A test device and method for testing the disaster threshold of multi-field coupling in water-rich tunnels

Abstract

This invention discloses a multi-field coupled disaster threshold testing device and method for water-rich tunnels. It integrates a triaxial loading module, a high-resolution CT diagnostic module, and a seepage erosion visualization and intelligent early warning module, enabling simultaneous analysis of microstructural evolution and macroscopic disaster behavior. The device employs a detachable transparent pressure chamber, compatible with standard tests and high-precision CT scans. The seepage erosion module quantifies particle initiation conditions through high-speed imaging and laser velocimetry. The testing method accurately determines the disaster thresholds for water inrush, mud inrush, and large deformation through multi-field parameter dynamic control and machine learning models. This invention provides key technical support for disaster prevention and control in water-rich tunnel engineering.

Description

Technical Field

[0001] This invention belongs to the field of geotechnical engineering testing technology, specifically relating to a multi-field coupled disaster threshold test device and test method for water-rich tunnels. Background Technology

[0002] Traditional triaxial testing equipment cannot synchronize data with microstructure evolution, making it difficult to capture dynamic changes in seepage paths. The equipment can only measure macroscopic mechanical parameters and lacks quantitative analysis of microscopic parameters such as porosity and coordination number.

[0003] Existing seepage erosion test devices mostly use constant confining pressure, which cannot simulate complex confining pressure conditions and is difficult to reflect the dynamic changes of confining pressure in actual engineering. The critical conditions for particle initiation are also difficult to quantify.

[0004] For disaster early warning systems, there is a heavy reliance on empirical formulas, a lack of intelligent criteria for multi-source data fusion, and existing early warning models have a high false alarm rate (>20%) and cannot output disaster probability in real time. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a multi-field coupling disaster threshold test device and method for water-rich tunnels. By integrating multi-field coupling loading, high-resolution CT diagnostic modules, and seepage erosion visualization and intelligent early warning technologies, it accurately determines the critical thresholds of disasters such as water inrush, mud inrush, and large deformation, providing a scientific basis for the safe construction and operation of tunnel engineering.

[0006] Therefore, the present invention adopts the following technical solution:

[0007] A multi-field coupled disaster threshold test device for water-rich tunnels includes a triaxial loading module, a high-resolution CT diagnostic module, a seepage erosion visualization module, and a data fusion and early warning module;

[0008] The triaxial loading module includes an axial servo loading unit, a dynamic osmotic pressure adjustment system consisting of a back pressure pump, a safety relief valve, and a three-way valve; a confining pressure control chamber; a sample holder is installed inside the confining pressure control chamber for placing the sample; the axial servo loading unit applies vertical pressure to the sample inside the confining pressure control chamber; the dynamic osmotic pressure adjustment system achieves hydraulic gradient control through a dual closed-loop PID controller, the back pressure pump pressurizes the water flow, and the three-way valve divides the water flow into upper and lower paths within the confining pressure control chamber, allowing seepage inside the sample; a drainage channel is installed in the middle of the confining pressure control chamber for drainage metering; the control objective is to establish a pressure difference ΔP between the two ends of the sample to achieve a hydraulic gradient;

[0009] The working method of the triaxial loading module is as follows: the axial servo loading unit applies a load to the sample inside the confining pressure control chamber, and the load sensor on the axial servo loading unit is used to monitor the pressure magnitude; the data terminal of the load sensor and the back pressure pump terminal are connected to the control box of the module; the control box transmits all data to the overall information processing and fusion terminal; when the triaxial loading reaches the dynamic threshold, liquid nitrogen quick-freezing fixation is performed.

[0010] High-resolution CT diagnostic module: includes a detachable transparent pressure chamber, a 3D rotating stage, and a microfocus CT scanner gantry; the microfocus CT scanner gantry is cubic in shape, with the transparent pressure chamber placed at the center inside the gantry, the 3D rotating stage placed at the bottom of the transparent pressure chamber, and a sample chamber placed on top of the 3D rotating stage. The sample chamber contains a sample that has been flash-frozen in liquid nitrogen. The 3D rotating stage is used to adjust the height of the sample chamber and rotate it; the left side of the transparent pressure chamber is for connecting the X-ray source, and the right side is the detector panel;

[0011] The high-resolution CT diagnostic module operates as follows: First, the sample is scanned and positioned. Then, the three-dimensional rotary stage is driven to lift the sample along the Z-axis to the CT scan focal plane. After positioning, the rotary stage rotates at a constant step angle, triggering the CT projection acquisition system to capture data at each step. For data synchronization, the system utilizes the CT scan trigger signal in conjunction with the triaxial loading module: when the real-time monitoring data reaches a preset dynamic threshold, the triaxial loading module immediately freezes the current load and aligns it via timestamps to achieve synchronous correlation between the sample's mechanical response data and microstructure evolution data. Regarding data synchronization, this invention employs a cross-module synchronization method of "threshold triggering—state freezing—off-site scanning—event registration." During triaxial loading, axial load, confining pressure, pore water pressure, seepage flow rate, permeability coefficient, and acoustic emission signals are continuously acquired by the real-time data acquisition system and timestamped by a unified clock module. When the real-time monitoring data reaches the preset dynamic threshold, the synchronization trigger simultaneously sends trigger signals to the triaxial loading control box, liquid nitrogen quick-freezing unit, data fusion terminal, and CT diagnostic module, generating a unique event number and recording the freeze trigger time t0. The triaxial loading module locks the current loading state at time t0, saves the mechanical-percolation-acoustic emission data within a preset time window before and after t0, and simultaneously initiates liquid nitrogen rapid freezing to fix the sample structure. After the sample is rapidly frozen, it is transferred to the high-resolution CT diagnostic module for scanning via a cryogenic transfer fixture with a unique sample number. The CT scan data is marked with the same event number, sample number, and scan start and end timestamps, and registered with the triaxial loading data corresponding to time t0. Since the sample has already fixed its pore structure and damage state through liquid nitrogen rapid freezing at time t0, the microstructural parameters acquired by CT represent the frozen state when the triaxial loading reaches the threshold, thus achieving synchronous correlation between the out-of-place CT scan data and the triaxial loading process data.

[0012] The seepage erosion visualization module is used for dynamic capture of particle transport and seepage path, including a transparent seepage tank, a high-speed camera system, a laser emitter, a confining pressure loading device, inlet and outlet water channels, and a sample chamber. After the high-resolution CT diagnostic module finishes working, the sample is transferred to the sample chamber. The transparent seepage tank has a built-in porous media support with flow guide holes. The confining pressure loading device uses an annular airbag to apply dynamic confining pressure of 50~500kPa. The high-speed camera system, in conjunction with the laser emitter, realizes the quantitative acquisition of particle transport trajectory and velocity.

[0013] The working method of the seepage erosion visualization module: The confining pressure loading device applies a dynamic confining pressure of 50~500kPa through the annular airbag. Water flows through the inlet and outlet channels and forms a stable seepage field with the porous media support. The high-speed camera system, together with the laser emitter, captures and quantifies the particle migration trajectory and migration speed in real time.

[0014] The data fusion and early warning module is used for multi-source data collaborative analysis, cross-module data registration, and intelligent disaster identification. It includes a multi-source sensor array, a real-time data acquisition system, a unified clock module, a synchronization trigger, a sample number identification unit, an event registration unit, an acoustic emission acquisition unit, and an LSTM-based intelligent early warning unit. The multi-source sensor array synchronously acquires parameters such as pressure, pore water pressure, and permeability coefficient. The acoustic emission acquisition unit acquires acoustic emission signals generated during sample damage evolution and extracts parameters such as acoustic emission energy, ring count, number of events, amplitude, duration, rise time, dominant frequency, and b-value. The unified clock module provides a unified time reference to each acquisition module. The synchronization trigger generates a unique event number and a freeze trigger timestamp when the monitored data reaches a preset dynamic threshold. The sample number identification unit records the one-to-one correspondence between the triaxially loaded sample and the CT scan sample before and after transfer. The event registration unit matches the mechanical-permeable-acoustic emission data of the triaxial loading stage with the CT scan data based on the event number, sample number, freeze trigger timestamp, and CT scan timestamp. The microstructure parameters obtained by scanning are correlated; the real-time data acquisition system is used to acquire, cache and transmit the above multi-source data; the LSTM-based intelligent early warning unit is used to fuse multi-parameter time series data and output a three-level disaster early warning signal.

[0015] Furthermore, the axial servo loading unit of the triaxial loading module includes a servo motor, a coupling, a ball screw, and a piston rod, used to apply an axial load of 0-50kN and an axial stiffness of 500N / μm; the confining pressure control cavity has a double-layer structure, with an outer layer of 304 stainless steel and an inner layer of polyurethane insulation.

[0016] Furthermore, the back pressure pump of the dynamic osmotic pressure regulation system is used to provide a stable water pressure of 0-2MPa; the dual closed-loop PID controller has an adjustment cycle of 1ms.

[0017] Furthermore, the transparent pressure chamber of the high-resolution CT diagnostic module is made of polycarbonate material with a light transmittance of ≥92% and an inner diameter of Φ150mm × height of 300mm; the top is a hinged, detachable top cover, which can be moved into the microfocus CT scanner rack as a whole.

[0018] Furthermore, the microfocus CT scanner has a spatial resolution of 3μm, a scanning slice thickness of 20μm, a three-dimensional rotating stage with a cross slide structure, a Z-axis travel of 100mm, a repeatability of ±2μm, and acquires projection data at a step angle of 0.5°.

[0019] Furthermore, the transparent permeation tank has a porous media support in the upper and lower layers, a sample chamber in the middle layer, and a water collection chamber in the bottom layer; it is made of acrylic material, and the tank wall has a double-layer hollow structure and is filled with silicone oil to eliminate optical distortion; the guide hole of the porous media support has an inclination angle of 30°, a pore diameter of Φ2mm, and a porosity of 35%±2%.

[0020] Furthermore, the acoustic emission acquisition unit includes a piezoelectric acoustic emission sensor, a water-resistant acoustic waveguide rod, a preamplifier, a bandpass filter, and an acoustic emission acquisition card. The piezoelectric acoustic emission sensor is installed on the outer wall of the confining pressure control cavity, the upper pressure head, the lower pressure head, or the waveguide rod at the end of the sample via an acoustic coupling agent. The acoustic emission signal is transmitted to the data fusion terminal after preamplification, filtering, and analog-to-digital conversion, and is used to obtain the number of acoustic emission events, ring count, amplitude, energy, duration, rise time, dominant frequency, RMS value, and b-value in real time.

[0021] A method for testing the disaster threshold of multi-field coupling in water-rich tunnels, comprising the aforementioned test device for testing the disaster threshold of multi-field coupling in water-rich tunnels, and including the following steps:

[0022] S1 Sample preparation: The moisture content of the sample was controlled at 12%~30% by liquid nitrogen freeze drying, and the CO2 back pressure saturation method was used to gradually pressurize to 2MPa so that the sample B value is ≥0.95;

[0023] S2 Multi-field Coupled Loading: Within the confining pressure range of 50~500kPa and hydraulic gradient range of 0.5~8.0, axial stress and seepage pressure are applied simultaneously, and the hydraulic gradient is increased stepwise in increments of 0.2, with each step maintained for 30 minutes until seepage stabilizes;

[0024] S3 Microstructure Dynamic Observation: Porosity, coordination number, and throat diameter distribution parameters are obtained through CT scanning to reconstruct a three-dimensional pore network model;

[0025] S4 Catastrophic Threshold Determination: Using a sudden change in the penetration coefficient Δk / Δt > 10⁻ 5 A sudden drop in s⁻¹, acoustic emission b-value <1.0, and a sudden increase in axial strain are used as joint criteria to identify the critical state of catastrophic events in the sample. When the triaxial loading module detects that any of the above indicators reaches a preset threshold, a synchronous trigger generates a unique event number and records the freezing trigger timestamp t0, while simultaneously controlling the liquid nitrogen quick-freezing unit to fix the sample state. Subsequently, the sample with the unique number is transferred to the high-resolution CT diagnostic module for scanning under low-temperature conditions, and the CT scan data is registered with the triaxial loading data corresponding to time t0 based on the sample number, event number, and scanning timestamp. The LSTM-based intelligent early warning unit integrates mechanical, seepage, acoustic emission, and CT microstructure parameters to output three levels of catastrophic warning signals: green, yellow, and red.

[0026] Furthermore, in step S1, liquid nitrogen freeze drying is carried out at -196℃ for 48 hours under vacuum conditions ≤10Pa, with glycerol added as an ice crystal growth inhibitor, and the sample structure deformation is controlled to ≤3%; CO2 back pressure saturation is applied at a gradient of 0.1MPa per hour.

[0027] Furthermore, when the CT scan is triggered in step S3, the loading system freezes the current load, the load fluctuation is ≤0.05% of the full scale, the sample is quickly frozen and fixed by spraying liquid nitrogen at -196℃ for 10s, and transferred in a constant temperature environment of -50℃ to avoid freeze-thaw damage to the microstructure.

[0028] Furthermore, in step S4, the LSTM intelligent early warning unit inputs six-dimensional time-series data including axial strain, confining pressure, pore pressure, permeability coefficient, acoustic emission energy, and b-value, and outputs three-level early warnings: green, yellow, and red. When the hydraulic gradient is ≥3.8 and Δk / Δt>10⁻ 5 A sudden water inrush warning is triggered when the damage threshold is ≥0.35 and the b value is <1.0. A large deformation warning is triggered when the damage threshold is ≥0.35 and the b value is <1.0.

[0029] The innovation of this invention lies in:

[0030] Multi-field coupled synergistic loading: Achieve dynamic coupling of seepage field, stress field, and damage field (error <5%).

[0031] Cross-scale observation technology: CT scans (3μm resolution) and macroscopic mechanical responses enable data fusion and collaborative operation.

[0032] Intelligent early warning system: LSTM model integrating acoustic emission, permeability coefficient and strain data (AUC≥0.93).

[0033] Quantitative Threshold Determination: Based on a dynamic threshold model using energy dissipation rate and bifurcation theory, the critical hydraulic gradient is output. =3.8) and damage threshold ( =0.35).

[0034] Specifically, the energy dissipation rate calculation model is based on a thermodynamic framework, defining the energy dissipation rate per unit volume: ,in For stress tensor, For plastic strain rate, This represents the seepage energy flux. When... The sudden increase exceeded the threshold When this occurs, the system enters an unstable phase; bifurcation theory is applied, and the Hill bifurcation condition is used to determine the system's stability: ,in Here is the tangent stiffness matrix. This represents the seepage coupling matrix. When the system approaches a bifurcation point, a small perturbation will cause a sudden change in the deformation mode, corresponding to the catastrophic threshold.

[0035] The beneficial effects of this invention due to the adoption of the above technical solutions are as follows:

[0036] This invention provides a multi-field coupling disaster threshold test device and method for water-rich tunnels. By integrating triaxial loading, CT scanning, seepage erosion monitoring, and intelligent early warning technologies, it can accurately simulate the multi-field coupling mechanism of seepage-stress-damage, and has the following significant advantages: In terms of accuracy, the accuracy of determining the critical gradient of water inrush is improved by 40%, and the early warning time for large deformation is advanced by 30 minutes; in terms of efficiency, more than 200 sets of multi-physics field parameters can be obtained in a single test, and the test efficiency is improved by 3 times; in terms of engineering application, engineering threshold parameters can be directly output, providing scientific and reliable technical support for the prevention and control of seepage disasters in water-rich strata tunnels. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the overall structure of a multi-field coupling disaster threshold test device for water-rich tunnels according to the present invention;

[0038] Figure 2 This is a schematic diagram of the three-axis loading module structure;

[0039] Figure 3 This is a schematic diagram of the high-resolution CT diagnostic module.

[0040] Figure 4 A schematic diagram of the visualization module structure for seepage erosion;

[0041] Figure 5 This is a schematic diagram of the vertical cross-sectional structure of the transparent pressure chamber;

[0042] Figure 6This is a schematic diagram of the vertical cross-sectional structure of the seepage erosion module;

[0043] Figure 7 This is a diagram of the data fusion system architecture.

[0044] Figure 8 A flowchart for the multi-field coupling disaster threshold test method in water-rich tunnels;

[0045] Figure 9 Diagram of LSTM network architecture;

[0046] Figure 10 This is a graph showing the confining pressure variation in Example 3;

[0047] Figure 11 This is a graph showing the hydraulic gradient variation in Example 3.

[0048] Figure 12 This is a graph showing the change in permeability coefficient in Example 3;

[0049] Figure 13 This is a graph showing the change in the acoustic emission b-value in Example 3.

[0050] Figure 14 This is a graph showing the change in porosity of CT in Example 3.

[0051] In the diagram: 10-Axial servo loading unit, 101-Servo motor, 102-Coupling, 103-Ball screw, 104-Piston rod, 11-Upper pressure head, 12-Lower pressure head, 13-Containing pressure control chamber, 14-Dynamic osmotic pressure adjustment system, 141-Back pressure pump, 142-Safety relief valve, 143-Three-way valve, 15-Base, 16-Control box, 17-Drainage channel, 21-Transparent pressure chamber, 211-Polycarbonate housing, 212-Quartz glass observation window, 213-Sensor interface (M6 threaded hole), 22-Rotary table spindle, 23-Three-dimensional rotary table 24-Sample chamber, 25-Microfocus CT scanner gantry, 26-X-ray source, 27-Quick-release interface, 28-Detector panel, 31-Transparent seepage tank, 32-Tank wall, 33-Laser emitter, 34-Porous media support, 341-Guide hole, 35-Hydraulic pipeline, 361-Water inlet channel, 362-Water outlet channel, 37-Sample chamber, 38-Confining pressure loading device, 381-Annular airbag, 39-High-speed camera system, 391-LED light source array, 392-Tripod, 40-Load sensor, 41-Synchronous trigger, 42-Early warning output interface. Detailed Implementation

[0052] The present invention will be further described below with reference to the accompanying drawings and specific embodiments:

[0053] A multi-field coupled disaster threshold testing device for water-rich tunnels includes the following modules: a triaxial loading module, a high-resolution CT diagnostic module, a seepage erosion visualization module, and a data fusion and early warning module. Detailed descriptions of each module are as follows:

[0054] I. Three-axis loading module

[0055] like Figure 2 As shown, the system includes an axial servo loading unit 10, a confining pressure control chamber 13, and a dynamic osmotic pressure adjustment system 14, supporting precise control of the hydraulic gradient from 0.5 to 8.0. The servo motor 101 has a rated torque of 12 N·m and an encoder resolution of 17 bits, communicating with the controller via the EtherCAT protocol. The ball screw 103 has a diameter of 32 mm, a lead of 5 mm, and an axial stiffness of 500 N / μm, and is fixed at both ends with angular contact bearings (P4 grade). The piston rod 104 is made of nitrided S45C steel with a surface hardness of HRC60. The confining pressure control chamber 13 has a double-layer structure: an outer layer of 304 stainless steel (10 mm thick) and an inner polyurethane insulation layer. The chamber is bolted to the base (8×M16). The dynamic osmotic pressure adjustment system 14 uses a dual closed-loop PID controller (adjustment period 1 ms), connecting the confining pressure control chamber 13 to the back pressure pump 141 (pressure fluctuation ≤ 0.01 kPa) via a three-way valve 143.

[0056] The back pressure pump 141 is responsible for providing a stable water pressure environment ranging from 0 to 2 MPa. The three-way valve 143 precisely distributes the pressure ratio between the upper and lower chambers of the loading system according to real-time PID control commands.

[0057] Specifically, in this invention, the servo motor 101 receives a control signal and drives the ball screw 103 through the coupling 102, converting the rotational motion into the axial displacement of the piston rod 104, applying a load of 0-50kN. The piston rod 104 applies vertical pressure to the confining pressure control chamber 13, where a sample is placed. During pressurization, the data from the control signal terminal received by the servo motor 101 and the data output by the annular airbag 40 are used to monitor and control the pressure magnitude. The data terminal of the annular airbag 40 and the back pressure pump 141 are connected to the control box 16 of this module. The control box 16 transmits all data to the main information processing and fusion terminal.

[0058] In the dynamic osmotic pressure regulation system 14, the back pressure pump 141 pressurizes the water flow, which is then divided into upper and lower paths in the confining pressure control chamber 13 via a three-way valve 143. The water flows inside the sample, and a drainage channel 17 is installed in the middle of the confining pressure control chamber 13 for drainage metering. The control objective is to establish a pressure difference ΔP across the sample, achieving a hydraulic gradient i = ΔP / L (where L is the sample height). The dynamic osmotic pressure regulation system 14 uses a dual-closed-loop PID controller to achieve hydraulic gradient control. The inner loop (pressure loop) adjusts the speed of the back pressure pump 141 to measure the pore water pressure; the outer loop (gradient loop) dynamically allocates the pressure ratio between the upper and lower chambers, setting the gradient to obtain the measured gradient. The osmotic pressure system simulates the seepage force of groundwater around the tunnel, which is the physical source of the hydraulic gradient.

[0059] The triaxial loading module of this invention includes the following installation steps:

[0060] Step S1, install base 15. Secure base 15, ensuring it is level.

[0061] Step S2: Install the confining pressure control cavity 13. Secure the cavity to the base 15 with bolts, ensuring a tight seal.

[0062] Step S3: Install the ball screw 103 and piston rod 104. The lower end of the ball screw 103 is machined with internal threads (M30×2), and the upper end of the piston rod 104 has corresponding external threads. The two are screwed together and secured with a lock nut to prevent loosening. The contact surfaces are coated with high-temperature grease to allow for slight deflection while transmitting axial loads.

[0063] Step S4: Install the servo motor 101 and coupling 102. The servo motor 101 is fixed on the bracket and connected to the ball screw 103 through the coupling 102.

[0064] Step S5: Install the annular airbag 40. Install the annular airbag 40 at the lower end of the piston rod 104 and connect the upper end to the ball screw 103.

[0065] Step S6: Install the upper and lower pressure heads. The upper pressure head 11 is connected to the lower end of the wireless annular airbag 40 via a ball joint. The ball joint allows for ±2° deflection to ensure uniform load distribution. The lower pressure head 12 is aligned with the base of the confining pressure control cavity 13 via a positioning pin and fixed with four M10 bolts. After installation, the coaxiality error of the upper and lower pressure heads is ≤0.05mm.

[0066] Step S7: Install the safety relief valve 142 and the control box 16. The safety relief valve 142 is connected to the confining pressure control chamber 13, and the control box 16 is fixed to the side of the frame.

[0067] Step S8, installation of the pressure relief system. Install the shock-absorbing pads, bolt the back pressure pump 141, connect the power supply and data signal lines, install the flange at the predetermined position in the confining pressure control chamber 13, conduct a sealing test, and finally lay the water conduit. A high-pressure rigid pipe is used from the outlet of the back pressure pump 141 to the inlet of the three-way valve 143, while PTFE hoses are used from the section from the three-way valve 143 to the confining pressure control chamber 13 and the drainage conduit. Ensure a tight seal at 2.0 MPa for 30 minutes.

[0068] Step S9: Connect the pipes and electrical wiring.

[0069] II. High-resolution CT diagnostic module

[0070] like Figure 3 As shown, it includes a detachable transparent pressure chamber 21 and a microfocus CT scanner to achieve high-precision capture of microstructures.

[0071] Specifically, in this invention, the transparent pressure chamber 21 is a cylinder with an inner diameter of Φ150mm × height of 300mm and a wall thickness of 15mm. The polycarbonate shell 211 (tensile strength ≥65MPa, light transmittance ≥92%) and the metal flange are sealed with a trapezoidal groove. The top of the transparent pressure chamber 21 is a hinged top cover, housing a sample chamber 24. It can be moved as a whole into the microfocus CT scanner frame 25 and fixed to the frame during scanning via a quick-release interface 27 (ISO160 standard). The rotary stage spindle 22 is a stepped shaft with dimensions of Φ40h6 × 200mm (M30 thread at the front end) and is made of 38CrMoAl nitrided steel (HV≥900). The sample chamber 24 is a cylindrical sleeve with dimensions of Φ148H7 × 150mm (inner diameter adapted to the sample) and is made of high-strength, high-light-transmittance polycarbonate. The microfocus CT scanner gantry 25 is a rectangular frame measuring 600×600×600mm, equipped with a quick-release interface 27, and is made of cast iron (with shock-absorbing design). The X-ray source 26 uses a 20mm tungsten target with an exit port, operating at 120kV and 200μA, with a focal spot size of 3μm. The detector panel 28 measures 300×300×50mm, with an effective area of ​​280×280mm (2048×2048 pixels, 16-bit dynamic range), and is made of cadmium telluride scintillator (50μm pixel size). The 3D rotary stage 23 is a cross-slide structure with a base measuring 200×200mm, a Z-axis travel of 100mm, and is made of aluminum alloy. It employs a cross-slide structure (repeatability ±2μm) and supports 360° continuous rotation of the sample, acquiring projection data at 0.5° / step angles during scanning. The quartz glass observation window 212 is a circular window with dimensions of Φ80mm × 5mm (embedded installation), and is made of fused silica (X-ray transmittance ≥99%).

[0072] The installation steps and methods for each component of the high-resolution CT diagnostic module are as follows:

[0073] The assembly sequence of the transparent pressure chamber 21 is as follows: install the quartz glass observation window 212, press in the fluororubber sealing ring, open the top cover of the pressure chamber, install the sample chamber 24, and lock the main shaft of the rotary table 22.

[0074] The three-dimensional rotary stage 23 is integrated with the transparent pressure chamber 21: the pressure chamber base is positioned, the three-dimensional rotary stage 23 is installed, and the main spindle is aligned and calibrated coaxially.

[0075] The microfocus CT scanner gantry 25 is docked with the transparent pressure chamber 21: the quick-release interface 27 of the CT gantry is opened and moved into the transparent pressure chamber 21, the quick-release interface of the pressure chamber is locked, and the X-ray source-detector is aligned.

[0076] Detector panel 28 installation: The laser tracker is positioned as a reference and protected (wearing an anti-static wrist strap and covering the detector panel surface with a dustproof film). The detector panel 28 is positioned on the microfocus CT scanner frame 25 (the distance between the center of the panel and the focal point of the X-ray source is 350±0.1mm). It is initially fixed with a pre-tightening positioning pin. The vacuum pump is started (vacuum degree -80kPa) for adsorption and fixation. Adsorption force is tested (a lateral force of 50N is applied, and the displacement is ≤2μm).

[0077] In this invention, the transparent pressure chamber 21 is fixed and removed from the microfocus CT scanner frame 25 via the quick-release interface 27; the sample chamber 24 has a hinged top cover that can be opened to directly insert the sample; the quartz glass observation window 212 is disassembled by removing the pressure ring (4×M6 screws) from the outside.

[0078] During the experiment, the three-dimensional rotary stage 23 operated according to the following procedure: First, the sample was scanned and positioned, then the Z-axis was driven to lift the sample to the CT scan focal plane. After positioning, the rotary stage rotated in 0.5-degree increments, triggering the CT projection acquisition system to capture data at each step. The tomographic scanning process was set with a slice thickness of 20 micrometers and a spatial resolution of 3 micrometers. A total of 720 projection data points were acquired through a complete 360-degree rotation, with each scan taking 60 seconds. For data synchronization, the system utilized the CT scan trigger signal (TTL 5-volt level) to work in conjunction with the loading system: when the real-time monitoring data reached a preset dynamic threshold, the loading system immediately froze the current load (ensuring load fluctuations did not exceed 0.05% of full scale), and through precise timestamp alignment technology, synchronized the sample's mechanical response data with its microstructure evolution data.

[0079] III. CT Image Processing and 3D Reconstruction Algorithms

[0080] The raw projection data acquired by the microfocus CT scanning system is converted into quantitative structural parameters using the following algorithm:

[0081] (1) Image reconstruction algorithm. A filtered back projection algorithm is used, and the projection data is preprocessed to reconstruct the acquired projection values. Perform logarithmic transformation and negative value correction. ,in For the detected value, The input is the no-load value; for filtering function selection, a Ram-Lak filter (ramp filter) is used for frequency domain filtering. ,in The cutoff frequency is set to the Nyquist frequency; for backprojection reconstruction, the filtered projection is backprojected along the ray path to the image space. ,in This is the filtered projection. Output: Reconstruct a cross-sectional grayscale image sequence with voxel size 3μm×3μm×20μm.

[0082] (2) Image segmentation and aperture recognition algorithm. Adaptive threshold segmentation (Otsu algorithm), calculate the image grayscale histogram, and find the threshold. To maximize the inter-class variance ,in , The proportions of two types of pixels, , The mean was used; morphological post-processing was performed, employing opening operations (erosion followed by dilation) to remove noise. ,in The structuring element is 3×3; pore marking is performed using a connected component marking algorithm (8-neighborhood) to distinguish independent pore regions and label them as follows. .

[0083] (3) Three-dimensional pore network extraction algorithm (based on the maximum sphere algorithm). Distance transformation: calculate the Euclidean distance transformation for the binary image. , representing the nearest distance from each pore voxel to the solid boundary; maximum sphere identification involves finding a local maximum point in the distance field, which serves as the center of the "maximum sphere," and its radius... The current local maximum radius of the pore. Throat identification and connection, traversing all maximum sphere pairs. Calculate the minimum distance along the line connecting their centers. ,in ,like If a throat connection exists between the two pores, then the network is constructed using the largest sphere as a node and the throat as an edge. The node attribute is coordinates. ,radius The side attribute is the throat length. equivalent radius ( (This refers to the cross-sectional area of ​​the throat).

[0084] (4) Formulas for calculating structural parameters. Porosity, ,in ; Coordination number Count the number of throats connected to each pore node and output a histogram of the distribution; throat diameter distribution. Statistical analysis of the equivalent diameter of all larynxes The frequency distribution is output at 5μm intervals.

[0085] (5) Data output and synchronization. Data is uploaded to the data fusion system in real time via the OPC UA protocol and is aligned with the timestamps of mechanical and seepage data (error <10ms).

[0086] Specifically, in this invention, the microfocus CT scanner gantry 25 surrounds the transparent pressure chamber 21, with the scanning axis coinciding with the sample axis. The microfocus CT scanner is an independent peripheral device, achieved through a detachable design. The transparent pressure chamber 21 serves as a carrier, with the internal sample cavity 24 holding the sample. It works collaboratively with the triaxial loading module via data transmission. The workflow is as follows: when the triaxial loading reaches a dynamic threshold, liquid nitrogen rapid freezing fixation is performed; the sample is then transferred to the CT module for high-resolution scanning and three-dimensional reconstruction comparison.

[0087] In this invention, the key technical system for inter-module collaborative work includes the following core elements: First, liquid nitrogen quick-freezing technology (spraying at -196℃ for 10s) is used, and glycerol is added as an ice crystal growth inhibitor to strictly control the deformation of the sample structure within the range of ≤3%, and ensure that the displacement field registration error is ≤0.1mm; Second, a special transfer fixture is designed, and a vacuum adsorption tray (maintaining a negative pressure of 0.1Pa) is used for operation in a constant temperature environment of -50℃ to effectively prevent the freeze-thaw effect from damaging the microstructure of the sample; At the same time, a high-precision CT positioning reference system is established, and a positioning accuracy of ≤0.01mm is achieved by embedding the sample into the positioning pin; Finally, a multi-sample sequence test scheme is implemented, and five groups of samples prepared from the same sample are loaded to strain levels of ε=5%, 10%, 15%, 20%, and 25% respectively and then frozen and scanned to construct a strain-porosity relationship database with engineering application value (as shown in Example 3), providing key data support for the study of seepage disaster mechanism.

[0088] IV. Seepage Erosion Module

[0089] like Figure 4 As shown, it includes a transparent seepage channel 31 and an annular airbag 39 to quantify the particle transport trajectory.

[0090] Specifically, the transparent seepage tank 31 is made of aerospace-grade acrylic with a light transmittance of ≥92% and a bending strength of 90MPa. It features a layered design: upper and lower layers are porous media supports 34, the middle layer is the sample chamber 37, and the bottom layer is the water collection chamber. Its dimensions are 500×200×200mm (length×width×height), and the wall thickness is 12mm (pressure resistance ≥0.5MPa). During installation, the tank is placed on the base, and the mounting holes are aligned using locating pins (tolerance ±0.02mm). Eight sets of M12 stainless steel bolts are tightened (torque 25N·m), and the sealing ring is made of fluororubber (pressure resistance 2MPa).

[0091] The tank wall 32 adopts a double-layer hollow design (outer layer thickness 5mm, inner layer thickness 7mm), with the interlayer filled with silicone oil (refractive index 1.403) to eliminate optical distortion. The inner wall is covered with an anti-scratch film (thickness 0.1mm), and after installation, the transmitted wavefront distortion is ≤λ / 4 (λ=532nm).

[0092] The annular airbag 39 is equipped with a camera featuring a 50mm macro lens. During installation, it is fixed to the side of the tank via a tripod 392, with the optical axis perpendicular to the center plane of the sample; it is connected to the GigE interface to the synchronous trigger 41 with a delay ≤10μs.

[0093] In this invention, the LED light source array 391 consists of 48 cool white LED units, featuring a standard color temperature of 5600K and an illuminance output of 20000 lux. Its 120° wide-angle diffusion characteristic ensures uniform illumination of the sample area. During installation, the array plate is precisely snapped into the top rail of the camera. A constant current drive mode (operating current 2A±1%) ensures the stability of the light source. Simultaneously, a high-efficiency heat dissipation system is configured, with the heat sink surface temperature strictly controlled within ≤40℃ during operation, guaranteeing the reliability and safety of long-term testing.

[0094] The laser emitter 33 uses a 532nm laser with a spot size of 5mm and a velocity measurement accuracy of ±0.1mm / s. Its optical axis forms an angle of 45°±0.5° with the center plane of the sample (calibrated by the laser positioning instrument). During installation, it is fixed within the wall 32 of the transparent permeation tank using precise positioning hole spacing, and the seal is made of fluororubber.

[0095] The porous media support 34 is CNC-carved from transparent high-strength polycarbonate (PC) or aerospace acrylic (PMMA) to ensure a light transmittance of ≥90% for easy optical observation. The surface is polished to reduce flow resistance. The guide holes 341 have an inclination angle of 30°±0.5°, with a diameter distribution of Φ2±0.05mm (70%) and Φ1mm (30%), and a porosity of 35%±2%. The support consists of two layers, upper and lower, which are inserted into the groove of the tank (gap ≤0.05mm) during installation. The bolt preload is 10N·m to ensure a leak-free seal with the tank wall 32.

[0096] The guide hole 341 is a tapered hole with an inlet diameter of 2mm and an outlet diameter of 5mm, with an inclination angle of 30°±0.5° and a surface roughness Ra≤0.4μm. During installation, the guide hole 341 is located inside the bracket, precisely drilled, and arranged at 10mm intervals.

[0097] The confining pressure loading device 38 is a ring-shaped airbag made of fluororubber, which surrounds the side wall of the sample chamber 37. The pressure is independently controlled by hydraulic lines 35. The confining pressure range is 50-500 kPa, and the response time is ≤5 s. During installation, the hydraulic lines 35 are connected to the interface on the side wall of the tank (quick-connect sealing joint).

[0098] The inlet and outlet channels 361 and 362 adopt a symmetrical dual-channel design with a channel inclination angle of 30° to match the guide hole of the support. The interface standard is pipe thread. The flow rate range is 0.1-10L / min, and the hydraulic gradient is controlled with an accuracy of ±0.01. During installation, the inlet channel is connected to the seepage pressure system, and the outlet is connected to the flow meter. The pipe slope is ≥5° (to prevent air bubble retention).

[0099] In this invention, the sample chamber 37 adopts a cylindrical design (dimensions Φ150mm×150mm). The bottom of the chamber is designed as a flange structure, connected to the porous media support 34 by six M8 bolts, providing both rigid fixation and detachability. A rubber sealing ring is installed between the flanges to ensure pressure resistance and water tightness. This chamber can accommodate samples with a diameter of 50-100mm, and the sample saturation is precisely controlled within the range of 30%-100%. The chamber matrix material is made of aerospace-grade polycarbonate (PC), which has both high light transmittance (≥92%) and excellent impact resistance, ensuring that the high-speed camera system can clearly capture the migration trajectory of particles with a diameter ≥5μm. During installation, the chamber is placed between the upper and lower layers of the porous media support 34: first, the lower support is positioned and fixed, then the chamber is inserted and the upper support is installed, and finally, vacuum silicone grease is evenly applied to the surface of the sealing ring, and a clamping force of ≥5kPa is applied to ensure sealing performance.

[0100] The synchronous trigger 41 uses an FPGA controller with a time resolution of 10ns and supports TTL / RS422 trigger signals. The wiring method is as follows: CH1 is connected to the high-speed camera (exposure trigger), CH2 is connected to the laser emitter, and CH3 is connected to the warning output interface of this module (to transmit data to the main information processing and fusion terminal).

[0101] Specifically, the workflow of the seepage erosion module is as follows:

[0102] Step S1, Water Inlet: Water flows evenly into the tank through the water inlet channel 361 with an inclination angle of 30°.

[0103] Step S2, seepage: the sample enters the sample chamber 37 through the porous medium support 34 to simulate groundwater seepage.

[0104] Step S3, confining pressure: the annular airbag 381 applies lateral pressure to simulate the stress of the surrounding rock in the tunnel.

[0105] Step S4, Observation: The annular airbag 39 records particle transport through the transparent groove wall 32.

[0106] Step S5, Water discharge: The seepage water is discharged through the symmetrical water discharge channel 362.

[0107] In this invention, the seepage erosion module specializes in observing seepage erosion effects through independent confining pressure loading and inclined flow channel design; it forms a micro-macro complementary relationship with the CT module, and a stress-seepage coupling with the triaxial module. The three modules achieve accurate determination of catastrophic threshold through data fusion.

[0108] Specifically, the collaborative work between the seepage erosion module and the high-resolution CT diagnostic module is reflected in three aspects. First, the two modules achieve effective data complementarity. The high-resolution CT diagnostic module focuses on capturing the dynamic evolution of the microscopic pore structure inside the sample, while the seepage erosion module simultaneously observes the particle migration trajectory on a macroscopic scale. Second, the two modules are physically separated. This isolation design effectively avoids electromagnetic interference caused by the high-voltage source of the off-site high-resolution CT module to the high-speed camera system of the seepage erosion module. Third, the two modules have a joint early warning function. When the high-resolution CT detects an abnormal change in porosity with a sudden increase, and the seepage erosion module also captures the key signal of particle migration, the system will combine these two key indicators to jointly trigger a red warning mechanism based on the LSTM algorithm, as shown in Example 3.

[0109] The differences between the seepage erosion module and the triaxial loading module in this invention are mainly reflected in three aspects. First, their core objectives differ: the triaxial loading module aims to simulate the complex stress state at the tunnel face, while the seepage erosion module focuses on simulating the seepage erosion effect of groundwater on the soil. Second, the loading methods differ: the triaxial loading module applies axial loads to achieve stress loading, while the seepage erosion module applies lateral confining pressure without applying axial loads. Third, the core data output by the modules also differ: the triaxial loading module mainly provides stress-strain curves and permeability coefficients, while the seepage erosion module outputs key parameters such as particle transport rate and critical initiation hydraulic gradient.

[0110] V. Data Fusion Module

[0111] It includes a multi-source sensor array and an LSTM-based intelligent early warning unit, which outputs the probability of disaster in real time.

[0112] The multi-source sensor array consists of: a laser velocimetry system integrated into the seepage erosion module, where the laser emitter 33 is synchronously triggered with the annular airbag 39, and the particle transport velocity is calculated using the particle image velocimetry (PIV) method. The data is transmitted to the data acquisition system via the synchronous trigger 41; an acoustic emission sensor, arranged on the inner wall of the confining pressure control cavity 13, employing a piezoelectric sensor (frequency range 20kHz–1MHz), to collect the acoustic emission event energy, count, and b-value in real time; and other sensors, including strain gauges (axial / circumferential), pore water pressure gauges, piezometers, and displacement sensors, which together constitute the multi-source sensor array.

[0113] Real-time data acquisition system: The OPC UA gateway is a 128-channel @ 24-bit ADC that supports Modbus TCP protocol conversion.

[0114] Caching strategy: 1MB / channel circular buffer, supporting 72-hour breakpoint resume.

[0115] Intelligent early warning unit: The LSTM model has 3 hidden layers (128-64 neurons), takes 6-dimensional time series data as input, and outputs three-level early warning signals.

[0116] Warning thresholds: probability of disaster P<0.3 (green), 0.3≤P<0.7 (yellow), P≥0.7 (red).

[0117] The principle of the intelligent early warning algorithm in this invention is as follows:

[0118] LSTM model architecture.

[0119] Specifically, the input layer consists of 6-dimensional time-series data (axial strain, confining pressure, pore pressure, permeability coefficient, acoustic emission energy, and b-value). The hidden layer is the first layer: 128 neurons, tanh activation, with dropout=0.2.

[0120] Second layer: 64 neurons, ReLU activation. Output layer: 3 neurons (corresponding to green / yellow / red alerts), softmax normalization, loss function: weighted cross-entropy. (Weight w = [0.2, 0.3, 0.5] to improve the sensitivity of red alerts)

[0121] Furthermore, the model was validated based on measured data. 200 sets of historical experimental data were used for training and evaluation (80% for model training and 20% for independent validation). The obtained LSTM early warning model has excellent performance, with an AUC value of 0.94 and an F1 score of 0.89.

[0122] The key algorithm formulas are annotated as follows:

[0123] Specifically, the LSTM cell is calculated as follows: , , , , , .in For the Gate of Oblivion For input gate, For output gate, This is the Hadamard product. The early warning probability is calculated as follows: , where z is the output value of the fully connected layer.

[0124] In this invention, the specific experimental method is as follows:

[0125] 1. Sample preparation

[0126] Specifically, the sample preparation process involves two key methods: liquid nitrogen freeze-drying and CO2 back-pressure saturation treatment. Liquid nitrogen freeze-drying involves continuously freezing the sample at -196 degrees Celsius for 48 hours under a vacuum of no more than 10 Pa. The purpose is to instantly freeze the pore water to prevent moisture migration and structural damage, ensuring the central region of the sample is completely frozen. Moisture is then removed through ice crystal sublimation, ultimately precisely controlling the sample's moisture content between 12% and 30%. CO2 back-pressure saturation treatment uses a gradient pressurization method of 0.1 MPa per hour to gradually increase the pressure to 2 MPa, requiring a B value of no less than 0.95, allowing the sample saturation to be adjustable within the range of 30% to 100%. This method effectively increases water saturation by replacing air in the pores with carbon dioxide, avoiding structural damage caused by sudden changes in excess pore water pressure, ensuring the sample reaches complete saturation, and simulating the actual water content conditions in different sections of tunnel engineering.

[0127] 2. Multi-field coupled loading

[0128] Specifically, it comprises two core components: confining pressure loading and hydraulic gradient control. The confining pressure loading component applies a confining pressure at a rate of 0.5 kPa within the range of 50 to 500 kPa, and satisfies… Its core purpose is to simulate the real geostress environment, reconstruct the in-situ stress state of the soil, and effectively limit the lateral deformation of the specimens that may occur during the experiment. The hydraulic gradient control stage sets the gradient within the range of 0.5 to 8.0, increasing it in increments of 0.2, with each pressure level maintained for 30 minutes to ensure seepage stability. The role of this stage is to improve the approximation accuracy of the critical hydraulic gradient to within 5%, ensuring that the seepage process reaches a stable state and covering the range of inrush gradients that may be encountered in typical tunnel engineering.

[0129] This invention forms a closed-loop technology: freeze-drying technology ensures that the structural characteristics of the sample are effectively preserved in its natural state; then, carbon dioxide back pressure saturation treatment is used to accurately simulate and control the groundwater occurrence conditions; on this basis, the complex stress field and seepage field environment of the tunnel is reproduced through the coupling effect of confining pressure loading and seepage control; finally, by using the gradual loading method of hydraulic gradient, the key threshold for seepage instability is accurately captured and determined.

[0130] 3. Data acquisition and processing.

[0131] Specifically, through high-resolution CT diagnosis, after triggering a CT scan under preset conditions, the system acquires internal structural information of the sample with a scan slice thickness of 20 μm, accurately extracts key parameters including porosity, coordination number, and throat diameter distribution, and reconstructs a three-dimensional pore network model of the sample based on these data. The data acquisition and data transmission software is well-known to those skilled in the art and will not be described in detail in this invention.

[0132] Furthermore, the catastrophic criterion includes the permeability coefficient mutation criterion and the acoustic emission b-value criterion. The permeability coefficient mutation criterion is as follows: The acoustic emission b-value criterion is

[0133] (when (An alert is triggered when the value is less than 1.0 and persists for 3 consecutive sampling periods).

[0134] In this invention, the intelligent early warning system uses an LSTM model for training, based on 200 sets of historical data, with 80% used for model training and 20% for validation. The loss function is weighted cross-entropy. ( =[0.2,0.3,0.5]). The model ultimately achieved excellent performance, with outputs including an AUC value as high as 0.94 and an F1 score of 0.89, and was able to complete real-time warning responses within 30 minutes.

[0135] Example:

[0136] Example 1: Device Implementation

[0137] The three-axis loading module includes an axial servo loading unit 10 with a maximum output load of 50kN and a control accuracy of ±0.1%FS; a confining pressure control chamber 13 with a pressure resistance of 2MPa and equipped with a temperature-compensated pressure sensor (accuracy ±0.05%FS); and a dynamic osmotic pressure adjustment system 14 supporting hydraulic gradients of 0.5~8.0 and equipped with a dual closed-loop PID controller (adjustment accuracy ±0.01kPa).

[0138] High-resolution CT diagnostic module: Transparent pressure chamber 21 made of polycarbonate material (pressure resistant ≥2MPa), inner diameter 150mm, equipped with quartz glass observation window 212 (diameter 80mm); Microfocus CT scanner with resolution 3μm and scanning slice thickness 20μm, achieving high-precision capture of microscopic conditions.

[0139] Seepage erosion module: transparent seepage channel 31 with dimensions of 500×200×200mm, built-in porous media support 34 (pore diameter Φ2mm, spacing 10mm); annular airbag 39 with a frame rate of 10,000fps, equipped with a laser speckle velocity measurement unit (accuracy ±0.1mm / s).

[0140] Data fusion module: The multi-source sensor array includes strain gauges (100Hz), piezometers (10Hz), and acoustic emission probes (specimen positioning accuracy ±0.5m); the real-time data acquisition system is based on the OPC UA protocol (delay <100ms) and supports multi-protocol conversion; the intelligent early warning unit is based on the LSTM algorithm (TensorFlow framework) and outputs three-level early warning signals (green / yellow / red).

[0141] Example 2: Method Implementation

[0142] Sample preparation: Take the sample, freeze-dry it with liquid nitrogen, and cut it into Φ50×100mm standard samples; use the CO2 back-pressure saturation method to make the sample B value ≥0.95, and control the initial saturation S. o =30%~100%.

[0143] Multi-field coupled loading: Apply confining pressure σ3 = 50~500 kPa, and increase the hydraulic gradient i = 0.5~8.0 at a rate of 0.5 kPa / s; simultaneously monitor the axial stress-strain curve and permeability coefficient. With pore water pressure .

[0144] Microstructure observation: Local CT scans (resolution 3 μm), slice thickness 20 μm were performed; a three-dimensional pore network model was reconstructed, and porosity was statistically analyzed. , coordination number Distribution of throat diameter .

[0145] The disaster threshold determination includes the criteria for sudden water inrush and large deformation, as detailed below:

[0146] Criterion for sudden water inrush: When the sudden change in permeability coefficient Δk / Δt > 10⁻ 5 s⁻¹ and hydraulic gradient ≥ When the value is 3.8, a sudden water inrush warning is triggered.

[0147] Large deformation criterion: when damage variable ≥ When the value of b = 0.35 and the acoustic emission b value < 1.0, a large deformation warning is triggered.

[0148] In this invention, the permeability coefficient mutation has a higher priority than the b value in the joint criterion; if both are triggered simultaneously, an early warning is initiated directly. When the catastrophic threshold is not reached, the system dynamically adjusts the loading parameters based on real-time data, forming a closed-loop control.

[0149] Example 3: Experimental Case

[0150] Experimental conditions: The experiment used loess samples taken from a certain location, with an initial moisture content of w=22% and a dry density of [missing information]. =1.65 g / cm³, B value = 0.97. The test loading path was set as follows: confining pressure The loading rate was increased from 200 kPa to 350 kPa at a rate of 50 kPa / h; simultaneously, the hydraulic gradient... The version was upgraded from 2.0 to 4.5 at a rate of 0.3 per hour.

[0151] The key data records in this embodiment are as follows:

[0152] Analysis of the results in this embodiment shows that when i=3.5, the rate of change of the permeability coefficient Δk / Δt=2.7×10⁻ 5 s⁻¹ (significantly exceeds the preset threshold of 1×10⁻) 5 Simultaneously, the b-value, characterizing the soil structural integrity, plummeted to 0.7, triggering a red alert. The concurrently performed high-resolution CT scan directly confirmed the emergency mechanism, showing that the sample porosity suddenly increased by 5.2%, and the proportion of pores with throat diameters >50 μm rose sharply from the initial 12% to 35%. This crucial data verified that significant connectivity had occurred within the internal seepage channels.

Claims

1. A multi-field coupled disaster threshold test device for water-rich tunnels, characterized in that, It includes a triaxial loading module, a high-resolution CT diagnostic module, a seepage and erosion visualization module, and a data fusion and early warning module; The triaxial loading module includes an axial servo loading unit (10), a dynamic osmotic pressure adjustment system (14) consisting of a back pressure pump (141), a safety relief valve (142), and a three-way valve (143); a confining pressure control chamber (13); a sample holder is provided inside the confining pressure control chamber (13), and the sample holder is used to place the sample; the axial servo loading unit (10) applies vertical pressure to the sample inside the confining pressure control chamber (13); the dynamic osmotic pressure adjustment system (14) realizes hydraulic gradient control through a dual closed-loop PID controller, the back pressure pump (141) pressurizes the water flow, and the water flow is divided into upper and lower paths in the confining pressure control chamber (13) through the three-way valve (143), the sample seeps inside, and a drainage channel (17) is arranged in the middle of the confining pressure control chamber (13) for drainage metering. The control objective is to establish a pressure difference ΔP at both ends of the sample to realize the hydraulic gradient; The working method of the triaxial loading module is as follows: the axial servo loading unit (10) applies a load to the sample inside the confining pressure control chamber (13), and the load sensor (40) on the axial servo loading unit (10) is used to monitor the pressure magnitude; the data terminal of the load sensor (40) and the back pressure pump (141) are connected to the control box (16) of the module; the control box (16) transmits all data to the total information processing and fusion terminal; when the triaxial loading reaches the dynamic threshold, liquid nitrogen quick-freezing fixation is performed; High-resolution CT diagnostic module: includes a detachable transparent pressure chamber (21), a three-dimensional rotating stage (23), and a microfocus CT scanner gantry (25); the microfocus CT scanner gantry (25) is cubic in shape, the transparent pressure chamber (21) is placed in the center of the microfocus CT scanner gantry (25), the three-dimensional rotating stage (23) is placed at the bottom of the transparent pressure chamber (21), and a sample chamber (24) is placed on the top of the three-dimensional rotating stage (23). The sample chamber (24) contains a sample that has been frozen and fixed with liquid nitrogen. The three-dimensional rotating stage (23) is used to adjust the height of the sample chamber (24) and drive the sample chamber (24) to rotate; the left side of the transparent pressure chamber (21) is used to connect to the X-ray source (26), and the right side is the detector panel (28). The working method of the high-resolution CT diagnostic module is as follows: First, the sample scanning and positioning are completed, and then the three-dimensional rotary stage (23) is driven to lift the sample to the CT scanning focal plane along the Z-axis. After positioning, the rotary stage rotates at a constant degree step angle and triggers the CT projection acquisition system to capture data at each step. In terms of data synchronization, the cross-module synchronization method of "threshold trigger - state freeze - off-site scanning - event registration" is adopted. During the triaxial loading process, the axial load, confining pressure, pore water pressure, seepage flow rate, permeability coefficient and acoustic emission signal are continuously acquired by the real-time data acquisition system and timestamped by the unified clock module. When the real-time monitoring data reaches the preset dynamic threshold, the synchronous trigger simultaneously sends trigger signals to the triaxial loading control box, liquid nitrogen quick-freezing unit, data fusion terminal, and CT diagnostic module, generating a unique event number and recording the freezing trigger time t0. The triaxial loading module locks the current loading state at time t0, saves the mechanical-percolation-acoustic emission data within the preset time window before and after t0, and simultaneously initiates liquid nitrogen quick-freezing to fix the sample structure. After the sample is quick-frozen, it is transferred to the high-resolution CT diagnostic module for scanning via a cryogenic transfer fixture with a unique sample number. The CT scan data is marked with the same event number, sample number, and scan start and end timestamps, and registered with the triaxial loading data corresponding to time t0. Since the sample has fixed its pore structure and damage state by liquid nitrogen quick-freezing at time t0, the microstructural parameters obtained by CT represent the frozen state when the triaxial loading reaches the threshold, thus achieving synchronous correlation between the off-site CT scan data and the triaxial loading process data. The seepage erosion visualization module is used for dynamic capture of particle transport and seepage path, including a transparent seepage tank (31), a high-speed camera system (39), a laser emitter (33), a confining pressure loading device (38), water inlet and outlet channels, and a sample chamber (37); after the high-resolution CT diagnostic module finishes working, the sample is transferred to the sample chamber (37). The transparent seepage tank (31) has a built-in porous medium support (34) with a flow guide hole. The confining pressure loading device (38) uses an annular airbag to apply dynamic confining pressure of 50~500kPa. The high-speed camera system (39) works with the laser emitter (33) to realize the quantitative acquisition of particle transport trajectory and velocity. The working method of the seepage erosion visualization module: The confining pressure loading device (38) applies a dynamic confining pressure of 50~500kPa through the annular airbag. The water flows through the inlet and outlet channels and the porous media support (34) to form a stable seepage field. The high-speed camera system, in conjunction with the laser emitter, captures and quantifies the particle migration trajectory and migration speed in real time. The data fusion and early warning module is used for multi-source data collaborative analysis, cross-module data registration, and intelligent disaster identification. It includes a multi-source sensor array, a real-time data acquisition system, a unified clock module, a synchronization trigger, a sample number identification unit, an event registration unit, an acoustic emission acquisition unit, and an LSTM-based intelligent early warning unit. The multi-source sensor array synchronously acquires pressure, pore water pressure, and permeability parameters. The acoustic emission acquisition unit acquires acoustic emission signals generated during sample damage evolution and extracts acoustic emission energy, ring count, event count, amplitude, duration, rise time, dominant frequency, and b-value parameters. The unified clock module provides a unified time reference to each acquisition module. The synchronization trigger generates a unique event number and a freeze trigger timestamp when the monitored data reaches a preset dynamic threshold. The sample number identification unit records the one-to-one correspondence between the triaxially loaded sample and the CT scan sample before and after transfer. The event registration unit matches the mechanical-permeable-acoustic emission data of the triaxial loading stage with the CT scan data based on the event number, sample number, freeze trigger timestamp, and CT scan timestamp. The microstructure parameters obtained by scanning are correlated; the real-time data acquisition system is used to acquire, cache and transmit the above multi-source data; the LSTM-based intelligent early warning unit is used to fuse multi-parameter time series data and output a three-level disaster early warning signal.

2. The multi-field coupled disaster threshold test device for water-rich tunnels according to claim 1, characterized in that, The axial servo loading unit (10) of the triaxial loading module includes a servo motor (101), a coupling (102), a ball screw (103) and a piston rod (104), which applies an axial load of 0-50kN through the piston rod (104); the confining pressure control cavity (13) is a double-layer structure, with the outer layer being 304 stainless steel and the inner layer being a polyurethane heat insulation layer.

3. The multi-field coupled disaster threshold test device for water-rich tunnels according to claim 1, characterized in that, The back pressure pump (141) of the dynamic osmotic pressure regulation system (14) is used to provide a stable water pressure of 0-2MPa; the dual closed-loop PID controller has an adjustment cycle of 1ms.

4. The multi-field coupled disaster threshold test device for water-rich tunnels according to claim 1, characterized in that, The transparent pressure chamber (21) of the high-resolution CT diagnostic module is made of polycarbonate material with a light transmittance of ≥92% and an inner diameter of Φ150mm × height of 300mm; the top is a hinged detachable top cover, which is moved into the microfocus CT scanner frame (25).

5. The multi-field coupled disaster threshold test device for water-rich tunnels according to claim 1, characterized in that, The microfocus CT scanner has a spatial resolution of 3μm, a scanning layer thickness of 20μm, a three-dimensional rotary stage (23) with a cross slide structure, a Z-axis travel of 100mm, a repeatability of ±2μm, and acquires projection data at a step angle of 0.5°.

6. The multi-field coupled disaster threshold test device for water-rich tunnels according to claim 1, characterized in that, The transparent seepage tank (31) has a porous medium support (34) on the upper and lower layers, a sample chamber (37) in the middle layer, and a water collection chamber at the bottom layer. It is made of acrylic material, and the tank wall is a double-layer hollow structure filled with silicone oil to eliminate optical distortion. The guide hole of the porous medium support (34) has an inclination angle of 30°, a diameter of Φ2mm, and a porosity of 35%±2%.

7. The multi-field coupled disaster threshold test device for water-rich tunnels according to claim 1, characterized in that, The acoustic emission acquisition unit includes a piezoelectric acoustic emission sensor, a water-resistant acoustic waveguide rod, a preamplifier, a bandpass filter, and an acoustic emission acquisition card. The piezoelectric acoustic emission sensor is installed on the outer wall of the confining pressure control cavity, the upper pressure head, the lower pressure head, or the waveguide rod at the end of the sample via an acoustic coupling agent. The acoustic emission signal is transmitted to the data fusion terminal after preamplification, filtering, and analog-to-digital conversion, and is used to obtain the number of acoustic emission events, ring count, amplitude, energy, duration, rise time, dominant frequency, RMS value, and b-value in real time.

8. A method for testing the disaster threshold of multi-field coupling in water-rich tunnels, characterized in that, The multi-field coupling disaster threshold test device for water-rich tunnels according to any one of claims 1-7 includes the following steps: S1 Sample preparation: The moisture content of the sample was controlled at 12%~30% by liquid nitrogen freeze drying, and the CO2 back pressure saturation method was used to gradually pressurize to 2MPa so that the sample B value is ≥0.95; S2 Multi-field Coupled Loading: Within the confining pressure range of 50~500kPa and hydraulic gradient range of 0.5~8.0, axial stress and seepage pressure are applied simultaneously, and the hydraulic gradient is increased stepwise in increments of 0.2, with each step maintained for 30 minutes until seepage stabilizes; S3 Microstructure Dynamic Observation: Porosity, coordination number, and throat diameter distribution parameters are obtained through CT scanning to reconstruct a three-dimensional pore network model; S4 Catastrophic Threshold Determination: Using a sudden change in the penetration coefficient Δk / Δt > 10⁻ 5 A sudden drop in s⁻¹, acoustic emission b-value <1.0, and a sudden increase in axial strain are used as joint criteria to identify the critical state of catastrophic events in the sample. When the triaxial loading module detects that any of the above indicators reaches a preset threshold, a synchronous trigger generates a unique event number and records the freezing trigger timestamp t0, while simultaneously controlling the liquid nitrogen quick-freezing unit to fix the sample state. Subsequently, the sample with the unique number is transferred to the high-resolution CT diagnostic module for scanning under low-temperature conditions, and the CT scan data is registered with the triaxial loading data corresponding to time t0 based on the sample number, event number, and scanning timestamp. The LSTM-based intelligent early warning unit integrates mechanical, seepage, acoustic emission, and CT microstructure parameters to output three levels of catastrophic early warning signals: green, yellow, and red.

9. The method for testing the disaster threshold of multi-field coupling in water-rich tunnels according to claim 8, characterized in that, In step S1, liquid nitrogen freeze drying was carried out at -196℃ for 48 hours under vacuum conditions ≤10Pa, with glycerol added as an ice crystal growth inhibitor, and the sample structure deformation was controlled to ≤3%; CO2 back pressure saturation was applied at a gradient of 0.1MPa per hour.

10. The method for testing the disaster threshold of multi-field coupling in water-rich tunnels according to claim 8, characterized in that, When the CT scan is triggered in step S3, the loading system freezes the current load, and the load fluctuation is ≤0.05% of the full scale. The sample is quickly frozen and fixed by spraying liquid nitrogen at -196℃ for 10s, and then transferred in a constant temperature environment of -50℃ to avoid freeze-thaw damage to the microstructure. In step S4, the LSTM intelligent early warning unit receives six-dimensional time-series data including axial strain, confining pressure, pore pressure, permeability coefficient, acoustic emission energy, and b-value, and outputs three levels of early warning: green, yellow, and red. When the hydraulic gradient is ≥3.8 and Δk / Δt>10⁻ 5 A sudden water inrush warning is triggered when the damage threshold is ≥0.35 and the b value is <1.

0. A large deformation warning is triggered when the damage threshold is ≥0.35 and the b value is <1.0.

Images