine intelligent tunneling regulation and control system and method based on muck digital imaging and a knowledge graph, and the system comprises a muck image collection module, an image processing unit, a knowledge graph reasoning module, a parameter reference suggestion generation module, a visual interaction module, and a feedback and iterative optimization module. The method comprises the following steps: acquiring an image of residue soil on a residue discharge belt of a shield tunneling machine in real time, dynamically identifying a stratum type and a change trend in combination with a knowledge graph, generating a reference regulation and control suggestion of tunneling parameters of the shield tunneling machine, and displaying the reference regulation and control suggestion on a screen. According to the method, a shield tunneling machine driver can scientifically and systematically adjust construction operation, the real-time performance and accuracy of stratum recognition are improved, cutter head abrasion and shield attitude offset caused by uneven hardness or boulder strata are reduced, and construction safety and tunneling efficiency are improved; the system carries out iterative optimization on the knowledge graph and the image recognition model according to construction feedback, the adaptability and reliability of the system under different working conditions are enhanced, and the construction economy and the engineering application value are improved.

Description

Technical Field

[0001] This invention relates to the field of tunnel boring machine (TBM) construction technology, and more specifically, to an intelligent tunneling control system and method for TBMs based on digital imaging of excavated soil and knowledge graphs. Background Technology

[0002] Shield tunneling, as one of the most important excavation methods in underground engineering, is widely used in projects such as subways, railway tunnels, municipal integrated pipe corridors, and urban underground space development. The shield machine continuously excavates underground and simultaneously assembles segments to form a lining, completing the construction of long-distance, large-diameter tunnels. However, when facing uneven geological formations, such as alternating layers of sand and clay, weathered rock and soft soil, or densely packed boulders, the shield machine faces significant challenges. These complex geological conditions require the equipment to have precise adaptive control capabilities. However, limited by the resolution and coverage of geological exploration technology, the geological data obtained before construction often fails to fully and accurately reflect the true situation of the excavation area. Local boulders or hard interlayers may be missed during the exploration, or the actual distribution and scale may deviate from the exploration report. When tunnel boring machines (TBMs) excavate into such geological sections, the cutterhead and cutters must withstand sudden high-intensity impact loads, leading to uneven stress, abnormal wear, and even equipment failures such as cutterhead cracking and main bearing damage. At the same time, the difference in stiffness between soft and hard strata can cause the TBM to lose control of its posture, further exacerbating over-excavation or under-excavation problems. This not only affects the quality of segment assembly but may also induce engineering risks such as excessive surface settlement, seriously restricting construction efficiency and project safety.

[0003] To address the challenges of tunneling through complex geological formations, current construction processes primarily rely on the experience and manual parameter adjustments of the tunnel boring machine (TBM) operator. By observing the color, moisture content, particle composition, and particle size distribution of the excavated soil discharged from the screw conveyor, and combining this with their accumulated engineering experience, the operator infers the geological type of the current tunneling area and manually adjusts key construction parameters such as thrust, cutterhead speed, soil chamber pressure, or grouting volume accordingly. However, this experience-based manual judgment method has significant limitations: First, it is highly subjective; different drivers may have significantly different judgments on strata with the same characteristics of excavated soil. For example, the judgment of the hardness of "excavated soil containing gravel" may differ due to experience differences, resulting in a lack of unified standards for parameter adjustment. Second, it lacks real-time capability; excavated soil observation and driver decision-making rely on manual inspection and experience association, while the tunneling process is continuous and high-speed. Manual judgment often lags behind actual geological changes and cannot respond promptly to sudden hard interlayers or weak layers. Third, it lacks systematicity; drivers can only make qualitative inferences based on single superficial information of the excavated soil, making it difficult to integrate multi-source heterogeneous information such as cutterhead torque, thrust fluctuations, attitude sensor data, and stratum resistivity for comprehensive analysis and scientific reasoning. This makes it impossible to achieve accurate quantitative identification and parameter optimization for complex strata conditions. These problems directly lead to low tunneling efficiency, increased equipment wear and tear, and higher construction risks in strata with uneven hardness or dense boulders, making it difficult to meet the requirements of modern urban underground engineering for high-quality, high-safety, and high-efficiency construction. Summary of the Invention

[0004] The present invention aims to overcome at least one defect in the prior art and provide an intelligent tunneling control system and method for tunnel boring machines based on digital imaging of slag and knowledge graph, which solves the problems of subjectivity, lack of systematicity and real-time performance of judgment based on human experience.

[0005] The technical solution adopted in this invention is an intelligent tunneling control system for tunnel boring machines based on digital imaging and knowledge graphs of excavated soil, comprising: The waste soil image acquisition module is used to capture waste soil in real time and obtain the original digital image sequence of waste soil; The image processing unit is used to denoise, enhance, segment, and extract features from the original digital images of slag and soil to obtain image features that characterize the properties of the strata. The knowledge graph reasoning module is used to compare and reason with the image features and the pre-constructed knowledge graph that associates the features of the slag, the type of stratum and the tunneling parameters, and to determine the type of stratum and its changing trend. The parameter reference suggestion generation module generates corresponding tunneling parameter reference suggestions based on the reasoning results and the real-time operating status of the tunnel boring machine. The visual interactive module displays reference suggestions, formation identification results, and risk warnings on the screen in real time; The feedback and iterative optimization module corrects and updates the knowledge graph and image recognition process based on the actual tunneling results, actual operations, and deviations from system suggestions during the construction process.

[0006] The muck image acquisition module uses a high-speed industrial camera to capture images, installed above the muck conveyor belt of the tunnel boring machine, to capture real-time images of the muck. The module obtains raw digital image sequences of the muck, which are then transmitted to the image processing unit for preprocessing. The knowledge graph reasoning module then compares and infers from the preprocessed images to determine the stratum type and trend changes. Based on the stratum type and trend results, the parameter reference suggestion generation module generates parameter reference suggestions. The visualization interaction module then displays the reference suggestions, stratum identification results, and risk warnings to the driver for reference. Finally, the feedback and iterative optimization module corrects and updates the knowledge graph and image recognition modules to improve the system's recognition accuracy and adaptability.

[0007] A method for intelligent tunneling control of tunnel boring machines based on digital imaging and knowledge graph of excavated soil, implemented based on the aforementioned intelligent tunneling control system for tunnel boring machines based on digital imaging and knowledge graph of excavated soil, includes the following steps: S1. Obtain real-time digital images of construction waste; S2. Preprocess the digital image of the slag and extract features to form characteristic parameters that characterize the properties of the strata; S3. Match and reason with the characteristic parameters of the slag and soil and the pre-constructed knowledge graph to identify the stratum type and change trend; S4. Generate corresponding tunneling parameter reference suggestions based on stratum type and change trend, and display them through a visualization interface; S5. Based on actual construction feedback, the knowledge graph and feature extraction process are corrected and iteratively optimized.

[0008] The above methods enable intelligent identification and analysis of excavated soil and geological information, providing drivers with scientific reference suggestions for tunneling parameters, significantly improving the safety and efficiency of shield tunneling, and are particularly suitable for shield tunneling scenarios with uneven soft and hard surfaces, isolated boulders, or complex geological conditions.

[0009] To ensure the spatiotemporal consistency of the data, step S1, acquiring the real-time digital image sequence of construction waste, includes: S11. Install the industrial camera above the muck conveyor belt of the tunnel boring machine, and set the acquisition frequency and exposure parameters of the industrial camera according to the muck discharge speed. S12. The industrial camera continuously captures images of the excavated soil during the tunnel boring machine's excavation process in real time, acquiring the original digital image sequence of the excavated soil, which is then sent to the image processing unit via the excavated soil image acquisition module. S13. Use a time-sampling strategy to control the frame rate of the image sequence and buffer the image data. S14. The original digital images of the construction waste are stored in the database in chronological order through the image data buffer, and images that are blurry, overexposed, or severely occluded are removed using a preliminary screening algorithm.

[0010] By controlling the frame rate, the excavated soil is photographed in real time. The image acquisition process is synchronized with the tunnel boring machine's excavation speed, ensuring the spatiotemporal consistency of the data.

[0011] To achieve data continuity and real-time performance, the time-based sampling strategy includes: adaptive frame rate adjustment based on the tunnel boring machine's (TBM) excavation speed, dynamically adjusting the image sampling frequency as the excavation speed changes; selecting key frames based on the rate of change of image content, calculating the differences in grayscale histograms, texture change rates, or edge distribution between adjacent frames, and retaining key frames exceeding a threshold; combining fixed-period sampling with abnormal event-triggered sampling, performing high-frequency acquisition when abrupt changes in the particle size distribution of slag or isolated rock signals are detected; and binding a timestamp and corresponding TBM operating parameters to each frame of image during the acquisition process.

[0012] The frame rate is adaptively adjusted based on the tunnel boring machine's tunneling speed, so that the image sampling frequency is dynamically adjusted according to the tunneling speed. Fixed-period sampling and abnormal event-triggered sampling are combined to ensure data continuity. During the sampling process, a timestamp and corresponding tunnel boring machine operating parameters are bound to each frame of image to realize the synchronous association between image sequence and tunneling conditions.

[0013] To improve the preprocessing effect, step S2 involves preprocessing the digital image of the slag and extracting features to form characteristic parameters representing the geological properties, including: S21. Perform noise filtering and illumination equalization on the original digital image of the slag and soil to obtain the preprocessed image. S22. Use an image segmentation algorithm to separate the soil particles from the background region in the preprocessed image to obtain the segmentation result; S23. Perform edge detection and contour tracking on the segmentation results to extract the boundary shape of the slag particles; S24. Use a convolutional neural network model to extract deep features from the slag image to obtain multi-scale color histograms, gray-level co-occurrence matrix texture features, particle size distribution parameters, and moisture content features corresponding to the surface wetness of the slag. S25. Normalize the multi-dimensional characteristics of the waste soil, and transform features of different dimensions into a unified scale through feature standardization methods to form structured feature data that matches the knowledge graph nodes, and construct a structured feature database that can be input into the knowledge graph reasoning module.

[0014] The noise filtering in the preprocessing step employs median filtering, Gaussian filtering, or bilateral filtering to eliminate particle scattering noise during the acquisition process. Illumination equalization includes histogram equalization or adaptive contrast enhancement to improve image clarity and local detail contrast. The image segmentation algorithms in the background separation step include the Otsu algorithm based on threshold segmentation, the Canny algorithm based on edge detection, or the U-Net semantic segmentation network based on deep learning. The convolutional neural network (CNN) model in the feature extraction step uses ResNet, VGGNet, or EfficientNet network structures to obtain multi-scale color histograms, gray-level co-occurrence matrix texture features, particle size distribution parameters, and moisture content features corresponding to the surface wetness of the slag. The feature normalization methods in the standardization step include Min-Max normalization and Z-score normalization to form structured feature data that can be matched with knowledge graph nodes for subsequent stratigraphic identification and inference.

[0015] To predict unknown strata and improve the accuracy of parameter suggestion generation, step S3 involves matching and reasoning the characteristic parameters of the slag heap with a pre-constructed knowledge graph to identify stratum types and trends, including: S31. Based on geological exploration data, historical shield tunneling construction data, typical stratum case library and slag image feature data, construct a knowledge graph in the form of triplets, wherein the triplets include slag characteristics, stratum type and parameter control suggestions. S32. Store the knowledge graph in a graph database; S33. Establish one-to-one, one-to-many, and many-to-many semantic relationships between soil feature entities, stratum entities, and parameter entities in the knowledge graph; S34. Use a rule-based reasoning algorithm to match the input feature vector of the slag and soil. S35. Use graph embedding models to learn representations of knowledge graphs.

[0016] The knowledge graph's soil feature nodes include attributes such as color, texture, particle size distribution, and moisture content; stratum type nodes include sand, clay, gravel, and mixed boulder layers; and parameter control suggestion nodes include cutterhead rotation speed, propulsion speed, and torque. The graph database storing the knowledge graph is Neo4j or OrientDB, supporting efficient storage and retrieval of entity nodes and relation edges. Multiple semantic relationships between different entities are established in the knowledge graph to reflect tunneling patterns under complex working conditions. The graph embedding model used for learning the knowledge graph is TransE, TransH, or GraphSAGE, which maps entities and relations to a low-dimensional vector space. Through joint training with historical tunneling data, implicit relationships are mined, improving the accuracy of predictions for unknown strata and parameter suggestion generation.

[0017] Step S34 involves using a rule-based reasoning algorithm to match the input slag feature vector. Specifically, this involves constructing a matching rule set based on forward chain reasoning and backward chain reasoning, deriving the stratum category, and calculating the confidence level corresponding to the stratum category.

[0018] To predict unknown geological formations.

[0019] To assist the driver in making informed decisions, step S4 generates corresponding tunneling parameter reference suggestions based on geological type and trend, and displays them through a visual interface, including: S41. Based on knowledge graph reasoning, the stratum type, stratum heterogeneity degree and future change trend are obtained. Combined with the real-time operating parameters of the tunnel boring machine, parameter reference suggestions are generated through multi-parameter fusion algorithm. S42. Conduct a risk level assessment on the aforementioned parameter reference recommendations; S43. The parameter reference suggestions, the formation type, the change trend and the risk level prompts are displayed in real time through a visual human-computer interaction interface.

[0020] The real-time operating parameters of the tunnel boring machine (TBM) include propulsion speed, cutterhead torque, cutterhead rotation speed, auger speed, and grouting volume. Multi-parameter fusion algorithms, including fuzzy logic reasoning, Bayesian network probabilistic reasoning, or random forest regression prediction, are used to determine parameter ranges under different geological conditions. In the real-time visualization step, the interactive interface includes two-dimensional or three-dimensional visualization modules to display schematic diagrams of the TBM's tunneling cross-section, muck distribution curves, and risk warning icons. The interface also allows the driver to switch between different views and zoom in or out on parameter ranges to assist in scientific decision-making. The system only provides decision-making references and risk warnings to the driver, without directly and automatically adjusting the TBM's operating parameters, thus ensuring the flexibility and safety of manual intervention.

[0021] The risk level assessment of the parameter reference recommendations in step S42 is specifically as follows: The risk level is based on the abrupt change rate of formation hardness, the probability of isolated boulders, and the fluctuation of soil moisture content. The risk is divided into low risk, medium risk, and high risk levels through a weighted scoring model, and specific risk warning information is given.

[0022] In order to conduct risk warnings.

[0023] To improve the system's recognition accuracy and parameter recommendation robustness in complex environments, step S5 involves revising and iteratively optimizing the knowledge graph and feature extraction process based on actual construction feedback, including: S51. Collect actual operating parameters of the tunnel boring machine, tunneling log, excavated soil discharge status and construction effect data; S52. Compare the construction results with the parameter reference suggestions generated by the system. Quantify the matching degree between the parameter reference suggestions and the actual operation results by comparing the differences between the mean square error, mean absolute deviation, correlation coefficient deviation and the construction results. S53. Based on the difference results, dynamically correct the correlation between the characteristics of slag soil, stratum type and parameter control suggestions in the knowledge graph, and optimize the weight of nodes and edges. S54. Optimize and update the image feature extraction process and the knowledge graph reasoning process using the corrected data.

[0024] Construction performance data includes the smoothness of excavated soil transportation, tool wear rate, propulsion posture deviation, and surface settlement. The construction performance is compared with the parameter reference recommendations, and the difference between the parameter deviation and the construction results is calculated to quantify the matching degree between the recommended values ​​and the actual operation results. Based on the difference results, the knowledge graph is corrected. Correction methods include adding new entity nodes, adjusting the weights of relation edges, and merging low-confidence entity relations, thereby enhancing the adaptability of the knowledge graph to complex working conditions. The corrected data is used to optimize and update the image feature extraction process and the knowledge graph reasoning process. Image feature extraction is based on transfer learning updates using a convolutional neural network (CNN). Through the above feedback and retraining process, the system achieves self-learning and continuous optimization, enabling the system to maintain high recognition accuracy and robustness of parameter recommendations under complex, dynamic, uneven, or mixed rock formations.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows: By acquiring real-time images of the excavated soil on the tunnel boring machine's (TBM) muck conveyor belt and combining this with a knowledge graph to dynamically identify the stratum type and its changing trends, the system generates reference control suggestions for TBM tunneling parameters. This allows the TBM operator to scientifically and systematically adjust construction operations. This method improves the real-time performance and accuracy of stratum identification, reduces cutterhead wear and TBM attitude deviation caused by uneven hardness or isolated rock formations, and enhances construction safety and tunneling efficiency. Simultaneously, the system can iteratively optimize the knowledge graph and image recognition model based on construction feedback, achieving self-learning and continuous improvement capabilities. This enhances the system's adaptability and reliability under different working conditions, reduces construction costs, and improves construction economy and engineering application value. Attached Figure Description

[0026] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the invention. To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0027] Figure 1 This is a structural diagram of the intelligent tunneling control system for tunnel boring machines based on digital imaging and knowledge graph of slag and soil, as described in this invention.

[0028] Figure 2 This is a flowchart of the intelligent tunneling control method for tunnel boring machines based on digital imaging and knowledge graph of slag and soil, as described in this invention.

[0029] Figure 3 This is a flowchart illustrating the acquisition of slag images in the intelligent tunneling control method for tunnel boring machines based on digital imaging and knowledge graphs of slag in this invention.

[0030] Figure 4 This is a flowchart of the preprocessing process for the intelligent tunneling control method for tunnel boring machines based on digital imaging and knowledge graph of slag soil, as described in this invention.

[0031] Figure 5 This is a flowchart illustrating the construction of the knowledge graph in the intelligent tunneling control method for tunnel boring machines based on digital imaging and knowledge graph of slag soil, as described in this invention.

[0032] Figure 6 This is a schematic diagram illustrating the integrated process of slag image acquisition, preprocessing, feature extraction, and knowledge graph reasoning in the intelligent tunneling control method for tunnel boring machines based on slag digital imaging and knowledge graph of the present invention.

[0033] Figure 7 This is a flowchart illustrating the parameter reference suggestions generated for the intelligent tunneling control method for tunnel boring machines based on digital imaging and knowledge graphs of slag in this invention.

[0034] Figure 8 This is a flowchart illustrating the iterative optimization of the intelligent tunneling control method for tunnel boring machines based on digital imaging and knowledge graphs of slag in this invention.

[0035] Figure 9 This is a schematic diagram illustrating the parameter reference suggestion generation, visualization, and feedback iterative optimization process of the intelligent tunneling control method for tunnel boring machines based on digital imaging and knowledge graph of slag soil according to the present invention. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0038] Example 1 like Figure 1 As shown in the figure, the intelligent tunneling control system for tunnel boring machines based on digital imaging and knowledge graph of slag and soil provided in this embodiment includes a slag and soil image acquisition module, an image processing unit, a knowledge graph reasoning module, a parameter reference suggestion generation module, a visualization interaction module, and a feedback and iterative optimization module. The system comprises several modules: a muck image acquisition module installed above the muck conveyor belt of the tunnel boring machine (TBM), which uses a high-speed industrial camera to capture real-time images of the muck to obtain digital images; an image processing unit that performs denoising, enhancement, segmentation, and feature extraction on the acquired muck images to obtain key image features that characterize the geological properties; a knowledge graph reasoning module that compares and reasons with a pre-constructed knowledge graph that correlates muck features, geological types, and tunneling parameters to identify geological types and trends; a parameter reference suggestion generation module that generates corresponding tunneling parameter reference suggestions based on the reasoning results and the real-time operating status of the TBM; a visualization interaction module that displays the reference suggestions, geological identification results, and risk warnings to the TBM operator in real time; and a feedback and iterative optimization module that corrects and updates the knowledge graph and image recognition process based on the actual tunneling results, operator operations, and deviations from system suggestions during construction to improve the system's recognition accuracy and adaptability.

[0039] Through the synergistic effect of the above modules, real-time acquisition, intelligent analysis, and parameter reference suggestions of muck information during tunnel boring machine excavation are achieved, thereby assisting the driver in scientifically controlling the operation of the tunnel boring machine and improving construction safety and tunneling efficiency.

[0040] Example 2 The intelligent tunneling control method for tunnel boring machines based on digital imaging and knowledge graph of excavated soil, implemented based on the intelligent tunneling control system for tunnel boring machines based on digital imaging and knowledge graph of excavated soil described in Example 1, includes the following steps: like Figure 3 and 6 As shown, S1, acquire real-time digital images of construction waste; S11. Install the industrial camera above the muck conveyor belt of the tunnel boring machine, and set the acquisition frequency and exposure parameters of the industrial camera according to the muck discharge speed. High-speed industrial cameras feature high frame rates, dustproof, waterproof, and shockproof characteristics. They are fixed with brackets and shock-absorbing devices to ensure stable operation even in the vibration and dusty environment of tunneling construction. The camera acquisition frequency and exposure parameters are set according to the excavated soil speed to ensure clear and continuous images, and the image quality in low-light environments is improved by using an external light source.

[0041] S12. The industrial camera continuously captures images of the excavated soil during the tunnel boring machine's excavation process in real time, acquiring the original digital image sequence of the excavated soil, which is then sent to the image processing unit via the excavated soil image acquisition module. S13. Use a time-sampling strategy to control the frame rate of the image sequence and buffer the image data. By controlling the frame rate, the excavated soil is photographed in real time. The image acquisition process is synchronized with the tunnel boring machine's excavation speed, ensuring the spatiotemporal consistency of the data.

[0042] In addition, the time-based sampling strategy includes: adaptive frame rate adjustment based on the tunnel boring machine's (TBM) tunneling speed, so that the image sampling frequency is dynamically adjusted according to the tunneling speed; key frame selection based on the rate of change of image content, by calculating the difference in grayscale histograms, texture change rate, or edge distribution differences between adjacent frames, and retaining key frames when the difference exceeds a set threshold; combining fixed-period sampling and abnormal event-triggered sampling to ensure data continuity and to perform high-frequency acquisition when abrupt changes in the particle size distribution of slag or isolated rock signals are detected; and binding a timestamp and corresponding TBM operating parameters to each frame of image during the sampling process to achieve synchronous association between the image sequence and the tunneling conditions.

[0043] S14. The original digital images of the construction waste are stored in the database in chronological order through the image data buffer, and images that are blurry, overexposed, or severely occluded are removed using a preliminary screening algorithm.

[0044] A preliminary screening algorithm is used to remove unqualified images, providing a high-quality data source for subsequent image processing and feature extraction.

[0045] like Figure 4 and 6 As shown, S2, the digital image of the slag and soil is preprocessed and features are extracted to form characteristic parameters representing the properties of the strata; S21. Perform noise filtering and illumination equalization on the original digital image of the construction waste to obtain a preprocessed image; this eliminates noise and interference from the construction environment and ensures that the characteristics of the construction waste particles can be clearly presented.

[0046] Noise filtering employs median filtering, Gaussian filtering, or bilateral filtering to eliminate particle scattering noise during the acquisition process; illumination equalization processing includes histogram equalization or adaptive contrast enhancement to improve image clarity and local detail contrast.

[0047] S22. Use an image segmentation algorithm to separate the soil particles from the background region in the preprocessed image to obtain the segmentation result; Image segmentation algorithms include the threshold-based Otsu algorithm, the edge-detection-based Canny algorithm, or the deep learning-based U-Net semantic segmentation network.

[0048] S23. Perform edge detection and contour tracking on the segmentation results to extract the boundary shape of the slag particles; S24. Use the convolutional neural network model CNN to extract deep features from the slag image to obtain multi-scale color histograms, gray-level co-occurrence matrix texture features, particle size distribution parameters and moisture content features corresponding to the wetness of the slag surface. The CNN model uses ResNet, VGGNet, or EfficientNet network structures to obtain multi-scale color histograms, gray-level co-occurrence matrix texture features, particle size distribution parameters, and moisture content features corresponding to the surface wetness of the slag.

[0049] S25. Normalize the multi-dimensional characteristics of slag and soil, and transform features of different dimensions into a unified scale through feature standardization methods to form structured feature data that matches knowledge graph nodes. Construct a structured feature database that can be input into the knowledge graph reasoning module for subsequent stratigraphic identification and reasoning.

[0050] Feature standardization methods include Min-Max normalization and Z-score standardization.

[0051] like Figure 5 and 6 As shown in Figure S3, the characteristic parameters of the slag and soil are matched and reasoned with the pre-constructed knowledge graph to identify the stratum type and change trend; S31. Based on geological exploration data, historical shield tunneling construction data, typical stratum case library and slag image feature data, construct a knowledge graph in the form of triplets, wherein the triplets include slag characteristics, stratum type and parameter control suggestions. The characteristics of the construction waste include attributes such as color, texture, particle size distribution, and moisture content; the stratum type includes sand, clay, gravel, and mixed boulder layers; and the parameter control suggestion includes cutterhead rotation speed, propulsion speed, and torque. The knowledge graph expresses entities and relationships in the form of nodes and edges, covering the characteristics of construction waste particles, stratum attributes, construction parameters, and their causal relationships.

[0052] S32. Store the knowledge graph in a graph database; The graph database can be Neo4j or OrientDB, which supports efficient storage and retrieval of entity nodes and relation edges.

[0053] S33. Establish one-to-one, one-to-many, and many-to-many semantic relationships between soil feature entities, stratum entities, and parameter entities in the knowledge graph to reflect the tunneling patterns under complex working conditions.

[0054] The real-time collected data on the characteristics of construction waste is integrated with existing knowledge in the knowledge graph, and the knowledge graph is dynamically expanded and optimized through an iterative update mechanism to enhance its adaptability to new or sudden geological conditions.

[0055] S34. Use a rule-based reasoning algorithm to match the input feature vector of the slag and soil. The inference algorithm includes a set of matching rules based on forward chain inference and backward chain inference, which are used to deduce possible stratigraphic categories and calculate their corresponding confidence scores. Based on the knowledge graph, stratigraphic identification and parameter recommendation rules are set, and combined with a hybrid inference method based on logical reasoning and machine learning, stratigraphic type identification and parameter control mode derivation under complex working conditions are realized.

[0056] S35. Use graph embedding models to learn representations of knowledge graphs.

[0057] Graph embedding (also called network embedding) is a process of mapping graph data, typically high-dimensional dense matrices, to low-dimensional dense vectors to address the problem of graph data being difficult to efficiently input into machine learning algorithms.

[0058] Graph embedding models, such as TransE, TransH, or GraphSAGE, are used to map entities and relationships to a low-dimensional vector space. Through joint training with historical tunneling data, implicit relationships are mined, improving the accuracy of predictions and parameter suggestions for unknown strata. The current slag feature vector is input into the knowledge graph inference module to obtain the corresponding strata identification results and their changing trends, and to generate suitable tunneling parameter suggestions for the tunnel boring machine, providing data support for subsequent control.

[0059] like Figure 7 and 9 As shown in Figure S4, corresponding tunneling parameter reference suggestions are generated based on the stratum type and change trend, and displayed through a visualization interface; S41. Based on knowledge graph reasoning, the stratum type, stratum heterogeneity degree and future change trend are obtained. Combined with the real-time operating parameters of the tunnel boring machine, parameter reference suggestions are generated through multi-parameter fusion algorithm. The real-time operating parameters of the tunnel boring machine include propulsion speed, cutterhead torque, cutterhead rotation speed, auger rotation speed, and grouting volume; the multi-parameter fusion algorithm includes fuzzy logic reasoning, Bayesian network probabilistic reasoning, or random forest regression prediction to determine parameter ranges under different geological conditions; combined with the real-time operating status of the tunnel boring machine, it generates control reference suggestions for key construction parameters such as cutterhead rotation speed, propulsion speed, earth pressure balance value, and auger rotation speed.

[0060] S42. Conduct a risk level assessment on the aforementioned parameter reference recommendations; The risk level is based on the abrupt change rate of formation hardness, the probability of isolated boulders, and the fluctuation of soil moisture content. The risk is divided into low risk, medium risk, and high risk levels through a weighted scoring model, and specific risk warning information is given to conduct risk early warning.

[0061] S43. Display the parameter reference suggestions, formation identification results and risk level prompts in real time through a visual human-computer interaction interface; The interactive interface includes two-dimensional or three-dimensional visualization modules to display schematics of the tunnel boring machine's (TBM) excavation cross-section, spoil distribution curves, and risk warning icons. The interface allows the operator to switch between different views and zoom in or out of parameter ranges to assist in informed decision-making. The system only provides decision-making references and risk warnings to the operator, without directly and automatically adjusting the TBM's operating parameters, thus ensuring the flexibility and safety of manual intervention. Based on the system's suggested recommendations, combined with practical construction experience and site conditions, the TBM operator independently decides whether to adjust the TBM's excavation parameters.

[0062] S44. Record the parameter suggestions provided by the system and the actual operation of the driver, and compare and analyze them with the tunneling results; provide data support for subsequent optimization of the knowledge graph and system iteration.

[0063] like Figure 8 and 9 As shown in Figure S5, the knowledge graph and feature extraction process are corrected and iteratively optimized based on actual construction feedback.

[0064] S51. Collect actual operating parameters of the tunnel boring machine, tunneling log, excavated soil discharge status and construction effect data; The construction effect data includes the smoothness of soil transportation, cutter wear rate, propulsion posture deviation and ground settlement; actual operation parameters include propulsion force, cutterhead torque, earth pressure change, slag discharge, cutter wear and construction log information.

[0065] S52. Compare the construction results with the parameter reference suggestions generated by the system. Quantify the matching degree between the parameter reference suggestions and the actual operation results by measuring the differences in mean square error, mean absolute deviation, correlation coefficient deviation and construction results. Also, evaluate the consistency between the tunneling results and the stratum identification results.

[0066] S53. Based on the difference results, dynamically correct the correlation between the characteristics of slag soil, stratum type and parameter control suggestions in the knowledge graph, and optimize the weight of nodes and edges to improve the inference accuracy of the system.

[0067] The correction methods include adding new entity nodes, adjusting the weights of relation edges, and merging low-confidence entity relations, thereby enhancing the adaptability of the knowledge graph to complex working conditions.

[0068] S54. Optimize and update the image feature extraction process and knowledge graph reasoning process using the corrected data; Image feature extraction is based on transfer learning updates using convolutional neural networks (CNNs), while inference is based on iterative optimization using graph neural networks (GNNs) or reinforcement learning. Through the above feedback and retraining processes, the system achieves self-learning and continuous optimization, enabling the system to maintain high recognition accuracy and robustness of parameter recommendations under complex, dynamic, heterogeneous soft-hard strata or strata containing isolated boulders.

[0069] Through the above steps, the method of this invention can acquire images of the excavated soil on the shield tunneling machine's muck conveyor belt in real time, and dynamically identify the stratum type and its changing trends by combining a knowledge graph. This generates reference control suggestions for the shield tunneling parameters, enabling the shield tunneling machine operator to scientifically and systematically adjust construction operations. This method improves the real-time performance and accuracy of stratum identification, reduces cutterhead wear and shield attitude deviation caused by uneven hardness or isolated rock strata, and enhances construction safety and tunneling efficiency. At the same time, the system can iteratively optimize the knowledge graph and image recognition model based on construction feedback, achieving self-learning and continuous improvement capabilities. This enhances the system's adaptability and reliability under different working conditions, significantly reduces construction costs, and improves construction economy and engineering application value.

[0070] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementation methods. Clearly, many modifications and variations can be made based on the content of this specification. The selection and specific description of these embodiments in this specification are intended to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A shield tunneling intelligent excavation regulation system based on digital imaging of slag and knowledge graph, characterized in that, The system comprises: a muck image acquisition module for real-time shooting of muck to obtain a raw muck digital image sequence; an image processing unit for denoising, enhancing, segmenting and feature extracting of the raw muck digital image to obtain image features representing stratum properties; a knowledge graph reasoning module for comparing and reasoning the image features with a pre-constructed knowledge graph of muck features, stratum types and tunneling parameters to determine stratum types and change trends; a parameter reference suggestion generation module for generating corresponding tunneling parameter reference suggestions based on the reasoning results and in combination with real-time running states of the shield machine; a visual interactive module for real-time display of the reference suggestions, stratum identification results and risk prompts on a screen; a feedback and iterative optimization module for correcting and updating the knowledge graph and the image recognition process according to deviations between actual tunneling results, actual operations and system suggestions in the construction process.

2. A shield tunneling intelligent excavation regulation method based on slag digital imaging and knowledge graph, characterized in that, The shield machine intelligent tunneling regulation system based on muck digital imaging and knowledge graph according to claim 1 comprises the following steps: S1, obtaining real-time muck digital images; S2, pre-processing the muck digital images and extracting features to form feature parameters representing stratum properties; S3, matching and reasoning the muck feature parameters with a pre-constructed knowledge graph to identify stratum types and change trends; S4, generating corresponding tunneling parameter reference suggestions based on the stratum types and change trends and displaying the suggestions through a visual interface; S5, correcting and iteratively optimizing the knowledge graph and the feature extraction process in combination with actual construction feedback.

3. The method according to claim 2, wherein, The step S1 of obtaining a real-time muck digital image sequence comprises: S11, installing an industrial camera above a muck discharge belt of the shield machine, setting industrial camera acquisition frequency and exposure parameters according to muck discharge speed; S12, continuously shooting muck in the tunneling process of the shield machine in real time by the industrial camera to obtain a raw muck digital image sequence, which is transmitted from the muck image acquisition module to the image processing unit; S13, using a time-permissive sampling strategy to control frame frequency of the image sequence and perform image data buffering; S14, storing the raw muck digital images in a database in chronological order through image data buffering and removing blurred, overexposed or severely occluded images by using a preliminary screening algorithm.

4. The method according to claim 3, wherein, The time-permissive sampling strategy comprises: frame rate self-adaptive adjustment based on the tunneling speed of the shield machine to dynamically adjust the image sampling frequency according to the tunneling speed; selecting key frames of image content change rate, calculating the difference of adjacent frame gray histogram, texture change rate or edge distribution difference, and retaining key frames exceeding the threshold; combining fixed period sampling and abnormal event triggered sampling to perform high frequency collection when muck particle size gradation mutation or boulder signals are detected; and binding a timestamp and corresponding shield machine running parameters for each frame of image during the collection process.

5. The method of claim 2, wherein the method further comprises: The step S2 of pre-processing the muck digital images and extracting features to form feature parameters representing stratum properties comprises: S21, performing noise filtering and light balancing on the raw muck digital images to obtain pre-processed images; S22, the slag particles and the background area are separated by using an image segmentation algorithm on the preprocessed image to obtain a segmentation result; S23, edge detection and contour tracking are performed on the segmentation result to extract the boundary shape of the slag particles; S24, a convolutional neural network model is used to extract deep features of the slag image to obtain multi-scale color histogram, gray level co-occurrence matrix texture features, particle size distribution parameters, and moisture content features corresponding to the moisture content of the slag surface; S25, the multi-dimensional slag features are normalized, different dimensional features are converted into a unified scale by a feature standardization method, structured feature data matching the knowledge graph nodes are formed, and a structured feature database inputting the knowledge graph reasoning module is constructed.

6. The method of claim 2, wherein the method further comprises: The step S3 of matching and reasoning the slag feature parameters with the pre-constructed knowledge graph to identify the formation type and the change trend comprises: S31, based on geological exploration data, historical shield construction data, a typical formation case library, and slag image feature data, a knowledge graph represented in the form of triples is constructed, the triples comprising slag features, formation types, and parameter control suggestions; S32, the knowledge graph is stored in a graph database; S33, one-to-one, one-to-many, and many-to-many semantic association relationships between slag feature entities, formation entities, and parameter entities are established in the knowledge graph; S34, a rule-based reasoning algorithm is used to match the input slag feature vector; S35, a graph embedding model is used to perform representation learning on the knowledge graph.

7. The method according to claim 6, wherein, The step S34 of matching the input slag feature vector by using a rule-based reasoning algorithm specifically comprises: constructing a matching rule set based on forward chain reasoning and reverse chain reasoning, deducing the formation category, and calculating the confidence degree corresponding to the formation category.

8. The method of claim 2, wherein the method further comprises: The step S4 of generating corresponding tunneling parameter reference suggestions based on the formation type and the change trend and displaying the suggestions through a visual interface comprises: S41, based on the knowledge graph reasoning, the formation type, the formation unevenness degree, and the future change trend are obtained, the real-time running parameters of the shield machine are combined, and parameter reference suggestions are generated through a multi-parameter fusion algorithm; S42, the parameter reference suggestions are evaluated in terms of risk level; S43, the parameter reference suggestions, the formation type, the change trend, and the risk level are displayed in real time through a visual human-computer interaction interface.

9. The method according to claim 8, wherein, The step S42 of evaluating the parameter reference suggestions in terms of risk level specifically comprises: The risk level is divided into low risk, medium risk, and high risk levels through a weighted scoring model based on the formation hardness mutation rate, the probability of appearing isolated stones, and the slag moisture content volatility, and specific risk prompt information is given.

10. The method of claim 2, wherein the method further comprises: The step S5 of correcting and iteratively optimizing the knowledge graph and the feature extraction process in combination with actual construction feedback comprises: S51, actual operation parameters of the shield machine, tunneling logs, slag discharge conditions, and construction effect data are collected; S52, difference comparison of construction effect and system generated parameter reference suggestion, through the difference of mean square error, mean absolute deviation, correlation coefficient deviation and construction result, the matching degree of parameter reference suggestion and actual operation result is quantified; S53, based on the difference result, the association relationship between the slag characteristics, stratum type and parameter control suggestion in the knowledge graph is dynamically corrected, and the weight of the node and the edge is optimized; S54, the image feature extraction process and the knowledge graph reasoning process are updated and optimized by using the corrected data.

Images