Multi-modal data-driven in-situ intelligent monitoring method for soil microstructure

Abstract

This invention relates to the field of geotechnical engineering investigation and monitoring technology, and discloses a multimodal data-driven in-situ intelligent monitoring method for soil microstructure. The method involves deploying a multimodal sensor array to synchronously acquire multimodal raw data; extracting feature vectors for each modality after preprocessing and spatiotemporal alignment; fusing the feature vectors and inputting them into a pre-trained interpretation model to output a quantitative description of the microstructure state; simultaneously calculating a stability judgment value (WD) based on the feature data, comparing it with a preset threshold, and outputting three levels of early warning information; finally, visualizing the results. This invention achieves time-synchronous acquisition of each sensor unit through a multi-channel synchronous data acquisition system, ensuring precise alignment of each modal data in the time dimension. Using the field-of-view center of the microstructure imaging unit as a spatial reference point, the known spatial geometric relationship between each sensor unit and the reference point maps each modal data to the same spatial grid, achieving precise registration in the spatial dimension.

Description

Technical Field

[0001] This invention relates to the field of geotechnical engineering investigation and monitoring technology, and more specifically to a multimodal data-driven in-situ intelligent monitoring method for soil microstructure. Background Technology

[0002] The macroscopic engineering mechanical properties of soil are ultimately a comprehensive reflection of the adjustment and evolution of its internal microscopic structural elements. Revealing the intrinsic mechanism of soil engineering properties at the microscopic level has become an important direction for the in-depth development of soil mechanics and geotechnical engineering. Accurately obtaining dynamic evolution information of soil microstructure has important theoretical significance and engineering application value for engineering practices such as slope stability assessment, roadbed settlement prediction, foundation bearing capacity analysis, and geological disaster early warning. Currently, the monitoring and characterization of soil microstructure mainly rely on the following technical means: First, microscopic testing techniques based on indoor experiments, including scanning electron microscopy analysis, mercury intrusion porosimetry, gas adsorption, etc. These methods can obtain high-precision microstructure images and quantitative parameters, and require on-site sampling and transportation to the laboratory for testing and analysis. Second, in-situ testing techniques, including geophysical methods such as resistivity tomography, seismic wave tomography, and ground-penetrating radar detection, as well as fixed-point monitoring techniques such as fiber optic grating sensing and time-domain reflectometer. In existing technologies, the data acquisition times of different sensing devices are not synchronized, and their spatial positioning is independent, making it impossible to establish a precise correspondence between data from different modes. This severely restricts the quality of multi-source information fusion and the accuracy of subsequent interpretation. Current soil microstructure stability assessments mostly rely on simple threshold judgments based on single physical indicators, failing to fully consider the multi-source and complex nature of microstructure disturbances. A single indicator is insufficient to comprehensively depict the true state of the microstructure, lacking a comprehensive quantitative judgment value to reflect the overall stability of the microstructure. Currently, only static snapshots or intermittent state records of the microstructure at a certain moment can be provided, lacking the ability to perceive the trend of microstructure deterioration. This makes it difficult for existing technologies to capture precursor signals of microstructure abrupt changes, failing to provide valuable early warning time for engineering rescue and emergency response. Summary of the Invention

[0003] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a multimodal data-driven in-situ intelligent monitoring method for soil microstructure to solve the technical problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a multimodal data-driven in-situ intelligent monitoring method for soil microstructure, comprising the following steps: Step S1: Deploy a multimodal sensor array in the target soil monitoring area; Step S2: The output signals of each sensing unit in step S1 are synchronously acquired through a multi-channel synchronous data acquisition system to obtain multimodal raw monitoring data. Step S3: Preprocess and align the multimodal raw monitoring data collected in step S2 in a spatiotemporal manner. The preprocessing includes noise reduction. Step S4: Extract features from the preprocessed data of each modality to obtain the feature vector of each modality; Step S5: Input the modal feature vectors extracted in step S4 into the multimodal feature fusion module to generate a unified multimodal fusion feature vector; Step S6: Input the multimodal fusion feature vector obtained in step S5 into the pre-trained microstructure state interpretation model, and output a quantitative description of the soil microstructure state. Step S7: Calculate the stability judgment value WD based on the data extracted in step S4, and compare and issue a warning by comparing the stability judgment value WD with the first judgment threshold Y1 and the second judgment threshold Y2. Step S8: Output and visualize the analysis data from step S5 and the early warning information output from step S7.

[0005] In a preferred embodiment, in step S1, the multimodal sensing array includes a microstructure imaging unit, an acoustic sensing unit, an electromagnetic sensing unit, a mechanical sensing unit, and a temperature and humidity sensing unit. The multimodal sensing array in step S1 is an integrated probe-type structure, with each sensing unit arranged in layers along the probe axis, achieving multimodal synchronous monitoring across the entire depth range through a single penetration.

[0006] In a preferred embodiment, in step S1, the microstructure imaging unit includes a miniature zero-distance camera and a microscopic imaging probe for acquiring image data of the soil's microstructure; the acoustic sensing unit includes an ultrasonic transducer or acoustic emission sensor for acquiring acoustic response signals inside the soil; the electromagnetic sensing unit includes a resistivity probe or dielectric constant sensor for acquiring electromagnetic response signals of the soil; the mechanical sensing unit includes a fiber optic strain sensor and a miniature pressure sensor for acquiring stress-strain response signals of the soil; and the temperature and humidity sensing unit includes a temperature sensor and a moisture sensor for acquiring environmental physical parameters of the soil.

[0007] In a preferred embodiment, in step S3, spatiotemporal alignment includes time alignment and spatial registration. Spatial registration is performed using the following method: taking the center of the field of view of the microstructure imaging unit as the spatial reference point, and mapping each modal data to the same spatial grid through the known spatial geometric relationship between each sensing unit and the reference point.

[0008] In a preferred embodiment, in step S4, the features extracted from the image modal data include porosity, particle size distribution, particle orientation, particle roundness, crack density, and crack geometric parameters; the features extracted from the acoustic modal data include sound wave velocity, attenuation coefficient, spectral characteristics, and acoustic emission event rate; the features extracted from the electromagnetic modal data include resistivity, dielectric constant, and their spatial distribution characteristics; the features extracted from the mechanical modal data include strain, stress, modulus, and creep characteristics; and the features extracted from the temperature and humidity modal data include temperature gradient, moisture content, and its rate of change.

[0009] In a preferred embodiment, in step S5, the multimodal feature fusion module adopts one of the following strategies: feature concatenation fusion strategy, attention mechanism fusion strategy, and physical constraint fusion strategy. The feature concatenation fusion strategy concatenates the feature vectors of each modality to form a high-dimensional fusion feature vector. The attention mechanism fusion strategy adopts a cross-modal attention mechanism and dynamically allocates fusion weights based on the contribution of each modality feature to the description of the current microstructure state. The physical constraint fusion strategy establishes physical constraint relationships between each modality feature based on the soil physical mechanism and introduces physical consistency constraints during the fusion process.

[0010] In a preferred embodiment, in step S6, the microstructure state interpretation model adopts a regression and classification model based on a deep neural network, which is trained by a large amount of labeled multimodal microstructure label data. The microstructure state interpretation model outputs microstructure type identification results, quantitative inversion values ​​of microstructure parameters, and predicted values ​​of microstructure evolution trends. The microstructure state interpretation model adopts a temporal neural network architecture, including LSTM, Transformer, and temporal convolutional networks.

[0011] In a preferred embodiment, in step S7, when calculating the stability judgment value WD, the formula for calculating the qualitative judgment value WD is as follows: In the formula, RD is the microstructure disturbance index, TH is the cementation degradation index, SL is the seepage stability index, k1, k2 and k3 are all weighting coefficients, and k1+k2+k3=1. The weighting coefficients k1, k2 and k3 are the contribution ratios of the microstructure disturbance index RD, the cementation degradation index TH and the seepage stability index SL.

[0012] In a preferred embodiment, in step S7, within the stability judgment value WD, the formula for calculating the microstructure perturbation index RD is as follows: In the formula, KX is the current change in porosity, CK is the initial porosity, KS is the current change in particle orientation entropy, CS is the initial particle orientation entropy, FM is the current change in fracture density, CM is the initial fracture density, α1, α2, and α3 are the corresponding normalized weighting coefficients, and the formula for calculating the cementation degradation index TH is: In the formula, ZB represents the current change in P-wave velocity, CB represents the initial P-wave velocity, SS represents the current change in acoustic attenuation coefficient, CB represents the initial acoustic attenuation coefficient, β1 and β2 are the corresponding normalized weighting coefficients, and the formula for calculating the seepage stability index SL is as follows: In the formula, DB is the current change in resistivity, CD is the initial resistivity, HS is the current change in volumetric water content, CH is the initial volumetric water content, and γ1 and γ2 are the corresponding normalized weighting coefficients.

[0013] In a preferred embodiment, in step S7, the calculated stability judgment value WD is compared with the corresponding first judgment threshold Y1 and second judgment threshold Y2, and the first judgment threshold Y1 < the second judgment threshold Y2. When the stability judgment value WD ≤ the first judgment threshold Y1, the soil microstructure is determined to be in a stable state and no warning is issued. When the first judgment threshold Y1 < the stability judgment value WD ≤ the second judgment threshold Y2, the soil microstructure is determined to be in a basically stable state and a yellow warning signal is issued. When the second judgment threshold Y2 < the stability judgment value WD, the soil microstructure is determined to be in an unstable state and a red warning signal is issued.

[0014] The technical effects and advantages of this invention are as follows: This invention achieves time-synchronous acquisition of each sensing unit through a multi-channel synchronous data acquisition system, ensuring precise alignment of each modal data in the time dimension. Using the field of view center of the microstructure imaging unit as the spatial reference point, the known spatial geometric relationship between each sensing unit and the reference point is used to map each modal data to the same spatial grid, achieving precise registration in the spatial dimension and constructing a unified spatiotemporal reference framework for multimodal data, providing a high-quality data foundation for subsequent feature-level fusion and intelligent interpretation. This invention extracts the microstructure disturbance index RD, cementation degradation index TH, and seepage stability index SL from multimodal data, and introduces weighting coefficients k1, k2, and k3 for weighted fusion to construct a comprehensive stability judgment value WD, thereby realizing a single-value quantitative characterization of soil microstructure stability. This invention constructs a complete dynamic trend judgment system based on multiple stability judgment values ​​WD calculated on continuous time series. It compares the comprehensive stability judgment value WD with the preset first judgment threshold Y1 and second judgment threshold Y2 in a hierarchical manner to construct a three-level quantitative evaluation and hierarchical early warning system, providing engineers with clear and operable differentiated handling basis. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the intelligent monitoring method of the present invention. Detailed Implementation

[0016] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the forms of the various structures described in the following embodiments are merely illustrative. The multimodal data-driven in-situ intelligent monitoring method for soil microstructure involved in the present invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Reference Figure 1 This invention provides a multimodal data-driven in-situ intelligent monitoring method for soil microstructure, comprising the following steps: Step S1: Deploy a multimodal sensor array in the target soil monitoring area; Step S2: The output signals of each sensing unit in step S1 are synchronously acquired through a multi-channel synchronous data acquisition system to obtain multimodal raw monitoring data. Step S3: Preprocess and align the multimodal raw monitoring data collected in step S2 in a spatiotemporal manner. The preprocessing includes noise reduction. Step S4: Extract features from the preprocessed data of each modality to obtain the feature vector of each modality; Step S5: Input the modal feature vectors extracted in step S4 into the multimodal feature fusion module to generate a unified multimodal fusion feature vector; Step S6: Input the multimodal fusion feature vector obtained in step S5 into the pre-trained microstructure state interpretation model, and output a quantitative description of the soil microstructure state. Step S7: Calculate the stability judgment value WD based on the data extracted in step S4, and compare and issue a warning by comparing the stability judgment value WD with the first judgment threshold Y1 and the second judgment threshold Y2. Step S8: Output and visualize the analysis data from step S5 and the early warning information output from step S7.

[0018] In this embodiment, the present application realizes an integrated in-situ intelligent monitoring solution that covers the entire chain from multimodal data acquisition, spatiotemporal alignment, feature extraction, intelligent fusion and interpretation to quantitative hierarchical early warning and visualization, overcoming the shortcomings of traditional methods that separate data acquisition, indoor analysis and manual interpretation.

[0019] Reference Figure 1In step S1, the multimodal sensing array includes a microstructure imaging unit, an acoustic sensing unit, an electromagnetic sensing unit, a mechanical sensing unit, and a temperature and humidity sensing unit. The multimodal sensing array in step S1 is an integrated probe-type structure, with each sensing unit arranged in layers along the probe axis. Multimodal synchronous monitoring across the entire depth range is achieved through a single penetration. In step S1, the microstructure imaging unit includes a miniature zero-distance camera and a microscopic imaging probe for acquiring image data of the soil's microstructure. The acoustic sensing unit includes an ultrasonic transducer or acoustic emission sensor for acquiring acoustic response signals within the soil. The electromagnetic sensing unit includes a resistivity probe or dielectric constant sensor for acquiring electromagnetic response signals of the soil. The mechanical sensing unit includes a fiber optic strain sensor and a miniature pressure sensor for acquiring stress-strain response signals of the soil. The temperature and humidity sensing unit includes a temperature sensor and a moisture sensor for acquiring environmental physical parameters of the soil.

[0020] In this embodiment, by specifying the specific equipment selection for the microstructure imaging unit, acoustic sensing unit, electromagnetic sensing unit, mechanical sensing unit, and temperature and humidity sensing unit, this invention allows for flexible configuration of sensing combinations according to different monitoring needs and site conditions, taking into account monitoring accuracy, environmental adaptability, and engineering costs, thus expanding the scope of application of this invention. The integrated probe-type structural design allows each sensing unit to be arranged in layers along the probe axis, enabling multimodal synchronous monitoring across the entire depth range through a single static penetration. Compared to the traditional method of layer-by-layer burial or multi-stage testing, this shortens deployment time, reduces construction disturbance, and ensures the temporal synchronicity of monitoring data at each depth. The microstructure imaging unit directly contacts the soil through a transparent observation window, achieving in-situ, non-destructive, and continuous imaging of the soil microstructure, avoiding microstructure distortion caused by stress release, moisture content changes, and structural disturbances during sampling.

[0021] Reference Figure 1 In step S3, spatiotemporal alignment includes time alignment and spatial registration. Spatial registration is performed using the following method: taking the center of the field of view of the microstructure imaging unit as the spatial reference point, the modal data are mapped to the same spatial grid through the known spatial geometric relationship between each sensing unit and the reference point.

[0022] In this embodiment, the application uses the center of the field of view of the microstructure imaging unit as the spatial reference point and performs spatial mapping using the known fixed geometric relationship between each sensing unit and the reference point. This is a low-cost, high-precision spatial registration scheme that does not require external positioning equipment and is suitable for underground or deep soil environments. Since the relative positions of each sensing unit in the probe structure are precisely calibrated during manufacturing, this spatial registration method can be stably reproduced across the entire depth range, ensuring the consistency of multimodal data on the three-dimensional spatial grid and laying a foundation for accuracy in the joint analysis of spatially corresponding features in subsequent feature fusion.

[0023] Reference Figure 1 In step S4, the features extracted from the image modal data include porosity, particle size distribution, particle orientation, particle roundness, crack density, and crack geometric parameters; the features extracted from the acoustic modal data include sound wave velocity, attenuation coefficient, spectral characteristics, and acoustic emission event rate; the features extracted from the electromagnetic modal data include resistivity, dielectric constant, and their spatial distribution characteristics; the features extracted from the mechanical modal data include strain, stress, modulus, and creep characteristics; and the features extracted from the temperature and humidity modal data include temperature gradient, moisture content, and its rate of change.

[0024] In this embodiment, the application extracts quantitative feature parameters with clear physical meaning and mutual complementarity for five modes: image, acoustic, electromagnetic, mechanical, and temperature and humidity, respectively, to construct a complete multi-physics field feature system for soil microstructure. Image features reflect the evolution of geometric shape, acoustic features reflect the changes in the mechanical properties of the medium, electromagnetic features reflect the state of the interface between pore fluid and solid phase, mechanical features reflect the macroscopic stress response, and temperature and humidity features reflect environmental driving factors. The five types of features synergistically depict the complete physical picture of microstructure evolution from different dimensions. Each feature extracted in this application has a clear physical meaning and engineering relevance, and can be directly used for the subsequent calculation of the stability judgment value WD.

[0025] Reference Figure 1 In step S5, the multimodal feature fusion module adopts one of the following strategies: feature concatenation fusion strategy, attention mechanism fusion strategy, and physical constraint fusion strategy. The feature concatenation fusion strategy concatenates the feature vectors of each modality to form a high-dimensional fusion feature vector. The attention mechanism fusion strategy adopts a cross-modal attention mechanism and dynamically allocates fusion weights based on the contribution of each modality feature to the description of the current microstructure state. The physical constraint fusion strategy establishes physical constraint relationships between each modality feature based on the soil physical mechanism and introduces physical consistency constraints during the fusion process.

[0026] In this application embodiment, three optional fusion strategies are provided: feature cascade fusion, attention mechanism fusion, and physical constraint fusion. This application can flexibly configure the most suitable fusion scheme according to the data characteristics and quality of specific application scenarios, taking into account both the versatility of the method and the adaptability of the scenario. Feature cascade fusion is simple to implement and computationally efficient, and is suitable for field monitoring scenarios with high real-time requirements. Attention mechanism fusion can automatically identify the information importance of different modal features at different times, and is suitable for complex environments where the data quality of each modality fluctuates greatly. Physical constraint fusion introduces basic physical laws of soil, such as mass conservation and energy conservation, as regularization constraints, which effectively suppresses overfitting and physical inconsistencies of the data-driven model under sparse sample conditions.

[0027] Reference Figure 1 In step S6, the microstructure state interpretation model adopts a regression and classification model based on a deep neural network, which is trained by a large amount of labeled multimodal microstructure label data. The microstructure state interpretation model outputs microstructure type identification results, quantitative inversion values ​​of microstructure parameters, and predicted values ​​of microstructure evolution trends. The microstructure state interpretation model adopts a temporal neural network architecture, including LSTM, Transformer, and temporal convolutional networks.

[0028] In this embodiment, a regression and classification model constructed using a deep neural network is trained on a large amount of labeled multimodal-microstructure label data. This achieves a high-precision end-to-end mapping from multimodal fusion features to semantic descriptions of microstructure states. Compared to traditional interpretation methods based on human experience, the interpretation speed is significantly improved, and the consistency and objectivity of the interpretation results are greatly enhanced. The interpretation model simultaneously outputs classification results, regression values, and predicted values, realizing a multi-level output of qualitative description of the current state, quantitative representation of key parameters, and prediction of future trends. This meets the needs of engineering monitoring for comprehensive and hierarchical information. LSTM, Transformer, and temporal convolutional networks are all mature temporal modeling architectures. Their selection can be flexibly determined according to the time span of the monitoring data, sampling frequency, and computing resources. LSTM is suitable for medium-length sequences, Transformer is suitable for long sequence dependency modeling, and temporal convolutional networks are suitable for high sampling rate real-time processing. The optional configuration of the three further enhances the deployment flexibility and computational efficiency adaptability of this application in different engineering scenarios.

[0029] Reference Figure 1 In step S7, when calculating the stability judgment value WD, the formula for calculating the qualitative judgment value WD is as follows: In the formula, RD is the microstructure disturbance index, TH is the cementation degradation index, SL is the seepage stability index, and k1, k2, and k3 are all weighting coefficients, with k1+k2+k3=1. The weighting coefficients k1, k2, and k3 represent the contribution ratios of the microstructure disturbance index RD, the cementation degradation index TH, and the seepage stability index SL. In step S7, within the stability judgment value WD, the formula for calculating the microstructure disturbance index RD is as follows: In the formula, KX is the current change in porosity, CK is the initial porosity, KS is the current change in particle orientation entropy, CS is the initial particle orientation entropy, FM is the current change in fracture density, CM is the initial fracture density, α1, α2, and α3 are the corresponding normalized weighting coefficients, and the formula for calculating the cementation degradation index TH is: In the formula, ZB represents the current change in P-wave velocity, CB represents the initial P-wave velocity, SS represents the current change in acoustic attenuation coefficient, CB represents the initial acoustic attenuation coefficient, β1 and β2 are the corresponding normalized weighting coefficients, and the formula for calculating the seepage stability index SL is as follows: In the formula, DB is the current change in resistivity, CD is the initial resistivity, HS is the current change in volumetric water content, CH is the initial volumetric water content, and γ1 and γ2 are the corresponding normalized weighting coefficients.

[0030] In this embodiment, the soil microstructure stability is quantified into a comprehensive judgment value WD. Through weighted fusion of three intermediate indicators—microstructure disturbance index RD, cementation degradation index TH, and seepage stability index SL—complex multimodal and multidimensional microstructure information is mapped into a scalar indicator with clear physical meaning and ease of engineering use. This solves the engineering technical problem of redundant multi-indicator information that is difficult to directly evaluate. By introducing weighting coefficients k1, k2, and k3, differentiated consideration of various influencing factors is achieved for different soil types and monitoring scenarios. The contribution ratios of RD, TH, and SL can be flexibly adjusted according to actual engineering conditions, enabling the evaluation model to have scenario adaptability. The R-index characterizes the overall disturbance degree of the soil skeleton in terms of geometry and spatial arrangement. Compared with single index discrimination, it can distinguish different disturbance modes such as "pore compaction-dominated", "particle rearrangement-dominated" and "crack propagation-dominated". The cementation degradation index TH reflects the overall stiffness of the soil and the energy dissipation inside the medium, realizing dual acoustic verification of the degradation of the cementation state between soil particles and reducing the risk of misjudgment by a single acoustic index. The seepage stability index SL reflects the pore fluid connectivity, pore liquid ion concentration sensitivity and water content changes, characterizing the development degree of seepage channels and hydraulic stability inside the soil, and providing a quantitative basis for the assessment of microstructural instability under hydrological conditions such as rainfall infiltration and water level fluctuations.

[0031] Reference Figure 1In step S7, the calculated stability judgment value WD is compared with the corresponding first judgment threshold Y1 and second judgment threshold Y2, and the first judgment threshold Y1 < the second judgment threshold Y2. When the stability judgment value WD ≤ the first judgment threshold Y1, the soil microstructure is determined to be in a stable state and no warning is issued. When the first judgment threshold Y1 < the stability judgment value WD ≤ the second judgment threshold Y2, the soil microstructure is determined to be in a basically stable state and a yellow warning signal is issued. When the second judgment threshold Y2 < the stability judgment value WD, the soil microstructure is determined to be in an unstable state and a red warning signal is issued.

[0032] In this embodiment, a three-level quantitative grading early warning system is established, which organically connects the qualitative grading requirements for soil stability in engineering practice with the quantitative data of multimodal monitoring. This provides standardized risk assessment and early warning schemes for engineering projects such as slopes, roadbeds, and foundations. Differentiated engineering response instructions are given according to the early warning level: no early warning corresponds to routine monitoring, yellow corresponds to intensive observation and key attention, and red corresponds to immediate work stoppage and emergency response. This ensures that the early warning results have clear operational guidance significance and improves on-site decision-making efficiency.

[0033] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. The units and algorithm steps of the various examples described in the embodiments can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0034] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0035] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0036] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multimodal data-driven in-situ intelligent monitoring method for soil microstructure, characterized in that: Includes the following steps: Step S1: Deploy a multimodal sensor array in the target soil monitoring area; Step S2: The output signals of each sensing unit in step S1 are synchronously acquired through a multi-channel synchronous data acquisition system to obtain multimodal raw monitoring data. Step S3: Preprocess and align the multimodal raw monitoring data collected in step S2 in a spatiotemporal manner. The preprocessing includes noise reduction. Step S4: Extract features from the preprocessed data of each modality to obtain the feature vector of each modality; Step S5: Input the modal feature vectors extracted in step S4 into the multimodal feature fusion module to generate a unified multimodal fusion feature vector; Step S6: Input the multimodal fusion feature vector obtained in step S5 into the pre-trained microstructure state interpretation model, and output a quantitative description of the soil microstructure state. Step S7: Calculate the stability judgment value WD based on the data extracted in step S4, and compare and issue a warning by comparing the stability judgment value WD with the first judgment threshold Y1 and the second judgment threshold Y2. Step S8: Output and visualize the analysis data from step S5 and the early warning information output from step S7.

2. The multimodal data-driven in-situ intelligent monitoring method for soil microstructure according to claim 1, characterized in that: In step S1, the multimodal sensing array includes a microstructure imaging unit, an acoustic sensing unit, an electromagnetic sensing unit, a mechanical sensing unit, and a temperature and humidity sensing unit. The multimodal sensing array in step S1 is an integrated probe-type structure, with each sensing unit arranged in layers along the probe axis. Multimodal synchronous monitoring across the entire depth range is achieved through a single penetration.

3. The multimodal data-driven in-situ intelligent monitoring method for soil microstructure according to claim 2, characterized in that: In step S1, the microstructure imaging unit includes a miniature zero-distance camera and a microscopic imaging probe for acquiring image data of the soil's microstructure; the acoustic sensing unit includes an ultrasonic transducer or acoustic emission sensor for acquiring acoustic response signals inside the soil; the electromagnetic sensing unit includes a resistivity probe or dielectric constant sensor for acquiring electromagnetic response signals of the soil; the mechanical sensing unit includes a fiber optic strain sensor and a miniature pressure sensor for acquiring stress-strain response signals of the soil; and the temperature and humidity sensing unit includes a temperature sensor and a moisture sensor for acquiring environmental physical parameters of the soil.

4. The multimodal data-driven in-situ intelligent monitoring method for soil microstructure according to claim 1, characterized in that: In step S3, spatiotemporal alignment includes time alignment and spatial registration. Spatial registration is performed using the following method: taking the center of the field of view of the microstructure imaging unit as the spatial reference point, and mapping each modal data to the same spatial grid through the known spatial geometric relationship between each sensing unit and the reference point.

5. The multimodal data-driven in-situ intelligent monitoring method for soil microstructure according to claim 1, characterized in that: In step S4, the features extracted from the image modal data include porosity, particle size distribution, particle orientation, particle roundness, crack density, and crack geometric parameters; the features extracted from the acoustic modal data include sound wave velocity, attenuation coefficient, spectral characteristics, and acoustic emission event rate; the features extracted from the electromagnetic modal data include resistivity, dielectric constant, and their spatial distribution characteristics; the features extracted from the mechanical modal data include strain, stress, modulus, and creep characteristics; and the features extracted from the temperature and humidity modal data include temperature gradient, moisture content, and its rate of change.

6. The multimodal data-driven in-situ intelligent monitoring method for soil microstructure according to claim 1, characterized in that: In step S5, the multimodal feature fusion module adopts one of the following strategies: feature concatenation fusion strategy, attention mechanism fusion strategy, and physical constraint fusion strategy. The feature concatenation fusion strategy concatenates the feature vectors of each modality to form a high-dimensional fusion feature vector. The attention mechanism fusion strategy adopts a cross-modal attention mechanism and dynamically allocates fusion weights based on the contribution of each modality feature to the description of the current microstructure state. The physical constraint fusion strategy establishes physical constraint relationships between each modality feature based on the soil physical mechanism and introduces physical consistency constraints during the fusion process.

7. The multimodal data-driven in-situ intelligent monitoring method for soil microstructure according to claim 5, characterized in that: In step S6, the microstructure state interpretation model adopts a regression and classification model based on a deep neural network, which is trained by a large amount of labeled multimodal microstructure label data. The microstructure state interpretation model outputs microstructure type identification results, quantitative inversion values ​​of microstructure parameters, and predicted values ​​of microstructure evolution trends. The microstructure state interpretation model adopts a temporal neural network architecture, including LSTM, Transformer, and temporal convolutional networks.

8. The multimodal data-driven in-situ intelligent monitoring method for soil microstructure according to claim 1, characterized in that: In step S7, when calculating the stability judgment value WD, the formula for calculating the qualitative judgment value WD is as follows: In the formula, RD is the microstructure disturbance index, TH is the cementation degradation index, SL is the seepage stability index, k1, k2 and k3 are all weighting coefficients, and k1+k2+k3=1. The weighting coefficients k1, k2 and k3 are the contribution ratios of the microstructure disturbance index RD, the cementation degradation index TH and the seepage stability index SL.

9. The multimodal data-driven in-situ intelligent monitoring method for soil microstructure according to claim 8, characterized in that: In step S7, within the stability judgment value WD, the formula for calculating the microstructure disturbance index RD is as follows: In the formula, KX is the current change in porosity, CK is the initial porosity, KS is the current change in particle orientation entropy, CS is the initial particle orientation entropy, FM is the current change in fracture density, CM is the initial fracture density, α1, α2, and α3 are the corresponding normalized weighting coefficients, and the formula for calculating the cementation degradation index TH is: In the formula, ZB represents the current change in P-wave velocity, CB represents the initial P-wave velocity, SS represents the current change in acoustic attenuation coefficient, CB represents the initial acoustic attenuation coefficient, β1 and β2 are the corresponding normalized weighting coefficients, and the formula for calculating the seepage stability index SL is as follows: In the formula, DB is the current change in resistivity, CD is the initial resistivity, HS is the current change in volumetric water content, CH is the initial volumetric water content, and γ1 and γ2 are the corresponding normalized weighting coefficients.

10. The multimodal data-driven in-situ intelligent monitoring method for soil microstructure according to claim 9, characterized in that: In step S7, the calculated stability judgment value WD is compared with the corresponding first judgment threshold Y1 and second judgment threshold Y2, and the first judgment threshold Y1 < the second judgment threshold Y2. When the stability judgment value WD ≤ the first judgment threshold Y1, the soil microstructure is determined to be in a stable state and no warning is issued. When the first judgment threshold Y1 < the stability judgment value WD ≤ the second judgment threshold Y2, the soil microstructure is determined to be in a basically stable state and a yellow warning signal is issued. When the second judgment threshold Y2 < the stability judgment value WD, the soil microstructure is determined to be in an unstable state and a red warning signal is issued.

Images