Residue soil property sensing and mud cake intelligent removing system and method for cutterhead for shield tunneling

Abstract

The invention belongs to the field of mud cake removal of shield tunneling machines, and discloses a shield cutterhead residue soil property sensing and intelligent mud cake removal system and method, and the system comprises a sensing module, a modifier spraying module, a regulation and control module and an intelligent terminal; the sensing module is used for collecting mechanical, chemical and thermodynamic signals in a soil bin of the shield tunneling machine in real time; the modifier spraying module is used for intelligently regulating and controlling the proportion of a modifier so as to inhibit mud cake adhesion; the regulation and control module is used for dynamically optimizing a modifier proportioning parameter and a cutterhead tunneling strategy according to the sensing data; and the intelligent terminal is used for displaying the mud cake distribution state and receiving a manual intervention instruction. According to the method, three-dimensional form dynamic analysis and accurate targeted removal of the mud cake can be realized.

Description

Technical Field

[0001] The present invention belongs to the field of shield machine mud cake removal, and particularly relates to a shield cutter head slag property sensing and mud cake intelligent removal system and method. Background Art

[0002] Shield tunneling is the core construction technology for underground projects such as urban subways and cross-river tunnels. The dynamic behavior of the cutterhead-soil interface directly determines the excavation efficiency and project safety. Especially in highly viscous strata, the rapid accumulation of mud cakes will cause a surge in cutterhead torque, a decrease in excavation efficiency, and even mechanical failure. This causes huge economic losses in shield machine downtime accidents every year. In the existing technical system, there are significant defects in soil property perception and mud cake prevention and control: First, the traditional cutterhead sensor network (such as patent CN112228042A) only uses discrete pressure or temperature sensors, which cannot simultaneously obtain multi-dimensional data of mechanical, chemical and thermal signals, resulting in low mud cake removal efficiency and high risks of manual operation; second, the existing control system adopts an open-loop control mode, which has a long response time, the ratio of foaming agent and viscosity reducer cannot be adjusted in real time, the injection positioning error is large, and it cannot match the dynamic working conditions of the cutterhead.

[0003] The above defects lead to two major chain problems: (1) It is difficult to construct a three-dimensional digital twin model of the mud cake with single-dimensional sensing data, and the sudden change of local adhesion force of the cutter disc cannot be warned; (2) The control algorithm does not establish a dynamic mapping relationship between torque-propulsion speed-injection volume, and the concentration ratio of the improver has a high deviation, resulting in an imbalance in the cutter disc load and environmental pollution.

[0004] Especially during construction in high-risk strata, manual judgment is highly subjective, making it difficult for existing technologies to dynamically analyze mud cake rheological data and accurately target and control mud cakes. This forces frequent openings and clearing of slag, leading to high construction interruption rates. Therefore, there is an urgent need to develop a next-generation shield mud cake control system that integrates multimodal sensing fusion and intelligent closed-loop control. Summary of the Invention

[0005] In order to solve the problems existing in the prior art, the present invention provides a shield cutter head slag property perception and mud cake intelligent removal system and method, which can realize dynamic analysis of the three-dimensional morphology of mud cake and precise targeted removal.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A shield cutterhead soil property sensing and mud cake intelligent removal system, the system comprising: a sensing module, an improver injection module, a control module and an intelligent terminal;

[0008] The sensing module is used to collect mechanical, chemical and thermodynamic signals in the soil bin of the shield machine in real time;

[0009] The modifier injection module is used to intelligently control the modifier ratio to inhibit mud cake adhesion;

[0010] The control module is used to dynamically optimize the improver ratio parameters and the cutterhead excavation strategy based on the perception data;

[0011] The intelligent terminal is used to display the mud cake distribution status and receive manual intervention instructions.

[0012] Preferably, the sensing module includes: a soil bin rheometer, a soil bin pressure sensor, a soil bin temperature sensor and a soil bin chemical sensor group;

[0013] The soil bin rheometer is installed in the middle of the inner wall of the soil bin and is used to measure the viscosity and yield stress of the slag through rotational shear and output rheological characteristic parameters in real time;

[0014] The soil bin pressure sensor array is arranged at the entrance area of the soil bin screw conveyor to monitor the axial pressure distribution gradient of the slag;

[0015] The soil bin temperature sensor is integrated into the soil bin partition to detect the friction temperature rise and heat conduction characteristics of the slag soil;

[0016] The soil bin chemical sensor group is located at the soil bin slag discharge port and is used for online analysis of the clay mineral content and pH value in the slag.

[0017] Preferably, the modifier injection module includes: a modifier formulation unit, a storage and delivery unit, and an injection execution unit;

[0018] The improver formula unit is composed of a viscosity reducer and a foaming agent, and is used to adjust the distribution ratio of each improver component in real time according to the sensing data to reduce adhesion;

[0019] The storage and delivery unit is used to deliver the viscosity reducing agent to the cutter head surface via a high-pressure pump;

[0020] The injection execution unit includes a circumferentially distributed micro injection valve group for dynamically adjusting the injection angle and coverage range.

[0021] Preferably, the control module includes: a data fusion unit, a decision model unit and an instruction generation unit;

[0022] The data fusion unit is used to synchronously process multi-source sensing signals and construct a three-dimensional digital twin of the mud cake;

[0023] The decision model unit is used to generate optimization instructions based on the correlation between the cutter head torque, propulsion speed and mud cake thickness;

[0024] The instruction generation unit is used to control the injection valve group to perform targeted injection and cutter disc parameter adjustment.

[0025] Preferably, the intelligent terminal includes: a visual interface and an interaction unit;

[0026] The visualization interface is used to display the mud cake thickness distribution thermal map and the cutterhead load status in real time;

[0027] The interaction unit is used to receive a manually input priority adjustment instruction or an emergency stop instruction.

[0028] The present invention also provides a shield cutterhead slag soil property sensing and mud cake intelligent removal method, which is implemented by applying the aforementioned shield cutterhead slag soil property sensing and mud cake intelligent removal system. The method includes:

[0029] Step S1: synchronously collect mechanical, chemical and thermodynamic signals inside the soil bin through multimodal sensors to obtain multi-source data;

[0030] Step S2: Input multi-source data into the deep learning model to dynamically analyze the three-dimensional morphology and adhesion distribution of the mud cake;

[0031] Step S3: Calculate the optimal injection amount and target position of the viscosity reducer according to the mud cake thickness and the cutter head load state;

[0032] Step S4: Control the micro-injection valve group to perform injection according to a dynamic time sequence, and adjust the cutter head speed and propulsion pressure at the same time;

[0033] Step S5: Closed-loop optimization of control parameters based on real-time feedback data until the mud cake thickness drops to a safe threshold.

[0034] Preferably, the step S4 includes:

[0035] Establish a dynamic matching model based on multi-source data;

[0036] According to the dynamic proportioning model, a polar coordinate nozzle matrix is constructed;

[0037] Based on the polar coordinate nozzle matrix, multi-reagent coordinated injection control is performed;

[0038] Based on the results of multi-reagent collaborative injection control, three-dimensional path planning and mechanical coordination are carried out;

[0039] Based on the structure of 3D path planning and mechanical collaboration, closed-loop performance evaluation and parameter iteration;

[0040] Among them, the multi-source data include: mud cake thickness, mud cake azimuth, soil particle size distribution, temperature change rate, and rheological index;

[0041] The dynamic proportioning model includes: foaming agent basic flow rate: Q_foam = K1*h^1.2*exp(-0.5*d50); viscosity reducer ratio coefficient: α_detergent = 1 / (1+exp(-0.3*(ΔT / Δt-2))); flushing water pressure: P_water = min(8,0.6*h+0.4*(K / K_threshold)^3);

[0042] Total control intensity factor: Γ = [0.7, 0.2, 0.1] * [h / 50; K / K_threshold; ΔT / Δt];

[0043] K1 is the coefficient dynamically matched through the geological parameter library;

[0044] The polar coordinate nozzle matrix includes:

[0045] The circumference of the cutter head is divided into 36 10° sectors;

[0046] Three groups of directional nozzles are deployed in each sector.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] 1. Multimodal perception fusion: Through the coordinated collection of mechanical, chemical, and visual signals and deep learning modeling, the purpose of real-time control of the distribution ratio of each group of improvers is achieved, with higher accuracy than traditional single sensors;

[0049] 2. Closed-loop intelligent control: short response time, high targeting accuracy, reduced frequency of manual intervention, and improved continuous tunneling efficiency;

[0050] 3. Transparency throughout the entire process: A visual interface displays the cutterhead status in real time, providing digital decision-making support for construction in complex strata. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0052] Figure 1 This is a schematic structural diagram of a shield cutterhead slag property perception and mud cake intelligent removal system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0054] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0055] Example 1

[0056] As can be seen from the background technology,

[0057] In view of the following defects in the existing technology: 1. The perception dimension of the rheological data of the slag in the soil bin is single, and the discrete sensor network cannot synchronously obtain multi-dimensional data of mechanical, chemical and thermal signals, resulting in large errors in the analysis of the slag properties; 2. The mud cake prevention and control system has a long response time, large injection positioning errors, and large deviations in the concentration ratio of the improver, resulting in an imbalance in the cutter head load and waste of resources. The present invention aims to provide a shield cutter head slag property perception and mud cake intelligent removal system. Specifically solve the following technical problems: ① Establish a full-domain perception system for the cutter head that integrates radar, micro-cameras and multi-physics field sensors to achieve the purpose of real-time perception of the rheological data of the slag in the soil bin; ② Construct an adaptive closed-loop control system based on a multi-objective decision-making model to intelligently adjust the improver ratio, shorten the response time, improve the precision of targeted injection, and achieve a reduction in the amount of slag improver and an improvement in continuous excavation efficiency.

[0058] like Figure 1 As shown, the present invention provides a shield cutter head slag property perception and mud cake intelligent removal system comprising: a perception module, an improver injection module, a control module and an intelligent terminal;

[0059] The sensing module is used to collect mechanical, chemical and thermodynamic signals in the soil bin in real time;

[0060] Improver injection module, used to intelligently control the improver ratio to inhibit mud cake adhesion;

[0061] A control module is used to dynamically optimize the modifier ratio parameters and cutterhead excavation strategy based on the sensing data;

[0062] Intelligent terminal, used to display the mud cake distribution status and receive manual intervention instructions.

[0063] In this embodiment, the sensing module includes: a soil bin rheometer, a soil bin pressure sensor, a soil bin temperature sensor, and a soil bin chemical sensor group;

[0064] The soil bin rheometer is installed in the middle of the inner wall of the soil bin. It measures the viscosity and yield stress of the soil through rotational shear and outputs rheological characteristic parameters in real time.

[0065] The soil bin pressure sensor array is arranged at the entrance area of the soil bin screw conveyor to monitor the axial pressure distribution gradient of the soil;

[0066] The soil bin temperature sensor is integrated into the soil bin partition to detect the friction temperature rise and heat conduction characteristics of the soil;

[0067] The soil bin chemical sensor group is located at the slag discharge port of the soil bin, and is used to online analyze the clay mineral content (such as the proportion of montmorillonite and illite) and pH value in the slag.

[0068] In this embodiment, the modifier injection module includes: a modifier formulation unit, a storage and delivery unit, and an injection execution unit;

[0069] The improver formula unit is composed of viscosity reducers, foaming agents, etc., and the distribution ratio of each improver component is adjusted in real time according to the sensing data to reduce adhesion;

[0070] The storage and delivery unit delivers the viscosity reducer to the cutter disc surface through a high-pressure pump;

[0071] The injection execution unit contains a circumferentially distributed micro-injection valve group that can dynamically adjust the injection angle and coverage range.

[0072] Specifically, the distribution ratio of each group of improvers is adjusted in real time according to the perception data, including:

[0073] Step 1: Coupling and matching of geological parameters

[0074] The geological characteristic vector G = [montmorillonite content, illite / kaolinite ratio, pore water mineralization] is established, and the geological adaptation coefficient K1 is calculated by fuzzy neural network, where K1 is calculated by Sigmoid activation function, and the weight matrix W_g is obtained by training 300 sets of different formation conditions.

[0075] Step 2: Multi-physics field coupling feedback control

[0076] Construct a real-time control parameter matrix Ψ(t), including the mud cake thickness h(t), temperature gradient The rheological index K(t) and pH deviation coefficient ζ_{pH}(t) are calculated, and a dynamic weight allocator is designed to realize the non-balanced weight dynamic allocation of parameters.

[0077] Step 3: Quantum Jet Control

[0078] The circumference of the cutter disc is divided into 36 quantum control units (QCEs). Each unit is equipped with three groups of high-frequency piezoelectric ceramic nozzles and micro eddy current sensors. The state superposition equation of the injection quantum is established, and nano-scale mixing coverage is achieved through dynamic calculation of the coefficients α (foaming agent), β (viscosity reducer), and γ (flushing water).

[0079] Step 4: Bionic Pulse Jet Sequence

[0080] Generates a jet waveform with biological nerve pulse characteristics, the frequency f_c is adaptively adjusted according to the resonance frequency of the mud cake, and the phase difference Used to eliminate injection blind areas.

[0081] Step 5: Metabolic Closed-Loop Optimization

[0082] A "metabolism-regeneration" model of the modifier is constructed. When the concentration decay rate is detected to exceed the threshold, the slow-release modifier injection protocol is triggered to maintain the continuity of the chemical action.

[0083] Dynamically adjust spray angle and coverage, including:

[0084] Step 1: Multi-level spatial grid division

[0085] (1) First-level partitioning: The cutterhead circumference is divided into 36 10° sectors (numbered N1-N36) for macro-positioning. The span angle of each sector is 10°±0.5°, corresponding to an arc length of 52.4 mm in the direction of the cutterhead circumference (calculated based on a 6 m diameter cutterhead). (2) Second-level partitioning: Each sector is divided into 57 hexagonal honeycomb units (side length 8 mm±0.2 mm). The entire cutterhead has a total of 36×57=2052 units, with a unit center spacing of 14 mm, for micro-injection control.

[0086] Step 2: Dynamic Angle Adjustment Mechanism

[0087] (1) Basic elevation angle setting: Calculate the initial injection elevation angle based on the mud cake thickness h (unit: mm) When h≤30mm, (Unit: degree) When h>30mm,

[0088] (2) Real-time deflection compensation: Dynamic fine-tuning of the nozzle elevation angle is achieved through the piezoelectric ceramic driver, and the adjustment amplitude Response time <5ms, positioning accuracy ±0.1°;

[0089] Step 3: Coverage range optimization control (1) Single nozzle coverage diameter D_n = 3.2×h^0.7 (unit: mm, h is the mud cake thickness) (2) Overlap rate control: adjacent nozzle jet overlap rate η = 22%×(1+0.05×K), where K is the rheological index (dimensionless, range 0.5-1.8) (3) Boundary correction algorithm: Enable the jet focusing mode in the 5% area of the cutter disc edge (radius 2.85-3.0m), and the coverage diameter is reduced to 65%±3% of the normal value.

[0090] In this embodiment, the control module includes: a data fusion unit, a decision model unit and an instruction generation unit;

[0091] The data fusion unit synchronously processes multi-source sensing signals and constructs a three-dimensional digital twin of the mud cake;

[0092] The decision model unit generates optimization instructions based on the correlation between cutterhead torque, propulsion speed and mud cake thickness;

[0093] The command generation unit controls the injection valve group to perform targeted injection and cutter disc parameter adjustment.

[0094] Specifically, the simultaneous processing of multi-source sensing signals and the construction of a 3D digital twin of the mud cake include:

[0095] Step 1: Spatiotemporal alignment of multi-source data

[0096] (1) Time synchronization: Using the IEEE 1588 precision time protocol, the clock deviation of each sensor is controlled within ±50μs, and a unified time scale system is established; (2) Spatial alignment: The spatial coordinates of each sensor are calibrated by a laser tracker, and the transformation matrix T_{sensor}∈SE(3) is constructed, with a positioning error of <0.2mm; (3) Feature extraction: The raw data (pressure, temperature, chemical signals) are decomposed by wavelet packets to extract a 16-dimensional feature vector F = [f1,...,f16], with a quantization accuracy of ±0.5% for each dimension.

[0097] Step 2: Dynamic Weight Fusion

[0098] (1) Constructing credibility indicators:

[0099] Mechanical confidence C_m = 1-exp(-0.3×ΔP / σ_P) (ΔP is the pressure gradient modulus, σ_P = 0.15 MPa); thermodynamic confidence C_t = tanh(0.5×ΔT / 5) (ΔT is the temperature change rate, unit: °C / min); chemical confidence C_c = 1 / (1 + 10^{pH-7});

[0100] Fusion rules:

[0101] Total fused data D = Σ(w_i × D_i), weight w_i = softmax([2C_m, 1.5C_t, C_c]); update frequency 200 Hz, delay < 5 ms.

[0102] Step 3: Self-repair of abnormal data

[0103] (1) Establish a sliding time window: the window length τ = 2s, storing 600 sets of historical data; (2) Anomaly detection: trigger repair when |D(t)-μ_τ|>3σ_τ (μ is the mean, σ is the standard deviation); (3) Repair algorithm: use bidirectional LSTM to predict replacement values, with a prediction error of <4.7%.

[0104] Generate optimization instructions based on the relationship between cutterhead torque, propulsion speed and mud cake thickness, including:

[0105] Step 1: Dynamic parameter set construction: (1) Input parameter set: cutter head torque T (unit kN·m, range 0-8500, accuracy ±0.3%); mud cake thickness distribution matrix H [36×57] (unit mm, resolution 0.1mm); temperature gradient vector (unit: °C / mm, range: 0-15); rheological index K (dimensionless, range: 0.5-2.2, according to ISO 3219); (2) environmental parameters: formation type code E (classes 1-8, classified by montmorillonite content); excavation speed v (unit: mm / min, range: 10-80).

[0106] Step 2: Multi-objective optimization model: (1) Objective function: Minimize: 0.4×(T / T_max)+0.3×(ΣH / 3600)+0.2×(ΔP / ΔP_max)+0.1×(Q / Q_max); Parameter description: T_max=8000kN·m (maximum design torque of the cutterhead); ΔP=injection pressure fluctuation amplitude (reference value 3MPa); Q=instantaneous flow rate of the modifier (safety threshold 120L / min). (2) Constraints: Cutterhead vibration acceleration ≤2.5m / s 2 (ISO 10816-3 standard); mud film residual thickness ≥ 0.3mm (to prevent cutter head wear); reagent mixing ratio error ≤ ± 4.7%.

[0107] Step 3: Instruction generation logic: (1) Dynamic priority allocation: When H_max>50mm, activate the "peak removal" mode and adjust the weights to: [0.6, 0.2, 0.1, 0.1]; when When the temperature is 0.05°, the "thermal shock protection" mode is activated and the temperature weight factor λ is increased to 1.5; (2) Nozzle control instructions: Generate a 36×57 control matrix, each element of which contains: injection pulse width PW_ij=12×(H_ij)^0.7×(1+0.05E) (unit: ms, range: 5-200); atomization angle θ_ij=35°+0.3×(H_ij-20) (dynamic adjustment is started when Hi_ij≥20mm); (3) Cutter motion instructions: speed ω=ω_0×(1-0.018×H_avg), H_avg is the average thickness (reference speed ω_0=1.2rpm); reciprocating swing amplitude A=min(5°,0.1×H_max).

[0108] Step 4: Real-time feedback optimization: (1) Establish performance evaluation indicators:

[0109] η=0.7×(ΔH / Δt)+0.2×exp(-|T-T_target|)+0.1×(1-Q_used / Q_limit); Parameter Description: ΔH / Δt: Mudcake thickness change rate (target ≥-0.8mm / s); T_target=0.7T_max (torque optimization target value); Q_limit=maximum reagent flow rate allowed by the current geological formation. (2) Dynamic parameter correction: Online calibration is performed every 15 seconds: If η<0.6: K_p*=1.25 (proportional coefficient gain adjustment) H_threshold+=2mm (thickness threshold increment) Else: The historical optimal parameter set is retained.

[0110] Control the injection valve group to perform targeted injection and cutterhead parameter adjustment, including:

[0111] Step 1: Adaptive pulse width modulation injection: (1) Dynamic adjustment of pulse frequency: basic frequency f0 = 50 Hz (corresponding to normal working conditions); when the mud cake thickness H ≥ 40 mm, start high frequency mode: f = 50 + 0.8 × (H - 40) (unit Hz, maximum 120 Hz); when the temperature gradient When the blasting mode is enabled, 10 100ms ultra-wide pulses are sent within 0.5s; (2) pressure gradient control: base pressure P_base = 1.2×H^0.5 (unit: MPa, H is the mud cake thickness in mm); superimposed oscillation pressure: ΔP = 0.3×sin(2πft) (f = 5Hz, to enhance the penetration effect); maximum pressure limit: P_max = 8.5MPa (safety threshold).

[0112] Step 2: Multi-nozzle coordinated phase control: (1) Phase difference optimization: adjacent nozzle injection start time difference Δt = 10 / (N×f) (N = the number of nozzles in the same ring, f is the current frequency); circumferential phase difference angle θ_phase = 360° / (n_ring×0.8) (n_ring = the number of currently activated ring nozzle groups); (2) Waveform interference control: generate a three-frequency composite waveform: carrier f_c = nozzle base frequency, modulation wave f_m1 = 0.8f_c, f_m2 = 1.2f_c; amplitude envelope A(t) = A_0×[1+0.3cos(2πf_m1t)+0.2cos(2πf_m2t)].

[0113] Step 3: Dynamic backwash adjustment: (1) Blockage warning mechanism: pressure difference monitoring: when ΔP_injector>1.8×P_base lasts for 5 seconds, it is judged as a sign of blockage; flow attenuation rate detection: when dQ / dt<-15%Q_nom (rated flow), the maintenance instruction is triggered; (2) Self-cleaning program: reverse pressure pulse: within 0.2 seconds, the pressure rises from P_base to 2P_base and then drops sharply to -0.5P_base; high-frequency vibration assistance: superimpose 200Hz mechanical vibration, amplitude 0.1mm, duration 3 seconds; f0 is the basic injection frequency; Δt is the phase time difference; θ_phase is the circumferential phase difference angle; A(t) is the dynamic amplitude envelope reverse pressure and self-cleaning pulse pressure.

[0114] In this embodiment, the intelligent terminal includes: a visual interface and an interaction unit;

[0115] The visual interface displays the mud cake thickness distribution heat map and cutterhead load status in real time;

[0116] The interactive unit receives a manually input priority adjustment instruction or an emergency stop instruction.

[0117] It should be noted that the embodiments of the present invention can be implemented in combination based on the system architecture shown in the accompanying drawings. The following describes the implementation process of the technical solution in detail in conjunction with the shield construction scenario:

[0118] 1. Tucang Multimodal Perception Module

[0119] A multi-source sensor network is deployed inside the soil bin to monitor the rheological properties and mineral composition of the soil in real time:

[0120] Soil bin rheometer: Installed in the middle of the inner wall of the soil bin, it uses a rotary shear rheological measurement unit to obtain the soil viscosity-shear rate curve by controlling the shear rate, and calculate the yield stress and thixotropy index;

[0121] Soil bin pressure sensor array: deployed at the screw conveyor inlet area, monitors the axial pressure gradient distribution of the soil and uses the Bingham fluid model to invert the fluidity index;

[0122] Soil bin temperature sensor: integrated in the soil bin partition to detect the friction temperature rise and heat conduction characteristics of the soil;

[0123] Soil bin chemical sensor group: located downstream of the slag discharge port, it analyzes clay mineral composition and cation exchange capacity in real time.

[0124] Furthermore: the rheometer communicates with the control module via an industrial bus, and when it detects that the yield stress exceeds a set threshold, a high-dose anti-adhesive agent injection mode is triggered;

[0125] The chemical sensor group dynamically adjusts the anti-adhesive agent formula according to the clay mineral content, giving priority to high-montmorillonite formations;

[0126] Temperature sensor data is used to modify the rheological model to avoid thermal decomposition of the anti-adhesive.

[0127] 2. Anti-adhesive dynamic proportioning and injection module

[0128] Anti-adhesive agent storage and mixing unit: including anionic surfactant storage tank, nonionic surfactant storage tank and bio-based stabilizer storage tank, which are mixed in real time by high-precision metering pumps;

[0129] Injection execution unit: The anti-sticking agent is delivered to the micro nozzle on the surface of the cutter disc by a high-pressure pump. The nozzle is driven by piezoelectric ceramics and distributed along the circumference of the cutter disc. The spray particle size and flow rate are adjustable.

[0130] Furthermore: when the chemical sensor detects an acidic environment, the mixing unit automatically adds a buffer to maintain the pH range within which the anti-adhesive agent works;

[0131] The atomization angle of the nozzle is controlled synchronously with the speed of the cutter disc to ensure that the anti-sticking agent covers without blind spots.

[0132] 3. Adaptive control module

[0133] Data fusion unit: Receives multi-source data from rheometers, pressure sensors, and chemical sensors, eliminates noise through filtering algorithms, and constructs a soil state matrix;

[0134] Multi-objective decision-making model: Based on a reinforcement learning algorithm, it takes cutterhead torque, propulsion speed, and soil fluidity index as inputs and outputs the optimized values of anti-adhesive agent injection parameters and cutterhead speed;

[0135] Actuator: drives the nozzle to achieve targeted injection, with a positioning error of less than 5mm.

[0136] 4. Human-computer interaction terminal

[0137] Visual interface: Real-time display of thermal map of soil rheological parameters, anti-adhesive agent consumption rate curve and cutterhead load balance index;

[0138] Interaction unit: supports manual input of priority instructions or activation of emergency protocols;

[0139] Data management module: archive construction data and generate suitability reports.

[0140] 5. Alternative Implementations

[0141] The rheometer can be replaced by a capillary rheometer instead of a rotational rheometer, and the shear rate range needs to be adjusted;

[0142] The control algorithm can be replaced by a genetic algorithm to solve the optimal injection strategy through multi-objective optimization.

[0143] 6. Technical effect verification

[0144] In formations with a clay mineral content of 45%, this system achieves the following performance improvements:

[0145] The accuracy of rheological sensing is improved to less than 3%;

[0146] The amount of anti-sticking agent used was reduced by 21%, and the torque fluctuation of the cutter head was reduced by 65%;

[0147] The biodegradation rate of the anti-adhesive agent exceeds 93%, and there is no heavy metal residue.

[0148] Example 2

[0149] The present invention also provides a shield cutterhead slag soil property sensing and mud cake intelligent removal method, which is implemented by applying the aforementioned shield cutterhead slag soil property sensing and mud cake intelligent removal system. The method includes:

[0150] Step S1: synchronously collect mechanical, chemical and thermodynamic signals inside the soil bin through multimodal sensors to obtain multi-source data;

[0151] Step S2: Input multi-source data into the deep learning model to dynamically analyze the three-dimensional morphology and adhesion distribution of the mud cake;

[0152] Step S3: Calculate the optimal injection amount and target position of the viscosity reducer according to the mud cake thickness and the cutter head load state;

[0153] Step S4: Control the micro-injection valve group to perform precise injection according to a dynamic timing sequence, while adjusting the cutter head speed and propulsion pressure;

[0154] Step S5: Closed-loop optimization of control parameters based on real-time feedback data until the mud cake thickness drops to a safe threshold.

[0155] In this embodiment, multi-source data is input into the deep learning model to dynamically analyze the three-dimensional shape and adhesion distribution of the mud cake, including: wherein the deep learning model is: multimodal feature pyramid fusion network

[0156] Step 1: Spatiotemporal alignment of multi-source data: Hardware synchronization: Using the PTPv2 protocol, the time deviation of the millimeter-wave radar (200Hz), infrared thermal imager (30Hz), and chemical sensor (10Hz) is controlled within ±80μs; Spatial registration: A unified coordinate system is established using a laser tracker, with a positioning error of <0.15mm.

[0157] Step 2: Feature Fusion and Enhancement: 1. Point cloud denoising; 2. Density clustering to remove outliers (neighborhood radius 3mm, minimum number of points 15); 3. Bilateral filtering to preserve edge features (σ_space = 2mm, σ_color = 0.1); 4. Multimodal fusion; 5. Temperature-viscosity mapping:

[0158] T(x,y)→η(x,y)=0.5exp(0.03ΔT); Chemical compensation: When pH < 6, the point cloud density of the montmorillonite region is weighted by ×1.3.

[0159] Step 3: 3D reconstruction and dynamic prediction: 1. Morphological generation; 2. Input point cloud data of DGCNN to generate basic mesh (number of vertices 5×10^5); Optimize surface continuity through Laplacian smoothing (5 iterations, λ=0.3); 3. Adhesion force calculation; 4. Micro contact model: A_ij=1.2×exp(-0.4d_ij)×(1+0.05ΔT_ij); d_ij: mud cake-cutter surface distance (unit: mm); ΔT_ij: local temperature difference (unit: ℃); Macro correction: introduce formation type coefficient E (1-8 categories), and the final adhesion force A'_ij=A_ij×(0.8+0.03E).

[0160] Step 4: Real-time visualization output: Thermodynamic overlay display: Areas with temperature gradients > 8°C / mm are marked as red warnings; Mechanical focus prompts: Areas with adhesion > 1.5 MPa are automatically enlarged and displayed (zoom ratio 3:1).

[0161] In this embodiment, the optimal injection amount and target position of the viscosity reducer are calculated based on the mud cake thickness and the cutterhead load state, including:

[0162] Step 1: Dynamic injection volume calculation: 1. Mud cake thickness correction model: When the mud cake thickness h≤40mm, the correction coefficient C h =0.8h 0.6 ; When h>40mm, C h =1.2h 0.4 (Thickness h unit: mm, accuracy ±0.1mm) Technical basis: Based on 500 sets of experimental calibration, thin mud cake (<40mm) is mainly based on penetration, and the correction index is 0.6; thick mud cake (>40mm) is mainly based on shear peeling, and the correction index is adjusted to 0.4. 2. Calculation of final injection volume: Qfinal =Q base ×K×(1-0.03pH)Parameter Description:Q final =1.5C h γT: Basic injection volume (L / min / m 2 ), where γT is a thermodynamic factor (range 0.8-1.3), which is dynamically calculated by the temperature change rate; K is a formation correction factor (0.7 for Class I, 1.0 for Class II, and 1.4 for Class III), which is graded according to the montmorillonite content; pH is the pH of the mud cake (4-9). Acidic conditions (pH < 5) trigger compensation, and the injection volume is increased by an additional 15%.

[0163] Step 2: Target Position Optimization Algorithm: 1. Priority Heat Map Generation: Parameter Description: h i is the thickness of the mud cake in the unit (mm); is the absolute value of temperature gradient (℃ / mm); =Clay mineral content percentage (0-100%). 2. Quantum Optimization Algorithm: Core Logic: Dynamically plans paths using pheromone weight (1.2) and heuristic weight (0.8), prioritizing coverage of high-priority cells. Emergency Response: When adhesion force is greater than 2.5 MPa, the spray volume is tripled and the 0.3 mm focusing nozzle is switched (normally 0.8 mm).

[0164] In this embodiment, controlling the micro-injection valve group to perform precise injection according to a dynamic time sequence while adjusting the cutter head rotation speed and propulsion pressure includes:

[0165] Step 1: Dynamic coupling adjustment model: 1. Speed reference setting: Base speed n0 = 1.5 × (1-0.008E) (unit: rpm, E is the formation hardness level 1-8), parameter description: When E = 1 (soft soil), the speed is 1.49 rpm, and when E = 8 (hard rock), it is reduced to 1.15 rpm to match the formation shear strength; the coefficient of 0.008 is calibrated based on 120 sets of formation-torque experiments. 2. Propulsion pressure compensation mechanism: Dynamic pressure P = P base +0.12×ΔT+0.07×ln(K)(Unit: MPa), Parameter Description: P base : Base pressure (18MPa for sand, 22MPa for clay, 25MPa for rock); ΔT: Cutterhead temperature change rate (°C / min, range 0-30); K: Rheological index (0.5-2.2, measured in real time by a modifier rheometer). Step 2: Multi-parameter collaborative optimization: 1. Vibration feedback suppression: When the vibration acceleration a>2.0m / s 2When: the speed is reduced by Δn=0.05×(a-2.0) (unit: rpm, minimum adjustment step 0.01rpm); the propulsion pressure is reduced synchronously by ΔP=0.8×Δn (unit: MPa). 2. Torque-thrust balance control: set the torque threshold Tmax=7500×(1+0.03C clay )(Unit: kN-m,C clay 3. Adjustment rules: When T>0.95T max : Propulsion pressure is reduced by 5%, speed is increased by 0.03rpm; when T<0.7T max : The propulsion pressure is increased by 3%, and the rotation speed is reduced by 0.02rpm. Step 3: Formation adaptive learning: 1. Dynamic learning rate adjustment: Initial learning rate α=0.2, updated every 10 minutes: α new =α old ×(1+0.1×η), η is the excavation efficiency of the previous cycle, ranging from 0 to 1; 2. Stratum memory library: stores 500 sets of historical stratigraphic parameters (hardness E, clay content C clay , temperature T); when matching similar formations, the optimal speed-pressure combination is directly called, and the response time is <0.3 seconds.

[0166] In this embodiment, closed-loop optimization of control parameters based on real-time feedback data until the mud cake thickness drops to a safe threshold includes:

[0167] Step 1: Real-time acquisition of multimodal data: Sensor configuration: Mud cake thickness array: 2048 piezoelectric film sensors (range 0-100 mm, accuracy ±0.05 mm, sampling rate 500 Hz); Mechanical feedback: 36-channel dynamic torque meter (range 0-10 kN·m, resolution 0.1% FS); Temperature monitoring: Infrared thermal imager (resolution 0.1°C, refresh rate 30 Hz); Data preprocessing: Outlier filtering: Sliding window Z-score detection (window length 2 s, threshold ±3σ); Spatiotemporal alignment: Laser positioning system (accuracy ±0.2 mm) synchronized with PTPv2 clock (deviation <50 μs).

[0168] Step 2: Dynamic Weighted Optimization Engine: Control Parameter Set: Core Parameters: Injection Pressure (12-35 MPa), Nozzle Elevation (30°-75°), Cutter Head Speed (0.8-1.5 rpm); Auxiliary Parameters: Modifier Flow (20-150 L / min), Vibration Suppression Gain (0.5-2.0); Optimization Rule Base: Thickness-Dominant Mode (h>25 mm): Prioritizes adjustment of injection pressure (step size ±1.2 MPa / time) and elevation (step size ±3° / time); Temperature-Dominant Mode (ΔT>10°C / min): Activate the cooling protocol, reduce the speed (0.03rpm / ℃) and increase the flow rate (5L / min / ℃); Adhesion-dominated mode (A>1.8MPa): switch to focused injection, intervene with a 0.3mm diameter nozzle, and increase the pressure to 28MPa; Reinforcement learning mechanism: State space: 6-dimensional vector (h, T, A, v, Q, n); Reward function: R=0.5Δh+0.3exp(-|T-45|)+0.2(Q / Q_max); Decision frequency: generate 3 sets of candidate parameters per second, and select the solution with the highest reward value.

[0169] Step 3: Closed-loop verification and iteration: Effect evaluation: Thickness attenuation rate: Δh / Δt ≥ -0.5mm / s (safety threshold h_safe = 10mm); Thermal stability: ΔT ≤ 8℃ / min (3 consecutive cycles of meeting the standard are considered stable); Parameter fine-tuning: If Δh / Δt>-0.3mm / s: the pressure increase step is increased to 1.5MPa / time, and the elevation angle adjustment range is expanded to ±5°; If local h>40mm: trigger regional enhanced injection, and the flow rate in the area is increased to 300% of the baseline value; Historical memory library: stores the optimal parameter combination (including labels such as formation type, ambient temperature, and tunneling speed); the matching accuracy of similar working conditions reaches 93%, and the parameter reuse response time is <0.2 seconds.

[0170] In this embodiment, multi-sensor data is collected and pre-processed synchronously:

[0171] Establish a multi-channel data acquisition module to synchronously receive the following signals:

[0172] Cross-rotating rheological blade data: The blade rotation resistance torque (T) and speed (n) are recorded through the embedded torque sensor, and the torque-speed dynamic curve is constructed to calculate the rheological index K = ΔT / Δn;

[0173] Temperature sensor array: 8-12 infrared temperature probes are arranged on the inner wall of the soil bin to record the temperature gradient distribution T(x,y,t);

[0174] Vibration sensor group: A three-axis accelerometer is installed on the back of the cutter head to collect vibration signals a_x(t), a_y(t), and a_z(t);

[0175] Ultrasonic sensor array: emits 20-200kHz ultrasonic pulses and receives reflected wave signals S_r(t) and transmitted wave signals S_t(t);

[0176] Perform data preprocessing:

[0177] Perform wavelet noise reduction on vibration signals to eliminate mechanical vibration interference;

[0178] Use matched filtering to process ultrasonic signals to enhance the effective echo signal-to-noise ratio;

[0179] Establish a spatiotemporal registration model to align the spatiotemporal coordinate systems of multiple sensors.

[0180] In this embodiment, rheological properties and temperature field are jointly analyzed:

[0181] Step 1: Synchronous Multi-Source Data Acquisition 1. Temperature Field Data: Utilizes a 64-channel infrared thermal imaging array (accuracy ±0.3°C, resolution 0.1°C / mm, sampling rate 100Hz); Spatial Coverage: 3m diameter area on the cutterhead surface, divided into a 36×57 grid; 2. Rheological Property Data: Installs a pressure-cooking composite sensor (range: shear stress 0-50kPa, normal stress 0-200kPa, accuracy ±0.8%); Real-time Calculation of the Rheological Index K: (τ is shear stress, is the shear rate, n is the power law exponent) Step 2: Dynamic coupling modeling: 1. Temperature-viscosity correction model: define the temperature-sensitive factor β T =0.05×(T-T0) (T0=45℃ is the reference temperature); modified rheological index: K′=K×(1+β T )(limit K′∈[0.5K,1.5K]). 2. Rheological-thermal coupling equation: heat generation rate (Unit: W / m 3 , with a coefficient of 0.12 calibrated by energy conservation experiments). Temperature prediction: T t+1 =T t +0.8qΔt-0.15(T t -T env )(T env =Ambient temperature, Δt = 0.1s). Step 3: Combined control strategy: 1. Double closed-loop feedback mechanism: Inner loop (rheological control): Adjust the modifier flow Q = 20 + 1.5K' (unit L / min, reference value 20L / min) every 0.5 seconds. Outer loop (temperature control): When T > 50°C, start the cooling spray (water volume W = 0.3 (T-50) 2 Unit L / min) 2. Dynamic compensation protocol: When β T >0.2 (i.e. T>49°C): rheological control weight is reduced by 30%; temperature compensation coefficient is increased to β T′=1.2β T When K′>1.8 (high rheological resistance): the pressure gradient increases from 0.5 MPa / s to 1.2 MPa / s; the high-frequency vibration mode is triggered (amplitude 0.3 mm, frequency 80 Hz).

[0182] Construct rheological state judgment matrix:

[0183] IF K>K_threshold ANDΔT / Δt>0.5℃ / min THEN

[0184] Marked as "high cake risk";

[0185] ELSE IF K>K_threshold ANDΔT / Δt<0.2℃ / min THEN

[0186] Labeled as "reunion risk";

[0187] ELSE

[0188] Labeled as "normal flow plastic state";

[0189] (K_threshold is dynamically adjusted according to the soil type).

[0190] The finite element heat conduction inversion algorithm is used to reconstruct the three-dimensional temperature cloud map in the soil bin according to the temperature field T(x,y,t) to locate the abnormal temperature rise area.

[0191] The specific reconstruction process includes:

[0192] Step 1: Multi-source data fusion modeling: Sensor configuration: Deploy 72 high-precision fiber Bragg grating temperature sensors (range 0-150°C, accuracy ±0.1°C, spatial resolution 5mm), arranged in a hexagonal array on the inner wall of the soil bin; synchronously collect speed sensor data (cutter head speed 0.8-1.5rpm, accuracy ±0.01rpm) and pressure data (range 0-35MPa, accuracy ±0.2%); Dynamic mesh generation: The basic mesh size is 3mm, and it is automatically encrypted to 0.5mm in areas with temperature gradients >5°C / mm; an unstructured tetrahedral mesh is used, and the number of nodes is dynamically adjusted (baseline 2 million units, maximum encryption to 12 million units).

[0193] Step 2: Heat conduction inversion calculation: Boundary condition setting: Cutter head friction heat source: Q_friction = 0.15×n×T (n is the speed rpm, T is the torque kN·m, and the coefficient 0.15 is calibrated by energy consumption experiments); mud-water convection heat transfer coefficient h = 120×(1+0.03v)^{0.8} (v is the mud-water flow rate m / s, ranging from 0.2-1.5); Inversion algorithm core: time step Δt = 0.05 seconds, implicit Euler method is used to solve the transient equation; speed-heat source coupling factor α = 0.8×(1-exp(-0.1n)) is introduced to enhance the time-varying effect of speed on temperature rise; convergence conditions are set: the temperature residual of adjacent iteration steps is <0.03℃ or the maximum number of iterations is 50 times.

[0194] Step 3: Abnormal temperature rise location: Feature extraction: Define the anomaly index A_i = (T_i-μ_z) / (σ_z+0.1) (μ_z is the mean of the 3×3×3 neighborhood, σ_z is the standard deviation); when A_i>3.5, it is marked as an abnormal area and triggers a level 3 alarm; 3D visualization: Dynamic adjustment of the temperature cloud map color scale: The base color scale is 25-85℃, and when an anomaly is detected, it is locally expanded to 20-100℃; Isothermal surface rendering: Generate a double-layer transparent isothermal surface of 55℃ (warning threshold) and 65℃ (danger threshold).

[0195] In this embodiment, vibration spectrum feature extraction and mud cake positioning:

[0196] Step 1: Multi-dimensional analysis of vibration signals: 1. Data acquisition: Install 12 three-axis accelerometers (range ±25g, frequency response 0.5-5000Hz, sampling rate 10kHz), arranged in a hexagonal array on the back of the cutter head; synchronously collect cutter head speed (0.8-1.5rpm, accuracy ±0.01rpm) and torque data (0-8500kN-m, accuracy ±0.3%); 2. Dynamic frequency band division: basic frequency band: 0-200Hz (structural vibration), 200-800Hz (tool friction), 800-2000Hz (mud cake stripping characteristics); adaptive adjustment: when the speed is >1.2rpm, the characteristic frequency band moves up 50Hz / every 0.1rpm increment. Step 2: Spectrum-mud cake state mapping: 1. Feature parameter extraction: main frequency energy ratio (R>1.8 when mud cake thickness>30mm); harmonic distortion rate (A n=The nth harmonic amplitude, when D>25%, it is determined to be an adhesion anomaly); 2. Spatial positioning algorithm: Establish a 36° phase difference positioning model, calculate the vibration source coordinates through the adjacent sensor signal delay (accuracy ±2μs); define the mud cake index C = 0.6R + 0.4D, and when C>2.1, it is marked as a high-risk area. Step 3: Real-time control feedback: 1. Dynamic adjustment of injection parameters: When R>1.8: the injection pressure in the corresponding area is increased to P=1.5P base (P base The reference pressure is 18-25MPa); when D>25%, high-frequency pulse mode is triggered (200Hz short pulse, pulse width 50ms, interval 20ms); 2. Cutter movement optimization: if more than 3 areas with C>2.5 are detected, the speed is reduced by 0.05rpm, and reciprocating oscillation is activated (amplitude 3°, frequency 0.2Hz).

[0197] Perform time-frequency analysis on vibration signals:

[0198] Obtain the time-frequency spectrum STFT(t,f) through short-time Fourier transform;

[0199] Extract feature parameters:

[0200] Main band energy ratio E_ratio = ∫{50-150 Hz}PSD(f)df / ∫{0-500 Hz}PSD(f)df;

[0201] Impact event frequency N_peak = count (da / dt>threshold);

[0202] Establish cutterhead angle-vibration feature mapping model:

[0203] θ = arctan(y / x);

[0204] F(θ)=∑[E_ratio(θ)×N_peak(θ)] / ∑θ;

[0205] When F(θ)>F_threshold, it is determined that mud cake accumulation exists in the θ±15° region.

[0206] In this embodiment, ultrasonic signal inversion and microstructure analysis:

[0207] Step 1: Multi-frequency composite ultrasonic detection: 1. Sensor configuration: dual-frequency ultrasonic array (1MHz low frequency penetrates the layer thickness, 5MHz high frequency captures details), 64 channels are arranged in a ring (spacing 12mm), and the transmission power is dynamically adjusted (10-100W, step 5W); synchronous reception of echo signals (sampling rate 100MHz, resolution 0.1ns), the trigger threshold is set to -40dB (to avoid environmental noise interference); 2. Signal feature extraction: frequency band division: 0.8-1.2MHz: crack feature frequency band (resolution ±0.05mm), 4.8-5.2MHz: clay mineral response frequency band (to identify montmorillonite / kaolinite differences); dynamic parameters: main frequency reflectivity R = 0.7×(E ref / E inc )(E ref is the reflected energy, E inc is the incident energy), attenuation slope S = ΔA / (0.1f) (ΔA is the attenuation at 10 cm depth, unit: dB / cm-MHz). Step 2: Microstructure inversion algorithm: 1. Fracture network modeling: define the fracture density D c =N / (0.1R 2 )(N is the number of abnormal reflection points in the 1MHz frequency band, R is the detection radius in cm); crack orientation angle θ = arctan(∑A y / ∑A x )(A x / A y ) is the X / Y polarization echo amplitude ratio; 2. Mineral component analysis: clay type identification; montmorillonite characteristics: 5MHz frequency band attenuation slope S>8.5dB / cm·MHz; kaolinite characteristics: S∈[6.0,8.5]dB / cm·MHz; quartz sand detection: 1MHz frequency band main frequency reflectivity R>0.65; 3. Dynamic tracking algorithm: time window length 0.5μs (corresponding to the microstructure change period), dynamically adjusted to 0.2μs (grain boundary reflection capture), establish a three-dimensional acoustic impedance matrix (256×256×128 voxels, resolution 0.05mm 3 ). Step 3: Real-time control feedback: 1. Injection strategy optimization: When D c >15: Activate high-pressure rotary jetting (pressure 28 MPa, nozzle speed 120 rpm) in the corresponding area to detect montmorillonite (S>8.5): trigger the alkaline modifier (pH=9.0, flow rate increased by 20%); 2. Cutter vibration suppression: Quartz sand enrichment area (R>0.65): reduce the speed by 0.03 rpm and limit the amplitude to within 0.15 mm.

[0208] Implement multi-scale waveform analysis:

[0209] Macrostructure tomography: Calculate the equivalent density ρ = K1·Δt by the travel time difference Δt of the transmitted wave 2 ;

[0210] Mesoscopic particle analysis: Extract the reflected wave envelope area A and attenuation coefficient α to establish a particle grading model:

[0211] d50 = K2 (A / α) ^ 0.5, (d50 is the average particle size, K2 is the calibration coefficient);

[0212] Microscopic contact testing: Analyzing ultrasonic phase distortion Inverse inter-particle contact stiffness;

[0213] Step 1: Multi-frequency ultrasonic excitation and phase acquisition: 1. Sensor configuration: 32-element circular phased array (center frequency 2MHz, bandwidth 0.5-4MHz, transmitting voltage ±120V), element spacing λ / 2 (λ=0.75mm), receiving end sampling rate 200MHz, phase resolution 0.01°, dynamic range 120dB; 2. Signal preprocessing: Dispersion compensation: Delay calibration is performed on the 1.5-2.5MHz frequency band (compensation amount Δt=0.12×F 2 μs, f is the frequency in MHz), noise suppression: apply adaptive notch filter (stopband width ±5kHz, attenuation depth -45dB). Step 2: Phase distortion feature extraction: 1. Dynamic reference establishment: calibrate the reference phase under no-load condition Accuracy ±0.05°, defined distortion (ΔT is the temperature drift, unit is ℃); 2. Nonlinear characteristic analysis: Extract the second harmonic phase distortion rate δ = ( Fundamental distortion, Second harmonic distortion), when δ>0.15, it is determined that the particle contact nonlinearity is enhanced. Step 3: Contact stiffness inversion algorithm: 1. Stiffness calculation model: define the equivalent stiffness ( is the saturation distortion), introduce the frequency compensation factor α=1+0.03(f-2) (f unit MHz, compensates for the dispersion effect); 2. Spatial distribution reconstruction: map the 64×64 unit grid stiffness value into a three-dimensional field (resolution 0.1mm 3 ), set the stiffness threshold: K<1.8×10 6 N / m 3 The area marked as weak contact is marked. Step 4: Real-time control feedback: 1. Vibration optimization: Trigger high-frequency micro-vibration (amplitude 50μm, frequency 200Hz, duration 3s) in weak contact areas (K < 1.8×10°). When the anisotropy index is > 0.25, adjust the vibration direction angle (in 15° steps). 2. Injection parameter adjustment: When the stiffness gradient is > 5×10°N / mm, increase the injection pressure in the corresponding area to 1.8P (P0 = 18MPa base pressure).

[0214] Calculation of mud cake thickness:

[0215] h=Σ[c·τ_i](i=1~n);

[0216] Where c is the sound velocity correction coefficient, and τ_i is the time delay of each reflection interface.

[0217] In this embodiment, multimodal data fusion and decision output:

[0218] Establishing the DS evidence theory fusion framework:

[0219] Define the identification framework Θ = {normal flow plastic state, mud cake, agglomeration and blockage};

[0220] Assign basic probability assignment function:

[0221] m1(Θ) comes from the rheological-temperature joint analysis results,

[0222] m2(Θ) comes from the vibration spectrum diagnosis results,

[0223] m3(Θ) comes from the ultrasonic inversion parameters;

[0224] Execute evidence synthesis rules:

[0225]

[0226] Formula expression:

[0227]

[0228]

[0229] Output the final decision:

[0230] When m_final (mud cake) > 0.7, the cleaning instruction is triggered;

[0231] The mud cake position θ, thickness h and particle grading curve are displayed simultaneously.

[0232] Among them, 1. Accurate calibration of mud cake position 0: 1. Position definition and measurement method: The mud cake position θ is directly measured by a high-precision cutter head angle encoder (24-bit absolute encoder, resolution ±0.0015°) and is defined as the central angle (0-360°) of the starting boundary of the mud cake in the cutter head rotation plane.

[0233] Position correction formula:

[0234] θ=θ encoder +Δθ calibrate

[0235] Where θ encoder : Encoder original angle value, Δθ calibrate: The boundary compensation of the laser displacement sensor (range 0-50mm, accuracy ±0.01mm) is calculated by the following formula:

[0236] Δθ calibrate =0.12×(d edge -d base )(d edge is the measured distance from the edge of the mud cake, d base = ( θ is the cutterhead reference radius). 2. Multi-sensor fusion verification: When the vibration sensor detects a sudden change in energy in a specific frequency band (800-1200Hz energy > 1.5 times the baseline), a secondary calibration of the θ value is triggered to ensure that the positioning error is < 0.5°. 2. Dynamic modeling of particle gradation curve: 1. Grading function expression: Using the improved fractal-S curve model:

[0237]

[0238] Where, d is the particle size (mm), the detection range is 0.001-20mm (real-time measurement by laser particle size analyzer); d 50 is the median particle size (mm), which is dynamically calibrated by the 50% point of the cumulative distribution; D is the fractal dimension (1.2-2.8), which reflects the uniformity of particle distribution. Calculate (R 80 、R 20 =The particle size ratio corresponding to the cumulative 80% and 20% values). 2. Real-time update mechanism: The gradation curve is updated every 5 seconds. When D>2.5, it is judged as poor gradation, triggering the injection of fine particle modifier (flow rate increased by 15%); 3. Characteristic particle size threshold: coarse particles dominate (d 50 >0.5mm): Activate high-pressure water jet mode (pressure>25MPa); fine particles dominate (d 50 <0.1mm): Enable ultrasonic dispersion device (frequency 28kHz, power 200W).

[0239] In this embodiment, the shield machine mud cake intelligent dynamic control algorithm: five steps form a closed-loop link of "modeling → mapping → execution → collaboration → optimization", which progresses step by step: Step 1 is the data basis, and the mud cake state model is constructed in real time; Step 2 maps the model into physical execution coordinates; Step 3 controls the precise action of multiple reagents; Step 4 coordinates the mechanical system to complete the action; Step 5 feedback evaluates and optimizes the model parameters of Step 1.

[0240] Step 1: Dynamic modeling of control parameters

[0241] Input status parameters:

[0242] Mud cake thickness: h (unit: mm)

[0243] Mud cake azimuth: θ (unit: °)

[0244] Soil particle size distribution d 50 (Unit: mm)

[0245] Temperature change rate: ΔT / Δt (unit: ℃ / min)

[0246] Rheological index: K (unit: N·m / rpm)

[0247] Establish a dynamic ratio model:

[0248] Foaming agent basic flow rate (L / min):

[0249] Q_foam=K1*h^1.2*exp(-0.5*d50);

[0250] Viscosity reducer ratio coefficient:

[0251] α_detergent=1 / (1+exp(-0.3*(ΔT / Δt-2)));

[0252] Flushing water pressure (MPa):

[0253] P_water=min(8,0.6*h+0.4*(K / K_threshold)^3);

[0254] Total control intensity factor:

[0255] Γ=[0.7,0.2,0.1]*[h / 50; K / K_threshold; ΔT / Δt];

[0256] (Coefficient K1 is dynamically matched through the geological parameter library);

[0257] Step 2: Position-nozzle precise mapping

[0258] Construct the polar coordinate nozzle matrix:

[0259] The cutter disc circumference is divided into 36 10° sectors (numbered N1-N36), and each sector is equipped with three sets of directional nozzles (foam / viscosity reducer / water are independently controlled); activation rules:

[0260]

[0261] Step 3: Multi-reagent coordinated injection control:

[0262] Timing control logic:

[0263] Stage 1 (0-20s): Pre-injection of foaming agent (softening mud cake structure)

[0264] set_pulse_mode(FOAM,5Hz,80%duty_cycle);

[0265] Stage 2 (20-45s): Alternating injection of viscosity reducer and water (chemical-mechanical combined action)

[0266]

[0267] Stage 3 (45-60s): high-pressure water directional flushing (peak pressure reaches 12MPa) ramp_pressure (8MPa→12MPa, 15s);

[0268] 2. Adaptive termination conditions:

[0269]

[0270] Step 4: 3D path planning and mechanical collaboration:

[0271] Cutter head speed coupling control:

[0272]

[0273] Screw propulsion compensation algorithm:

[0274] def thrust_compensation(h):

[0275] F_adjust=0.12*h*(1+0.05*sin(2πt / 60))#Superimposed periodic fluctuation component

[0276] set_thrust(F_base+F_adjust)

[0277] Step 5: Closed-loop performance evaluation and parameter iteration:

[0278] Establish the performance evaluation function: η = 0.6*(Δh / h_initial)+0.3*exp(-Δt / 10)+0.1*(1-E_ratio_new / E_ratio_old)

[0279] Parameter self-optimization mechanism:

[0280]

[0281] In this embodiment, the deep learning model in step S2 is a rheological-chemical coupling model that uses a convolutional neural network to extract the spatial characteristics of the mud cake and combines it with a long-short-term memory network to predict the thickness evolution trend. Specifically, the rheological-chemical coupling model in step S2 uses a Kalman filter to fuse data and a deep neural network to predict the adhesion evolution trend.

[0282] In this embodiment, the injection timing in step S4 is synchronized with the angular velocity of the cutter disc to ensure that the viscosity reducer covers all blind spots. Specifically, the injection particle size (50-200 μm) and atomization angle (30°-90°) in step S4 are dynamically adjusted based on the cutter disc rotation speed (0.5-2 rpm).

[0283] In this embodiment, step S5 uses a reinforcement learning algorithm to iteratively optimize the control strategy to reduce construction energy consumption and improver waste. Specifically, step S5 dynamically calculates the objective function weight coefficient using an entropy weight method to optimize comprehensive energy consumption.

[0284] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A shield cutterhead soil property sensing and mud cake intelligent removal system, characterized by: The system includes: a sensing module, a modifier injection module, a control module and an intelligent terminal; The sensing module is used to collect mechanical, chemical and thermodynamic signals in the soil bin of the shield machine in real time; The modifier injection module is used to intelligently control the modifier ratio to inhibit mud cake adhesion; The control module is used to dynamically optimize the improver ratio parameters and the cutterhead excavation strategy based on the perception data; The intelligent terminal is used to display the mud cake distribution status and receive manual intervention instructions.

2. The system according to claim 1, wherein: The sensing module includes: a soil bin rheometer, a soil bin pressure sensor, a soil bin temperature sensor and a soil bin chemical sensor group; The soil bin rheometer is installed in the middle of the inner wall of the soil bin and is used to measure the viscosity and yield stress of the slag through rotational shear and output rheological characteristic parameters in real time; The soil bin pressure sensor array is arranged at the entrance area of the soil bin screw conveyor to monitor the axial pressure distribution gradient of the slag; The soil bin temperature sensor is integrated into the soil bin partition to detect the friction temperature rise and heat conduction characteristics of the slag soil; The soil bin chemical sensor group is located at the soil bin slag discharge port and is used for online analysis of the clay mineral content and pH value in the slag.

3. The system according to claim 1, wherein: The modifier injection module includes: a modifier formulation unit, a storage and delivery unit, and an injection execution unit; The improver formula unit is composed of a viscosity reducer and a foaming agent, and is used to adjust the distribution ratio of each improver component in real time according to the sensing data to reduce adhesion; The storage and delivery unit is used to deliver the viscosity reducing agent to the cutter head surface via a high-pressure pump; The injection execution unit includes a circumferentially distributed micro injection valve group for dynamically adjusting the injection angle and coverage range.

4. The system according to claim 1, wherein: The control module includes: a data fusion unit, a decision model unit and an instruction generation unit; The data fusion unit is used to synchronously process multi-source sensing signals and construct a three-dimensional digital twin of the mud cake; The decision model unit is used to generate optimization instructions based on the correlation between the cutter head torque, propulsion speed and mud cake thickness; The instruction generation unit is used to control the injection valve group to perform targeted injection and cutter disc parameter adjustment.

5. The system according to claim 1, wherein: The intelligent terminal includes: a visual interface and an interaction unit; The visualization interface is used to display the mud cake thickness distribution thermal map and the cutterhead load status in real time; The interaction unit is used to receive a manually input priority adjustment instruction or an emergency stop instruction.

6. A shield cutterhead slag property sensing and mud cake intelligent removal method, which is implemented by using the shield cutterhead slag property sensing and mud cake intelligent removal system according to any one of claims 1 to 5, characterized in that: The method comprises: Step S1: synchronously collect mechanical, chemical and thermodynamic signals inside the soil bin through multimodal sensors to obtain multi-source data; Step S2: Input multi-source data into the deep learning model to dynamically analyze the three-dimensional morphology and adhesion distribution of the mud cake; Step S3: Calculate the optimal injection amount and target position of the viscosity reducer according to the mud cake thickness and the cutter head load state; Step S4: Control the micro-injection valve group to perform injection according to a dynamic time sequence, and adjust the cutter head speed and propulsion pressure at the same time; Step S5: Closed-loop optimization of control parameters based on real-time feedback data until the mud cake thickness drops to a safe threshold.

7. The method according to claim 6, characterized in that The step S4 comprises: Establish a dynamic matching model based on multi-source data; According to the dynamic proportioning model, a polar coordinate nozzle matrix is constructed; Based on the polar coordinate nozzle matrix, multi-reagent coordinated injection control is performed; Based on the results of multi-reagent collaborative injection control, three-dimensional path planning and mechanical coordination are carried out; Based on the structure of 3D path planning and mechanical collaboration, closed-loop performance evaluation and parameter iteration; Among them, the multi-source data include: mud cake thickness, mud cake azimuth, soil particle size distribution, temperature change rate, and rheological index; The dynamic proportioning model includes: foaming agent basic flow rate: Q_foam = K1*h^1.2*exp(-0.5*d50); viscosity reducer ratio coefficient: α_detergent = 1 / (1+exp(-0.3*(ΔT / Δt-2))); flushing water pressure: P_water = min(8,0.6*h+0.4*(K / K_threshold)^3); Total control intensity factor: Γ = [0.7, 0.2, 0.1] * [h / 50; K / K_threshold; ΔT / Δt]; K1 is the coefficient dynamically matched through the geological parameter library; The polar coordinate nozzle matrix includes: The circumference of the cutter head is divided into 36 10° sectors; Three groups of directional nozzles are deployed in each sector.

Images