A mode conversion system and method for a multi-mode shield machine in composite ground

Abstract

The present application relates to the technical field of tunneling equipment, in particular to a mode conversion system of a multi-mode shield tunneling machine in composite stratum, comprising a prediction layer, a perception layer, a data bus layer, a decision layer and an execution layer: the prediction layer obtains geological and hydrological information in front through TSP advanced geological prediction and an infrared water detector; the perception layer integrates multiple sensors to collect tunneling parameters and processes them through a data fusion algorithm; the data bus layer realizes spatio-temporal alignment of data based on a distributed stream processing platform; the decision layer generates conversion instructions using a CNN + fuzzy logic + reinforcement learning hybrid model; the execution layer drives an execution mechanism to complete EPB / SPB / TBM mode conversion and ensures safety through feedback correction. The present application realizes adaptive decision-making and seamless execution of mode conversion, significantly shortens conversion time, reduces tool wear, and improves composite stratum construction efficiency and safety.

Description

Technical Field

[0001] This invention relates to the field of tunnel boring equipment technology, and in particular to a mode conversion system and method for multi-mode shield tunneling machines in complex strata. Background Technology

[0002] As urban underground space development extends to areas with complex geological conditions, multi-mode tunnel boring machines (TBMs) have become core equipment due to their ability to adapt to complex strata such as alternating soft and hard rocks, karst caves, and fault zones. Multi-mode TBMs are mainly divided into two categories: dual-mode TBMs and triple-mode TBMs. Dual-mode TBMs are further divided into Earth Pressure Balance (EPB) / Hard Rock TBM dual-mode TBMs, EPB / Slurry Balance (SPB) dual-mode TBMs, and SPB / Hard Rock TBM dual-mode TBMs. The EPB / SPB dual-mode TBM combines two pressure control modes: earth pressure balance and slurry balance. Mode switching is achieved through a muck removal system. In low-permeability strata such as clay and silt, the excavation face is balanced by the soil chamber pressure; in high-permeability or high-water-pressure strata such as sand and gravel layers, the excavation face is stabilized by slurry circulation. The EPB / TBM dual-mode shield tunneling machine combines earth pressure balance (EPB) and hard rock tunneling capabilities. Mode switching is achieved by changing the cutterhead or adjusting the thrust system. The EPB mode is used for soft rock, weathered rock, and soft soil composite strata, while the TBM mode is used for full-section hard rock strata. The SPB / TBM dual-mode shield tunneling machine integrates slurry circulation and hard rock breaking functions, suitable for strata alternating between high water pressure hard rock and permeable soft soil. The tri-mode shield tunneling machine further integrates three modes based on the dual-mode design, achieving fully adaptable tunneling in extremely complex strata, and is compatible with slurry circulation, spiral muck removal, and roller cutter rock breaking systems within limited space. However, existing technologies mainly have the following shortcomings:

[0003] 1. Traditional multi-mode shield tunneling mode conversion relies on preset geological parameters, manual experience-based judgment, or offline simulation. This cannot adapt to the dynamic changes in complex strata, resulting in response lag and the risk of misjudgment. Statistics show that the human misjudgment rate reaches 18%. 2. Although existing online multi-mode shield tunneling has shortened the conversion time, the muck removal system has a fixed layout, lacks geological adaptability, and is prone to abnormal cutter wear. 3. The mode conversion decision is not closed-loop: grouting reinforcement ensures conversion safety, but it is not linked with real-time strata parameters, resulting in the risk of lag. 4. The mechanical system coordination during the mode conversion process is poor, posing safety hazards such as hydraulic shock and seal failure.

[0004] Therefore, there is an urgent need to develop a full-link adaptive system that integrates multi-source sensing, dynamic decision-making, and mechanical linkage to solve the real-time and security issues of mode switching in complex strata. Summary of the Invention

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution.

[0006] Design a mode conversion system for a multi-mode tunnel boring machine in composite strata, including...

[0007] Prediction layer: includes an advanced detection module, used to acquire geological structural features and hydrological information within a preset range in front of the tunnel boring machine;

[0008] The perception layer includes a real-time monitoring module, which is used to monitor and collect the operating parameters and working condition data of the tunnel boring machine in real time, and process the real-time monitoring data through a data fusion algorithm.

[0009] Data bus layer: Based on a distributed stream processing platform, the advanced detection data of the prediction layer and the real-time monitoring data of the perception layer are aligned and fused in the time and spatial dimensions.

[0010] Decision-making level: Based on spatiotemporally aligned advanced detection data and real-time monitoring data, a hybrid model integrating convolutional neural network (CNN) for geological feature extraction, fuzzy logic control, and reinforcement learning is used to generate mode conversion decision instructions;

[0011] Execution layer: Responds to decision commands, drives the tunnel boring machine's actuators to complete the transition between multiple modes, and dynamically adjusts execution parameters through closed-loop feedback.

[0012] Preferably, the advanced detection module includes an advanced geological prediction device and an infrared water detector. The advanced geological prediction device can detect at least 100m ahead, and the infrared water detector can identify fault water content information within a range of at least 30m ahead.

[0013] Preferably, the real-time monitoring module includes a cutterhead torque sensor, a feed speed sensor, a soil chamber pressure sensor, and a seepage monitoring instrument;

[0014] The data fusion algorithm includes wavelet transform denoising algorithm and Kalman filter algorithm.

[0015] The wavelet transform denoising algorithm adaptively decomposes the sensor signal through wavelet hierarchical denoising, and the number of decomposition layers is dynamically selected according to the signal bandwidth.

[0016] The Kalman filter algorithm fuses data from multiple sensors and introduces a sensor reliability weighting factor for state estimation.

[0017] An online compensation model for the drift error of the earth chamber pressure sensor was established.

[0018] Preferably, the data bus layer uses a distributed stream processing platform to achieve data integration. The distributed stream processing platform is the Apache Kafka platform, and data alignment is achieved in the following ways:

[0019] Data standardization: Real-time monitoring data is processed using the Protobuf protocol, and advanced detection data is processed using the Avro+GeoJSON protocol;

[0020] Time alignment: Time is unified across all devices via an NTP server, with a time error of less than 1ms;

[0021] Spatial alignment: Establish the transformation relationship between the local coordinate system of the tunnel boring machine and the global coordinate system of the project to achieve spatial location matching between monitoring data and geological data.

[0022] Preferably, the algorithm for the hybrid model includes:

[0023] Geological feature extraction model: It adopts a dual-branch processing structure to process advanced geological exploration image data and water-bearing identification parameters separately, and outputs stratigraphic classification results, permeability coefficient and risk area information;

[0024] Fuzzy logic control: A dynamic rule base is constructed based on the denoised real-time monitoring data, which includes cutterhead torque, feed speed, soil chamber pressure, and water seepage.

[0025] Reinforcement learning: Taking land subsidence increment, energy consumption and equipment health as optimization targets, the control strategy is adjusted through reward function, and mode switching instructions and execution parameters are output.

[0026] Preferably, the execution layer includes an instruction parsing module, an action execution module, a security monitoring module, and a feedback correction module;

[0027] in,

[0028] The instruction parsing module is used to decompose decision instructions into a sequence of coordinated actions across multiple systems;

[0029] The motion execution module is used to drive the coordinated action of the hydraulic system, grouting system and soil removal system, and to realize the conversion between earth pressure balance mode, slurry balance mode and hard rock tunneling mode.

[0030] The security monitoring module is used to verify the execution status in real time and trigger abnormal rollback or emergency shutdown.

[0031] The feedback correction module dynamically adjusts the execution parameters based on sensor closed-loop feedback.

[0032] Preferably, the advanced geological prediction equipment uses seismic wave method to detect the structural state and integrity of the surrounding rock, and the infrared water detector identifies hidden water sources by detecting the distortion of the infrared radiation field of the geological body. The two are matched with structural anomaly areas and water-bearing anomaly areas through a complementary data fusion mechanism.

[0033] Preferably, in the dual-branch processing structure of the geological feature extraction model, one branch processes seismic wave image data and the other branch processes infrared parameters; the reinforcement learning training process combines expert experience data, and the action exploration range is limited to within the engineering safety threshold.

[0034] Preferably, the mode switching process includes the coordinated action of screw conveyor locking, mud pump pressure regulation and shield tail seal dynamic control; the abnormal rollback strategy of the safety monitoring module is dynamically triggered according to the seepage volume and soil chamber pressure deviation.

[0035] A method for mode conversion of a multi-mode tunnel boring machine in composite strata includes the following steps:

[0036] Step S1: Obtain geological structure features and hydrological information within a preset range in front of the tunnel boring machine using advanced detection equipment;

[0037] Step S2: Collect operating parameters and working condition data of the tunnel boring machine during the tunneling process through multiple real-time monitoring sensors, and process the operating parameters and working condition data using a data fusion algorithm;

[0038] Step S3: Based on the distributed stream processing platform, align and fuse the geological and hydrological data obtained in step S1 with the real-time monitoring data processed in step S2 in both time and spatial dimensions.

[0039] Step S4: Based on the aligned and fused data from Step S3, a hybrid algorithm combining geological feature extraction model, fuzzy logic control, and reinforcement learning is used to generate mode switching decision instructions and corresponding execution parameters.

[0040] Step S5: According to the mode conversion decision command, drive the shield machine's actuator to complete the conversion between earth pressure balance mode, slurry balance mode and hard rock tunneling mode, and dynamically adjust the execution parameters through sensor closed-loop feedback. At the same time, perform safety monitoring on the conversion process and trigger emergency response in abnormal situations.

[0041] The beneficial effects of this invention are as follows:

[0042] 1. This invention enables advanced detection of geological structures up to 100m ahead and water content of faults up to 30m ahead through the TSP advanced geological prediction equipment and infrared water detector in the prediction layer. Combined with real-time monitoring by sensors such as cutterhead torque, propulsion speed, and soil chamber pressure in the sensing layer, and then precise spatiotemporal alignment through the data bus layer, it provides comprehensive data support for decision-making. The hybrid model that integrates CNN geological feature extraction and reinforcement learning can dynamically identify stratigraphic changes and generate the optimal conversion strategy, reducing the conversion time from the traditional 35 minutes to less than 18 minutes.

[0043] 2. This invention employs a data fusion algorithm combining Kalman filtering and wavelet transform, effectively improving the accuracy of monitoring data and reducing misjudgments caused by data noise; the decision-making mechanism combining fuzzy logic and reinforcement learning can dynamically adjust parameters such as cutterhead torque and soil chamber pressure according to geological characteristics, reducing tool wear rate by more than 42%; the safety monitoring and feedback correction module of the execution layer, through abnormal rollback and emergency shutdown mechanisms, controls surface settlement to 3.2mm (lower than the standard value of 8mm), significantly improving construction safety.

[0044] 3. This invention constructs a full-process adaptive closed-loop control system of "advanced prediction - real-time monitoring - intelligent decision-making - precise execution", breaking through the reliance on traditional manual experience. Through multi-source data fusion and intelligent algorithm autonomous optimization of mode conversion strategy, it adapts to the dynamic changes of complex strata. At the same time, the coordinated linkage of modular actuators (such as screw conveyor interlocking, mud pump pressure regulation, etc.) realizes seamless conversion of EPB / SPB / TBM modes, thereby improving construction efficiency.

[0045] 4. The mode conversion system of this invention can accurately identify different strata types such as clay, sand, and hard rock, as well as water-rich risk areas, and dynamically match the optimal tunneling mode. It performs excellently in extremely complex strata such as water-rich faults and alternating soft and hard rocks, and not only reduces energy consumption (e.g., the energy consumption of the mud circulation system is reduced to 1.8 kW·h / m), but also... 3 This method reduces costs by 33% compared to traditional methods, and also reduces rework and equipment maintenance costs, making it of significant engineering application value. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the overall system architecture of the present invention;

[0047] Figure 2 This is a schematic diagram of the data fusion processing process;

[0048] Figure 3 This is a flowchart of data processing at the data bus layer;

[0049] Figure 4 It is a collaborative decision-making flowchart;

[0050] Figure 5 This is a flowchart of the execution layer processing; Detailed Implementation

[0051] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0052] Example 1

[0053] A mode conversion system for multi-mode tunnel boring machines in complex geological formations, such as Figures 1 to 5 As shown, including

[0054] Prediction Layer: This layer includes an advanced detection module for acquiring geological structural features and hydrological information within a preset range ahead of the tunnel boring machine (TBM). The advanced detection module includes advanced geological prediction equipment and an infrared water detector. The advanced geological prediction equipment can detect up to 100m ahead, and the infrared water detector can identify fault water content information within at least 30m ahead. The advanced geological prediction equipment uses seismic wave methods to detect the state and integrity of the surrounding rock structure; its model is TSP-SK. The infrared water detector is an HW-304 tunnel infrared water detector, which identifies hidden water sources by detecting the distortion of the infrared radiation field of the geological body; its resolution is 0.05mW / cm² in H mode. 2 M mode 0.07mW / cm 2 Operating temperature: 0-40℃, humidity ≤80%. The two are matched with structural anomaly areas and water-bearing anomaly areas through a complementary data fusion mechanism.

[0055] The perception layer includes a real-time monitoring module, which is used to monitor and collect the operating parameters and working condition data of the tunnel boring machine in real time, and process the real-time monitoring data through a data fusion algorithm. The real-time monitoring module includes a cutterhead torque sensor, a propulsion speed sensor, a soil chamber pressure sensor, and a seepage monitoring instrument. The data fusion algorithm includes a wavelet transform denoising algorithm and a Kalman filter algorithm. Wavelet transform is a time-frequency analysis tool that decomposes the signal into multiple resolutions through basis functions of different scales, and can accurately locate the frequency band distribution of noise and effective signals.

[0056] Kalman filtering is mainly used for denoising and state estimation of real-time sensor data, and for correcting sensor drift. It ensures the real-time accuracy of the input data of the control algorithm. For example, when a sudden increase in torque is detected, an emergency deceleration is triggered immediately, which can significantly improve data quality and achieve efficient fusion and state estimation of multi-source sensors.

[0057] The wavelet transform denoising algorithm adaptively decomposes the sensor signal through wavelet hierarchical denoising, with the number of decomposition layers dynamically selected based on the signal bandwidth; the Kalman filter algorithm fuses multi-sensor data and introduces sensor reliability weighting factors for state estimation; and an online compensation model for the drift error of the earth chamber pressure sensor is established. Specifically, this includes:

[0058] Time synchronization: Configure GPS synchronization clock modules for all sensors, with timestamp accuracy ≤1ms;

[0059] Spatial mapping: Establishing the transformation relationship between the tunnel boring machine coordinate system (local) and the global coordinate system of the project:

[0060]

[0061] Where S is the tunneling mileage and θ is the shield machine attitude angle.

[0062] Wavelet hierarchical denoising, with adaptive selection of the number of decomposition layers, dynamically determined by the signal bandwidth J:

[0063]

[0064] Where fs is the sampling frequency and fc is the cutoff frequency of the effective components of the signal.

[0065] Improved threshold function: A piecewise adaptive thresholding method is used to handle high-frequency coefficients.

[0066]

[0067] Where, σ j Let N be the estimated standard deviation of the noise at layer j, and N be the signal length.

[0068] Multi-sensor Kalman fusion, state vector design

[0069] X k =[T k ,P k V k Q k ] T

[0070] Where T represents torque, P represents earth chamber pressure, V represents propulsion speed, and Q represents the optimal estimate of seepage volume.

[0071] The observation model was improved by introducing a sensor reliability weighting factor α. i :

[0072]

[0073] Among them, z k H represents the observed and measured values ​​at time point k; k This represents the observation or measurement matrix at time point k; x k It is the system state vector at time point k; T represents k ,P k V k Q k The four noises; v i,k α represents the noise of the i-th sensor at time point k; i Dynamically adjust based on the sensor's historical failure rate.

[0074] Online drift error compensation, drift model: Establishing drift compensation equations for the soil chamber pressure sensor:

[0075] P corrected =P raw -β·t

[0076] Where, P corrected Indicates the pressure value after compensation; P raw represents the original pressure value; β represents the drift coefficient, estimated online using the sliding window least squares method; t represents the time interval between the current time and the calibration time.

[0077] Data Bus Layer: Based on a distributed stream processing platform, the predictive layer's advance detection data and the perception layer's real-time monitoring data are aligned and fused in both time and spatial dimensions. The data bus layer uses a distributed stream processing platform, specifically Apache Kafka, for data integration. This platform supports high-throughput, low-latency real-time data transmission and is commonly used to build data pipelines and message buses. Data alignment is achieved through the following methods:

[0078] Data standardization: Real-time monitoring data is processed using the Protobuf protocol, and advanced detection data is processed using the Avro+GeoJSON protocol;

[0079] Time alignment: Time is unified across all devices via an NTP server, with a time error of less than 1ms; a high-precision timestamp is embedded in the data header for time alignment.

[0080] Spatial alignment: Establish the transformation relationship between the local coordinate system of the tunnel boring machine and the global coordinate system of the project to achieve spatial location matching between monitoring data and geological data.

[0081] Specifically, the spatial coordinate mapping relationship during spatial alignment is as follows: the real-time sensor position (relative coordinate system of the tunnel boring machine) is converted into the engineering absolute coordinate system that is located in real time through the tunnel boring machine guidance system; the advanced geological features (geographic coordinate system WGS84) are converted into the engineering coordinate system through the Proj4 library.

[0082] Based on the Apache Kafka platform, spatiotemporal consistency fusion of real-time monitoring data and advanced detection data is achieved through spatiotemporal composite key partitioning, stream processing window alignment, and dynamic coordinate mapping.

[0083] The spatiotemporal composite key includes ring number, timestamp, and geological feature type. The partitioning allocation strategy is dynamically calculated based on the ring number hash value.

[0084] The stream processing window is configured with a sliding time window and a watermarking mechanism to tolerate out-of-order data delays.

[0085] Using Kafka Streams state storage, and storing geological features in a geospatial database, enables real-time querying during stream processing.

[0086] The Kafka cluster optimization described above can be partitioned according to the tunnel boring machine ID or ring number to ensure that data in the same spatial location is in the same partition.

[0087] Decision layer: Based on spatiotemporally aligned advanced detection data and real-time monitoring data, a hybrid model integrating convolutional neural network (CNN) for geological feature extraction, fuzzy logic control, and reinforcement learning generates mode conversion decision instructions; the algorithm of the hybrid model includes:

[0088] Geological feature extraction model: For multimodal geological feature extraction, it analyzes aligned seismic waveforms and infrared thermal imaging data to extract features such as stratigraphic classification and permeability.

[0089] Input advanced geological prediction images (TSP seismic wave waveforms) and infrared water exploration parameters (temperature gradient matrix (ΔT / Δx) output by infrared thermal imager).

[0090] A dual-branch processing structure is adopted to process advanced geological exploration image data and water-bearing identification parameters separately, and output stratigraphic classification results, permeability coefficient and risk area information. The stratigraphic classification is clay, sand and hard rock.

[0091] Output permeability coefficient regression k value (×10) -5 The seismic feature extraction model uses a dual-branch processing structure: one branch processes seismic wave image data, and the other processes infrared parameters. The model is used to segment risk areas and identify water-rich fissures and cavities. It also analyzes relevant data to identify the condition of the tunnel face ahead.

[0092] Fuzzy logic control: A dynamic rule base is constructed based on the denoised real-time monitoring data, which includes cutterhead torque T (range 0~2500kN·m), feed speed V (range 0~60mm / min), soil chamber pressure P (range 0~0.6MPa) and seepage rate Q (range 0~50L / min);

[0093] Among them, fuzzy logic control converts the current precise perception layer data into fuzzy linguistic values, then defines the mapping relationship between numerical values ​​and fuzzy sets, calculates the output fuzzy set based on the input fuzzy values ​​and the rule base, and then performs defuzzification to convert the fuzzy output into precise control quantities.

[0094] Among them, the cutterhead torque is directly affected by the hardness, density and friction characteristics of the stratum, and can be used as one of the important bases for judging the timing of mode conversion of multi-mode shield tunneling machines. When the torque is continuously monitored to exceed the current mode threshold for a certain period of time, the system can issue an early warning or automatically trigger the mode conversion logic.

[0095] Among them, the advance speed is a direct feedback of the hardness of the strata. For example, the advance speed is usually higher in soft soil strata, while the advance speed decreases significantly in hard rock or composite strata.

[0096] Among them, the soil pressure needs to be dynamically matched with the soil and water pressure of the stratum to prevent the excavation face from becoming unstable, and can be used as one of the core parameters for the multi-mode shield tunneling mode conversion judgment.

[0097] Among them, the seepage rate parameter mainly reflects the real-time feedback of the stratum permeability and groundwater status, as well as the prediction of construction risks. Low seepage rate indicates that the stratum permeability is low, and it is mostly clay, silty clay, etc. High seepage rate indicates that the stratum permeability is high, and it is mostly sand, gravel, etc.

[0098] Fuzzy design: The membership function is optimized using an asymmetric Gaussian function to adapt to the dynamic range of parameters.

[0099]

[0100] Where x is the input variable, representing the original parameters to be fuzzified (cutterhead torque, feed speed, soil chamber pressure, seepage rate); σ L and σ R The left and right standard deviations (σ) L =σ R ), which controls the steepness of the curves on both sides of the center point; c is the center point of the membership function, and t is the segmentation threshold, which divides the application interval of the left and right standard deviations.

[0101] Dynamic rule base: Set 25 initial rules and optimize weights online through reinforcement learning.

[0102] Reinforcement learning: Targeting land subsidence increment, energy consumption, and equipment health, it adjusts control strategies through a reward function, outputting mode-switching commands and execution parameters. The reinforcement learning training process incorporates expert experience data, and the scope of action exploration is limited to engineering safety thresholds.

[0103] The specific optimization of reinforcement learning strategies is as follows:

[0104] State space: s t = [CNN feature vector (32-dimensional), blurring parameters (6-dimensional), historical action sequence (5 steps)]

[0105] Action space: Discrete actions: {EPB mode, SPB mode, TBM mode}

[0106] Continuous operation: mud pressure setting (0.2~0.6MPa), foam injection rate (0~50L / min).

[0107] Step S47, Set the reward function:

[0108]

[0109] Among them, R t V represents the reward value at time t. t V represents the propulsion speed. max ΔS represents the maximum propulsion speed. t S represents the increase in surface subsidence. th E is the settlement threshold. t For energy consumption, E max For the maximum allowable energy consumption, H t The equipment status is rated (0-100%), and the shield tunneling maintenance personnel evaluate and score it daily, with 100% indicating the best condition.

[0110] Training algorithm: Proximal policy optimization (PPO) combined with expert experience replay, with expert data accounting for 30% (historical operation records), and the action exploration range is limited to ±10% of the engineering safety threshold.

[0111] Multimodal decision fusion

[0112] Emergency layer: Fuzzy logic directly responds to a sudden increase in seepage (>30L / min) or torque exceeding the limit (>2000kN·m), triggering a mode switch;

[0113] Optimization layer: Reinforcement learning generates long-term strategies (such as planning tool changes 50 rounds in advance);

[0114] Human-machine collaboration: In abnormal operating conditions, push instructions with a confidence level of <80% to human confirmation.

[0115] Execution layer: Responds to decision commands, drives the tunnel boring machine's actuators to complete the transition between multiple modes, and dynamically adjusts execution parameters through closed-loop feedback. The execution layer includes a command parsing module, an action execution module, a safety monitoring module, and a feedback correction module.

[0116] The instruction parsing module is used to decompose decision instructions (mode switching, parameter adjustment) into a multi-system collaborative action sequence.

[0117] The action execution module is used to drive the hydraulic system, grouting system and soil removal system to work together, and to realize the conversion between earth pressure balance mode, slurry balance mode and hard rock tunneling mode; the mode conversion process includes the coordinated action of screw conveyor locking, mud pump pressure regulation and shield tail seal dynamic control.

[0118] The safety monitoring module is used to verify the execution status in real time and trigger abnormal rollback or emergency shutdown; the feedback correction module dynamically adjusts the execution parameters based on sensor closed-loop feedback. Furthermore, the abnormal rollback strategy of the safety monitoring module is dynamically triggered based on seepage volume and soil pressure deviation.

[0119] Example 2

[0120] A method for mode conversion of a multi-mode tunnel boring machine in composite strata includes the following steps:

[0121] Step S2: Collect operating parameters and working condition data of the tunnel boring machine during the tunneling process through multiple real-time monitoring sensors, and process the operating parameters and working condition data using a data fusion algorithm;

[0122] Step S3: Based on the distributed stream processing platform, align and fuse the geological and hydrological data obtained in step S1 with the real-time monitoring data processed in step S2 in both time and spatial dimensions.

[0123] Step S4: Based on the aligned and fused data from Step S3, a hybrid algorithm combining geological feature extraction model, fuzzy logic control, and reinforcement learning is used to generate mode switching decision instructions and corresponding execution parameters.

[0124] Step S5: According to the mode conversion decision command, drive the shield machine's actuator to complete the conversion between earth pressure balance mode, slurry balance mode and hard rock tunneling mode, and dynamically adjust the execution parameters through sensor closed-loop feedback. At the same time, perform safety monitoring on the conversion process and trigger emergency response in abnormal situations.

[0125] This invention enables conversion between Earth Pressure Balance (EPB), Slurry Balance (SPB), and Hard Rock Tunneling (TBM) modes:

[0126] I. Conversion from Earth Pressure Balance (EPB) to Slurry Balance (SPB) Mode

[0127] Application scenarios:

[0128] When the tunnel boring machine reached mileage SK12+356, the advanced detection module detected a highly permeable sand layer (permeability coefficient k = 2.3 × 10⁻⁶) 30m ahead. -5 The TSP waveform (m / s) showed that the formation integrity coefficient suddenly dropped from 0.85 to 0.52, and the infrared water detector detected that the water content of the fault zone reached 18%.

[0129] Implementation steps:

[0130] Step S1, data trigger (timestamp 2023-08-15T09:23:15.356Z)

[0131] Real-time data from the sensing layer: the cutterhead torque suddenly increased to 1850 kN·m (EPB mode threshold 1800 kN·m), the soil chamber pressure fluctuated by ±0.12 MPa (exceeding the limit by ±0.05 MPa), and the seepage rate increased from 8 L / min to 32 L / min;

[0132] Step S2, Data Fusion: After wavelet denoising, J=5-level decomposition, Kalman filtering is used to compensate for the pressure drift β=0.003MPa / h in the earth chamber;

[0133] Step S3, Decision generation (response time < 800ms);

[0134] CNN output: Sand layer probability 92%, permeability coefficient k = 2.1 × 10⁻⁶ -4 m / s (error <8%);

[0135] Reinforcement learning strategy: Reward function Rt = 0.72 (historical best Rmax = 0.81), generating action sequences:

[0136]

[0137] Step S4, execute control: screw conveyor lockout (time 2 min 15 s, pressure gradient 0.05 MPa / min),

[0138] Mud pump set startup (flow rate Q = 420 m³ / h) 3 / h, pipe diameter Φ250mm), shield tail seal dynamic adjustment (inject sealing grease 1.2kg / m, pressure 0.25MPa);

[0139] Step S5, Safety Monitoring: Closed-loop control of earth chamber pressure: PID parameters Kp = 2.5, Ki = 0.8, Kd = 0.3.

[0140] Abnormal rollback trigger condition: Switch back to EPB mode when seepage rate > 45L / min for 30 seconds.

[0141] Technical effects: The mode conversion time was reduced from the traditional 35 minutes to 18 minutes, the cutter wear rate was reduced by 42%, and the surface settlement was controlled at 3.2 mm (lower than the standard value of 8 mm). The reduction in cutter wear rate was derived from the following "Comparison Table of Cutter Wear Rate Before and After Mode Conversion": Specifically, the arithmetic mean of the reduction rate of the volumetric wear rate of 62 cutters was calculated, ultimately showing that the cutter wear rate was reduced by approximately 42% after the mode conversion. Specifically, the volumetric wear rate of the cutter rock breaking refers to the radial wear generated by the cutter excavating a unit volume of rock. The calculation method is as follows: v i =Q i / π(R i 2 -R i-1 2 )L, where v i The rock-breaking volumetric wear rate of the i-th cutter on the cutterhead is expressed in mm / m. 3 Q i R represents the cumulative wear of the i-th hob, in mm. iR represents the installation radius of the i-th hob, in meters (m). i-1 The radius of the previous cutter head is in meters (m); L is the tunneling distance of the tunnel boring machine (TBM) in meters (m). The higher wear on cutters #59, #61, and #62 is mainly due to more severe wear on the edge cutters, while the lower wear on the edge cutter #61 may be due to errors in manual measurement data.

[0142]

[0143]

[0144] II. Conversion from Hard Rock Tunneling (TBM) to Slurry Balance (SPB) Mode

[0145] Application scenarios:

[0146] When tunneling reached mileage SK14+215 and crossed a water-rich fault zone, the TSP showed that the RQD value of the rock mass 50m ahead dropped sharply from 85% to 32%. The infrared water detector detected a water temperature gradient ΔT / Δx = 0.8℃ / m (threshold 0.5℃ / m).

[0147] Implementation steps:

[0148] Step S1, Data Fusion: Advanced Detection Data: Fault width 4.8m, permeability coefficient k = 5.6 × 10⁻⁶ -5 m / s,

[0149] Real-time sensor: The vibration frequency of the cutter head suddenly increased from 58Hz to 82Hz, and the feed speed decreased from 18mm / min to 6mm / min;

[0150] Step S2, Decision Optimization: CNN Feature Vector: [0.87, 0.12, 0.01] (Probability distribution of hard rock / fault breccia / clay), Fuzzy Rule Activation: Rule #17 (High vibration + low propulsion speed → mud wall protection requirement 0.89), Reinforcement Learning Output: Mud viscosity set to 22s (Madrick funnel), injection pressure 0.42MPa;

[0151] Step S3, execution process:

[0152] Tool turret mode switching: The hob retraction angle is adjusted from 75° to 30°, and the tool load is controlled evenly.

[0153] Slag removal system conversion: belt conveyor shutdown → mud circulation system startup (time 9 min 23 s);

[0154] Synchronous grouting parameters: water-cement ratio 0.45:1, grouting pressure 0.28MPa (35% higher than normal);

[0155] Step S4, Safety Protection: Dynamic Limiting of Cutter Head Torque: The maximum value is limited to 85% of the rated torque (2200kN·m);

[0156] Emergency grouting system pre-start: 30m³ of grout in reserve 3 Ready to serve at any time.

[0157] Technical benefits: The energy consumption of the mud circulation system is lower than that of the traditional method (2.7 kW·h / m). 3 Reduced to 1.8 kWh / m 3 .

[0158] III. Conversion from Hard Rock Tunneling (TBM) to Earth Pressure Balance (EPB) Mode

[0159] Application scenarios:

[0160] When tunneling reached mileage SK15+780, advance detection showed that a clay interlayer (2.3m thick, 24% water content) appeared 18m ahead, and the cutterhead torque in TBM mode remained below 800kN·m.

[0161] Implementation steps:

[0162] Step S1, Data Preprocessing: Wavelet Denoising: db8 wavelet basis is selected, decomposition level J=4, signal-to-noise ratio is improved to 38dB; Kalman Fusion: State vector X_k=[785kN·m,0.11MPa,52mm / min,5L / min]^T;

[0163] Step S2, Hybrid Decision Making: CNN Multi-Task Output: Clay Probability 89%, Permeability Coefficient k = 3.2 × 10⁻⁶ -6 m / s;

[0164] Fuzzy control: membership degree μ_T(785) = 0.67 (TBM mode mismatch > 60%); reinforcement learning reward: R_t = 0.65 (mainly from the propulsion speed gain term);

[0165] Step S3, actuator linkage: cutterhead conversion: the roller cutter retracts across its entire cross-section (takes 6 min 10 s), and the scraper extension increases to 120 mm; soil chamber pressure build-up: the screw conveyor speed accelerates from 0 rpm to 8 rpm (gradient 0.5 rpm / s). 2 Foam system parameters: injection rate 12L / min, expansion ratio 25:1;

[0166] Step S4, Dynamic Feedback: Earth pressure balance control: Pressure setpoint P_set = 0.18MPa (dynamically calculated based on burial depth); Correction mechanism: The cutterhead torque deviation is checked every 30 seconds, and if it exceeds ±5%, the parameters are re-optimized.

[0167] Technical effect: The time for establishing soil pressure during mode conversion is reduced from 8 minutes in the traditional method to 3 minutes. Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A mode conversion system for a multi-mode tunnel boring machine in composite strata, characterized in that, include Prediction layer: includes an advanced detection module, used to acquire geological structural features and hydrological information within a preset range in front of the tunnel boring machine; The perception layer includes a real-time monitoring module, which is used to monitor and collect the operating parameters and working condition data of the tunnel boring machine in real time, and process the real-time monitoring data through a data fusion algorithm. Data bus layer: Based on a distributed stream processing platform, the advanced detection data of the prediction layer and the real-time monitoring data of the perception layer are aligned and fused in the time and spatial dimensions. Decision-making level: Based on spatiotemporally aligned advanced detection data and real-time monitoring data, a hybrid model integrating convolutional neural network (CNN) for geological feature extraction, fuzzy logic control, and reinforcement learning is used to generate mode conversion decision instructions; Execution layer: Responds to decision commands, drives the tunnel boring machine's actuators to complete the transition between multiple modes, and dynamically adjusts execution parameters through closed-loop feedback; The real-time monitoring module includes a cutterhead torque sensor, a feed speed sensor, a soil chamber pressure sensor, and a seepage monitoring instrument. The data fusion algorithm includes wavelet transform denoising algorithm and Kalman filter algorithm. The wavelet transform denoising algorithm adaptively decomposes the sensor signal through wavelet hierarchical denoising, and the number of decomposition layers is dynamically selected according to the signal bandwidth. The Kalman filter algorithm fuses multi-sensor data and introduces sensor reliability weighting factors for state estimation; an online compensation model for the drift error of the earth chamber pressure sensor is established. The data bus layer uses a distributed stream processing platform, namely Apache Kafka, to achieve data integration, and data alignment is achieved through the following methods: Data standardization: Real-time monitoring data is processed using the Protobuf protocol, and advanced detection data is processed using the Avro+GeoJSON protocol; Time alignment: Time is unified across all devices via an NTP server, with a time error of less than 1ms; Spatial alignment: Establish the transformation relationship between the local coordinate system of the tunnel boring machine and the global coordinate system of the project to achieve spatial location matching between monitoring data and geological data; The algorithm for the hybrid model includes: Geological feature extraction model: It adopts a dual-branch processing structure to process advanced geological exploration image data and water-bearing identification parameters separately, and outputs stratigraphic classification results, permeability coefficient and risk area information; Fuzzy logic control: A dynamic rule base is constructed based on the denoised real-time monitoring data, which includes cutterhead torque, feed speed, soil chamber pressure, and water seepage. Reinforcement learning: Taking land subsidence increment, energy consumption and equipment health as optimization targets, the control strategy is adjusted through reward function, and the output mode switching instructions and execution parameters are output.

2. The mode conversion system for multi-mode tunnel boring machines in composite strata as described in claim 1, characterized in that: The advanced detection module includes advanced geological prediction equipment and infrared water detector. The advanced geological prediction equipment can detect at least 100m ahead, and the infrared water detector can identify fault water content information within a range of at least 30m ahead.

3. The mode conversion system for multi-mode tunnel boring machines in composite strata as described in claim 1, characterized in that: The execution layer includes an instruction parsing module, an action execution module, a security monitoring module, and a feedback correction module; in, The instruction parsing module is used to decompose decision instructions into a sequence of coordinated actions across multiple systems; The motion execution module is used to drive the coordinated action of the hydraulic system, grouting system and soil removal system, and to realize the conversion between earth pressure balance mode, slurry balance mode and hard rock tunneling mode. The security monitoring module is used to verify the execution status in real time and trigger abnormal rollback or emergency shutdown. The feedback correction module dynamically adjusts the execution parameters based on sensor closed-loop feedback.

4. The mode conversion system for multi-mode tunnel boring machines in composite strata as described in claim 2, characterized in that: The advanced geological prediction equipment uses seismic wave method to detect the structural state and integrity of surrounding rock, and the infrared water detector identifies hidden water sources by detecting the distortion of the infrared radiation field of the geological body. The two use a complementary data fusion mechanism to match structural anomaly areas and water-bearing anomaly areas.

5. The mode conversion system for multi-mode tunnel boring machines in composite strata as described in claim 1, characterized in that: In the dual-branch processing structure of the geological feature extraction model, one branch processes seismic wave image data, and the other branch processes infrared parameters. The reinforcement learning and training process incorporates expert experience data, and the scope of action exploration is limited to within the engineering safety threshold.

6. The mode conversion system for multi-mode tunnel boring machines in composite strata as described in claim 3, characterized in that: The conversion process between the earth pressure balance mode, slurry balance mode and hard rock tunneling mode includes the coordinated action of screw conveyor interlocking, mud pump pressure regulation and shield tail seal dynamic control; the abnormal rollback strategy of the safety monitoring module is dynamically triggered according to the seepage volume and soil chamber pressure deviation.

7. A mode conversion method for a multi-mode tunnel boring machine in composite strata, implemented based on the mode conversion system for a multi-mode tunnel boring machine in composite strata according to any one of claims 1 to 6, characterized in that: Includes the following steps: Step S1: Obtain geological structure features and hydrological information within a preset range in front of the tunnel boring machine using advanced detection equipment; Step S2: Collect operating parameters and working condition data of the tunnel boring machine during the tunneling process through multiple real-time monitoring sensors, and process the operating parameters and working condition data using a data fusion algorithm; Step S3: Based on the distributed stream processing platform, align and fuse the geological and hydrological data obtained in step S1 with the real-time monitoring data processed in step S2 in both time and spatial dimensions. Step S4: Based on the aligned and fused data from Step S3, a hybrid algorithm combining geological feature extraction model, fuzzy logic control, and reinforcement learning is used to generate mode switching decision instructions and corresponding execution parameters. Step S5: According to the mode conversion decision command, drive the shield machine's actuator to complete the conversion between earth pressure balance mode, slurry balance mode and hard rock tunneling mode, and dynamically adjust the execution parameters through sensor closed-loop feedback. At the same time, perform safety monitoring on the conversion process and trigger emergency response in abnormal situations.

Images