Intelligent adjusting system and method for soil layer penetrating power of shield tunneling machine

Abstract

The invention discloses an intelligent adjusting system and method for soil layer penetrating power of a shield tunneling machine. The intelligent adjusting system comprises a sensor array, wherein the sensor array is annularly and uniformly distributed around a cutter head of the shield tunneling machine; the risk level module is used for calculating a similarity matrix between section rings by adopting a Gaussian kernel function, dividing performance communities through a Louvain algorithm and outputting geological risk levels and corresponding coordinates; and the intelligent decision module integrates a multilayer convolutional neural network, fuses sensing data and a network analysis result in real time, and generates dynamic instructions of the cutterhead rotating speed, the propelling speed and the grouting pressure. The shield tunneling machine soil layer penetrating power intelligent adjusting system achieves real-time monitoring and intelligent analysis of multi-dimensional information such as soil layer hardness, pore water pressure, ground stress and temperature by integrating a high-precision multi-dimensional sensing module, a risk level module and an intelligent decision module. The system can automatically adjust the cutterhead rotating speed, the propelling speed and the grouting pressure of the shield tunneling machine according to the geological risk grade.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel boring equipment, and particularly to an intelligent adjustment system and method for the soil penetration force of a shield machine. Background Art

[0002] The construction of a shield machine is an advanced underground tunnel construction method. The construction of a shield machine refers to the method of using a shield machine to bore underground, and at the same time, under the protection of the shield, excavation and lining operations are safely carried out inside the machine to construct a tunnel. The working principle of a shield machine is to excavate soil while advancing along the axis of the tunnel through a cylindrical component (shield). The construction process of a shield machine generally includes links such as initial operation preparation, shield starting and arriving, shield tunneling, lining assembly, grouting, and waterproofing.

[0003] Traditional shield machine construction has many limitations. To solve these problems, modern researchers and engineers have begun to explore innovative technical means. Traditional shield machine construction mainly relies on manual experience and judgment, lacking real-time monitoring and comprehensive analysis of multi-dimensional information such as soil properties, geological conditions, and the working state of the shield machine. This results in limited construction efficiency, high safety risks, and difficulty in ensuring construction accuracy and efficiency.

[0004] To overcome these limitations, researchers and engineers have begun to combine a variety of sensors with advanced artificial intelligence algorithms to develop intelligent systems capable of comprehensively analyzing multi-dimensional information. These intelligent systems can, through integrated devices such as ground penetrating radar, accelerometers, and pressure sensors, real-time monitor key indicators such as the hardness, water content, and stress distribution of the soil layer, as well as important data such as the working state of the shield machine.

[0005] The patent application with the publication number CN110185461A discloses a shield machine for real-time detecting tunnel geological conditions. This shield machine can detect the tunnel geological conditions in real time, and its components include: the shield machine main body, which is responsible for carrying out the excavation work of the tunnel; the detection system, which is specifically used for detecting the geological conditions of the tunnel; and the data transmission system, which undertakes the task of sending the detected data. This shield machine has achieved a multi-dimensional improvement in geological condition detection by integrating an electro-magnetic wave pulse detection device, an induced polarization detection device, and a seismic wave detection device, thus significantly improving the accuracy of detection. However, although these technologies have improved the construction efficiency and accuracy to a certain extent, they lack the ability of intelligent decision-making support and automatic adjustment of construction parameters.

[0006] Therefore, there is an urgent need for an intelligent system that can comprehensively analyze multi-dimensional information, make intelligent decisions based on the collected data, and automatically adjust construction parameters. Such an intelligent system can not only significantly improve construction efficiency and accuracy, reduce safety risks, but also bring about a revolution in the field of shield machine construction. Summary of the Invention

[0007] In view of the problems existing in the construction of existing shield machines, it has become an urgent technical problem to develop a smart adjustment system for the soil penetration force of shield machines that can real-time monitor multi-dimensional information of soil layers, intelligently analyze the geological risk level, and automatically adjust construction parameters.

[0008] The purpose of the present invention is achieved through the following technical solutions:

[0009] A smart adjustment system for the soil penetration force of a shield machine, comprising:

[0010] A high-precision multi-dimensional sensing module, including piezoelectric soil hardness sensors (0 - 100 MPa, accuracy 0.01 MPa), silicon piezoresistive pore water pressure sensors (0 - 10 MPa, accuracy 0.05 kPa), fiber Bragg grating in-situ stress sensors (accuracy 0.1 MPa), and platinum resistance temperature sensors (-40°C to 120°C, accuracy 0.1°C) distributed in a circular array. The sensor array is evenly distributed in a ring around the cutter head of the shield machine, with a sensor spacing of 150 mm and a protection level of IP68;

[0011] A risk level module, which calculates the similarity matrix between segment rings using the Gaussian kernel function, divides the performance community through the Louvain algorithm, and outputs the geological risk level and corresponding coordinates;

[0012] A smart decision-making module, integrating a multi-layer convolutional neural network (input layer → three convolutional layers → fully connected layer → output layer), real-time fuses the sensing data and the network analysis results, and generates dynamic instructions for the cutter head rotation speed (0 - 10 rpm), propulsion speed (0 - 100 mm / min), and grouting pressure (0 - 2 MPa).

[0013] As a preferred mode, the working process of the risk level module includes:

[0014] Mapping the shield segment rings into network nodes, calculating the similarity matrix between segment rings, and the formula is:

[0015]

[0016] X i is the multi-dimensional feature vector of the segment ring, and σ is a multiple of the standard deviation of the feature vector;

[0017] Generating a binary adjacency matrix through thresholding;

[0018] Analyzing the network topology based on the node degree, average shortest path length, and clustering coefficient, and dividing high-risk communities.

[0019] As a preferred mode, the inputs of the smart decision-making module include:

[0020] Real-time sensing data, including soil layer hardness, pore water pressure, ground stress, and temperature;

[0021] Risk level labels output by the risk level module;

[0022] Historical geological parameter library;

[0023] The output is a dynamic adjustment instruction, including: cutter head rotation speed (accuracy 0.01 rpm), propulsion speed (accuracy 0.1 mm / min), grouting pressure (accuracy 0.01 MPa).

[0024] As a preferred method, it also includes an edge-cloud computing collaboration module. The edge node processes real-time control instructions, and the cloud platform performs complex network modeling and deep learning model training;

[0025] In the edge-cloud computing collaboration module: Deployed locally on the shield machine, directly connected to sensors using an industrial-grade Ethernet architecture; Integrated with an artificial intelligence-assisted decision-making module to predict geological risks in the next 10 minutes.

[0026] As a preferred method, it also includes a blockchain security module, which encrypts sensor data using AES-256 sharding, transmits instructions through a quantum key distribution (QKD) channel, and stores operation logs on the consortium chain to support audit traceability;

[0027] The blockchain security module includes:

[0028] Stored in distributed nodes, a single node failure does not affect the overall integrity;

[0029] Using the PBFT consensus algorithm, the transaction confirmation time < 1 second.

[0030] As a preferred method, it also includes a three-level pressure protection mechanism:

[0031] When 4 Mpa > soil chamber pressure ≥ 3 MPa, trigger a first-level warning and reduce the propulsion speed to 50 mm / min;

[0032] When 5 Mpa > soil chamber pressure ≥ 4 MPa, activate the second-level protection, stop the cutter head rotation and increase the grouting pressure;

[0033] When the soil chamber pressure ≥ 5 MPa, enter the safety mode, cut off the power and start the emergency drainage system.

[0034] As a preferred method, the neural network training process of the intelligent decision-making module includes:

[0035] Input historical geological data and construction parameters;

[0036] The Adam optimizer and ReLU activation function are adopted, and the loss function is a combination of mean squared error and cross entropy. As a preferred mode, the system is equipped with an adaptive working condition optimization function:

[0037] Soft soil layer: the cutter head rotation speed ≤ 5 rpm, and the grouting pressure ≥ 1.2 MPa;

[0038] Hard rock layer: the cutter head rotation speed ≥ 8 rpm, and the propulsion speed ≤ 60 mm / min;

[0039] Water-rich layer: increase the injection amount of sealing grease, and increase the drainage frequency to 2 times / minute.

[0040] The present invention also provides an intelligent adjustment method for the soil penetration force of a shield machine, including:

[0041] Data acquisition: Real-time obtain data of soil layer hardness, pore water pressure, ground stress and temperature through a ring-shaped sensor array;

[0042] Network modeling: Construct a complex network of shield segments and divide geological risk communities;

[0043] Intelligent decision-making: Integrate real-time data and network analysis results to generate control instructions;

[0044] Dynamic adjustment: Adjust the cutter head rotation speed, propulsion speed and grouting pressure based on the fuzzy PID algorithm.

[0045] As a preferred mode, the network modeling step includes:

[0046] Calculate the similarity matrix between segments, and the threshold is selected as similarity ≥ 0.7;

[0047] Divide communities based on the Louvain algorithm, and communities with modularity ≥ 0.3 are determined as independent risk communities;

[0048] Output the coordinates of risk communities and the recommended increment of grouting pressure.

[0049] The present invention has at least the following beneficial effects: The intelligent adjustment system for the soil penetration force of the shield machine realizes real-time monitoring and intelligent analysis of multi-dimensional information such as soil layer hardness, pore water pressure, ground stress and temperature by integrating high-precision multi-dimensional sensing modules, risk level modules and intelligent decision-making modules. The system can automatically adjust the cutter head rotation speed, propulsion speed and grouting pressure of the shield machine according to the geological risk level, thereby significantly improving the construction efficiency and safety of the shield machine under complex geological conditions. In addition, the system also has high adaptability and flexibility, can cope with construction challenges under different geological conditions, reduce construction costs and improve project quality. Description of the Drawings

[0050] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for the embodiments will be briefly introduced below. It should be understood that the following drawings only show the embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0051] Figure 1 It is the overall framework diagram for the intelligent adjustment of the soil penetration force of the shield machine;

[0052] Figure 2 It is the working flow chart of the intelligent control system. Specific embodiments

[0053] The technical solutions of the present invention will be further described in detail below with reference to the accompanying drawings, but the protection scope of the present invention is not limited to the following.

[0054] In the following, different embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that the present disclosure is not limited to the specific forms disclosed herein, but rather, the present disclosure should be construed as covering different changes, equivalents, and / or substitutions of the embodiments of the present disclosure. When describing the drawings, like reference numerals may be used to indicate like components.

[0055] In the present disclosure, terms are used to describe specific embodiments and do not limit the present disclosure. As used herein, the singular form is intended to also include the plural form unless the context clearly indicates otherwise. In the specification, it should be understood that the terms "comprising" or "having" indicate the presence of features, numbers, steps, operations, structural elements, components, or combinations thereof, without precluding the presence or addition of one or more other features, numbers, steps, operations, structural elements, components, or combinations thereof.

[0056] It should be understood that specific details are provided in the following description to facilitate a complete understanding of the exemplary embodiments. However, those of ordinary skill in the art should understand that the exemplary embodiments can be implemented without these specific details. For example, a system can be shown in a block diagram to avoid obscuring the example with unnecessary details. In other instances, well-known processes, structures, and techniques can be shown without unnecessary details to avoid obscuring the example.

[0057] An intelligent adjustment system for the soil penetration force of a shield machine includes:

[0058] High-precision multi-dimensional sensing module, including piezoelectric soil hardness sensors (0 - 100 MPa, accuracy 0.01 MPa) distributed in a circular array, silicon piezoresistive pore water pressure sensors (0 - 10 MPa, accuracy 0.05 kPa), fiber Bragg grating in-situ stress sensors (accuracy 0.1 MPa), and platinum resistance temperature sensors (-40°C to 120°C, accuracy 0.1°C). The sensor array is evenly distributed in a circular shape around the cutter head of the shield machine, with a sensor spacing of 150 mm and an IP68 protection level;

[0059] Risk level module, which calculates the similarity matrix between segment rings using the Gaussian kernel function, divides the performance community through the Louvain algorithm, and outputs the geological risk level and corresponding coordinates;

[0060] Intelligent decision-making module, integrating a five-layer convolutional neural network (input layer → three convolutional layers → fully connected layer → output layer), real-time fuses the sensing data and network analysis results, and generates dynamic instructions for the cutter head rotation speed (0 - 10 rpm), propulsion speed (0 - 100 mm / min), and grouting pressure (0 - 2 MPa).

[0061] Figure 1 It is the overall framework diagram for the intelligent adjustment of the soil penetration force of the shield machine. In this embodiment, a high-precision multi-dimensional sensing module is adopted. By integrating a variety of high-precision sensors distributed in a circular array, real-time monitoring of multi-dimensional parameters such as soil hardness, pore water pressure, in-situ stress, and temperature in the working environment of the shield machine is realized. The high precision and circular uniform distribution of the sensors ensure the accuracy and comprehensiveness of the data, providing strong support for the safe and efficient operation of the shield machine.

[0062] The risk level module uses the Gaussian kernel function and the Louvain algorithm, which can accurately calculate the similarity between segment rings and divide the performance community, output the geological risk level and corresponding coordinates, providing a scientific basis for the construction planning and risk control of the shield machine.

[0063] The intelligent decision-making module, through integrating a five-layer convolutional neural network, real-time fuses the sensing data and network analysis results, and can intelligently generate dynamic instructions for the cutter head rotation speed, propulsion speed, and grouting pressure according to the real-time working conditions, effectively improving the automation level and construction efficiency of the shield machine, and reducing the risks and costs of manual operation.

[0064] In summary, through the comprehensive application of modules such as high-precision multi-dimensional sensing, complex network modeling, and intelligent decision-making, this system significantly improves the intelligent level and construction efficiency of the shield machine, providing strong guarantee for the safe and efficient operation of the shield machine.

[0065] In a preferred embodiment, the working process of the risk level module includes:

[0066] Map the shield segment rings to network nodes and calculate the similarity matrix between segment rings. The formula is:

[0067]

[0068] X i is the multidimensional eigenvector of the segment ring, σ is the multiple of the standard deviation of the eigenvector;

[0069] Generate a binary adjacency matrix by thresholding, retaining edges with similarity ≥ 0.7;

[0070] The network topology is analyzed based on node degree, average shortest path length and clustering coefficient, and high-risk communities are divided. The edge network is a network structure constructed by calculating the similarity matrix between shield construction segment rings. Node: Each shield segment ring is mapped to a network node, and the similarity of multidimensional parameters (geological conditions, mechanical parameters, construction efficiency, etc.) between segment rings is calculated by Gaussian kernel function. If the similarity exceeds the threshold (such as ≥0.7), an edge is established between the nodes to indicate the correlation of their construction performance. The constructed network has small-world characteristics (high clustering coefficient + short average path length), which can identify the community structure of construction performance and reflect the area of ​​sudden change in geological conditions. The complex network modeling (edge ​​network) of shield construction is integrated into the intelligent control system, and the dynamic mapping of geological risks is realized through community division, which goes beyond the traditional control mode that only relies on real-time sensor data.

[0071] The content of the risk level module includes:

[0072] Data preparation and feature extraction

[0073] Input data: multi-dimensional feature vector of shield segment ring, each segment ring contains the following parameters:

[0074] Geological parameters: soil hardness (MPa), pore water pressure (kPa), ground stress (MPa), temperature (℃) Mechanical parameters: cutter head speed (rpm), propulsion speed (mm / min), grouting pressure (MPa), torque (kN·m) Construction parameters: penetration rate (m / h), utilization rate (%), daily excavation volume (m / d)

[0075] Data preprocessing:

[0076] Standardization: Z-score standardization is performed on each parameter.

[0077] Outlier filtering: Use the 3σ principle to eliminate abnormal segment ring data.

[0078] Similarity calculation (Gaussian kernel function)

[0079] formula: X i ,X j: The normalized eigenvectors of segment rings i and j.

[0080] ||X i -X j ||: Euclidean distance, calculating the difference in multi-dimensional parameters between two segment rings.

[0081] σ: Gaussian kernel bandwidth, initially taking a value of 1.5 times the standard deviation of all eigenvectors, used to balance noise suppression and sensitivity.

[0082] Implementation steps:

[0083] Calculate the Euclidean distance: Traverse all pairs of segment rings to generate a fully connected similarity matrix.

[0084] Dynamically adjust the value of σ: In water-rich sand layers or hard rock layers, dynamically reduce σ according to the parameter volatility (e.g., σ = 1.0 times the standard deviation) to enhance the sensitivity to key parameters. Dynamically optimize the similarity threshold of the edge network according to the geological conditions (e.g., in the water-rich layer, σ = 1.0 times the standard deviation) to enhance the sensitivity to key parameters and reduce the false alarm rate of risk identification.

[0085] Output of the similarity matrix: Obtain an N×N matrix (N is the total number of segment rings), and the range of matrix elements is 0 - 1.

[0086] Network construction and thresholding

[0087] Generate the binary adjacency matrix:

[0088] Threshold selection: Retain the edges with similarity ≥ 0.7 to ensure network connectivity and filter noise.

[0089] Logical verification: If the network density is too low (e.g., the proportion of the number of edges < 5%), then lower the threshold to 0.6 to ensure the retention of key associations.

[0090] Network topology analysis indicators:

[0091] Node degree: The number of direct associations of a segment ring, reflecting its representativeness of geological conditions.

[0092] Clustering coefficient: Measuring the tightness of the local network, a high clustering coefficient (> 0.5) indicates stable regional geological conditions.

[0093] Average shortest path length: If the path length ≤ 3, it indicates that the network has the small-world property and is suitable for community division. Community division (Louvain algorithm), algorithm process:

[0094] Modularity optimization: Maximize the modularity Q value (objective function):

[0095]

[0096] A ij: Adjacency matrix element. If there is an edge between nodes i and j (i.e., the similarity between segment rings ≥ threshold), A ij is 1, otherwise 0; k i , k j : Degrees of nodes i and j; m represents the total weight of all edges in the network (for an unweighted network, i.e., the total number of edges). In the complex network of shield tunneling construction, the edges are defined by a binary adjacency matrix where the similarity between segment rings exceeds the threshold (e.g., ≥0.7), so m is the total number of edges that meet the condition. c i , c j represents the community number to which nodes i and j belong. After partitioning by the Louvain algorithm, each segment ring is assigned to a specific community (such as Community 1, Community 2, etc.). δ(c i , c j ): Indicator function, which is 1 if i and j belong to the same community, otherwise 0.

[0097] Iterative merging:

[0098] Phase 1: Local optimization. Assign each node to an adjacent community to maximize the Q value. Traverse each node in the network, calculate the gain in the modularity Q value when moving the node to an adjacent community. If the gain is positive, move the node to the community that maximizes ΔQ; otherwise, keep the node in the current community. When the movement of all nodes cannot improve the Q value, Phase 1 ends.

[0099] Phase 2: Aggregate nodes in the same community into supernodes, construct a new network, and repeat Phase 1 until the Q value converges. Merge the nodes in the same community into "supernodes" to generate a new simplified network. All nodes within each community are merged into one supernode, and the edge weights between supernodes are the sum of the weights of the edges across communities in the original network. Each community corresponds to a supernode. If there is an edge from community X to community Y in the original network, the edge weight between the corresponding supernodes X and Y in the new network is the sum of these edges. Use the newly generated network as the input and return to Phase 1 for further optimization until the modularity Q value no longer changes significantly (converges).

[0100] Risky community identification:

[0101] Criteria for high-risk communities: The average pore water pressure within the community ≥ 5 kPa, or the ground stress fluctuation ≥ 2 MPa. The community modularity Q > 0.3 (ensuring strong internal connections within the community) and contains at least 3 segment rings (excluding isolated noise points).

[0102] First, identify high-risk communities through complex network modeling (Gaussian kernel similarity calculation → Louvain community partitioning), and extract their geographical coordinates and risk levels. Risk level mapping and coordinate output.

[0103] Risk level classification:

[0104] Low risk: Community modularity Q < 0.2, small parameter fluctuations, maintain current parameters (cutter head rotation speed ≤ 8 rpm, grouting pressure 1.0 MPa).

[0105] Medium risk Q ∈ [0.2, 0.3), pore water pressure 3 - 5 kPa, reduce the propulsion speed to 70 mm / min, increase the grouting volume by 10%.

[0106] High risk Q ≥ 0.3, sudden change in in-situ stress ≥ 3 Mpa, enable hard rock mode (rotation speed ≥ 10 rpm), activate the third-level warning.

[0107] Coordinate mapping method:

[0108] Spatial coordinate calculation: Based on the construction mileage of the segment ring and the three-dimensional geological model, map high-risk communities to specific geographical coordinates. Visualization output: Mark the risk area in the GIS platform, overlay the real-time position of the shield machine, and generate a dynamic risk heat map.

[0109] Implementation case: A subway tunnel passes through an alternating formation of water-rich sand layer and hard rock, with a total length of 1.2 km. Similarity analysis: At mileage K8 + 200, the similarity matrix shows that the similarity between segment rings drops suddenly from 0.8 to 0.4, triggering a warning. Community division: The Louvain algorithm identifies 3 high-risk communities, corresponding to fault fracture zones (pore water pressure 6.2 kPa, in-situ stress fluctuation 4.5 MPa). Construction adjustment: The system automatically switches to the water-rich mode, raises the grouting pressure to 1.8 MPa, reduces the cutter head rotation speed to 6 rpm, and successfully controls the ground settlement within 8 mm. High-precision correlation analysis: The Gaussian kernel function quantifies the multi-dimensional parameter differences between segment rings, avoiding single-index deviation. Dynamic risk identification: The Louvain algorithm automatically divides communities, and combines geological parameter thresholds to achieve dynamic calibration of risk levels. The risk coordinates are linked with construction parameters, providing real-time obstacle avoidance and parameter optimization suggestions for the shield machine.

[0110] In a preferred embodiment, the inputs of the intelligent decision-making module include:

[0111] Real-time sensing data, including soil hardness, pore water pressure, in-situ stress, and temperature;

[0112] Risk level labels (low / medium / high risk) output by the risk level module;

[0113] Historical geological parameter library (10,000 groups of data);

[0114] The output is a dynamic adjustment instruction, including: cutter head rotation speed (accuracy 0.01 rpm), propulsion speed (accuracy 0.1 mm / min), grouting pressure (accuracy 0.01 MPa).

[0115] The edge network integrates multi-dimensional data such as geological parameters, mechanical status, and construction efficiency, quantifies the correlation through the Gaussian kernel function, provides structured input for the deep learning model, and improves the decision-making accuracy.

[0116] In a preferred embodiment, it further includes an edge-cloud computing collaboration module. The edge node processes real-time control instructions, and the cloud platform executes complex network modeling and deep learning model training.

[0117] In the edge-cloud computing collaboration module: It is deployed locally on the shield machine, directly connected to sensors using an industrial-grade Ethernet architecture; an artificial intelligence-assisted decision-making module is integrated, which can predict geological risks in the next 10 minutes.

[0118] The following is an explanation of this embodiment.

[0119] The data input layer includes:

[0120] Real-time sensing data: Through the nanoscale sensor array (piezoelectric soil hardness sensor, silicon piezoresistive pore water pressure sensor, fiber Bragg grating ground stress sensor, platinum resistance temperature sensor) in the shield machine patent, soil layer parameters (hardness 0 - 100 MPa, pore water pressure 0 - 10 MPa, ground stress 0 - 50 MPa, temperature -40°C to 120°C) are collected at a frequency of 1000 Hz.

[0121] Installation location of the piezoelectric soil hardness sensor: Arranged in a circular array at the cutting surface of the front end of the shield machine cutter head. 6 sensors are evenly distributed in each ring section, with a spacing of 150 mm, directly contacting the soil layer to measure the cutting resistance in real time. Installation method: Embedded inside the cutter head tool base, with the sensor probe exposed 1 - 2 mm, protected by a tungsten carbide protective cover, and the pressure resistance level is 100 MPa. Fixed through a pre-set threaded interface to ensure close contact with the soil layer. Data collection: Trigger a measurement every 10 cm of advancement, with a frequency of 1000 Hz, and transmit it to the edge node through optical fiber, and the soil hardness (0 - 100 MPa, accuracy 0.01 MPa) is fed back in real time.

[0122] Installation location of the silicon piezoresistive pore water pressure sensor: Distributed on the inner wall of the soil chamber and the shield tail seal area. 4 sensors are configured in each ring section to monitor the dynamic distribution of pore water pressure. Installation method: Adopt embedded installation, with the sensor diaphragm flush with the inner wall of the soil chamber to avoid mud deposition. The protective housing is made of Hastelloy, and the working temperature is -40°C to 120°C. Data collection: Monitor the water pressure change of 0 - 10 MPa in real time (accuracy 0.05 kPa), transmit through industrial Ethernet (10 Gbps), and synchronously fuse with the ground stress data.

[0123] Installation location of fiber Bragg grating ground stress sensor: Arranged circumferentially along the shield machine shell, 8 measuring points are installed in each ring section, covering key stress-bearing parts such as the cutter head support ring and the propulsion cylinder base. Installation method: The fiber Bragg gratings are connected in series and parallel to form a mesh structure, embedded in a special alloy protective layer, with a pressure resistance of 100 MPa. Encapsulated with epoxy resin, shear strength ≥ 50 MPa. Data acquisition: Real-time monitoring of ground stress (0 - 50 MPa, accuracy 0.1 MPa) through wavelength demodulation technology, and data fusion algorithm to eliminate mechanical vibration interference.

[0124] Installation location of platinum resistance temperature sensor: High-temperature areas such as the cutter head drive motor bearing, main hydraulic pump station, and soil chamber sealing area, 3 groups of sensors are deployed in each ring section. Installation method: Embedded in the equipment body using a threaded sleeve, the temperature measurement end is 2 mm away from the heat source surface, response time ≤ 0.5 s. Protection level IP68, resistant to chemical corrosion. Data acquisition: Real-time monitoring of temperature changes from -40°C to 120°C (accuracy 0.1°C), and the data is used to correct the temperature drift error of other sensors.

[0125] Ring array layout: All sensors are distributed in a ring at a spacing of 150 mm, forming a 360° full-coverage monitoring network to ensure no blind spots.

[0126] Historical data: Call the geological parameter library stored in the cloud, which contains construction parameter optimization models for different strata. Input of complex network: Map each shield segment ring to a network node, and calculate the similarity between segment rings through the Gaussian kernel function.

[0127] Edge computing node, including:

[0128] Real-time control: Deployed locally on the shield machine, directly connected to the sensors through industrial-grade Ethernet (up to 10 Gbps), response time ≤ 50 ms. Execute the fuzzy PID control algorithm to dynamically adjust the cutter head speed (0 - 10 rpm), propulsion speed (0 - 100 mm / min), and grouting pressure (0 - 2 MPa). Trigger a three-level early warning mechanism (reduce speed when the soil chamber pressure ≥ 3 MPa, stop the cutter head when ≥ 4 MPa, and emergency drainage when ≥ 5 MPa). Data preprocessing: Filter and normalize the sensing data to generate a 128-dimensional feature vector.

[0129] Cloud computing platform, including:

[0130] Complex network modeling: Construct a complex network for the shield segment ring, and divide performance communities through the Louvain algorithm (communities with modularity Q>0.3 are determined as high-risk communities). Analyze network topology metrics (node degree, clustering coefficient, path length) to identify geological mutation regions (such as fault zones and water-rich layers). Deep learning model: A five-layer convolutional neural network (3 convolutional layers + 2 fully connected layers), with the input layer fusing real-time data and risk labels output by the complex network. The Adam optimizer (learning rate 0.001) and ReLU activation function are used, and the loss function is a combination of MSE (construction parameter error) and cross-entropy (risk classification). Risk prediction: Predict geological risks within the next 10 minutes through time series analysis (LSTM model). Dynamically optimize PID parameters: Proportional coefficient Kp = 0.5 - 2.0, integral coefficient Ki = 0.1 - 0.5, derivative coefficient Kd = 0.05 - 0.2.

[0131] The present invention adopts real-time data acquisition and edge processing: The sensor array collects data at a frequency of 1000Hz, and the edge node completes data normalization and feature extraction within 50ms. Trigger fuzzy PID control: For example, when the sudden change in in-situ stress ≥ 3MPa, automatically increase the cutter head speed to more than 8rpm.

[0132] Complex network modeling and risk identification (cloud): Similarity matrix construction: Calculate the Gaussian kernel similarity between segment rings, and threshold to generate a binary adjacency matrix (edges are retained if similarity ≥ 0.7). Community detection: Divide communities through the Louvain algorithm, and mark high-risk communities (pore water pressure ≥ 5kPa or in-situ stress fluctuation ≥ 2MPa). Risk mapping: Map community coordinates (such as K12+350) to a 3D geological model to generate a risk heat map. Deep learning decision generation: Input: Real-time feature vector + complex network risk label + historical geological parameters. Output: Construction parameter optimization instructions (such as increasing the grouting pressure in the water-rich layer to 1.8MPa). Risk response strategies (such as predicting the risk of the fault zone 30 minutes later and switching to the hard rock mode in advance).

[0133] Edge-cloud collaborative feedback: The cloud sends optimization parameters and risk strategies to the edge node to update the local control model. The edge node performs dynamic adjustment and at the same time feeds back the actual construction data to the cloud for model iterative training.

[0134] In a preferred embodiment, it further includes a blockchain security module, which encrypts sensor data using AES-256 sharding, transmits instructions through a quantum key distribution channel, and stores operation logs in the consortium chain to support audit traceability;

[0135] The blockchain security module includes:

[0136] Stored in distributed nodes, a single node failure does not affect the overall integrity;

[0137] With the PBFT consensus algorithm, the transaction confirmation time is less than 1 second.

[0138] In a preferred embodiment, it further includes a three - level pressure protection mechanism:

[0139] Detection location and sensor configuration. Sensor type: A silicon piezoresistive pore water pressure sensor (measurement range 0 - 10 MPa, accuracy 0.05 kPa) is adopted and integrated on the inner wall of the soil bin to be used for real - time monitoring of the composite pressure of soil pressure and water pressure in the soil bin. Supplementary with fiber Bragg grating in - situ stress sensors (accuracy 0.1 MPa), which are arranged on the front wall of the soil bin and the shield tail seal area to monitor the overall stress distribution of the soil bin. Front wall area: 6 sensors are arranged in a circular array with a spacing of 200 mm to monitor the direct soil pressure behind the cutter head. Side walls and top: 12 sensors are evenly distributed to cover the entire circumference of the soil bin to prevent local pressure anomalies. Shield tail seal area: 4 redundant sensors are configured to monitor the seal pressure and link with the grouting system. The soil bin pressure is the comprehensive value of soil stress and pore water pressure (the comprehensive value of soil bin pressure is the algebraic sum of effective soil stress and pore water pressure).

[0140] When 4 Mpa > soil bin pressure ≥ 3 MPa, a first - level early warning is triggered, and the propulsion speed is reduced to 50 mm / min;

[0141] When 5 Mpa > soil bin pressure ≥ 4 MPa, a second - level protection is started, the cutter head rotation is stopped, and grouting is pressurized to 1.5 MPa;

[0142] When the soil bin pressure ≥ 5 MPa, it enters the safety mode, the power is cut off, and the emergency drainage system is started.

[0143] In a preferred embodiment, the neural network training process of the intelligent decision - making module includes:

[0144] Input historical geological data and construction parameters, with the label being the actual construction efficiency and failure rate; the risk community label output by the edge network is used as an additional input feature of the convolutional neural network (CNN) to optimize the generation of control instructions. For example, the accuracy of the grouting pressure instruction in a high - risk community is improved from ±0.1% to ±0.05%.

[0145] The Adam optimizer (learning rate 0.001) and ReLU activation function are adopted, and the loss function is a combination of mean square error and cross - entropy.

[0146] Data preparation and feature fusion:

[0147] Input data:

[0148] Historical geological data: including soil layer hardness (0 - 100 MPa), pore water pressure (0 - 10 MPa), ground stress (0 - 50 MPa), and temperature (-40°C to 120°C), collected by a high-precision multi-dimensional sensing module, and the data is standardized to a mean of 0 and a variance of 1.

[0149] Construction parameters: parameters such as cutter head rotation speed (0 - 10 rpm), propulsion speed (0 - 100 mm / min), grouting pressure (0 - 2 MPa), etc., extracted according to the operation log.

[0150] Risk community labels: community classifications (low / medium / high risk) output from the risk level module, divided by the Louvain algorithm, and communities with modularity Q > 0.3 are marked as high risk.

[0151] Label data:

[0152] Construction efficiency labels: penetration rate (m / h), utilization rate (%), and daily tunneling volume (m / d).

[0153] Failure rate labels: tool wear rate (mm / h), number of hydraulic system failures, and proportion of downtime. Input layer design: Multi-modal feature splicing: Original feature vector (dimension 128): Geological data and construction parameters are compressed through a fully connected layer (512 nodes). Risk feature vector (dimension 64): Risk community labels are encoded as continuous vectors through an embedding layer. Fusion method: The original features and risk features are spliced into a 192-dimensional input vector and reduced to 128 dimensions through a 1×1 convolutional kernel.

[0154] Five-layer convolutional neural network: Risk feature guidance: An attention mechanism is introduced after convolutional layer 2, and risk community labels dynamically weight the feature maps through a gated unit, increasing the feature weights of high-risk communities. Feature fusion verification: Through the community similarity matrix output by the risk level module (calculated by the Gaussian kernel function), the feature distribution is constrained during backpropagation to ensure that the feature distance of the same type of community ≤ 0.1.

[0155] Training process, optimizer configuration: Adam optimizer: Learning rate 0.001, β1 = 0.9, β2 = 0.999, ε = 1e -8 。Dynamic learning rate: When the validation loss does not decrease for 3 consecutive epochs, the learning rate decays to 0.5 times the original value.

[0156] Training steps:

[0157] Forward propagation: The input features generate prediction parameters through the CNN, and the MSE loss is calculated. Backward propagation: When the gradient is backpropagated, risk community labels adjust the weights of the convolutional layer through a gated unit, and the feature gradient of high-risk communities is amplified by 1.2 times. Parameter update: The Adam optimizer updates the weights, and a model snapshot is saved every 1000 steps.

[0158] Precision improvement mechanism, refined control guided by risk communities: High-risk communities: In the loss function of the grouting pressure command, the error term is multiplied by a coefficient of 1.5, forcing the model to converge to a higher precision (±0.05%). Medium- and low-risk communities: Standard loss weights are adopted to maintain a precision of ±0.1%.

[0159] Real-time feedback optimization: After each tunneling is completed, actual construction data (such as soil pressure fluctuations, cutter wear) is transmitted back to the training set, triggering incremental learning: updating the weights of the fully connected layer while keeping the parameters of the convolutional layer unchanged. New data from high-risk communities is added to the training set with a weight of 2 times.

[0160] Fuzzy PID control of shield machine: The output parameters of the CNN are used as the PID set values, and dynamic adjustment is achieved through a closed-loop control with a response time ≤ 50 ms. The risk community labels are dynamically updated through Gaussian kernel similarity to ensure the spatio-temporal consistency of the input features.

[0161] In a preferred embodiment, the system is equipped with an adaptive working condition optimization function:

[0162] Soft soil layer: Cutter head rotation speed ≤ 5 rpm, grouting pressure ≥ 1.2 MPa;

[0163] Hard rock layer: Cutter head rotation speed ≥ 8 rpm, propulsion speed ≤ 60 mm / min;

[0164] Water-rich layer: The injection amount of sealing grease is increased by 30%, and the drainage frequency is increased to 2 times / minute.

[0165] In a preferred embodiment, the system supports digital twin monitoring:

[0166] The three-dimensional visualization interface displays the shield attitude (deviation ≤ ±1 mm), soil layer stress distribution, and the location of risk communities in real time;

[0167] The AR device superimposes virtual parameters onto the actual scene and supports remote operation.

[0168] In a preferred embodiment, the system integrates a reinforcement learning mechanism:

[0169] Adopt a deep Q network (5 convolutional layers + 3 fully connected layers), and dynamically optimize the PID parameters according to the construction feedback;

[0170] The model is automatically updated with the tunneling data per kilometer, and the parameter optimization period ≤ 24 hours.

[0171] An intelligent adjustment method for the soil layer penetration force of a shield machine, comprising:

[0172] Data acquisition: Real-time acquisition of soil layer hardness, pore water pressure, ground stress, and temperature data through a ring sensor array;

[0173] Network modeling: Construct a complex network for the shield segment ring and divide geological risk communities;

[0174] Intelligent decision-making: Integrate real-time data and network analysis results to generate control instructions;

[0175] Dynamic adjustment: Adjust the cutter head speed, propulsion speed, and grouting pressure based on the fuzzy PID algorithm; The specific process can be seen in Figure 2 。

[0176] Safety protection: Trigger a three-level pressure warning and store key operation logs on the blockchain.

[0177] In a preferred embodiment, the network modeling step includes:

[0178] Calculate the similarity matrix between segment rings, and the threshold is selected as similarity ≥ 0.7;

[0179] Divide communities based on the Louvain algorithm, and communities with modularity ≥ 0.3 are determined as independent risk communities;

[0180] Output the coordinates of risk communities and the recommended increment of grouting pressure (0.2 - 0.5 MPa).

[0181] In a preferred embodiment, the intelligent decision-making step includes:

[0182] Normalize real-time data into a 128-dimensional feature vector;

[0183] Extract features through a convolutional neural network (convolution kernel 3×3, stride 1);

[0184] Generate PID control parameters in combination with risk level labels.

[0185] In a preferred embodiment, the dynamic adjustment step includes:

[0186] Dynamic adjustment of fuzzy PID parameters: Kp = 0.5 - 2.0, Ki = 0.1 - 0.5, Kd = 0.05 - 0.2;

[0187] Response time ≤ 50 ms, adjustment accuracy ±0.1%.

[0188] In a preferred embodiment, the safety protection step includes: Encrypt sensor data in fragments (AES-256); Transmit control instructions through a quantum key channel; Store operation logs on the blockchain to support audit and traceability. This embodiment enhances data security, effectively prevents data from being illegally obtained or tampered with during transmission and storage, and ensures the confidentiality and integrity of data.

[0189] Although the preferred embodiments of the present invention have been described, additional changes and modifications can be made to these embodiments by those skilled in the art once they learn the basic inventive concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the present invention. The above description is only for the preferred embodiments of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent adjustment system for soil penetration of a shield machine, characterized in that: include: High-precision multi-dimensional sensing module, including piezoelectric soil hardness sensors, silicon piezoresistive pore water pressure sensors, fiber Bragg grating ground stress sensors and platinum resistance temperature sensors distributed in a ring array. The sensor array is evenly distributed in a ring around the shield machine cutterhead. The risk level module uses the Gaussian kernel function to calculate the similarity matrix between segment rings, divides the performance community through the Louvain algorithm, and outputs the geological risk level and corresponding coordinates; The intelligent decision-making module integrates a multi-layer convolutional neural network to fuse sensor data and network analysis results in real time to generate dynamic instructions for cutter head speed, propulsion speed, and grouting pressure.

2. According to claim 1, a shield machine soil penetration force intelligent adjustment system is characterized in that: The workflow of the risk level module includes: Map the shield segment rings to network nodes and calculate the similarity matrix between segment rings. The formula is: X i is the multidimensional eigenvector of the segment ring, σ is the multiple of the standard deviation of the eigenvector; Generate a binary adjacency matrix by thresholding; The network topology is analyzed based on node degree, average shortest path length and clustering coefficient, and high-risk communities are divided.

3. The intelligent adjustment system for soil penetration of a shield machine according to claim 1 is characterized in that: The input of the intelligent decision-making module includes: Real-time sensor data, including soil hardness, pore water pressure, geostress and temperature; The risk level label output by the risk level module; Historical geological parameter database; The output is dynamic adjustment instructions, including: cutter head speed, propulsion speed, and grouting pressure.

4. The intelligent adjustment system for soil penetration of a shield machine according to claim 1 is characterized in that: It also includes an edge-cloud computing collaboration module, where edge nodes process real-time control instructions and the cloud platform performs complex network modeling and deep learning model training; In the edge-cloud computing collaborative module: it is deployed locally on the shield machine and directly connected to the sensor using an industrial-grade Ethernet architecture; it integrates an artificial intelligence-assisted decision-making module to predict geological risks in the next 10 minutes.

5. The intelligent adjustment system for soil penetration of a shield machine according to claim 1 is characterized in that: It also includes a blockchain security module, which uses AES-256 sharding to encrypt sensor data and transmits instructions through a quantum key distribution channel; the blockchain security module includes: storage in distributed nodes, single node failure does not affect the overall integrity; using the PBFT consensus algorithm, the transaction confirmation time is <1 second.

6. The shield machine soil penetration force intelligent adjustment system according to claim 1 is characterized in that: Also includes a three-level pressure protection mechanism: 4Mpa>When the soil bin pressure is ≥3MPa, the first-level warning is triggered and the advancement speed is reduced to 50mm / min; 5Mpa>soil bin pressure≥4MPaStart secondary protection, stop the cutter head rotation and increase pressure for grouting; When the soil bin pressure is ≥5MPa, the system will enter safety mode, cut off power and start the emergency drainage system.

7. The shield machine soil penetration force intelligent adjustment system according to claim 1, characterized in that: The neural network training process of the intelligent decision-making module includes: inputting historical geological data and construction parameters; using Adam optimizer and ReLU activation function, and the loss function is a combination of mean square error and cross entropy.

8. The shield machine soil penetration intelligent adjustment system according to claim 1, characterized in that: The system is equipped with adaptive working condition optimization function: Soft soil layer: cutter head speed ≤5rpm, grouting pressure ≥1.2MPa; Hard rock formation: cutter head speed ≥ 8rpm, advance speed ≤ 60mm / min; Water-rich layer: The amount of sealing grease injected is increased and the drainage frequency is increased to 2 times / minute.

9. A shield machine soil penetration intelligent adjustment method, characterized in that: Data acquisition: Real-time acquisition of soil hardness, pore water pressure, ground stress and temperature data through a circular sensor array; Network modeling (risk level module): construct a complex network of shield segments and divide geological risk communities; Intelligent decision-making: Integrate real-time data and network analysis results to generate control instructions; Dynamic adjustment: adjust the cutter head speed, propulsion speed and grouting pressure based on fuzzy PID algorithm.

10. The method for intelligently adjusting soil penetration force of a shield machine according to claim 9, characterized in that: The network modeling step includes: Calculate the segment-ring similarity matrix, and select the threshold as similarity ≥ 0.7; The community was divided based on the Louvain algorithm, and the modularity ≥ 0.3 was determined as an independent risk community; Output the risk community coordinates and recommended grouting pressure increment.

Images