A method and system for co-evolution analysis of physical effects of pile-supported reinforced embankments

Abstract

This invention relates to the field of intelligent testing technology in geotechnical engineering, and in particular to a method and system for the co-evolution analysis of physical effects in pile-supported reinforced embankments. The method includes the following steps: fitting a pile-supported reinforced embankment using a model box, soil samples, and reinforcing materials to construct test conditions; acquiring multi-source test datasets through multi-modal data acquisition using a sensor array; performing data preprocessing and feature extraction to obtain soil arching effect feature sets and tensile membrane effect feature sets; constructing a machine learning model and obtaining a machine learning training model using model training data; acquiring the physical effect feature set to be analyzed, and combining it with the machine learning training model to obtain the co-evolution analysis results of physical effects. This invention enables controllable simulation of the degradation of soil arching effect and the enhancement of tensile membrane effect, and provides a quantitative evaluation of the structural stress and deformation response under different reinforcement stiffness conditions, offering experimental and predictive basis for the rational selection and optimized design of reinforcement materials.

Description

Technical Field

[0001] This invention relates to the field of intelligent testing technology in geotechnical engineering, and in particular to a method and system for analyzing the synergistic evolution of physical effects of pile-supported reinforced embankments. Background Technology

[0002] Pile-supported reinforced embankments are widely used in soft soil foundations, high embankment subgrades, and uneven settlement control projects. Their bearing mechanism primarily relies on the synergistic effect between the piles, pile caps, the soil between the piles, and the reinforcing materials. Specifically, the soil arching effect formed by the embankment soil under differential settlement between piles, and the tensile membrane effect generated by the reinforcing materials under deformation constraints, jointly participate in the transfer of the superstructure load to the piles, and are key factors determining the structural safety and economy. In actual engineering, due to factors such as the filling process, operational loads, and changes in groundwater levels, the soil between piles often undergoes significant differential settlement and seepage softening processes, leading to a marked time-varying or even degrading soil arching effect. During this process, the reinforcing materials may gradually transform from auxiliary load-bearing components into primary load-bearing components, with a significant increase in their stress level and deformation state, and may even enter an unfavorable stress or instability stage.

[0003] In existing design and experimental research methods for pile-supported reinforced embankments, the selection of stiffness parameters for reinforcing materials is mostly based on empirical formulas, recommended values ​​in specifications, or static assumptions. These methods typically assume that the soil arching effect is stable during service, making it difficult to fully consider the synergistic evolution and mutual transformation of the soil arching effect and the tensile membrane effect under complex conditions such as differential deformation and seepage. Therefore, the following problems commonly exist in engineering practice: on the one hand, selecting excessively high stiffness in the reinforcing materials leads to low material utilization efficiency and increased project costs; on the other hand, selecting excessively low stiffness in the reinforcing materials can easily result in rapid stress increase, strain concentration, and even structural safety risks under unfavorable conditions. Furthermore, existing model test devices mostly employ single settlement or single load loading methods, making it difficult to construct realistic conditions involving multi-zone differential settlement, seepage, and the coupling effect of external loads. Simultaneously, traditional experimental methods for evaluating the soil arching effect and the tensile membrane effect rely heavily on indirect inference, lacking systematic means for simultaneous observation, quantitative analysis, and predictable evolution trends, making it difficult to provide a reliable basis for the rational selection and optimized design of the stiffness of reinforcing materials.

[0004] Therefore, there is an urgent need for an experimental device and method that can synergistically observe, quantitatively characterize, and intelligently predict the soil arching effect and tensile membrane effect in pile-supported reinforced embankments under controlled differential deformation and seepage conditions, in order to solve the technical problem of the lack of specificity and verifiability in the selection of stiffness of reinforcement materials in existing projects. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for analyzing the synergistic evolution of physical effects in pile-supported reinforced embankments.

[0006] To achieve the above objectives, in a first aspect, this invention provides a method for the co-evolution analysis of physical effects in pile-supported reinforced embankments. The method includes the following steps: fitting a pile-supported reinforced embankment using a model box, soil samples, and reinforcing materials, and constructing test conditions based on multi-factor coupling effects; based on the test conditions, using a sensor array to collect multi-modal data from the pile-supported reinforced embankment to obtain a multi-source test dataset; performing data preprocessing and feature extraction on the multi-source test dataset to obtain a soil arching effect feature set and a tensile membrane effect feature set; constructing a machine learning model based on the soil arching effect feature set and the tensile membrane effect feature set, and obtaining a machine learning training model using model training data; obtaining the physical effect feature set to be analyzed, and combining it with the machine learning training model to obtain the physical effect co-evolution analysis results. This invention provides scientific data support for engineering design, construction, and operation and maintenance, helps improve the structural stability and bearing capacity of pile-supported reinforced embankments, reduces the risk of defects, and extends the service life of the project, thus having significant engineering application value.

[0007] Optionally, the process of fitting a pile-supported reinforced embankment using a model box, soil samples, and reinforcing materials, and constructing test conditions based on multi-factor coupling effects, includes: layering the soil samples inside the model box and simultaneously placing the reinforcing materials at preset heights to fit the pile-supported reinforced embankment; using a multi-zone differential deformation induction module to form differential settlement conditions; obtaining external load conditions through an external load application device; simulating seepage conditions based on a seepage loading module; and combining the differential settlement conditions, external load conditions, and seepage conditions as multi-factor coupling effects to construct the test conditions. This invention solves the problem of the one-sidedness of single-condition test data, providing real and comprehensive basic data for subsequent data acquisition and analysis, and ensuring the reliability and accuracy of subsequent analysis results.

[0008] Optionally, the differential settlement conditions include: stepped settlement patterns, linear gradient settlement patterns, settlement basin-type settlement patterns, and cyclic settlement patterns. This invention breaks through the limitations of traditional single-settlement-pattern analysis, accurately capturing the evolution characteristics of physical effects under different settlement patterns, providing a basis for optimizing embankment structural design, and improving the settlement resistance of pile-supported reinforced embankments.

[0009] Optionally, the step of acquiring multi-modal data of the pile-supported reinforced embankment using a sensor array based on the test conditions to obtain a multi-source test dataset includes: using earth pressure sensors, displacement sensors, and reinforcement strain sensors as the sensor array to form a multi-modal sensing and monitoring module; acquiring synchronous monitoring data of the pile-supported reinforced embankment, including earth pressure data, displacement data, and reinforcement strain data, through the multi-modal sensing and monitoring module based on the test conditions; and collecting the synchronous monitoring data according to a unified time base and adding time stamps to form the multi-source test dataset. This invention solves the problems of scattered data acquisition, disordered timing, and single dimension in traditional methods, laying a high-quality data foundation for subsequent data processing and feature extraction, and improving analysis efficiency and accuracy.

[0010] Optionally, the step of preprocessing and extracting features from the multi-source experimental dataset to obtain the soil arching effect feature set and the tensile membrane effect feature set includes: preprocessing the multi-source experimental dataset to obtain multi-source experimental processed data, wherein the data preprocessing includes time alignment, noise reduction, and normalization operations; extracting feature parameters based on the multi-source experimental processed data and obtaining the spatiotemporal variation patterns of the feature parameters; and combining and organizing the feature parameters according to the spatiotemporal variation patterns to construct the soil arching effect feature set and the tensile membrane effect feature set. This invention effectively avoids analytical biases caused by interference from the original data, ensures the scientific validity of the soil arching effect and tensile membrane effect feature sets, provides high-quality input for machine learning model construction, and improves the reliability of subsequent analysis.

[0011] Optionally, the step of extracting feature parameters based on the multi-source test data includes: quantifying the tensile membrane effect using the multi-source test data to obtain the grid tension and the vertical component of the tensile membrane; obtaining the pile cap pressure value, the inter-pile pressure value, and the average pressure value based on the multi-source test data; quantifying the soil arching effect to obtain the soil arching effect strength index; combining the soil arching effect to obtain the contributing factors of the tensile membrane effect of the grid; and using the grid tension, the vertical component of the tensile membrane, the soil arching effect strength index, and the contributing factors as the feature parameters. This invention overcomes the limitations of traditional qualitative analysis, providing a more accurate and intuitive characterization of physical effects, facilitating the study of the co-evolution mechanism of physical effects, and providing core support for reliable analytical results output by the model.

[0012] Optionally, the step of constructing a machine learning model based on the soil arching effect feature set and the tensile membrane effect feature set, and obtaining a machine learning training model through model training data, includes: using the soil arching effect feature set and the tensile membrane effect feature set as model input, and using the feature index of the co-evolution of physical effects as model output to establish the machine learning model; acquiring historical experimental data or multiple sets of experimental working condition data as model training data, and training the machine learning model to obtain the machine learning training model. This invention avoids the drawbacks of traditional analysis methods being time-consuming and highly subjective, improves analysis efficiency and prediction accuracy, provides a reference for engineering decision-making, and adapts to the needs of rapid engineering response.

[0013] Optionally, the machine learning model includes a temporal neural network model and a physical constraint learning model. This invention addresses the problem that a single model cannot simultaneously consider dynamism and rationality, enabling model predictions to both conform to actual data patterns and the essence of engineering physics, thus significantly improving the reliability and applicability of the model's output results.

[0014] Optionally, obtaining the physical effect feature set to be analyzed and combining it with the machine learning training model to obtain the physical effect co-evolution analysis result includes: using the soil arching effect feature set and the tensile membrane effect feature set under real-time or given test conditions as the physical effect feature set to be analyzed; and based on the physical effect feature set to be analyzed, performing state prediction on the pile-supported reinforced embankment according to the machine learning training model to obtain the physical effect co-evolution analysis result. This invention enables dynamic prediction and accurate assessment of engineering states, facilitating timely detection of potential risks and implementation of preventative measures, breaking the passive situation of traditional post-analysis, and providing efficient support for dynamic engineering control and safety assurance.

[0015] Secondly, this invention provides a system for the co-evolution analysis of physical effects of pile-supported reinforced embankments. The system executes the method for co-evolution analysis of physical effects of pile-supported reinforced embankments provided by this invention. The system includes input devices, output devices, a processor, and a memory, which are interconnected. The memory stores a computer program, which includes program instructions, and the processor is configured to call the program instructions. This invention, through high-performance hardware collaboration, enables the engineering implementation and convenient application of the analysis method, adapting to the needs of various scenarios such as engineering sites and laboratories, and improving work efficiency. Attached Figure Description

[0016] Figure 1 This is a flowchart of a method for analyzing the synergistic evolution of physical effects in a pile-supported reinforced embankment according to an embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram of the multi-field coupling model test device according to an embodiment of the present invention;

[0018] Figure 3 This is a schematic diagram of the internal structure of the model box according to an embodiment of the present invention;

[0019] Figure 4 This is a framework diagram of a physical effect co-evolution analysis system for pile-supported reinforced embankments according to an embodiment of the present invention;

[0020] Explanation of reference numerals in the attached drawings: Vertical guide mechanism 11, displacement limit protection gauge 12, vertical servo loading plate 13, loading force limit protection gauge 14, loading gantry structure 15, vertical servo loading actuator 16, moving roller 21, linear guide rail 22, horizontal guide rail trolley 23, water storage tank 31, pressure regulating valve 32, water inlet 33, water outlet 34, overflow regulating device 35, rainwater outlet 36, sand and soil collection device 37, porous permeable plate 38, filter screen 39, model box 41, multi-zone differential deformation induction module 42, movable door servo motor 43, movable door displacement gauge 44, reinforced material bracket 51, reinforced material anchoring device 52, soil pressure sensor 53, reinforced material 54, distributed optical fiber 55, multimodal sensing and monitoring module 56, reinforcement strain sensor 57, data analysis and prediction platform 61, feedback control interface 62. Detailed Implementation

[0021] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0022] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0023] Please see Figure 1One embodiment of the present invention provides a method for co-evolution analysis of physical effects of pile-supported reinforced embankments, the method comprising the following steps:

[0024] S1. The pile-supported reinforced embankment was fitted using model box 41, soil sample and reinforcement material 54, and the test conditions were constructed based on the coupling effect of multiple factors.

[0025] Please see Figure 2 The diagram shows a schematic of a multi-field coupled model test device, including a vertical guide mechanism 11, a displacement limit protection gauge 12, a vertical servo loading plate 13, a loading force limit protection gauge 14, a loading gantry structure 15, a vertical servo loading actuator 16, a moving roller 21, a linear guide rail 22, a horizontal guide rail trolley 23, a water storage tank 31, a pressure regulating valve 32, a water inlet 33, a water outlet 34, an overflow regulating device 35, a rainwater outlet 36, a sand collection device 37, a model box 41, a movable door servo motor 43, a movable door displacement gauge 44, a data analysis and prediction platform 61, and a feedback control interface 62.

[0026] Please see Figure 3 The diagram shows the internal structure of the model box; the components include: porous permeable plate 38, filter screen 39, multi-zone differential deformation induction module 42, reinforced material bracket 51, reinforced material anchoring device 52, soil pressure sensor 53, reinforced material 54, distributed optical fiber 55, multimodal sensing and monitoring module 56, and reinforcement strain sensor 57.

[0027] In this embodiment, soil samples are laid in the model box 41 and reinforcement material 54 is arranged to fit the pile-supported reinforced embankment. The preset differential settlement condition is formed by the multi-zone differential deformation induction module 42. At the same time, external loads and seepage loading conditions are applied as needed. The test condition is constructed based on the multi-factor coupling effect, and then a multi-field coupled model test is constructed.

[0028] It should be noted that multi-directional controllable seepage boundaries are set on the sides and bottom of the model box 41. Seepage loading is applied to the soil by adjusting the seepage boundary conditions. The seepage loading methods include variable head seepage, local leakage loading, and pulsed or step seepage loading. Furthermore, seepage loading can be applied synchronously or alternately with differential deformation loading and external loads as needed for the experiment, so as to simulate the influence of groundwater level changes, local leakage, or sudden seepage on the evolution of soil arching effect in actual engineering.

[0029] In this embodiment, the model box 41 is not a traditional closed test container, but rather the core load-bearing structure for multi-physics coupling tests. It needs to simultaneously meet the following requirements: soil filling and stability constraints; sealing and boundary control requirements for multi-directional seepage loading; structural adaptation requirements for multi-directional loading, especially moving load loading methods; and requirements for internal sensor layout, pipeline lead-out, and test repeatability. Based on the above requirements, the structural composition and function of the model box 41 were modularized and functionally integrated.

[0030] Specifically, the model box 41 includes: a main body structure that provides stable spatial constraints for the soil sample; bearing the soil's own weight, applied loads, and seepage pressure; ensuring that the box as a whole does not undergo significant deformation during the test, thereby avoiding distortion of boundary conditions; the structure consists of a bottom load-bearing frame; front, rear, left, and right side walls; and an open top structure. The main body of the box preferably uses a high-strength steel structure or an aluminum alloy frame as the external load-bearing skeleton, with a rigid transparent observation plate installed on the inner side.

[0031] The sidewall functional components of model box 41 adopt a detachable modular structure. Each sidewall includes: a rigid load-bearing plate; a replaceable seepage boundary plate; and a sealing structure (sealing ring or flexible sealing gasket). A top loading and moving load adaptation structure and a bottom interface and sealing connection structure work together to form model box 41 that meets the requirements of reinforced soil arching effect, seepage, and multi-directional loading tests. A rigid transparent observation plate can be used for visual observation or digital image correlation analysis during the test process.

[0032] The bottom interface and sealing connection structure (reserved for differential deformation and seepage) of the model box 41 is designed as follows: an interface structure that mates with the multi-zone differential deformation induction module 42; including flange, positioning groove or bolt connection structure. This ensures the mechanical continuity and watertightness between the model box 41 and the bottom multi-zone differential deformation induction module 42; and provides sealing and load-bearing functions.

[0033] In an optional embodiment, to prevent soil erosion or pipeline breakage during the operation of the multi-zone differential deformation induction module 42 and seepage loading, a flexible sealing and pipeline anti-damage mechanism is provided between the bottom of the model box 41 and the deformation unit.

[0034] Specifically, the flexible sealing mechanism includes a corrugated skirt flexible sealing structure and a multi-layered laminated sidewall sliding seal. The corrugated skirt flexible sealing structure is located between the edge of the bearing panel of each independent deformation unit and the bottom fixed base of the model box 41. It uses a high-elasticity, water-pressure-resistant rubber corrugated skirt, with its upper and lower flanges fastened to the bearing panel and the base partition, respectively. The unfolding of the corrugations compensates for the vertical displacement difference between adjacent units, preventing internal pressure water leakage. For the deformation unit that is close to the side wall of the model box 41, its bearing panel side is inlaid with a polytetrafluoroethylene wear-resistant slider and a multi-layered lip silicone sealing strip, which slides close to the side wall to maintain lateral watertightness. In addition, the bottom water supply pipeline of the seepage loading module adopts a settlement adaptive expansion structure. The water supply pipe uses a high-pressure braided hose or a spiral PU expansion pipe, and is arranged in a "U" or "S" redundant pattern in the installation cavity below the deformation unit to absorb the tensile amount when the deformation unit reaches its maximum settlement stroke, avoiding pipeline rupture caused by rigid tension.

[0035] In this embodiment, a pile cap model and the soil area between piles are arranged in the model box 41. Soil samples are laid in layers according to the design, and a layer of reinforcing material 54 is simultaneously arranged at a preset height. The laying position and anchorage length are controlled by the reinforcing laying structure. The number of layers, the layer spacing and the initial stress state of the reinforcing material 54 are set by the reinforcing laying and stress control module (no initial tension is applied to simulate the conventional construction state).

[0036] In this embodiment, the reinforcement laying and stress control module is set inside the model box 41. It includes a laying structure for positioning the spatial position of the reinforcement material 54 and a control structure for adjusting the initial stress state of the reinforcement material 54, so as to realize the controllable setting of the number of layers, spacing and initial tension state of the reinforcement material 54.

[0037] Specifically, the reinforcement laying and stress control module is used to controllably set the spatial laying state and initial stress state of the reinforcement material 54 in multi-field coupled model tests, so as to simulate the stress and deformation response of different reinforcement structure forms under different settlement conditions.

[0038] The laying structure is used to limit and support the number of layers, vertical spacing, and planar position of the reinforcing material 54. The laying structure is installed on the inner wall of the model box 41 or on an independent support component, and has several positioning layers along the vertical direction. Each positioning layer supports one layer of reinforcing material 54, thus achieving layered laying of the reinforcing material 54. By changing the number of positioning layers and their spacing, the number of layers and the spacing between layers of reinforcing material 54 can be adjusted. The laying structure also includes guide or limiting components to restrict the planar position of the reinforcing material 54, ensuring that the reinforcing material 54 maintains a preset laying direction and anchorage length during the laying process, thereby guaranteeing the consistency and repeatability of the spatial arrangement of each reinforcing layer.

[0039] The control structure is used to apply a preset initial tension force or initial deformation state to the reinforcing material 54 before or during backfilling. The control structure includes a tensioning force application component and a locking component. The tensioning force application component adjusts the tension of the reinforcing material 54 through displacement control or force control to achieve the target initial stress state. After the target state is achieved, the locking component fixes the ends of the reinforcing material 54 to maintain the initial stress state of the reinforcing material 54 during subsequent backfilling and differential settlement loading.

[0040] In an optional embodiment, each of the 54 layers of reinforcing material is provided with an independent control structure, thereby realizing independent adjustment of the initial tension state of different reinforcing layers to simulate the situation of layered construction or stress difference of different reinforcing layers in engineering.

[0041] Through the above-mentioned reinforcement laying and stress control module, the number of layers, spacing, spatial position and initial stress state of the reinforcement material 54 can be parametrically controlled in multi-field coupled model tests. This allows the mechanical behavior of the reinforced structure under differential settlement conditions to be systematically studied under controllable conditions, thereby improving the guiding significance of the model test results for the design of reinforced structures and disaster prevention in actual engineering.

[0042] After the soil samples and reinforcement material 54 are laid, multiple spatially independent deformation units are set at the bottom of the model box 41, corresponding to the pile cap area and the pile inter-area area respectively; and the multi-zone differential deformation induction module 42 is activated to gradually apply settlement to the pile inter-area area, so that the deformation units of different zones at the bottom of the model box 41 independently generate vertical displacements of different amplitudes according to preset displacement parameters, thereby forming a differential settlement condition between the pile inter-area area and the pile cap area inside the soil.

[0043] It should be noted that each deformation unit includes at least: an actuator + transmission / limiting + reset / locking. An electric lead screw / cylinder is used, which offers advantages such as high displacement control accuracy, closed-loop capability, and convenient programmed loading paths, and is suitable for situations requiring complex settlement curves (gradual, linear, sinusoidal, holding, cyclic). The actuator displacement is stably transmitted to the panel through transmission and connecting components. A universal fine-tuning joint / ball joint is used between the actuator end and the panel to absorb minor assembly errors and avoid jamming; an anti-loosening structure (double nuts, anti-reverse washers) is installed to ensure no drift during long-term testing. To achieve "controllable displacement," each unit is equipped with at least displacement measurement, and force measurement is added if necessary.

[0044] In this embodiment, the multi-zone differential deformation induction module 42 is set at the bottom of the model box 41 and includes multiple deformation units that are spatially independent. Each deformation unit can independently generate controllable displacement in the vertical direction to form differential settlement in different areas within the model box 41, which is used to simulate the differential deformation conditions of soil under pipeline or pile-supported embankment conditions.

[0045] The multi-zone differential deformation induction module 42 includes: a support frame installed at the bottom of the model box 41, several independent deformation units arranged along the plane, a zone bearing panel corresponding to the deformation unit, and a sealing and isolation structure; wherein, each deformation unit includes a vertical guide mechanism 11, a controllable displacement actuator, a displacement sensor, and a limit protection component (displacement limit protector 12, loading force limit protector 14). The actuator drives the corresponding bearing panel to generate independent and controllable vertical displacement, so that the soil in the model box 41 forms preset differential settlement / heave boundary conditions in different zones; the control system is used to apply different displacement-time curves to each deformation unit and perform closed-loop correction, thereby simulating the differential deformation environment of soil under working conditions such as pipelines or pile-supported embankments.

[0046] Structural Layer: Load-bearing and zoned force-transfer components; bottom mounting base / frame provides rigid support and positioning reference for each deformation unit, bearing the soil's self-weight and loading reaction force. Zoned load-bearing panel / force-transfer plate function: evenly transfers the actuator displacement to the overlying soil sample; avoids localized penetration caused by point loading. Zoned isolation and guiding component function: ensures spatial independence and unidirectional (vertical) displacement of each deformation unit, reducing crosstalk and lateral sway between adjacent units.

[0047] Control Layer: Multi-unit collaboration and settlement field generation; Multi-channel controller function: Independently assign displacement-time curves to each unit; Supports synchronous, grouped, and phase difference loading. Supports at least N axes (N = number of deformable units); Supports interpolation / synchronous triggering; Data recording frequency meets experimental requirements (e.g., 10Hz-100Hz or higher, depending on dynamic needs). Displacement control mode (main mode): Each unit tracks the target displacement curve; Force control / force limiting mode (auxiliary): Sets the maximum reaction force threshold based on displacement control; Hybrid mode: Some units are displacement controlled, some units are force controlled, used to simulate "pile top stiffness differences," etc.

[0048] In this embodiment, multi-field coupled model tests are carried out sequentially according to different differential settlement conditions, including: stepped settlement mode, which mainly involves phased settlement in the pile-to-pile area while the pile cap area remains relatively stable, used to simulate the rapid settlement of the soil between piles in the early stage of filling; linear gradient settlement mode, which mainly forms a continuous settlement gradient along the embankment transversely, used to simulate gradual settlement under uneven foundation conditions; settlement basin settlement mode, which mainly forms the maximum settlement in the middle of the pile span and gradually decreases on both sides, used to simulate typical arch-shaped stress conditions; and cyclic settlement mode, used to simulate repeated deformation of the foundation caused by repeated traffic loads, frost heave-thaw settlement cycles, or periodic water level changes.

[0049] Settlement field generation methods include: stepped type: 5mm settlement in area A, 10mm settlement in area B, and 0mm settlement in area C; linear gradient type: settlement slope is formed along the pipeline axis; settlement basin type (Gaussian / parabolic fitting): the largest settlement in the middle and gradually smaller settlement on both sides; cyclic type: simulating traffic load / freeze-thaw settlement cycle.

[0050] The settlement field generation module is set within the control system of the multi-zone differential deformation induction module 42. It automatically generates the target vertical displacement-time curve for each deformation unit based on a preset settlement distribution pattern, spatial zoning information, and time loading strategy. This curve is then output to the corresponding deformation unit actuator, thereby forming a controllable and repeatable differential settlement field within the model box 41. This module achieves settlement mode parameterization, spatial mapping, time history control, and multi-unit collaborative output, providing standardized settlement boundary conditions for conditions such as pipelines, pile-supported embankments, and uneven deformation of soft foundations. It includes the following sub-modules:

[0051] The settlement mode selection submodule is used to select and configure the type of the target settlement field and its control parameters. Supported settlement modes include at least: stepped settlement mode, linear gradient settlement mode, settlement basin settlement mode, and cyclic settlement mode; users can input the settlement mode number and corresponding parameters through the control interface or preset scripts.

[0052] Spatial partitioning mapping submodule: used to establish the mapping relationship between the coordinates of the model box 41 plane and the deformation element number.

[0053] The displacement calculation and interpolation submodule is used to divide the model region into multiple discrete partitions, assigning different constant settlement amounts to different partitions; and to define the spatial center coordinates for each deformable element. , Based on the direction of the pipeline axis or the embankment axis, the main settlement direction is defined, and the value of the settlement function in the continuous space is discretized and mapped to each deformation unit, generating a spatial weighting coefficient or target settlement amplitude for each deformation unit.

[0054] Typical Settlement Field Generation Submodule: The stepped settlement field is used to simulate working conditions where there are significant zonal differences in the foundation, such as localized weak areas, uneven backfilling, or areas with abrupt changes in pile-soil stiffness. Generation Rules: The bottom of the model box 41 is divided into several settlement zones A, B, C,...; all deformation elements within each zone are subject to the same target settlement value. Applicable Working Conditions: Pipelines crossing the boundary between soft and hard foundations and differential settlement between the pile area and the soil between piles in pile-supported embankments.

[0055] A submodule for generating linear gradient settlement fields; linear gradient settlement fields are used to simulate differential settlement that gradually changes along the pipeline axis or the longitudinal direction of the embankment, such as conditions of gradual changes in foundation stiffness or changes in fill height. A coordinate axis is established along a specified direction (e.g., the pipeline axis direction), and the settlement varies linearly along this direction, satisfying the following relationship:

[0056]

[0057] in, Position coordinates Settlement at the location, For position coordinates, This is the maximum settlement reference value. The slope of the settlement change. These are the coordinates of the reference point.

[0058] The submodule for generating settlement basin-shaped settlement fields (Gaussian / parabolic) is used to simulate localized concentrated settlement phenomena, such as consolidation settlement in soft soil areas, underground cavities, and pipeline suspension conditions.

[0059] Gaussian settlement basins satisfy the following relationship:

[0060]

[0061] in, For the first The settlement of each deformation unit. This is the maximum settlement reference value. It is an exponential function. For the first The position coordinates of each deformation element The location of the settlement center. This refers to the parameters representing the range of settlement impact.

[0062] Parabolic settlement basin-shaped settlement field, suitable for It satisfies the following relationship:

[0063]

[0064] in, For the first The settlement of each deformation unit. This is the maximum settlement reference value. It is an exponential function. For the first The position coordinates of each deformation element The location of the settlement center. The radius of influence of the settling basin.

[0065] Cyclic settlement fields are used to simulate repeated foundation deformation caused by repeated traffic loads, frost heave-thaw settlement cycles, or periodic water level changes. A time-cycle function is superimposed on any type of foundation settlement field (stepped, gradient, or settlement basin type) and satisfies the following relationship:

[0066]

[0067] in, For the first Each deformable element at time... The target settlement, For a moment, For the first The static settlement of the foundation of each deformation unit For cyclic amplitude, Angular frequency, This represents the phase difference.

[0068] It should be noted that cyclic settlement fields are suitable for pipelines and embankments under repeated loads on highways and railways and repeated effects of frost heave and thawing settlement. Combined settlement fields can also be used as needed.

[0069] The aforementioned settlement field generation module can transform complex differential settlement conditions of foundations into displacement control commands executable by multi-deformation units, enabling refined and repeatable simulation of settlement field morphology, amplitude, and time history. This provides reliable experimental boundary conditions for the study of the mechanical response of structures such as pipelines and pile-supported embankments under differential deformation conditions.

[0070] In this embodiment, the time loading history generation submodule is used to uniformly generate and control the time evolution path of the differential settlement loading process based on the preset spatial settlement field, so as to realize the real simulation of the occurrence and development of settlement over time in the project.

[0071] Specifically, the time-based loading history generation submodule manages the settlement loading process in stages, dividing the overall loading process into an initial stage, a loading stage, a holding stage, and a cyclic stage, and generating corresponding settlement displacement-time functions within each stage. By setting the loading path function, the settlement loading can vary with time in a linear, graded, exponential, or periodic manner, thereby simulating typical engineering processes such as construction filling, soft soil consolidation, long-term service, and repeated loading.

[0072] Furthermore, regarding the first The vertical settlement of each deformation element over time can be uniformly represented by the loading history, satisfying the following relationship:

[0073]

[0074] in, For the first Each deformable element at time... The target settlement, The spatial settlement amplitude is determined by the settlement field generation function module. This is a function for loading time history.

[0075] It should be noted that the time loading history function has a value range of [0,1] and is used to describe the evolution of settlement over time.

[0076] The evolution pattern of the loading history function differs under different loading modes, and its main expression satisfies the following relationship:

[0077]

[0078] in, For loading the time history function, For a moment, The duration of the loading phase.

[0079] The time-load history function is applicable to linear loading functions (uniform settlement), where the settlement gradually develops at a constant rate and remains unchanged after reaching the final settlement.

[0080] The graded step loading function (staged settlement) is used to simulate the development of phased filling or abrupt settlement. Its time loading history function satisfies the following relationship:

[0081]

[0082] in, For loading the time history function, The total amount for the different levels, For hierarchical indexing, For the first Grade settlement ratio, For Heaviside step function, For a moment, For the first Level loading start time.

[0083] It should be noted that the settlement occurs gradually in several "steps", and a holding time can be set after each loading step.

[0084] The exponential loading function (soft soil consolidation settlement) satisfies the following relationship in its time loading history function:

[0085]

[0086] in, For loading the time history function, It is an exponential function. These are consolidation control parameters. For a moment.

[0087] It should be noted that the consolidation control parameters reflect the settlement development rate.

[0088] For a cyclic loading function (repeated settling), its time loading history function satisfies the following relationship:

[0089]

[0090] in, For loading the time history function, For cyclic amplitude, Angular frequency, For a moment, This represents the phase difference.

[0091] In this embodiment, the settlement loading process of each deformation unit is defined by multiplying the spatial settlement amplitude by a time loading function, wherein the time loading function can be in the form of a linear, stepwise, exponential, or periodic function, thereby achieving parameterized control of differential settlement in the time dimension.

[0092] Meanwhile, the time-based loading history generation submodule controls the settlement loading rate and single-step displacement increment, discretizing the continuous loading path into multiple quasi-static displacement steps. This avoids introducing non-engineering responses due to excessively rapid loading, ensuring the stability and repeatability of multi-field coupled model tests. After reaching the target settlement value, it can maintain the displacement state of each deformation unit and continuously record the structural and soil responses to simulate the long-term evolution of engineering settlement. Furthermore, it supports superimposing periodic or repetitive displacement histories on the foundation settlement loading path to simulate engineering conditions such as repeated traffic loads and frost heave-thaw settlement cycles, providing temporal loading conditions for studying structural fatigue damage and the cumulative evolution of disasters.

[0093] Through the above methods, the time loading history generation submodule realizes the parameterization, controllability and repeatability of differential settlement conditions in the time dimension, and works in conjunction with the settlement field generation functional module to enable multi-field coupled model tests to simultaneously reflect the engineering characteristics of settlement in spatial distribution and temporal evolution.

[0094] In an optional embodiment, the output and correction submodule is used to convert the target displacement-time function of each deformation unit formed by the time loading history generation submodule into executable control commands and output them to the multi-zone differential deformation induction module 42. At the same time, it performs real-time monitoring and deviation correction of the actual displacement during the execution process to ensure the accuracy and stability of settlement loading.

[0095] The output and correction submodule includes a control command output unit and a feedback correction unit. These units accurately apply the target settlement loading history of each deformation unit generated by the time-based loading history generation submodule to the multi-zone differential deformation induction module 42, ensuring consistency between the actual loading process and the preset loading history. The command output unit converts the target settlement amount into control commands for the corresponding actuators based on the target displacement-time function of each deformation unit, and drives each deformation unit to generate vertical displacement through multi-channel output. The control commands can be motor drive signals, actuator stroke commands, or other control signals matched to the actuators, thereby realizing the physical execution of the target loading history. The feedback correction unit collects the actual displacement of each deformation unit in real time during the loading process and compares the collected actual displacement with the target displacement to obtain the loading deviation. When the deviation exceeds a preset threshold, the control commands of the command output unit are dynamically corrected to compensate for the influence of actuator errors, assembly errors, and soil reaction force changes on the loading process, thus forming a closed-loop control.

[0096] Through the above methods, the output and correction submodule can effectively compensate for the influence of actuator errors, assembly errors, and soil reaction force changes on the loading process, ensuring that the actual settlement process of each deformation unit is consistent with the preset loading history, and improving the accuracy and repeatability of differential settlement simulation.

[0097] The time-based loading history generation submodule generates the target displacement-time history for differential settlement loading and belongs to the loading strategy generation layer. The output and correction submodule converts the target displacement-time history into executable control commands and performs real-time corrections, belonging to the loading execution and control layer. These two submodules are functionally independent yet work collaboratively, enabling accurate and stable application of differential settlement across multiple zones according to a preset time history, thus improving the accuracy, stability, and repeatability of differential settlement simulation in multi-field coupled model experiments.

[0098] In an optional embodiment, the random field type generates a partitioned displacement sequence according to a set correlation length to simulate natural weak inhomogeneity.

[0099] Specifically, the random field-based settlement generation method is based on random field theory, treating differential settlement of the foundation as a random variable field with statistical characteristics in space, rather than a deterministic function. The settlement of each deformation unit is not independent and random, but has spatial correlation; by setting the correlation length, the smoothness or abruptness of settlement in space is controlled; while maintaining the consistency of the overall settlement statistical characteristics, local random disturbances are introduced, thereby more realistically reflecting the conditions of natural soft foundations.

[0100] Furthermore, given that the bottom of the model box 41 is divided into N deformation elements, the following is defined: : No. Spatial coordinates of each deformable element; : No. The target settlement of each deformation unit; Mean settlement; : Settlement standard deviation; : Spatial correlation length (correlation scale); where correlation length is a key control parameter used to reflect the spatial scale of the influence range of soft foundation.

[0101] Random field-type settlement satisfies the following relationship:

[0102]

[0103] in, For the first The target settlement of each deformation unit This is the average settlement value. For the standard deviation of settlement, It is a spatial random field with a mean of 0 and a variance of 1.

[0104] By employing a stochastic field-based settlement generation method, controlled spatial randomness can be introduced into multi-field coupled model tests, freeing differential settlement from being confined to idealized models. This allows for: more realistic simulation of natural soft foundation conditions; revelation of the structural response characteristics under the most unfavorable uneven settlement conditions; improved representativeness of model test results for actual engineering disasters; and provision of more conservative and reliable experimental evidence for structural safety assessment.

[0105] During the process of forming differential settlement conditions, vertical or planar loads are applied to the soil or structure inside the model box 41 through an external load application device as required by the test, so as to simulate the action of the superstructure load or the filling load as the external load conditions.

[0106] It should be noted that the top loading and moving load adaptation structure (multi-directional loading core) supports the simulation of moving load conditions such as vehicle loads and construction equipment loads; in conjunction with bottom differential deformation and seepage loading, it simulates more realistic engineering service conditions; the top opening and loading interface, the top of the model box 41 is set as an open structure, with the following set on top: top bearing beam or frame; loading interface connected to external load application device, this structure is used to bear the force from vertical loading device or moving load system.

[0107] The moving load guiding structure, to realize the moving load loading method, includes a linear guide rail or guide groove on the top of the model box 41 along the loading direction; a moving roller 21 and a horizontal guide rail trolley 23 or loading platform that cooperate with the guide rail. The moving loading device can move back and forth or unidirectionally along the guide rail on the top of the model box 41. The guiding structure ensures stability and repeatability during the loading process and avoids loading deviation.

[0108] In an optional embodiment, the external load application device is configured as a top moving load loading module for simulating the dynamic movement and cyclic loading of vehicles or construction machinery.

[0109] The top moving load loading module includes: a vertical servo loading plate 13, a loading gantry structure 15, a horizontal drive assembly, a vertical servo loading actuator 16, and a contour loading head. The loading gantry structure 15 is fixed to both sides of the top of the model box 41, with double-line high-precision linear guide rails 22 laid parallel on its crossbeams. The horizontal drive assembly uses a movable door servo motor 43 in conjunction with a ball screw transmission mechanism to drive the horizontal guide rail trolley 23 to perform horizontal reciprocating motion along the linear guide rail 22, simulating different equivalent scaled-down vehicle speeds. The vertical servo loading actuator 16 is vertically mounted below the horizontal guide rail trolley 23, and has built-in high-frequency force and displacement sensors, configured to output constant static load, sinusoidal dynamic load, or random waveform dynamic load. The contour loading head is connected to the output end of the vertical servo loading actuator 16, and its bottom is wrapped with a polyurethane rubber pad to equivalently represent the ground pressure and contact area of ​​an actual tire or track. During the test, the control system simultaneously issued horizontal position commands and vertical force control commands, forming a "position-force" dual closed-loop control through real-time feedback from the force sensor, creating continuous and dynamic moving load boundary conditions on the embankment surface.

[0110] Meanwhile, by applying seepage loading or local leakage conditions with controllable water head conditions to the side and / or bottom of the model box 41 through the seepage loading module, a unidirectional or multidirectional seepage field is formed inside the soil to simulate the influence of groundwater level changes or seepage softening on the soil arching effect, thus forming a seepage action condition.

[0111] The system integrates seepage functionality, enabling lateral, localized, or multi-directional seepage loading on the soil. At least a portion of the sidewalls are configured as seepage loading sidewalls, internally equipped with seepage channels or cavities. An outlet 34 connects to an external water supply system (connecting to a sand collection device 37). Replaceable seepage media plates (such as porous seepage plates 38 and filter screens 39) are included. The system enables lateral, localized, or multi-directional seepage loading on the soil. Different seepage boundary conditions can be achieved by replacing the seepage media plates or controlling the inlet and outlet water pressures. The detachable structure facilitates changing the seepage scheme or performing maintenance without damaging the main body of the test chamber. A sealing structure ensures no leakage occurs under seepage loading conditions, maintaining test boundary stability.

[0112] In this embodiment, the seepage loading module is set on the side and bottom of the model box 41 and is used to apply seepage loading with multi-directional and variable boundary conditions to the soil to simulate the evolution process of the reinforced soil arching effect under seepage.

[0113] The seepage loading module is used to apply seepage loading with multidirectional and variable boundary conditions to the soil during model tests to simulate the influence of seepage inside the reinforced soil on the stress transfer and arching effect evolution process under engineering conditions such as groundwater level changes and seepage field evolution.

[0114] The seepage loading module includes a seepage supply structure, a seepage distribution and control structure, and a seepage boundary adjustment structure. These structures work together to achieve controllable settings for the seepage loading direction, seepage intensity, and seepage boundary conditions. The seepage supply structure provides a stable or adjustable water source to the model tank 41. The seepage supply structure includes a water storage tank 31, a water supply pipeline, a pressure regulating valve 32, an overflow regulating device 35, and a rainwater outlet 36. By adjusting the water supply height, water supply pressure, or water supply flow rate, the seepage loading can meet the head requirements of different test conditions.

[0115] In an optional embodiment, the seepage supply structure creates a stable head difference by changing the water supply level, thereby forming a controlled seepage field within the soil. The seepage distribution and control structure is located on the side or bottom of the model box 41 to uniformly or in a predetermined manner introduce seepage water into the soil. The seepage distribution and control structure may include a porous seepage plate 38, a filter screen 39, or a distributed inlet interface 33 to ensure uniform water flow into the soil during seepage loading, avoiding unexpected disturbances to the soil structure caused by localized concentrated seepage. By setting multiple seepage inlets at different locations on the model box 41, seepage loading methods from bottom to top, from side to inside, or in a multi-directional combination can be achieved. The seepage boundary adjustment structure is used to regulate the seepage boundary conditions of the soil. The seepage boundary adjustment structure may be located on the sidewall or top of the model box 41 opposite to the inlet side and includes an outlet 34, an overflow outlet, or an adjustable water level device to control the outflow conditions of seepage within the soil. By adjusting the water outlet height or opening and closing different water outlet channels, constant head boundary, variable head boundary, or drainage-non-drainage boundary conditions can be formed, thereby simulating engineering scenarios such as groundwater level rise and fall, and changes in seepage channels.

[0116] In an optional embodiment, the seepage loading module can work in conjunction with the time loading history generation submodule to adjust the seepage boundary conditions in a phased or continuous manner, so that the seepage field changes over time, thereby simulating the evolution of seepage.

[0117] In an optional embodiment, the seepage loading module can work in conjunction with the multi-zone differential deformation induction module 42 and the reinforcement laying and stress control module to cause stress redistribution in the soil under the coupling effect of differential settlement and seepage, thereby promoting the formation, development, or attenuation of the reinforced soil arching effect. By setting the above-mentioned seepage loading module, the influence of seepage on the stress transmission path and arching effect evolution of the reinforced soil can be reproduced in the model test. This provides a controllable loading method for analyzing the stability and disaster mechanism of reinforced soil structures under seepage conditions, thereby improving the representativeness and interpretability of the model test results for actual engineering conditions.

[0118] In an optional embodiment, the order of the seepage loading module and the multi-zone differential deformation induction module 42 can be as follows: simulating various working conditions as needed, the synchronous mode is mainly achieved by applying an intensity equal to and simultaneously increasing the settlement amplitude of the seepage loading module; while the staggered mode is to apply settlement first and then seepage, which is used to simulate sudden leakage conditions.

[0119] By combining differential settlement, external load, and seepage conditions, a test condition for pile-supported reinforced embankments under multi-factor coupling is constructed to simulate the formation and evolution of the reinforced soil arch effect under the combined action of differential deformation and seepage conditions in actual engineering.

[0120] S2. Based on the test conditions, a multi-modal data acquisition method is used to collect multi-source test datasets from the pile-supported reinforced embankment using a sensor array.

[0121] In this embodiment, a multi-type sensor array is set in the model box 41, including: earth pressure sensor 53 deployed in the key stress area of ​​the soil, displacement sensor for monitoring soil deformation, and reinforcement strain sensor 57 (such as distributed optical fiber 55 or multi-point strain gauge) deployed on the reinforcement material 54; a multi-modal sensing and monitoring module 56 is formed based on the sensor array.

[0122] The side wall of the model box 41 is also equipped with a reinforcing material bracket 51 and a reinforcing material anchoring device 52.

[0123] Specifically, earth pressure sensors 53 are deployed at the arch foot, arch top, pile cap, or pipe top to acquire soil pressure data at key locations during the formation of the soil arching effect. Displacement sensors are used to collect soil settlement and lateral displacement data. Reinforcement strain sensors 57 are used to acquire reinforcement strain data of the reinforcement material 54 during loading, and to monitor the stress and strain evolution of the reinforcement material 54 under differential settlement and seepage, so as to characterize the tensile membrane effect of the reinforcement material 54 and its possible strain concentration or abnormal stress state.

[0124] Displacement sensors: LVDT movable door displacement gauge 44 / magnetostrictive displacement sensor / grating ruler, used for real-time feedback of the vertical displacement of each unit, forming closed-loop control and data recording. Sensors should be arranged as coaxially as possible with the direction of execution; if space is limited, a side-mounted LVDT + linkage structure can be used, but gap-free operation must be ensured.

[0125] Earth pressure sensor 53: A thin-type pressure load cell / weighing module, located between the actuator and the panel. It is used to obtain the reaction force in each zone, facilitating the analysis of "pile-soil load sharing and stress redistribution phenomena," while also providing a safe load limit and a reliable reference standard for practical engineering applications.

[0126] Condition monitoring; temperature (motor / driver), current / torque, limit switch status, used for fault diagnosis and test safety.

[0127] In this embodiment, the multimodal sensing and monitoring module 56 includes an earth pressure sensor 53, a displacement sensor, and a reinforcement strain sensor 57 deployed inside the soil or on the reinforcement material 54, for synchronously collecting the soil stress state, deformation information, and the stress response data of the reinforcement material 54; the multimodal sensing and monitoring module 56 is set inside the model box 41, for synchronously monitoring and collecting data on the soil stress state, deformation behavior, and stress response of the reinforcement material 54 during the model test, so as to achieve a comprehensive characterization of the mechanical response of the soil-reinforced structure system under differential settlement conditions.

[0128] The various sensors in the multimodal sensing and monitoring module 56 are used to acquire information on internal stress changes in the soil, structural or soil deformation, and the stress and deformation response of the reinforcing material 54.

[0129] Earth pressure sensor 53 is used to monitor the stress distribution and evolution of soil under its own weight, superstructure load, and differential settlement. Earth pressure sensor 53 can be embedded at different depths and planar locations within the soil, or placed at the interface between the soil and reinforcement material 54 or structural components to obtain information on earth pressure changes at key locations. By simultaneously acquiring earth pressure data from multiple measuring points, the development characteristics of stress redistribution and arching effect within the soil under differential settlement conditions can be analyzed.

[0130] Displacement sensors are used to monitor the deformation of soil or structure in the vertical and horizontal directions. Displacement sensors can be deployed on the inner wall of the model box 41, on the soil surface, at key structural locations, or near the reinforcing material 54 to measure overall settlement, differential settlement, and local deformation. Displacement sensors can be contact or non-contact types, and their measurement direction corresponds to the main deformation direction of interest in the model test, thus accurately reflecting the deformation response of the soil and structure during loading.

[0131] The strain sensor 57 is used to monitor the strain changes of the reinforcing material 54 during laying and loading, thereby characterizing the stress state of the reinforcing material 54. The strain sensor 57 can be deployed at different locations along the length of the reinforcing material 54 and works in conjunction with the reinforcement laying and stress control module to obtain the strain evolution law of the reinforcing material 54 under initial tension, backfilling, and differential settlement. Through analysis of the strain data, the internal force changes of the reinforcing material 54 and its constraint effect on soil deformation can be further calculated.

[0132] In an optional embodiment, the multimodal sensing monitoring module 56 synchronously acquires and records signals from various sensors through a unified data acquisition system to ensure the consistency of different physical quantities over time, thereby enabling correlation analysis between soil stress, deformation, and the stress response of the reinforcing material 54. Through comprehensive processing of multi-source monitoring data, the coordinated deformation and stress mechanism of the soil-reinforced system under differential settlement conditions is revealed. The reinforcement strain sensor 57 is used to monitor the stress and strain evolution process of the reinforcing material 54 under differential settlement and seepage, to characterize the tensile membrane effect of the reinforcing material 54 and its possible strain concentration or abnormal stress states.

[0133] By setting up the multimodal sensing and monitoring module 56, comprehensive monitoring of the soil stress state, deformation behavior, and stress response of the reinforcing material 54 can be achieved in the model test. This provides reliable data support for analyzing the working mechanism and disaster evolution process of the reinforced structure under differential settlement conditions, and improves the integrity and engineering interpretation capability of the model test results.

[0134] In this embodiment, the multimodal sensing and monitoring module 56 is used to simultaneously collect soil pressure data, displacement data, and reinforcement strain data to form a multi-source test dataset containing time information.

[0135] Specifically, under differential settlement conditions, external loads, and seepage loading conditions, the multimodal sensing monitoring module 56 is used to synchronously monitor the soil and reinforcing material 54 in the model box 41 to obtain synchronous monitoring data. Soil pressure data reflecting the stress state of the soil, displacement data reflecting the deformation characteristics of the soil or structure, and strain data of the reinforcing material 54 reflecting the stress response are collected respectively.

[0136] Furthermore, the synchronous monitoring data are all collected according to a unified time benchmark and a time stamp is added, thereby forming a multi-source test dataset containing time information, which is used to characterize the stress and deformation evolution process of reinforced soil under differential deformation and seepage conditions, and form a multi-modal feature set characterizing the soil arching effect.

[0137] S3. Perform data preprocessing and feature extraction on the multi-source test dataset to obtain the soil arch effect feature set and the tensile membrane effect feature set.

[0138] In this embodiment, the multi-source test dataset is preprocessed to obtain multi-source test processed data. The data preprocessing includes time alignment of earth pressure data, displacement data and reinforcement strain data from different sources based on a unified time reference to eliminate the impact of differences in sampling frequency and sampling start time of different sensors on data synchronization. The multi-source test data is also subjected to noise processing and normalization to reduce the impact of environmental disturbances, test errors and dimensional differences on the data analysis results.

[0139] After data preprocessing, characteristic parameters that reflect changes in soil internal stress distribution, differential deformation characteristics, stress state of reinforced material 54, and seepage conditions are extracted from the multi-source test data. Based on the spatiotemporal variation of the characteristic parameters, including their spatial distribution and temporal evolution, the characteristic parameters are combined and organized to construct a soil arch effect characteristic set and a tensile membrane effect characteristic set to characterize the formation and evolution of soil arch effect under differential deformation and seepage conditions.

[0140] It should be noted that the physical effect co-evolution mechanism revealed in this invention is manifested as follows: In the initial stage of differential settlement, the settlement of the soil between piles causes the principal stress to deflect within the soil, and the soil arching effect dominates, bearing most of the upper load; as differential settlement continues or seepage softening is introduced, the shear stress at the soil arch foot gradually reaches the soil's shear strength limit, and the soil arching effect begins to progressively degrade; at this time, the reinforcing material 54 undergoes vertical flexural deformation under the downward pressure of the soil between piles, and its stress state gradually changes from the initial horizontal tension to a catenary-type tension membrane stress mode. In this co-evolution process of "degradation-reinforcement," the growth rate of the vertical component force of the tension membrane and the decay rate of the soil arching effect strength index show a highly negative correlation time series characteristic. By extracting the critical strain and pressure redistribution characteristics in this dynamic transformation process, core physical basis is provided for subsequent machine learning models to capture early signals of structural instability.

[0141] In this embodiment, to obtain the aforementioned spatiotemporal variation patterns, the spatial coordinate reference of the model test and the mapping relationship between the measuring points are first established: a global coordinate system O-XY (Z is vertical) is established on the plane of the model box 41, and the placement positions of the earth pressure sensor 53, the movable gate displacement gauge 44, and the strain gauge are mapped to coordinate points. , , The data is labeled according to functional zones such as pile cap area, pile inter-pile area, arch foot area, and arch top area. At the same time, the differential deformation loading process, seepage loading process and external load process are uniformly mapped to a unified time reference to form a three-dimensional aligned data base of "time-space-working condition".

[0142] Furthermore, the spatial distribution variation patterns are obtained, including: First, based on functional zoning, the earth pressure data, displacement data, and reinforcement strain data within each moment or preset time window are statistically analyzed to obtain zoning characteristics such as the pressure value of the pile cap area, the pressure value of the inter-pile area, and their differences and ratios; Second, based on the zoning statistics, the discrete measuring point values ​​are spatially reconstructed to obtain the spatial field, including the pressure field, settlement field, and strain field. The spatial reconstruction can use any one of inverse distance weighted interpolation, spline interpolation, or kriging interpolation; Finally, spatial morphological descriptive quantities are extracted from the spatial field, including the coefficient of variation of pressure / strain, the magnitude of the spatial gradient, and the migration of the pressure peak position over time, to characterize the concentration and spatial transfer characteristics of the arched stress zone.

[0143] Furthermore, the temporal evolution patterns are obtained, including: First, smoothing and denoising the sequences of each characteristic parameter, and calculating the evolution rate and acceleration through first-order and / or second-order differences to characterize the degradation rate of the soil arching effect and the enhancement rate of the tensile membrane effect; Second, identifying key stages in the evolution process based on abrupt change point detection to obtain key moments such as the starting point of soil arch degradation and the connection point of the tensile membrane. The abrupt change point detection can use any one of the following: sliding window statistical test, Cumulative Sum (CUSUM), or Pruned Exact Linear Time (PELT); Finally, obtaining the connection lag time between the soil arching effect intensity index and the vertical component of the tensile membrane through cross-correlation or lag correlation analysis, thereby quantifying the co-evolutionary relationship between "degradation-connection".

[0144] In this embodiment, the feature parameters are combined and organized to construct the soil arching effect feature set and the tensile membrane effect feature set, including: First, statistical summarization of each feature parameter within a preset time window, where the statistics within the window include at least one or more of the mean, maximum, minimum, final value, and slope; Second, missing data repair, dimension unification, and normalization of the statistical results within the window; Subsequently, under the same time window index, the feature parameters characterizing the degree of load concentration on the pile cap and the arch distribution pattern, along with their evolution rates, are combined to form a soil arching effect feature vector, thus forming the soil arching effect feature set; Simultaneously, the grid tension, the vertical component of the tensile membrane, the degree of strain concentration, and their evolution rates are combined to form a tensile membrane effect feature vector, thus forming a tensile membrane effect feature set; Finally, the differential deformation amplitude, seepage head / flow rate, and external load level, among other working condition parameters, are output as labels or covariates synchronously with the feature vectors for subsequent machine learning model training and risk prediction.

[0145] In this embodiment, the tensile membrane effect is quantitatively characterized based on the strain and geometric deformation relationship of the reinforced material 54, and the soil arching effect intensity is characterized by the change of soil pressure between the pile cap and the pile, thereby realizing the analysis of the coupled evolution process of the two.

[0146] Specifically, by processing data from multi-source experiments, the tensile membrane effect is quantified to obtain the grid tension and the vertical component of the tensile membrane, thereby transforming the "membrane effect" from a "concept" into a measurable quantity that can be directly inverted from "strain + displacement".

[0147] The tensile force of the above-mentioned grille satisfies the following relationship:

[0148]

[0149] in, The tensile force per unit width or equivalent cross-section of the grid. Let be the elastic modulus of the grid. This represents the equivalent cross-sectional area of ​​the grille per unit width. This is the strain data for the reinforcing steel.

[0150] The vertical component of the tensile membrane above satisfies the following relationship:

[0151]

[0152] in, For the vertical component of the tension membrane, The tensile force per unit width or equivalent cross-section of the grid. The angle between the grid and the pile is the horizontal angle.

[0153] Specifically, based on multi-source test data, the soil pressure sensor 53 obtains the pressure values ​​in the pile cap area, the inter-pile area, and the average pressure value. The soil arching effect is quantified to obtain a soil arching effect strength index, which is used to express the degree of load concentration in the pile cap area. The strength of the arch can be directly reflected in the numerical change of the soil arching effect strength index. The soil arching effect strength index satisfies the following relationship:

[0154]

[0155] in, As an index of soil arching effect strength, This represents the pressure value in the pile cap area. This represents the pressure value in the inter-pile zone. This represents the average pressure value.

[0156] In this embodiment, the contributing factors of the tensile membrane effect of the grid are obtained by combining the soil arching effect; in the coupled case, the corresponding contributing factors are calculated by the coupling effect of the tensile membrane effect of the grid and the soil arching effect.

[0157] First, the total vertical load is obtained based on the load transferred from the grid to the top of the adjacent piles on both sides and the load transferred to the top of the adjacent pile caps through the soil arching effect, satisfying the following relationship:

[0158]

[0159] in, For the total vertical load, The load transferred from the grid to the top of the adjacent piles on both sides This is to transfer the load to the top of the adjacent pile cap through the soil arching effect.

[0160] It should be noted that the load transferred from the grid to the top of the adjacent piles can be obtained by measuring and converting the load through strain gauges; the load transferred to the top of the adjacent pile cap through the soil arching effect can be measured by the soil pressure sensor 53 installed above the grid.

[0161] Secondly, based on the load transferred from the grid to the piles on both sides and the total vertical load, the contributing factors of the tension membrane effect of the grid satisfy the following relationship:

[0162]

[0163] in, As contributing factors, The load transferred from the grid to the top of the adjacent piles on both sides This represents the total vertical load.

[0164] It should be noted that during the seepage process, the seepage damages the structure of the soil arch, so the grid is forced to "take over" more loads. Therefore, this provides good guidance for selecting grids with reasonable stiffness.

[0165] S4. Construct a machine learning model based on the soil arch effect feature set and the tensile membrane effect feature set, and obtain a machine learning training model through the model training data.

[0166] In this embodiment, based on the soil arching effect feature set, feature parameters reflecting differential deformation conditions, reinforcement parameters, and seepage loading conditions are selected as model inputs, and feature indices characterizing the state or evolution response of the soil arching effect are used as model outputs to construct a machine learning model for describing the relationship between differential deformation conditions, reinforcement parameters, seepage conditions, and soil arching effect response.

[0167] In this embodiment, to ensure that the model input can accurately reflect the influence mechanism of differential deformation, reinforcement, and seepage on the soil arching effect, and to improve the model's generalization ability and interpretability, a hierarchical selection strategy of "candidate parameter library construction - screening criterion constraints - statistical testing and iterative optimization" is adopted for the selection of feature parameters, including:

[0168] 1) Candidate Parameter Library Construction: First, a candidate input parameter library is constructed from the test condition setting parameters and multi-source test processing data, including but not limited to differential deformation condition parameters, reinforcement parameters, seepage loading condition parameters, and initial state parameters, among which:

[0169] Differential deformation condition parameters include settlement pattern type (stepped, linear gradient, settlement basin, cyclic), maximum differential settlement, differential settlement development rate, settlement ratio between pile cap area and pile inter-pile area, and settlement loading stage identifier.

[0170] The reinforcement parameters include the equivalent axial stiffness of the reinforcement material 54, the number of reinforcement layers, the layer spacing, the anchorage length, the laying height, and the initial tension state or initial pre-strain.

[0171] The parameters for seepage loading conditions include head difference, seepage gradient, seepage flow rate or seepage pressure, seepage duration, loading sequence identifier (settlement before seepage / synchronous loading / staggered loading), and seepage boundary type identifier (constant head / variable head / drainage-no-drainage).

[0172] Initial state parameters include initial soil moisture content, initial void ratio or dry density, initial effective stress level, and test filling height, which are used to characterize differences in initial conditions.

[0173] 2) Screening Criterion Constraints: When screening feature parameters from the candidate input parameter library, the following criteria must be met:

[0174] 2.1) Physical correlation criterion: The characteristic parameters should have a clear mechanical or hydraulic correlation with the formation, development or degradation process of the soil arching effect, and can be used to characterize the changes in load transfer path, the degree of differential deformation or the degree of seepage softening.

[0175] 2.2) Measurable and controllable criteria: The characteristic parameters should be directly obtainable or reliably inverted by the test loading system or the sensing and monitoring system, and should be controllably set during the test condition construction stage in order to achieve model training and engineering reproducibility;

[0176] 2.3) Robustness criterion: The characteristic parameter has a high signal-to-noise ratio under different experimental repetitions or different acquisition frequencies, and is not sensitive to local missing values ​​or outliers;

[0177] 2.4) Non-redundancy criterion: Feature parameters should not be highly collinear to avoid duplication of input information leading to model instability; when there are strongly correlated parameters, retain those that have clearer physical meaning or contribute more to prediction.

[0178] 2.5) Consistency and constraint criteria: The characteristic parameters should be consistent with the test condition labels and physical boundary conditions, and can be used together with physical constraint terms in model training to ensure that the prediction results conform to common sense in engineering (e.g., reasonable range of pressure and displacement, range of contribution rate, etc.).

[0179] 3) Statistical testing and iterative optimization: Based on the above screening criteria, the feature parameters are determined according to the following steps:

[0180] 3.1) Correlation pre-screening: Calculate the correlation index between each candidate input parameter and the target output index (such as soil arching effect strength index, pile cap pressure value, inter-pile pressure value and its evolution rate, etc.). The correlation index includes Pearson correlation coefficient, mutual information or rank correlation coefficient; remove parameters that are almost unrelated to the output or are stable and unchanged.

[0181] 3.2) Collinearity test: Calculate the variance inflation factor or correlation matrix for the pre-screened input parameters. If there are parameter groups with collinearity exceeding the preset threshold, retain the representative parameters according to the priority of "more controllable / more interpretable / more stable", and use the remaining parameters as alternatives.

[0182] 3.3) Grouping and Combination: Input parameters are grouped into three categories: differential deformation, reinforcement, and seepage. Within each group, key parameters that can characterize "amplitude-rate-stage" are retained, and time-series inputs are represented by time window statistics or serialization. For continuous time-series parameters, statistical measures such as mean, maximum, minimum, and rate of change are extracted within a preset time window as input features to reduce noise and preserve time information.

[0183] 3.4) Importance Assessment and Iterative Optimization: The initial model is trained using cross-validation, and the contribution of each input parameter to the prediction accuracy is evaluated using feature importance methods (including but not limited to permutation importance, Shapley Additive exPlanations (SHAP) quantification of contribution, or ablation experiments). Parameters with low contribution and that do not meet the robustness requirements are removed or replaced until the model's error on the validation set converges and stabilizes.

[0184] 3.5) Final Input Set Consolidation: The set of input parameters obtained after filtering is denoted as... The samples are stored together with the test condition labels as training samples; among them, categorical parameters are input using one-hot encoding or ordinal encoding, and continuous parameters are normalized before input.

[0185] By using the above-mentioned selection criteria and screening steps for feature parameters, the model input includes controllable working condition information (differential deformation, reinforcement, seepage) and retains the key variables most sensitive to the response to the soil arching effect, thereby improving the prediction accuracy and engineering interpretability of the machine learning model for the state and evolution response of the soil arching effect.

[0186] In this embodiment, a mapping relationship between differential deformation conditions, reinforcement parameters, seepage loading conditions, soil arching effect, and tensile membrane response is established through machine learning modeling steps, thereby enabling the prediction of the evolution state of soil arching effect and potential instability risk under different test conditions.

[0187] Specifically, machine learning models (such as temporal neural network models and physical constraint learning models) are used to establish a mapping relationship between "differential deformation, reinforcement parameters, seepage, soil arch response and tensile membrane response"; to predict the instability risk of the reinforced arch under stress state and the degree of soil arch formation; the model output results are used to guide subsequent loading paths and predict key failure stages.

[0188] In this embodiment, in order to achieve a unified mapping of "differential deformation - reinforcement parameters - seepage - soil arch response and tensile membrane response" and to adapt to the dynamic characteristics of the soil arch effect and tensile membrane effect evolving over time, the machine learning model is defined as a temporal neural network model.

[0189] 1) Mathematical definitions of input and output:

[0190] In this embodiment, using a time window or sampling time under a unified time reference as an index, the characteristic parameters reflecting differential deformation conditions, reinforcement parameters, and seepage loading conditions are used to form the model input vector, and the state / evolution response of the soil arching effect and the tensile membrane effect are used to form the model output vector; satisfying the following relationship:

[0191]

[0192]

[0193] in, For the model input vector, For differential settlement, For differential settlement development rate, To enhance the equivalent axial stiffness, The number of layers of reinforcing material 54 to be laid. This represents the total height of the embankment fill. The laying height of the reinforcing material 54, The anchorage length of the reinforcing material 54 is given. Due to seepage head difference, For seepage gradient, This is the seepage flow rate. The duration of seepage loading, Indicates transpose. Output vectors for the model. As an index of soil arching effect strength, For a moment The predicted value of the pile cap area pressure, For a moment The predicted value of the inter-pile pressure. For the tension of the grille, For the vertical component of the tension membrane, Factors contributing to the tensile membrane effect This serves as an indicator or probability output for the risk of arch instability.

[0194] When using the time window input method, continuous The input composition at each time point The model output is a sequence. This establishes a time-series mapping that satisfies the following relationship:

[0195]

[0196] in, Output sequence for the model, For a moment, For the temporal neural network model to be trained, For model parameters, Input the sequence to the model.

[0197] 2) State update expression for a temporal neural network model:

[0198] In this embodiment, the temporal neural network model is implemented using a recurrent neural network, a gated recurrent unit, a long short-term memory network, or an attention mechanism temporal model. Its hidden state update and output satisfy the following relationship:

[0199]

[0200]

[0201] in, For a moment The hidden state, For state update functions, For a moment The hidden state, For a moment The input vector, To predict the output, This is the function for reading out.

[0202] By recursively updating the hidden states, the model can learn the temporal dependencies of soil arch formation-stabilization-degradation and tensile membrane effect take-off, enabling early prediction of key stages.

[0203] In this embodiment, to accurately capture the long-term dependence of soil arching effect degradation and the hysteresis characteristics of tensile membrane effect take-off, the temporal neural network model preferably adopts a long short-term memory network combined with a multilayer perceptron; the specific expressions and internal logic of the state update function and readout function are as follows:

[0204] First, the state update function simulates the "memory" and "accumulation" processes of physical effects within the embankment (e.g., the accumulation of differential settlement and the gradual decay of soil strength) by introducing cell states. At time... The state update process is jointly controlled by the forget gate, input gate, and output gate.

[0205] The forget gate is used to determine how much of the physical state information accumulated in the previous time step is retained, satisfying the following relationship:

[0206]

[0207] in, For a moment The forget gate vector, It is the Sigmoid activation function. This is the weight matrix. For a moment The hidden state, For a moment The input vector, This is a bias term.

[0208] The input gate and candidate cell states are used to determine the degree to which new inputs (such as increased sedimentation or changes in seepage head) update the overall physical state at the current moment, satisfying the following relationship:

[0209]

[0210]

[0211] in, For a moment The input gate vector, It is the Sigmoid activation function. This is the weight matrix. For a moment The hidden state, For a moment The input vector, For bias terms, Candidate cell state, It is the hyperbolic tangent activation function.

[0212] Cell state updates (representing the accumulation of intrinsic physical memory in the co-evolution of the soil arching effect and the tensile membrane effect) satisfy the following relationship:

[0213]

[0214] in, This represents the updated cell state. For a moment The forgetting gate, ⊙ represents the Hadama product, For a moment The state of the cells, For a moment The input gate, This represents the candidate cell state.

[0215] The generation of the output gate and the current hidden state satisfy the following relationship:

[0216]

[0217]

[0218] in, For a moment The output gate vector, It is the Sigmoid activation function. This is the weight matrix. For a moment The hidden state, For a moment The input vector, For bias terms, For a moment The hidden state, ⊙ represents the Hadama product. The hyperbolic tangent activation function is used. This represents the updated cell state.

[0219] Secondly, the readout function employs a multilayer perceptron network with non-linear activation to hide the high-dimensional state. Mapped to specific physical effect analysis results (i.e., predicted output) ), satisfying the following relationship:

[0220]

[0221] in, To predict the output, This is the weight matrix of the output layer. It is a linear rectified activation function. The weight matrix for the hidden mapping layer, For a moment The hidden state, To hide the bias term of the mapping layer, This is the bias term for the output layer.

[0222] Through the recursive update and nonlinear readout of the aforementioned hidden states, the model can not only learn the complex temporal dependencies of soil arch formation-stabilization-degradation and tensile membrane effect, but also implicitly characterize the cumulative process of soil internal damage or stress redistribution, thereby enabling early prediction of key failure stages.

[0223] 3) Construction of the loss function for the physical constraint learning model:

[0224] In this embodiment, to avoid non-physical predictions caused by relying solely on data fitting, and to ensure that the model output satisfies the basic mechanical constraints of the soil arching effect and the tensile membrane effect, a physical constraint term is introduced during the training phase. The objective function of the physical constraint learning model is constructed, satisfying the following relationship:

[0225]

[0226] in, This indicates minimizing the loss. For data fitting loss, These are the weighting coefficients. For boundary constraint loss, For consistency constraint loss, Loss due to monotonicity or trend constraint. The loss is due to constraints related to coupling relationships.

[0227] Data fitting loss: used to measure the predicted output Error between them.

[0228] Boundary constraint loss: used to constrain the range of values ​​for physical quantities, such as ensuring that contributing factors meet certain conditions. Risk output satisfies The tensile force and its vertical component are non-negative, and their forms satisfy the following relationship:

[0229]

[0230] in, For boundary constraint loss, For a moment, For out-of-bounds penalty function, For the tension of the grille, For the vertical component of the tension membrane, Factors contributing to the tensile membrane effect This serves as an indicator or probability output for the risk of arch instability.

[0231] Consistency constraint loss: used to embed known feature computational relationships to ensure consistency between the prediction results and measurable quantities, satisfying the following relationship:

[0232]

[0233] in, For consistency constraint loss, For a moment, For the vertical component of the tension membrane, For the tension of the grille, The included angle.

[0234] It should be noted that the included angle is at time. The real-time angle between the grid and the horizontal plane (or the horizontal plane at the top of the pile) after the grid is deformed under stress is obtained by inverting the grid geometric deformation data and vertical displacement data acquired by the multimodal sensing and monitoring module 56.

[0235] Monotonicity trend constraint loss: Used to characterize common engineering trends in this application scenario, such as the decreasing trend of soil arch strength and the increasing trend of tensile membrane response during the stages of enhanced differential settlement or enhanced seepage softening (can be triggered by "stage" or "condition"), satisfying the following relationship:

[0236]

[0237] in, Loss due to monotonicity or trend constraint. For a moment, This represents the set of indices for the degradation stage. For out-of-bounds penalty function, As an index of soil arching effect strength, For the time-varying quantity, Indicates the set of indexes for the takeover enhancement phase. For the tension of the grille.

[0238] It should be noted that the out-of-bounds penalty function employs a penalty mechanism based on modified linear units. Its core logic is to impose a penalty value greater than zero on predicted values ​​that do not conform to physical realities, while outputting zero for predicted values ​​that conform to physical laws. The specific expressions for different types of physical constraint variables are defined as follows:

[0239] First, regarding the lower bound constraint (applied to the grid tension T(t)≥0 and the vertical component of the tensile membrane force). (t)≥0): The out-of-bounds penalty function is defined as a one-sided penalty for negative values, satisfying the following relationship:

[0240]

[0241] in, For out-of-bounds penalty functions, This indicates taking the maximum value. These are the input variables for the function.

[0242] According to the above expression, when the model predicts a negative tensile force that violates the laws of physics, a quadratic loss penalty will be generated, forcing the model to correct the prediction direction.

[0243] Second, regarding interval constraints (applied to contributing factors γ(t)∈[0,1] and risk indicators) (t)∈[0,1]), then the out-of-bounds penalty function is defined as a two-sided out-of-bounds penalty, satisfying the following relationship:

[0244]

[0245] in, For out-of-bounds penalty functions, This indicates taking the maximum value. These are the input variables for the function.

[0246] This expression ensures that the dimensionless scaling factor and probability output are strictly limited to a physically and mathematically reasonable range of 0 to 1.

[0247] Third, for monotonicity or trend constraints (applied to...) (difference variable in the data)

[0248] During the degradation stage The soil arching effect strength index should show a decreasing trend; during the takeover stage The grid tension should show an upward trend. At this time, the variable passed to the boundary penalty function is actually the increment that violates the evolutionary trend. Therefore, the boundary penalty function is defined as a one-sided penalty for positive values, satisfying the following relationship:

[0249]

[0250] in, For out-of-bounds penalty functions, This indicates taking the maximum value. These are the input variables for the function.

[0251] It should be noted that the above out-of-bounds penalty function expressions all use the square form. The purpose of this is to ensure that the first derivative of the loss function is continuous at the boundary (such as x=0 or x=1), thereby avoiding gradient explosion or oscillation during the training process of the neural network and significantly improving the convergence stability of the physical constraint learning model.

[0252] Coupling constraint loss: Used to explicitly express the synergistic / substitution relationship between "soil arch degradation and tensile membrane splice," enabling the model to output prediction results that satisfy the coupling law. For example, the correlation constraint between the decrease in soil arch strength and the increase in contributing factors satisfies the following relationship:

[0253]

[0254] in, For coupling constraint loss, For a moment, As an index of soil arching effect strength, This is the coupling ratio coefficient. This is for predicting the output.

[0255] 4) Output definition of the arch instability risk indicator (used for "predicting critical failure stages"):

[0256] In this embodiment, to predict the risk of arch instability and the critical failure stage, the risk output is defined as a continuous value or a probability value. Taking the probability form as an example, the model output satisfies the following relationship:

[0257]

[0258] in, Indicates time Output the probability of instability / entering an unfavorable phase. For the Sigmoid function, This is the transpose of the weight matrix of the risk classification layer (fully connected layer). For a moment The hidden state, This is the bias term for the risk classification layer.

[0259] It should be noted that setting a high-risk threshold ,when The system determines when a high-risk phase has been entered and marks the moment when the threshold is first exceeded as the predicted moment of the critical damage phase.

[0260] In this embodiment, compared with the general time-series prediction model, the time-series neural network model of the present invention has at least the following improvements for the mechanism chain of "differential deformation-seepage softening-soil arch degradation-tensioned membrane connection":

[0261] First, dual-effect joint multi-task output: the soil arch effect index and the tensile membrane effect index are used as outputs at the same time, so that the model can learn the synergistic / substitution evolution law within the same framework, avoiding the problem that the single-effect model cannot explain the coupling transformation.

[0262] Second, introduce physical consistency constraints: explicitly embed physical constraint terms such as the relationship between the vertical component of the tensile membrane and the grid tension, and the range of values ​​of contributing factors into the loss function to reduce the risk of overfitting under small sample experimental data conditions and improve the physical interpretability of the prediction results;

[0263] Third, stage-triggered trend constraints and risk threshold determination: Based on the stage markers of settlement loading and seepage loading or the results of change point detection, different constraints (degradation / connection stage) are triggered to achieve early identification of critical failure stages and output risk indicators that can be used for engineering decision-making.

[0264] Fourth, closed-loop application for loading path guidance: The risk indicators output by the model and the prediction results of key stages are used to guide the subsequent loading path (e.g., controlling the settlement increment, seepage gradient increment or holding time in the next stage), thereby realizing a closed-loop test strategy of "prediction-adjustment-reprediction", improving test efficiency and enhancing the ability to capture adverse working conditions.

[0265] Through the above mathematical expressions and physical constraints, this embodiment can establish a mapping relationship between differential deformation conditions, reinforcement parameters, seepage conditions and soil arch / tension membrane response under the coupling effect of multiple factors, realize the prediction of the degree of soil arch formation, reinforcement stress state and arch instability risk, and use the prediction results to guide the subsequent loading path setting and key failure stage identification.

[0266] Furthermore, historical experimental data or multiple sets of experimental working condition data are used as model training data to train the machine learning model, thereby enabling the machine learning model to learn the evolution law of soil arching effect under the coupling effect of multiple factors, thus obtaining modeling results that can predict the state of soil arching effect under given experimental working conditions.

[0267] In an optional embodiment, the data analysis and prediction platform 61 is electrically connected to the multimodal sensing and monitoring module 56, which is used to perform time-series alignment and feature extraction on the collected multi-source data, construct a soil arching effect feature set, and establish a mapping relationship between differential deformation conditions, reinforcement parameters, seepage conditions and soil arching effect response based on a machine learning model, thereby realizing the prediction of the evolution state and instability risk of the soil arching effect.

[0268] Specifically, the data analysis and prediction platform 61 includes a data processing unit, a feature construction unit, and a prediction analysis unit. The data processing unit performs time alignment and preprocessing on multi-source data from the earth pressure sensor 53, displacement sensor, and reinforcement strain sensor 57 to ensure consistency across different data types in the time dimension. The feature construction unit extracts key parameters reflecting the stress and deformation characteristics of the soil and the stress state of the reinforcement material 54 from the processed data and constructs a feature set characterizing the soil arching effect. The prediction analysis unit establishes a correlation model between differential deformation conditions, reinforcement parameters, seepage conditions, and the soil arching effect response based on the feature set, thereby predicting the evolution trend and instability risk of the soil arching effect. A feedback control interface 62 is also provided on the data analysis and prediction platform 61.

[0269] The data analysis and prediction platform 61 can transform the multi-source monitoring information obtained from model tests into quantitative evaluation results of the soil arching effect, providing data support and prediction basis for analyzing the stability of reinforced soil structures under differential deformation and seepage conditions, thereby improving the engineering application value of the test results.

[0270] S5. Obtain the feature set of the physical effects to be analyzed, and combine it with the machine learning training model to obtain the results of the co-evolution analysis of physical effects.

[0271] In this embodiment, after completing the machine learning modeling step, the soil arching effect feature set obtained under real-time or given test conditions is used as the physical effect feature set to be analyzed. The feature set is input into the trained machine learning model to predict the evolution state of the soil arching effect under the combined action of differential deformation conditions, reinforcement parameters and seepage loading conditions, and output the corresponding soil arching effect response results to characterize the degree of formation, evolution trend or potential instability state of the soil arching effect.

[0272] Specifically, based on the mapping relationship, machine learning training models are used to predict the degree of soil arch formation, the stress state of the reinforced material 54, and the risk of soil arch effect instability under given test conditions, and the corresponding prediction results are output as physical effect co-evolution analysis results, which can be used to guide the setting of subsequent differential deformation loading paths, or to identify the stages in which key failures may occur during the evolution of the soil arch effect.

[0273] S6. Feedback is provided on the results of the co-evolution analysis of the physical effects.

[0274] In this embodiment, based on the prediction results obtained in the prediction step, the evolution state or risk assessment results of the soil arching effect are fed back to the test system for adjusting the differential deformation loading parameters, reinforcement parameters, or seepage loading conditions, or for analyzing and recording the stress and deformation response of the reinforced soil arching effect under test conditions, thereby realizing the result verification, parameter correction, or optimization setting of subsequent test conditions in the test process.

[0275] In an optional embodiment, the aim is to study the influence of different stiffness conditions of the reinforcing material 54 on the degree of tension membrane effect, the ability to maintain the soil arch effect, and the structural stress safety, thereby providing experimental basis for the rational selection of the stiffness of the reinforcing material 54.

[0276] Within the model box 41, the soil filling method, pile cap arrangement, and settlement loading mode are kept consistent. Three groups of reinforcing materials 54 with different equivalent stiffness are selected, representing low, medium, and high stiffness reinforcing materials 54, respectively. The number of layers, spacing, and anchoring method of each group of reinforcing materials 54 are kept consistent, only the equivalent axial stiffness parameter is changed. Under the action of the differential settlement induction module, a basin-shaped differential settlement condition is applied to the inter-pile area, while the pile cap area is kept relatively stable to simulate the stress state of a typical pile-supported reinforced embankment. No seepage conditions are applied during the settlement loading process to highlight the influence of reinforcement stiffness on the stress mechanism. The strain variation law of the reinforcing material 54 under various stiffness conditions with settlement development is collected through the multimodal sensing monitoring module 56; at the same time, the redistribution characteristics of soil pressure in the pile cap area and the inter-pile area, as well as the overall settlement and local deformation characteristics of the embankment, are collected. Based on the collected data, the contribution rate of the tension membrane effect, the contribution rate of the soil arching effect, and the strain concentration index of the reinforcing material 54 are constructed. Meanwhile, the collected results were compared under different reinforcement stiffness conditions: the timing of the tension membrane effect initiation; and the rate of soil arching degradation under loading and seepage conditions; and the stress growth rate and potential adverse stress stages of the reinforced material were evaluated.

[0277] Compared with existing technologies, this invention has at least the following beneficial effects: it can realistically construct multi-zone differential settlement and seepage coupling conditions at the model scale, closely resembling the actual stress environment of pile-supported reinforced embankments; and through multi-modal sensing monitoring, it achieves simultaneous observation and quantitative characterization of the soil arching effect and the tensile membrane effect of the reinforcement material; it can further reveal the synergistic evolution law of soil arching effect degradation and tensile membrane effect enhancement under different reinforcement stiffness conditions; and, in line with current trends, it can use machine learning models to predict key unfavorable stress stages and potential instability risks; and it provides experimental verification and intelligent analysis basis for the rational selection and optimization design of the stiffness parameters of the reinforcement material, avoiding material waste or safety hazards caused by improper stiffness selection in engineering.

[0278] Please see Figure 4In an optional embodiment, the present invention provides a physical effect co-evolution analysis system for pile-supported reinforced embankments. The system includes an input device, an output device, a processor, and a memory, all interconnected. The memory stores a computer program comprising program instructions. The processor is configured to invoke the program instructions to execute specific steps as described in the embodiments of the physical effect co-evolution analysis method for pile-supported reinforced embankments provided by the present invention. The physical effect co-evolution analysis system for pile-supported reinforced embankments provided by the present invention is structurally complete, objective, and stable, enhancing the overall applicability and practical application capability of the present invention.

[0279] In an optional embodiment, the present invention provides a multi-field coupled model test system for pile-supported reinforced embankments, comprising: a model box 41 main body for accommodating soil samples and reinforcing materials 54; a multi-zone differential deformation induction module 42 for forming preset differential settlement conditions; a reinforcement laying and stress control module for controlling the number of layers, spacing, and initial tension state of the reinforcing materials 54; a seepage loading module for applying seepage loading with multi-directional and variable boundary conditions; a multi-modal sensing and monitoring module 56 for synchronously acquiring soil pressure, displacement, and strain data of the reinforcing materials 54; and a data analysis and prediction platform 61 for constructing a feature set of soil arching effect and tensile membrane effect, and establishing a mapping relationship between differential deformation, reinforcement parameters, and structural response based on a machine learning model.

[0280] During the multi-field coupled model test, the following data were simultaneously collected by the multimodal sensing and monitoring module 56: earth pressure data in the pile cap area, arch foot, arch crown, and inter-pile area; vertical settlement displacement data on the embankment surface and inter-pile area; and strain data in the middle span of the reinforced material 54 and near the pile cap. This forms a multi-source test dataset containing time information. Based on the collected data, a machine learning platform was used to construct soil arching effect strength indices and tensile membrane effect characteristic parameters. Analysis was conducted under different settlement modes: the initial and stable stages of soil arching effect formation were quantified; the hierarchical characteristics of the tensile membrane effect during soil arching effect degradation were evaluated; and the synergistic or substitution relationship between soil arching effect and tensile membrane effect was assessed from the above embodiment. Based on this embodiment, the fundamental laws governing the synergistic evolution of soil arching effect and grid tensile membrane effect under differential settlement can be established.

[0281] In summary, the present invention provides a method and system for analyzing the synergistic evolution of physical effects in pile-supported reinforced embankments. Addressing existing research on pile-supported reinforced embankments, this invention grapples with challenges such as the difficulty in constructing realistic differential deformation conditions, the difficulty in quantifying the synergistic evolution of soil arching and the tensile membrane effect of the reinforcing material, and the lack of experimental basis for selecting the stiffness parameter of the reinforcing material. By constructing multi-zone differential settlement conditions and simultaneously applying seepage and external loads, this invention achieves controllable simulation of the degradation of the soil arching effect and the enhancement of the tensile membrane effect. Combined with multimodal sensor data and machine learning analysis methods, it enables quantitative evaluation of the structural stress and deformation response under different reinforcement stiffness conditions, thus providing experimental and predictive basis for the rational selection and optimized design of the stiffness of the reinforcing material. The method of this invention is easy to understand and convenient for engineering application, providing a theoretical foundation and technical support for the further development of intelligent testing technology in geotechnical engineering.

[0282] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for analyzing the synergistic evolution of physical effects in pile-supported reinforced embankments, characterized in that, Includes the following steps: The test conditions were constructed by fitting a pile-supported reinforced embankment using a model box, soil samples, and reinforcement materials, and based on the coupling effect of multiple factors. Based on the test conditions, a multi-source test dataset was obtained by using a sensor array to collect multi-modal data of the pile-supported reinforced embankment. Data preprocessing and feature extraction were performed on the multi-source experimental dataset to obtain the soil arching effect feature set and the tensile membrane effect feature set; A machine learning model is constructed based on the soil arching effect feature set and the tensile membrane effect feature set, and a machine learning training model is obtained through the model training data. Obtain the feature set of the physical effects to be analyzed, and combine it with the machine learning training model to obtain the results of the co-evolution analysis of physical effects; The test involves fitting a pile-supported reinforced embankment using a model box, soil samples, and reinforcing materials, and constructing test conditions based on the coupling effects of multiple factors, including: The soil samples are arranged in layers inside the model box, and the reinforcing material is simultaneously arranged at a preset height position to fit the pile-supported reinforced embankment. Differential settlement conditions are generated by using a multi-zone differential deformation induction module; External load conditions are obtained through an external load application device; Simulation of seepage conditions based on seepage loading module; The differential settlement condition, the external load condition, and the seepage condition are treated as multiple factors coupled together and combined to construct the test condition. The process of preprocessing and extracting features from the multi-source experimental dataset to obtain the soil arching effect feature set and the tensile membrane effect feature set includes: The multi-source experimental dataset is preprocessed to obtain multi-source experimental processed data. The data preprocessing includes time alignment, noise reduction, and normalization. Based on the multi-source experimental data, feature parameters are extracted, and the spatiotemporal variation patterns of the feature parameters are obtained; The feature parameters are combined and organized according to the spatiotemporal variation law to construct the soil arching effect feature set and the tensile membrane effect feature set; The extraction of feature parameters based on the multi-source experimental data includes: The tensile membrane effect is quantified by processing the multi-source test data to obtain the grid tension and the vertical component of the tensile membrane force; Based on the multi-source test data, the pressure values ​​of the pile cap area, the inter-pile area, and the average pressure value are obtained, and the soil arching effect is quantified to obtain the soil arching effect strength index. Based on the aforementioned soil arching effect, the contributing factors of the tensile membrane effect of the grid are obtained; The grid tension, the vertical component of the tensile membrane, the soil arching effect strength index, and the contributing factors are used as the characteristic parameters; The step of constructing a machine learning model based on the soil arching effect feature set and the tensile membrane effect feature set, and obtaining a machine learning training model through model training data, includes: The soil arch effect feature set and the tensile membrane effect feature set are used as model inputs, and the feature index of the co-evolution of physical effects is used as model output to establish the machine learning model. Historical test data or multiple sets of test condition data are obtained as training data for the model, and the machine learning model is trained to obtain the machine learning training model. The machine learning model includes: The machine learning model includes a temporal neural network model and a physical constraint learning model.

2. The method for co-evolution analysis of physical effects of pile-supported reinforced embankments according to claim 1, characterized in that, The differential settlement conditions include: The differential settlement conditions include stepped settlement mode, linear gradient settlement mode, settlement basin settlement mode, and cyclic settlement mode.

3. The method for co-evolution analysis of physical effects of pile-supported reinforced embankments according to claim 1, characterized in that, Based on the aforementioned test conditions, a multi-source test dataset is obtained by using a sensor array to collect multi-modal data of the pile-supported reinforced embankment, including: Earth pressure sensors, displacement sensors, and reinforcement strain sensors are used as the sensor array to form a multimodal sensing and monitoring module. Based on the test conditions, synchronous monitoring data of the pile-supported reinforced embankment was obtained through the multimodal sensing and monitoring module, including earth pressure data, displacement data and reinforcement strain data. The synchronous monitoring data is collected according to a unified time base and an additional time identifier is added to form the multi-source experimental dataset.

4. The method for co-evolution analysis of physical effects of pile-supported reinforced embankments according to claim 1, characterized in that, The process of obtaining the feature set of physical effects to be analyzed and combining it with the machine learning training model to obtain the results of the co-evolution analysis of physical effects includes: The soil arching effect feature set and the tensile membrane effect feature set under the real-time or given test conditions are used as the physical effect feature set to be analyzed. Based on the physical effect feature set to be analyzed, the state prediction of the pile-supported reinforced embankment is performed according to the machine learning training model to obtain the results of the physical effect co-evolution analysis.

5. A physical effect co-evolution analysis system for pile-supported reinforced embankments, characterized in that, The system includes an input device, an output device, a processor, and a memory, which are interconnected. The memory stores a computer program, which includes program instructions. The processor is configured to invoke the program instructions to execute the physical effect co-evolution analysis method for pile-supported reinforced embankments as described in any one of claims 1-4.

Images