A frozen soil thermal hydraulic property testing system and method based on multi-sensing and data physical fusion

Abstract

The application discloses a frozen soil thermal hydraulic response test system and method based on multi-sensing and data physical fusion, which integrates ultra-weak fiber grating, active heating optical fiber, micro dielectric constant sensor and optical frequency domain reflection technology to construct a temperature, moisture, ice content and strain synchronous monitoring network for the freezing process; the method is to construct a multi-field fusion data set based on thermal disturbance lag correction, temperature-strain decoupling and moisture-strain residual item compensation mechanism, and extract key criterion features of ice lens growth rate, frost heaving force evolution and shear deformation change rate; in combination with a physical information neural network (PINN) embedded with a thermal hydraulic control equation, risk identification and grade determination of freezing abnormal behavior are realized. The application supports indoor heating and water pressure linkage regulation, realizes a complete closed loop of multi-source sensing-risk identification-response verification under model driving, and breaks through the bottleneck of low multi-physical field coupling identification accuracy, large parameter cross interference and insufficient real-time risk identification of existing monitoring technologies.

Description

A multi-sensor and data physics fusion testing system and method for testing the hydrodynamic properties of frozen soil Technical Field

[0001] This invention relates to the fields of geotechnical engineering and frozen soil monitoring technology, and in particular to a multi-sensor and data physics fusion system and method for testing the thermo-hydraulic-mechanical properties of frozen soil. Background Technology

[0002] Permafrost, a four-phase porous medium composed of solid minerals (soil particles), unfrozen water, ice crystals, and gaseous inclusions, exhibits significant nonlinearity and spatiotemporal heterogeneity in its thermo-hydraulic-mechanical multi-field coupled behavior. The latent heat exchange during the ice-water phase transition, the evolution of permeability characteristics induced by water migration, and the degradation of mechanical parameters caused by frost heave and thaw settlement constitute the main physical mechanisms controlling permafrost stability. These multi-field interactions directly affect the response process of foundation structures in cold-region engineering projects (such as roadbeds, slopes, and tunnels). Therefore, constructing a high-resolution indoor experimental system capable of analyzing the evolution characteristics of the entire frost heave process is the core foundation for conducting risk criterion extraction and engineering early warning model construction.

[0003] Currently, commonly used permafrost model testing methods mainly employ separate sensors such as thermocouples, TDR probes, and resistance strain gauges to measure temperature, moisture, and strain separately. However, these systems have three significant limitations in terms of monitoring dimensions: First, the spatial distribution of points is sparse and the sensing dimensions are discrete, making it impossible to provide information on temperature gradients, ice content distribution, or strain field evolution at a continuous profile scale; second, temperature changes can significantly interfere with the measurement results of strain and moisture content, such as the coupling of thermal expansion effects with frost heave stress and the dielectric response of thermal pulse disturbances, which can easily lead to the accumulation of parameter inversion errors; third, monitoring parameters are often processed independently, lacking the ability to fuse multi-field data and identify coupling behaviors, making it difficult to reveal non-steady-state evolution mechanisms such as ice lens growth and freezing front advancement.

[0004] To enhance monitoring capabilities, distributed fiber optic sensing technologies (such as Optical Frequency Domain Reflectometry (OFDR) and Ultra-weak Fiber Bragg Grid (uFBG) arrays) have been gradually introduced into permafrost experiments and in-situ monitoring in recent years. These technologies offer high spatial resolution, flexible deployment, and simultaneous temperature and strain sensing capabilities, enabling continuous, multi-parameter acquisition of permafrost processes to a certain extent. However, existing distributed monitoring systems still face three key technical bottlenecks: First, differences in hardware principles, sampling frequencies, and interface protocols among multi-source data result in spatiotemporal asynchrony of temperature, moisture, and strain data, making unified fusion analysis difficult. Second, there is significant cross-interference among multiple physical fields, especially in the phase transition temperature range (-5℃ to 0℃), where thermal strain and mechanical strain are severely mixed, leading to significant errors in traditional linear decoupling methods and inaccurate frost heave stress inversion. Third, there is a lack of a unified physical index system; existing systems are mostly limited to local coupling of "temperature-strain" or "temperature-moisture," failing to construct comprehensive response parameters reflecting freezing risk states and hindering system identification and early warning modeling of the frost heave evolution process.

[0005] At the data analysis level, current research exhibits a trend of a binary separation between "algorithm" and "physics." On the one hand, purely data-driven models based on algorithms such as neural networks, support vector machines, and random forests, while possessing the ability to automatically extract statistical regularities and adaptively model, lack physical explanation of ice-water phase transition mechanisms, freezing control factors, and boundary conditions. Furthermore, their predictive generalization ability is poor under non-training data conditions (such as sudden cooling or heterogeneous strata), making it difficult to use the results to construct engineering-usable risk thresholds. On the other hand, traditional partial differential equation models, such as the COMSOL multi-field coupled numerical model, while capable of describing heat diffusion, moisture migration, and stress evolution mechanisms, rely on fine meshes, complex initial and boundary conditions, and a large number of physical property parameters, resulting in high computational costs and difficulty in responding to high-frequency sensor data in real time, thus limiting their practicality in engineering applications. Therefore, current models either overemphasize statistics and lack physical consistency, or are highly physicalized and lack system responsiveness; a unified and integrated modeling framework is lacking between the two.

[0006] Furthermore, in the process of mapping indoor models to engineering applications, existing testing platforms generally lack an effective mechanism to transfer indoor calibration parameters to field conditions. Although indoor tests offer high precision and controllability, their boundary conditions and thermodynamic processes differ significantly from the actual foundation environment. This makes it difficult to directly apply the frost heave criteria established in the tests to the construction of engineering early warning models, thus limiting the system's versatility and promotional value.

[0007] In summary, to achieve accurate perception, coupled identification, and engineering prediction of multi-physics processes in permafrost, it is not only necessary to have the ability to acquire full profiles, high resolution, and multiple parameters simultaneously, but also to build a modeling mechanism that integrates physical consistency and system responsiveness, and to establish a three-in-one technical path of "indoor calibration - physical inversion - engineering mapping" to open up the key links of the entire process from perception to identification to prediction. Summary of the Invention

[0008] Objective: To address the problems of existing indoor monitoring and mechanism identification technologies for frozen soil, such as dispersed multiphysics measurement systems, poor parameter synchronization, difficulty in correcting thermal coupling interference, and lack of transferable risk criteria, this invention proposes a multi-sensor and data-physical fusion testing system and method for testing the thermo-hydraulic-mechanical properties of frozen soil. This system integrates multi-source sensing, decoupled coupling parameters, and multi-sensor data fusion. It is used for mechanism research and risk criterion extraction in frost heave deformation processes in cold-region engineering projects such as roadbeds, culverts, and slopes. The testing system integrates ultra-weak fiber gratings (uFBG), actively heated optical fibers (AHFO), miniature dielectric constant sensors, and optical frequency domain reflectance cables (OFDR) to form a multi-source collaborative monitoring network. It combines physical information neural networks (PINN) for frost heave mechanism analysis and freezing risk identification. By constructing an integrated path of "laboratory calibration—multiphysics modeling—risk index generation," it achieves full-profile identification and prediction of temperature, moisture migration, ice content evolution, and deformation mechanisms during the freezing process, providing high-resolution, physically consistent supporting data for stability assessment and early warning strategies in cold-region engineering projects.

[0009] Technical solution: The multi-sensor and data-physical fusion testing system for the thermo-hydraulic-mechanical properties of frozen soil of this invention includes a temperature field monitoring module, a moisture content and ice content monitoring module, a deformation monitoring module, and a data processing module; the moisture content and ice content monitoring module includes a uFBG array and a dielectric constant sensor; the deformation monitoring module adopts double-helix and parallel-arranged bending-resistant single-mode optical fibers; the data processing module includes a temperature compensation unit, a data-physical fusion analysis unit, and an early warning unit.

[0010] The early warning unit includes a physical equation library containing pre-stored equations for frozen soil heat conduction, water migration, and frost heave mechanics, as well as a weight adjustment module that balances the weights of measured data errors and physical equation residuals in PINN.

[0011] The multi-sensor and data physics fusion method for testing the hydrodynamic properties of frozen soil includes the following steps:

[0012] Step (1): Heating optical fiber is laid along the guide groove, strain sensing optical cable is laid along the vertical profile of the frozen soil model in a spiral trajectory, and strain sensing optical cable is laid horizontally and parallel in the ice lens distribution layer to form a three-dimensional monitoring network through vertical connection optical fiber.

[0013] Step (2): Under the stable state of permafrost, a set of short-duration thermal pulses with power Q are applied through the uFBG array and AHFO. The temperature response curves T(z,t) at each point along the thermal pulse propagation path are recorded. Based on the one-dimensional unsteady-state heat conduction model, the inversion equation is constructed:

[0014]

[0015] Where k(z) is the soil thermal conductivity at depth z to be inverted, ρ is the soil density, c is the soil specific heat, and T(z,t) is the temperature response during the propagation of the heat pulse.

[0016] The thermal conductivity of frozen soil is the combined thermal conductivity between unfrozen water, ice, and soil particles, and is modeled as follows:

[0017] k = k s (1-θ w -θ i )+k w θ+k i θ i

[0018] Where, k s k is the thermal conductivity of the solid particles. w Let k be the thermal conductivity of water. i θ is the thermal conductivity of ice; w The volumetric content of unfrozen water is measured by dielectric constant; θ i Let be the ice volume content, and be the quantity to be inverted; combine the measured values ​​with the above formula to solve for θ. i .

[0019] The latent heat of phase change released during freezing causes a delay in the temperature response, affecting the accuracy of thermal conductivity inversion. This invention introduces a "hysteresis factor" Δt in the temperature response before and after thermal pulse excitation. lag As a correction term, the final ice content is obtained.

[0020]

[0021] Where μ is the latent heat interference coefficient of phase transition, which is calibrated experimentally; t pulse This represents the duration of the thermal pulse.

[0022] The total water content θ of the frozen soil was obtained:

[0023] Establish a temperature-strain decoupling model:

[0024] By integrating OFDR strain data, uFBG temperature data, and AHFO moisture content data, and applying a temperature-strain decoupling algorithm, the influence of thermally induced strain and non-temperature strain noise is eliminated, yielding the strain value ε at each point within the soil. mechanical The temperature-strain decoupling algorithm process is as follows:

[0025] Total strain ε measured by a distributed fiber optic sensing system in permafrost fiber for:

[0026] ε fiber =ε mechanical +ε thermal +ε residual

[0027] Where, ε fiber ε is the total strain measured based on OFDR. mechanical ε represents the strain generated by the actual stress and deformation of the soil. thermal ε represents thermal expansion / contraction caused by temperature changes. residual Other non-mechanical error items include stress disturbances caused by the release of latent heat from freezing and fiber attachment errors.

[0028] During the freezing process, volume redistribution occurs due to the migration of unfrozen water, capillary action, and freezing suction. In fiber optic measurements, this manifests as "strain noise not caused by temperature," expressed as:

[0029]

[0030] in, The rate of water redistribution during the freezing stage corresponds to the residual stress during the "water absorption-freezing-expansion" process. γ represents the local moisture content gradient, corresponding to the capillary water migration trend; γ and δ are empirical fitting parameters, determined through calibration experiments.

[0031] Considering the nonlinearity of temperature-strain response in frozen soil, and incorporating the influence of thermal diffusion field based on AHFO-based moisture content and temperature monitoring results, the following improved model is proposed:

[0032]

[0033] Where α(θ,T) is the thermal expansion coefficient of the optical fiber, which changes dynamically with water content and temperature; β(θ,T) is the nonlinear strain correction term in the phase transition region, which is the temperature gradient influence correction coefficient, reflecting the strain disturbance caused by local thermal diffusion; at the same time, the temperature disturbance field is generated by active heating through AHFO, and the thermal conductivity of the frozen soil is inverted to dynamically update β; ΔT=T(z,t)-T0(z) represents the temperature change at a certain location; The local heat flux density variation is determined by temperature measurement inversion using uFBG.

[0034] Combining the above parts, the temperature-strain decoupling model is as follows:

[0035]

[0036] By fusing high-precision physical parameter data, a data-physics fusion analysis unit is used to identify and extract core indicators with engineering significance during the freezing process. These include frost heave strain and frost heave amount extracted through coordinate transformation and integration methods; ice lens growth criteria formed by the coupling of ice content and temperature gradient; potential rupture surfaces identified by the rate of change of shear strain; the freezing front advancement rate reflected by the second-order thermal diffusivity derivative; and frost heave force extracted based on an elastoplastic constitutive model. Finally, a standardized response vector and intensity index are constructed to achieve a unified quantitative expression of freezing activity under multi-parameter coupling, providing input for risk prediction models.

[0037] The frost heave deformation is obtained by integrating the deformation field data. The calculation method for frost heave deformation data is as follows: based on the geometric relationship of the spiral winding layout of the strain sensing optical cable, the strain ε obtained by measuring the optical fiber after decoupling temperature and strain is transformed by a coordinate transformation matrix. mechanical Decomposed into vertical strain ε z Radial strain ε r and shear strain γ zr :

[0038]

[0039] in, The helix angle of the strain-sensing optical cable. Shear strain indicates the uneven deformation caused by ice lenses in frozen soil.

[0040] The frost heave ΔH is expressed through the vertical strain ε. z The result was obtained by integrating along the frozen soil profile height H:

[0041]

[0042] Furthermore, the deformation rate is expressed as the overall deformation rate v based on the amount of frost heave. ΔH The calculation shows that:

[0043]

[0044] Where ΔH(t2) and ΔH(t1) are the frost heave amounts at times t2 and t2, respectively.

[0045] By combining moisture and temperature field data, an expression for the initiation condition of ice lenses is constructed. When the ice content growth rate is higher than the thermal gradient conduction rate, regions of rapid local ice crystal development are identified.

[0046]

[0047] Wherein, η is the freezing response factor (which needs to be determined through experimental fitting), and the region that meets this criterion is considered a high-risk location for rapid ice crystal accumulation and growth.

[0048] Analysis of shear deformation evolution:

[0049] Shear strain γ zr It reflects the asymmetric deformation induced by non-uniform frost heave or the development of ice lenses, and its rate of change over time is... Characterizing the potential "fracture surface" initiation of the structure, when γ appears in multiple consecutive time steps. zr High-frequency fluctuations or sudden changes (thresholds such as >0.001h) -1 If the condition is as described above, it is determined to be a local shear concentration phenomenon, and the risk assessment module needs to be entered.

[0050] Analysis of local thermal diffusion rate:

[0051] Temperature perturbations are used to obtain the thermal diffusion rate of the temperature field through thermal impulse response inversion, and the unsteady behavior of thermal diffusion is further analyzed:

[0052]

[0053] The significantly enhanced thermal diffusion rate indicates a faster advance of the freezing front and an evolution of thermal disturbance from a steady state to an unstable state, which is a precursor to a sudden change in the freezing state.

[0054] Inversion of frost heave force:

[0055] Based on soil strain data, strain is converted into stress components using a frozen soil elastoplastic constitutive model.

[0056] Linear elastic stage:

[0057]

[0058] Where E is the elastic modulus, v is Poisson's ratio, and ε φ For circumferential strain, ε φ =ε r / r, where r is the radial coordinate, σ z 弹性 σ r 弹性 σ φ 弹性These represent the vertical stress, radial stress, and circumferential stress of the soil in the linear elastic stage, respectively.

[0059] When the strain exceeds the elastic limit, the soil enters the plastic stage, at which point the elastic stress no longer represents the actual stress state. The Mohr-Coulomb criterion is introduced for plastic correction, adjusting the vertical stress to obtain the soil's vertical stress response.

[0060]

[0061] Where, Δε p The plastic strain increment is determined through iterative calculation.

[0062] Vertical stress σ z Integrating along the height H of the frozen soil profile yields the frost heave force F per unit area.

[0063]

[0064] This parameter, along with shear strain and the ice lens criterion, is used for anomaly identification as a direct indicator of "structural instability".

[0065] Constructing a multi-parameter comprehensive response index:

[0066] The above key parameters are standardized to construct a unified response vector freeze-thaw response index system. After normalizing each parameter to eliminate dimensional differences, the response vector is constructed as follows:

[0067]

[0068] Based on this, a freezing evolution intensity index S(z,t) is defined to intuitively reflect the freezing response intensity and risk trend of the current monitoring unit:

[0069]

[0070] Where, ω i The weighting coefficients for each physical quantity are obtained through field experiments, historical data fitting, or training using machine learning methods, or by using the equal weighting method as the initial scheme.

[0071] This feature vector serves as the input basis for subsequent prediction modules and is dynamically matched and adaptively learned with historical data or field measurement results during the neural network training process.

[0072] A Physics-Informed Neural Network (PINN) is used as the freezing anomaly prediction model. The state feature vector X(z,t) is used as input, and the residuals of three types of physical equations—heat diffusion control, moisture migration control, and strain evolution control—are embedded as constraint terms. The total loss function is expressed as follows:

[0073]

[0074] in, Used to fit training samples with historical observation data; It includes residual terms from the water migration control equation, the thermal diffusion control equation, and the decoupled mechanical strain evolution equation; λ1 is used to constrain the output continuity of the model in the time and space dimensions; λ2 is the physical consistency coefficient and λ1 is the regularization balance coefficient.

[0075] The network output is the corresponding risk index R(z,t)∈[0,1], which represents the degree to which the freezing behavior of the current spatiotemporal unit deviates from the normal coupling state.

[0076] Based on the PINN output results and combined with physical anomaly criteria, a risk level judgment table is constructed:

[0077] Establish a multi-level criterion based on the risk index R(z,t):

[0078]

[0079] Meanwhile, the testing system of this invention incorporates an anomaly identification index as an auxiliary condition for risk index determination. This anomaly index is the freeze evolution intensity index S(z,t). When it continuously rises or exceeds an empirical threshold, such as 0.7, it serves as one of the early warning trigger conditions and supports the linkage between risk level classification and response mechanism.

[0080] The linkage response includes: adjusting the sampling frequency, generating an alarm report, and triggering active control measures of the heating / pressure control device.

[0081] The training of the Physical Information Neural Network (PINN) used includes:

[0082] Input data: Temperature, moisture, and deformation time series data within a time window of 1 hour;

[0083] Physical constraints: By embedding the permafrost heat conduction equation and the water migration equation, the residuals of the equations are used as loss function terms;

[0084] Output results: Frost heave risk level, probability of location of freeze-thaw front.

[0085] Working principle: The test system of the present invention integrates ultra-weak fiber gratings, actively heated optical fibers, micro dielectric constant sensors, and optical frequency domain reflectometry technology to construct a synchronous monitoring network for temperature, moisture content, ice content, and strain during the freezing process. Based on the mechanisms of thermal disturbance lag correction, temperature-strain decoupling, and moisture-strain residual term compensation, a multi-field fusion data set is constructed, and key criterion features such as the growth rate of ice lenses, the evolution of frost heaving forces, and the change rate of shear deformation are extracted. Further combined with a physics-informed neural network (PINN) embedded with thermo-hydro-mechanical control equations, the risk identification and level assessment of freezing abnormal behaviors are realized. The system supports the联动调节 of indoor heating and water pressure, realizes a complete closed-loop of multi-source sensing-risk judgment-response verification driven by the model, and supports the联动调节 of indoor heating and water pressure, constructs a complete closed-loop process from perception, analysis to prediction, breaks through the bottlenecks of existing monitoring technologies such as low accuracy in multi-physical field coupling identification, large parameter cross-interference, and insufficient real-time risk identification, and has physical consistency, transferability, and engineering promotion value.

[0086] Through an intelligent prediction mechanism under data-driven and physical consistency constraints, the early identification and response control of potential anomalies during the freezing process of frozen soil are realized. In the indoor model test platform, the system integrates adjustable temperature control, water pressure, and sampling devices to simulate the联动响应 process of the engineering system in a high-risk state. For example, after identifying a high risk of freezing failure, the system automatically adjusts the top heating unit, water head supply device, or sampling frequency to verify the response mechanism driven by the multi-parameter risk index. The联动控制 mechanism of the present invention is currently applied in the test system to verify the risk regulation strategy in actual engineering.

[0087] Beneficial effects: Compared with the prior art, the present invention has the following advantages:

[0088] (1) The multi-source high-resolution monitoring ability of the present invention is strong: integrating uFBG, OFDR, AHFO, and dielectric probes to achieve full-profile synchronous sensing of temperature, moisture content, ice content, and strain during the freezing process.

[0089] (2) The coupling error can be compensated and the measurement accuracy is high: constructing a non-linear decoupling model of temperature-strain-water content to effectively correct thermal-induced disturbances and parameter cross errors.

[0090] (3) The ability to identify the freezing mechanism is strong: key indicators such as frost heave, shear strain, and ice lens development can be extracted to reveal the complex freezing evolution process.

[0091] (4) It has the function of risk prediction: introducing a PINN model to fuse thermo-hydro-mechanical control equations, outputting a freezing risk index, and realizing an intelligent early warning with strong physical consistency.

[0092] (5) The technical path is complete and scalable: a closed-loop system has been constructed from the perception of experimental parameters, identification of freezing mechanism, prediction of risk criteria to verification of engineering application. It has a complete data-model-application transformation process and supports the quantitative control and mechanism transfer of frost heave risk in cold region engineering. Attached Figure Description

[0093] Figure 1 is a schematic diagram of the structure of the multi-sensor and data physics fusion testing system for hydrodynamic properties of frozen soil according to the present invention.

[0094] Figure 2 is a cross-sectional view of the layout of the multi-sensor and data physics fusion testing system for hydrodynamic properties of frozen soil according to the present invention.

[0095] Figure 3 is a flowchart of the test method for hydrodynamic properties of frozen soil based on multi-sensor and data physics fusion of the present invention.

[0096] Figure 4 shows the soil temperature distribution during the freeze-thaw process according to an embodiment of the present invention;

[0097] Figure 5. Curves of unfrozen water content and ice content in soil at different depths during the freeze-thaw process according to an embodiment of the present invention;

[0098] Figure 6. Soil deformation distribution during the freeze-thaw process according to an embodiment of the present invention;

[0099] Figure 7 shows the soil displacement curve during the freeze-thaw process according to an embodiment of the present invention;

[0100] Figure 8 shows the result of the ice lens body generation position recognition in an embodiment of the present invention. Detailed Implementation

[0101] As shown in Figures 1 and 2, in this embodiment of the invention, the temperature field monitoring module of the multi-sensor and data physics fusion frozen soil hydrodynamic property testing system consists of a uFBG array 8 with internal heating function, containing 10 grating points with a grating spacing of 5 cm, a temperature measurement range of -30℃ to 30℃, and an accuracy of ±0.1℃. The uFBG array 8 with internal heating function and a miniature probe-type dielectric constant sensor 9 (2 mm in diameter) constitute the water content and ice content monitoring module. The uFBG array generates a temperature perturbation field through controllable thermal pulses, simultaneously measuring the temperature gradient distribution and heat conduction rate. The miniature probe-type dielectric constant sensor 9 measures at a frequency of 50 MHz with an accuracy of ±1%.

[0102] The deformation monitoring module uses a double-helix arrangement and a parallel, closely spaced arrangement of the first single-mode fiber 5 and the second single-mode fiber 7 (G.657.A2) to resist bending, with a spatial resolution of 1 mm and a strain range of ±5000 με.

[0103] The data processing module includes a temperature compensation unit that dynamically corrects deformation monitoring errors in real time based on OFDR self-sensing temperature and uFBG temperature data; a data-physical fusion analysis unit with built-in physical information neural network (PINN) that performs spatiotemporal matching of temperature, moisture, and deformation data and outputs multi-field coupling analysis results of frozen soil freeze-thaw state; and an early warning unit that triggers an early warning signal when the predicted value of frost heave exceeds the indoor calibration threshold or when the freeze-thaw front intrudes into the preset safety range.

[0104] The data-physical fusion analysis unit also includes:

[0105] Physical equation library: Pre-stored equations for heat conduction in frozen soil, water migration, and frost heave mechanics;

[0106] Adaptive weight adjustment module: dynamically balances the weights of the measured data error and the physical equation residual of PINN.

[0107] The multi-sensor and data-physical fusion method for testing the hydrodynamic properties of frozen soil in this embodiment includes the following steps:

[0108] Step 1. Frozen soil model preparation, with simultaneous installation and integration of the testing system. Specific sub-steps are as follows:

[0109] Model container design: Acrylic glass cylinder 6 (30cm in diameter, 60m in height). The model container is placed in a constant temperature chamber 10. The bottom of the cylinder 6 is supplied with constant pressure water head by a water head control device 3. The cold plate on the top of the model is connected to a high-precision temperature control box 1 to control the temperature of the top of the soil column.

[0110] 1.1 Soil filling: Unsaturated silty clay (moisture content 15%), compacted in layers to a density of 1.5 g / cm³. 3 Pre-embed guide grooves during the soil preparation stage to facilitate subsequent sensor installation, including:

[0111] 1) uFBG guide grooves: arranged along the axial direction with an axial spacing of 5cm;

[0112] 2) OFDR spiral guide groove: spiral angle 30°, pitch 10cm;

[0113] 3) Dielectric probe hole positions: spacing 5cm, the depth of the measuring point corresponds to the depth of the uFBG grating.

[0114] 1.2 Sensor Installation: Sensors are installed according to the design scheme and the characteristics of each sensor acquisition module, as follows:

[0115] 1) uFBG array: Fibers with internal heating function are laid along the guide groove, and the uFBG grid points are aligned with the guide groove markings; the fiber is connected to the heating control device 2.

[0116] 2) Strain sensing optical cable: The strain sensing optical cable adopts optical frequency domain reflection (OFDR) technology. The strain sensing optical cable is laid out in a bidirectional spiral strain decoupling design. The strain sensing optical cable is laid out along the vertical profile of the frozen soil model in a spiral trajectory with a spiral angle of 30°-60° and a pitch of 20cm±5cm, forming a strain field covering the entire profile of the frozen soil.

[0117] The bidirectional helical strain decoupling design, based on a coordinate transformation matrix, separates the vertical, radial, and shear strain components. Simultaneously, strain-sensing optical cables are horizontally deployed at 10cm intervals along key permafrost layers, such as ice lens distribution layers. Each layer is connected in series via vertically linked optical fibers, forming a three-dimensional monitoring network. The strain-sensing optical cables are covered with a flexible frost-heave-resistant protective layer to ensure the lossless transmission of frost heave deformation.

[0118] 3) Dielectric probe: The dielectric probe is inserted into the preset hole to ensure close contact with the soil.

[0119] 1.3 System Integration: All sensors are connected to the central data acquisition unit 11 to complete the integration of the multi-sensor collaborative acquisition module and to perform debugging; the data acquisition unit 11, the head control device 3, the heating control device 2, and the high-precision temperature control box 1 are all controlled by the central control computer 12.

[0120] Step 2. Perform calibration tests.

[0121] (2.1) Thermal pulse hysteresis response calibration test

[0122] Sample preparation: Select several groups of frozen soil with known total water content and keep them in a constant freezing state (e.g., -5℃) in a constant temperature chamber to ensure the coexistence of frozen and unfrozen water. Measure the unfrozen water content using a dielectric constant probe.

[0123] Apply heat pulse: Apply unit heat (1W power) for a duration through the built-in heating function of uFBG to induce local temperature disturbance.

[0124] Temperature response acquisition: A high-sensitivity temperature fiber optic grating is deployed near the heating point to record the temperature change curves before and after thermal excitation. The delay time for the temperature to reach a steady-state trend in the response curve is extracted, i.e., the time offset of the temperature change lags behind the thermal pulse input.

[0125] Data processing and coefficient inversion: The latent heat interference coefficient μ of phase change is derived using the following formula, and the average value of data from multiple sets of experiments is taken:

[0126]

[0127] in, This is the final ice content (obtained by subtracting the unfrozen water content from the total water content).

[0128] (2.2) Calibration test of residual strain induced by water migration in frozen soil

[0129] The soil column is frozen from top to bottom, and the temperature control system controls the cooling rate to -1℃ / h.

[0130] Moisture migration is slow in the 1–2 hours before freezing, which suggests that: At this stage, it is believed that: ε residual =0, that is, ε mechanical =ε fiber -ε thermal Based on the known temperature change (uFBG temperature measurement), the strain baseline of the reference section is obtained by substituting it into the temperature-strain decoupling model.

[0131] During the high migration period, 3–5 hours after freezing, θ was collected, from which θ was calculated. ε established using the reference segment mechanical and ε thermal ε residual =ε fiber -ε thermal -ε mechanical .

[0132] ε was collected at multiple time points and measurement points during the active freezing period. residual , γ and δ are obtained by applying the least squares method for inversion.

[0133] (2.3) Calibration test of ice lens germination response factor

[0134] The soil column is frozen from top to bottom, and the temperature control system controls the cooling rate to be about -1℃ / h. During the freezing process, the temperature and ice content are collected every 5 minutes.

[0135] Calculated based on ice content time series data Temperature gradient is calculated based on the difference calculated using uFBG temperature data.

[0136] In the later stages of the freezing experiment, the location of ice lens growth was identified using image processing. The freezing response factor corresponding to the location of the ice lens formation was then calculated.

[0137]

[0138] Multiple sets of experiments were conducted, and the median or lower safety limit was taken as the model threshold coefficient.

[0139] Step 3. Soil column freeze-thaw test, data acquisition: Start the high-precision temperature control box 1 and sequentially perform the freeze-thaw process (-20℃, 68.5 hours) - thawing (-20℃ → 20℃, rate 1℃ / h). The temperature control system sends a synchronization signal to align the timestamps of OFDR, uFBG, AHFO, and digital image acquisition data (error ≤1ms). The multi-sensor collaborative acquisition module begins synchronous data acquisition, and the acquired parameters include key parameters of multi-field hydrodynamic parameters of frozen soil, which are dynamically displayed in real time through the data visualization platform. The parameters include:

[0140] (1) Temperature: uFBG collects temperature distribution at a frequency of 1Hz.

[0141] (2) Deformation: The deformation monitoring module uses OFDR technology to monitor the internal strain value ε of the soil in real time. fiber (Sampling rate 1kHz).

[0142] (3) Moisture: Periodic thermal pulses (power 1W, duration 5s) are applied to the internally heated uFBG array 8. The AHFO thermal pulse is triggered every 30min, and the dielectric spectrum data measured by the micro probe dielectric constant sensor 9 is read synchronously.

[0143] Step 4. Obtain multi-field data of frozen soil hydrodynamics:

[0144] (1) Based on thermal pulse thermal conductivity inversion and ice content calibration, obtain moisture data (ice content). Unfrozen water content θ w Total moisture content (θ);

[0145] (2) Calculation of the true soil strain ε based on the temperature-strain decoupling model mechanical .

[0146] Step 5. Construct a freeze-thaw multi-parameter coupling recognition unit

[0147] Based on the multi-field data from step 4, frost heave strain, frost heave amount, and deformation rate are extracted through coordinate transformation and integration methods. The ice lens growth criterion is based on the coupling of ice content and temperature gradient. The shear strain change rate is obtained based on shear strain. The second-order thermal diffusion derivative is calculated based on temperature data. Frost heave force is extracted based on the elastoplastic constitutive model.

[0148] Based on the above parameters, a standardized response vector and intensity index are constructed to achieve a unified quantitative expression of freezing activity under multi-parameter coupling, providing input for risk prediction models.

[0149] Step 6. Intelligent risk identification and engineering early warning for permafrost

[0150] In this embodiment, a risk identification and engineering mapping module based on the key parameters of multi-field hydrodynamics acquired by the monitoring system in steps 1-6 is constructed. This module uses the co-evolutionary characteristics of temperature, moisture, and strain during the freezing process as input, embedding three types of control equations—thermal diffusion, moisture migration, and strain evolution—as physical consistency constraints, and outputs a risk index R(z,t) reflecting the degree to which the freezing process deviates from a steady state. The PINN model was trained and its parameters tuned by combining historical evolution data from monitoring points and risk label samples during the experiment, and freezing risk levels were output at different experimental stages.

[0151] Model results show that the risk index R(z,t) can effectively identify regions of abnormal freezing evolution, and its numerical distribution is in good consistency with the formation location of ice lenses and areas of high shear strain concentration. Combining the model output with actual observations, a three-level criterion table of "low-medium-high risk" was constructed for intelligent hierarchical identification of structural responses and control strategies during the freezing process. This risk identification and engineering mapping module verifies the feasibility and practicality of the data-physical fusion identification method in freezing behavior prediction and risk mapping.

[0152] In this embodiment, a series of characteristic parameters with clear physical meaning obtained by the present invention are applied, and the abnormal evolution of the frozen soil freezing process is analyzed in depth. Figures 3 to 8 show the strain of soil, water, and temperature-strain decoupling, respectively, and demonstrate the internal temperature transfer, water migration, ice-water phase transition, freeze-thaw front changes, tensile deformation development caused by ice lens formation, and frost heave and thaw settlement processes during the freeze-thaw process.

[0153] Further, based on the risk identification model constructed in step 6, the risk level of each spatiotemporal unit was dynamically output during the freezing process. The results showed that in the first 35 hours after freezing, the risk index of all monitoring points was below 0.4, indicating a stable system. From the 40th hour onwards, the risk index of some deep units gradually increased, reaching a peak around the 52nd hour, with the maximum risk index exceeding 0.85. This period also corresponds to the stage of sudden increase in freezing rate and accelerated mechanical strain, indicating that the model can accurately capture unsteady freezing behavior and achieve early identification.

[0154] The system further tracked and analyzed the risk level evolution of typical units, finding that high-risk units exhibited synchronous fluctuations in freezing parameters 1–2 hours before entering the abnormal stage, such as sudden changes in strain rate, a sudden increase in freezing expansion, and an increased increase in ice content. This result verifies that the method of the present invention has the ability to respond in advance and has high physical consistency in the identification of freezing anomalies.

[0155] In summary, this embodiment demonstrates that the present invention, by constructing a multi-sensor and data physics fusion-based permafrost hydrodynamic property testing system, integrates high-resolution synchronous sensing technologies for temperature, moisture, ice content, and three-dimensional deformation. Combined with a multi-physics fusion model based on temperature-strain decoupling and moisture disturbance compensation, it improves the extraction accuracy of key parameters in the freezing process. Furthermore, a physical information neural network embedded with the thermo-hydraulic-mechanical control equations is introduced to achieve intelligent identification and risk level assessment of ice lens formation, frost heave anomaly development, and shear concentration unsteady processes. The system's constructed "perception-identification-prediction-mapping" technical closed loop verifies its ability to express the linkage between freezing evolution and engineering response mechanisms in an indoor simulation platform. It possesses physical consistency, real-time identification capabilities, and engineering migration adaptability, providing a scalable technical path and theoretical support for the safety assessment and risk warning of infrastructure in cold regions.

Claims

1. A method for testing the hydrodynamic properties of frozen soil by integrating multiple sensors and data physics, characterized in that: The steps include: Step (1), laying heating optical fibers along the guide groove, laying strain sensing optical cables along the vertical profile of the frozen soil model in a spiral trajectory, laying strain sensing optical cables horizontally and parallel in the ice lens distribution layer, and then connecting them in series with vertical connecting optical fibers to form a three-dimensional monitoring network; Step (2), selecting frozen soil with known total water content and measuring the unfrozen water content; applying a set of thermal pulses through the uFBG array and AHFO to construct the inversion equation: ;in, For the depth to be inverted The thermal conductivity of the soil at that location, For soil density, For soil specific heat, The temperature response during the propagation of the heat pulse; determined by the hysteresis factor. The final ice content was obtained. ： ;in, The latent heat interference coefficient of phase change. The duration of the thermal pulse was used to determine the total water content of the frozen soil. ： The temperature-strain decoupling algorithm was used to obtain the strain values ​​at various points within the soil. ： The strain obtained by fiber optic measurement after decoupling temperature and strain through coordinate transformation matrix. Decomposed into vertical strain ε z Radial strain ε r and shear strain γ zr : ;in, The helix angle of the strain-sensing optical cable is measured by vertical strain. Along the height of the frozen soil profile Calculate frost heave by integration ： ; Calculate the deformation rate based on frost heave. ： ;in, for The amount of frost heave at any given moment; for Fissure at any given time; identification of areas with rapid local ice crystal development: ,in, The freezing response factor; the vertical stress The frost heave force F per unit area is obtained by integrating along the height of the frozen soil profile: Step (3), construct the response vector: Define the freezing evolution intensity index. As a condition for triggering an early warning: ;in, The weighting coefficient for the physical quantity.

2. The method for testing the hydrodynamic properties of frozen soil using multi-sensor and data physics fusion as described in claim 1, characterized in that: In step (2), the strain is converted into stress components, and the linear elastic stage is as follows: Where E is the elastic modulus, Poisson's ratio, For circumferential strain, r is the radial coordinate. 、 、 These represent the vertical stress, radial stress, and circumferential stress of the soil in the linear elastic stage, respectively.

3. The method for testing the hydrodynamic properties of frozen soil using multi-sensor and data physics fusion as described in claim 2, characterized in that: In step (2), when the strain exceeds the elastic limit, the Mohr-Coulomb criterion is introduced for plastic correction, based on the plastic strain increment. Obtain the final stress response: 。 4. The method for testing the hydrodynamic properties of frozen soil using multi-sensor and data physics fusion as described in claim 1, characterized in that: In step (2), the total strain measured by the distributed fiber optic sensing system in the permafrost... for: ;in, The total strain is based on OFDR measurements. The strain is the actual deformation of the soil under stress. Thermal expansion / contraction caused by temperature changes For other non-mechanical error terms.

5. The method for testing the hydrodynamic properties of frozen soil using multi-sensor and data physics fusion as described in claim 1, characterized in that: In step (2), ;in, The coefficient of thermal expansion of optical fiber. This is the strain nonlinearity correction term for the phase transition region. The total strain is measured based on OFDR.

6. The method for testing the hydrodynamic properties of frozen soil using multi-sensor and data physics fusion as described in claim 4, characterized in that: In step (2), ;in, The coefficient of thermal expansion of optical fiber; This is a nonlinear strain correction term for the phase transition region; , indicating temperature change; This represents a localized change in heat flux density.

7. The method for testing the hydrodynamic properties of frozen soil using multi-sensor and data physics fusion as described in claim 1, characterized in that: In step (3), a physical information neural network is used as the freezing anomaly prediction model, with state feature vectors as the basis. For the input, construct the total loss function: ;in, This is used to fit the training samples to historical observation data; It includes the control equations for moisture migration, thermal diffusion, and residual terms of the decoupled mechanical strain evolution equation; This is used to constrain the continuity of the model's output in both time and space dimensions; The physical consistency coefficient. This is the regular balance coefficient.

8. The method for testing the hydrodynamic properties of frozen soil by multi-sensor and data physics fusion according to claim 1, characterized in that; In step (3), a risk level judgment table is constructed based on the output of the physical information neural network PINN and the physical anomaly criteria.

9. A system employing the multi-sensor and data-physical fusion method for testing the hydrodynamic properties of frozen soil as described in claim 1, characterized in that: include: The system includes a temperature field monitoring module, a moisture content and ice content monitoring module, a deformation monitoring module, and a data processing module. The moisture content and ice content monitoring module includes a uFBG array (8) and a dielectric constant sensor (9). The deformation monitoring module uses double-helix and parallel-laid anti-bending single-mode optical fibers. The data processing module includes a temperature compensation unit, a data physical fusion analysis unit, and an early warning unit with a built-in physical information neural network.

10. The multi-sensor and data physics fusion testing system for the hydrodynamic properties of frozen soil according to claim 9, characterized in that: The early warning unit includes a physical equation library containing pre-stored equations for frozen soil heat conduction, water migration, and frost heave mechanics, as well as a weight adjustment module that balances the weights of measured data errors and physical equation residuals in PINN.

Images