A karst cave pile foundation construction slurry leakage early warning management system based on slurry monitoring

Abstract

This invention discloses a grout leakage early warning management system for karst pile foundation construction based on mud monitoring. The invention relates to the field of pile foundation construction technology in karst areas, and includes: a multi-source dynamic mud sensing unit for real-time acquisition of multiple dynamic state parameters of mud within the pile hole; and a coupled behavior modeling and risk prediction unit connected to the multi-source dynamic mud sensing unit, used to output grout leakage risk prediction information based on the multiple dynamic state parameters and a pre-stored geological model through a built-in analysis model. This grout leakage early warning management system for karst pile foundation construction based on mud monitoring constructs a risk identification mechanism that surpasses traditional single pressure and flow monitoring. Its multi-layered verification logic, through cross-comparison and confidence fusion of model prediction results, real-time geological data, and adjacent hole conditions, effectively filters out interference signals, improves the reliability and accuracy of early warning signals, and makes the early identification of grout leakage risks more sensitive.

Description

Technical Field

[0001] This invention relates to the field of pile foundation construction technology in karst areas, specifically to a grout leakage early warning and management system for karst cave pile foundation construction based on mud monitoring. Background Technology

[0002] Currently, industry monitoring and early warning of grout leakage in pile foundation construction largely relies on manual experience, periodic borehole exploration, or threshold alarms for single parameters. For example, patent publication CN120429940B describes a method and system for early warning of grout leakage in karst pile foundations. This method acquires geological data through ground-penetrating radar, electrical resistivity tomography, and borehole exploration, and constructs a three-dimensional model based on pressure and flow monitoring for risk prediction and early warning. This technical solution achieves multi-source data fusion and visualized monitoring to a certain extent, improving the systematic nature of grout leakage early warning. However, the aforementioned existing technologies still have several limitations: First, their monitoring relies on a pre-set fixed sensor network, which is insufficient for real-time sensing of the dynamic behavior of mud during construction, such as changes in flow rate, flow direction, and viscosity, making it difficult to capture early signs of sudden or gradual grout leakage; second, the system focuses on the pre-analysis of geological structure and grouting parameters, lacking continuous, online monitoring and intelligent analysis of the state of the mud itself during grouting, such as sand content, consistency, and pressure pulsation, resulting in limited timeliness and accuracy of early warnings; third, existing solutions mostly focus on passive responses of "early warning-disposal," failing to form an adaptive adjustment mechanism for construction processes based on real-time feedback of mud properties, and lacking specificity for emergency response under complex karst networks and dynamic hydrological conditions.

[0003] Therefore, how to break through the excessive reliance on static geological data in existing technologies and build an intelligent slurry leakage management system that takes real-time mud monitoring as its core, can deeply integrate fluid dynamic characteristics and geological models, and realize the entire process from early warning to adaptive regulation has become an urgent technical problem to be solved in the field of safety control of karst pile foundation construction. Summary of the Invention

[0004] The purpose of this invention is to provide a grout leakage early warning and management system for karst pile foundation construction based on mud monitoring, so as to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a grout leakage early warning management system for karst cave pile foundation construction based on mud monitoring, comprising: The mud multi-source dynamic sensing unit is used to collect multiple dynamic state parameters of mud in the pile hole in real time. The coupled behavior modeling and risk prediction unit is connected to the mud multi-source dynamic sensing unit and is used to output mud leakage risk prediction information based on the multiple dynamic state parameters and the pre-stored geological model through the built-in analysis model. A multi-level verification and early warning decision unit is connected to the coupled behavior modeling and risk prediction unit. It is used to verify the leakage risk prediction information based on multi-level logic and generate graded early warning instructions based on the verification results. The adaptive control unit is connected to the multi-layer verification and early warning decision unit and the on-site grouting equipment, respectively. It is used to call the pre-stored control strategy library to output construction parameter adjustment instructions to the on-site grouting equipment according to the received early warning instructions, and to verify the effect based on the adjusted mud state parameters.

[0006] Furthermore, the multi-source dynamic sensing unit for mud includes: Pressure sensors are deployed around the pile and at key nodes of the pre-designed karst channels to monitor mud pressure. A flow sensor, installed in the grouting pipeline, is used to monitor the grouting flow rate; The mud property sensor group, set in the mud circulation loop, includes a viscosity sensor for online monitoring of mud viscosity and a sand content sensor for monitoring the content of solid particles in the mud.

[0007] Furthermore, the mud multi-source dynamic sensing unit also includes a borehole visual monitoring module, which is used to acquire mud surface images at the pile borehole opening and calculate the mud surface height change rate through an image recognition algorithm.

[0008] Furthermore, the analysis model built into the coupled behavior modeling and risk prediction unit is a fusion model, which is constructed by embedding the mud fluid dynamics equation as a constraint condition into a time-series neural network; the coupled behavior modeling and risk prediction unit is configured to: spatially correlate the multiple dynamic state parameter sequences acquired in real time with the geological model, input them into the fusion model, and output a risk prediction map containing the probability of mud leakage and the spatial distribution of potential mud leakage areas.

[0009] Furthermore, the multi-layered verification and early warning decision unit is configured to execute a verification decision process including the following steps: S1: Receive the grout leakage risk prediction information, perform a self-check of the first layer model output, determine whether the predicted grout leakage probability continuously exceeds the first threshold to reach a preset number of cycles, and simultaneously determine whether there are any abrupt changes in the multiple dynamic state parameters that exceed the corresponding change threshold. S2: If both conditions in S1 are met simultaneously, a primary warning is triggered and the second layer of multi-source data cross-validation is initiated; the second layer of multi-source data cross-validation includes: spatially comparing the potential grout leakage area with real-time geological survey data, and analyzing the correlation of changes in mud state parameters of adjacent pile holes. S3: Based on the verification results of S2, calculate the comprehensive early warning confidence level through evidence fusion algorithm, and determine the final early warning level according to the range to which the comprehensive early warning confidence level belongs. The early warning level includes at least observation level, early warning level and action level.

[0010] Furthermore, the multi-layer verification and early warning decision unit is also configured to: after determining the early warning level to be an early warning level, initiate the third-layer virtual scenario simulation; the virtual scenario simulation includes: simulating virtual adjustments to at least one grouting construction parameter in the geological model, and simulating the changes in mud state after adjustment based on the fusion model; if the simulation results show a decrease in the grout leakage risk index, then generating a construction parameter adjustment suggestion corresponding to the virtual adjustment, and attaching it to the early warning command.

[0011] Furthermore, the adaptive control unit includes a control strategy knowledge base and a closed-loop verification module; The control strategy knowledge base pre-stores the mapping relationship between abnormal patterns of mud state parameters under different geological scenarios and recommended grouting construction parameter adjustment strategies. The adaptive control unit is configured to: match or generate construction parameter adjustment instructions from the control strategy knowledge base according to the level and content of the early warning instruction, and send them to the on-site grouting equipment; The closed-loop verification module is configured to: after the construction parameter adjustment command is executed, continuously monitor the mud state parameters within a subsequent predetermined time period; if the risk indicators are detected to be continuously decreasing and tending to stabilize, generate a verification pass signal and feed it back to the multi-layer verification and early warning decision unit to trigger a downgrade of the early warning level.

[0012] Furthermore, the control strategy knowledge base has a self-learning function and is configured to: record the abnormal pattern of mud state parameters that triggers adjustment in each early warning event, the executed construction parameter adjustment command, and the verification result of the closed-loop verification module; when the same or similar abnormal pattern reappears, the adjustment strategy with a historically passed verification result is given priority recommendation.

[0013] Furthermore, it also includes a three-dimensional visualization interactive terminal, which is connected to the coupled behavior modeling and risk prediction unit and the multi-layer verification and early warning decision-making unit, and is used to dynamically display the geological model, the real-time spatial distribution cloud map of mud state parameters, the risk prediction map, and the current early warning level and control status.

[0014] Furthermore, the early warning management process constructed by the system forms a closed loop. The process is as follows: the mud multi-source dynamic sensing unit acquires mud dynamic data in real time, the coupled behavior modeling and risk prediction unit analyzes the data to obtain an initial risk signal, the multi-layer verification and early warning decision unit performs multiple filtering and confirmation and outputs graded instructions, and finally the adaptive control unit performs targeted adjustments and verifies the effect. The adjusted state is collected again by the sensing unit and enters the next cycle, thereby realizing fully automated control from risk perception, intelligent decision-making to execution feedback.

[0015] This invention provides a grout leakage early warning and management system for karst cave pile foundation construction based on mud monitoring. It has the following beneficial effects: This grout leakage early warning management system for karst pile foundation construction, based on mud monitoring, introduces real-time sensing of multi-dimensional dynamic properties such as mud viscosity and sand content, and combines this with geological models for fluid behavior coupling modeling, constructing a risk identification mechanism that surpasses traditional single pressure and flow monitoring. Its multi-layered verification logic, through cross-comparison and confidence fusion of model prediction results, real-time geological data, and adjacent borehole conditions, effectively filters out interference signals, improving the reliability and accuracy of early warning signals, and making early identification of grout leakage risks more acute.

[0016] This grout leakage early warning and management system for karst pile foundation construction based on mud monitoring establishes a complete management loop, from intelligent early warning to adaptive parameter control, and further optimizes strategies through closed-loop verification and case learning. This enables continuous dynamic intervention and autonomous optimization of the construction process. This closed-loop mechanism not only provides precise guidance for handling risks when they occur, but also enhances the system's adaptability to complex working conditions through accumulated practical knowledge. This strengthens overall control over construction quality and safety throughout the entire process, reducing reliance on manual experience. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the early warning decision-making logic of a grout leakage early warning management system for karst pile foundation construction based on mud monitoring, according to the present invention. Figure 2 This is an adaptive control closed-loop logic diagram of a grout leakage early warning management system for karst pile foundation construction based on mud monitoring, according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 and Figure 2 This invention provides a technical solution: a grout leakage early warning management system for karst cave pile foundation construction based on mud monitoring, comprising: The mud multi-source dynamic sensing unit is used to collect multiple dynamic state parameters of mud in the pile hole in real time. The coupled behavior modeling and risk prediction unit is connected to the mud multi-source dynamic sensing unit. It is used to output mud leakage risk prediction information based on multiple dynamic state parameters and pre-stored geological models through the built-in analysis model. The multi-level verification and early warning decision unit, connected to the coupled behavior modeling and risk prediction unit, is used to verify the leakage risk prediction information based on multi-level logic and generate graded early warning instructions based on the verification results. The adaptive control unit is connected to the multi-layer verification and early warning decision unit and the on-site grouting equipment. It is used to call the pre-stored control strategy library to output construction parameter adjustment instructions to the on-site grouting equipment according to the received early warning instructions, and to verify the effect based on the adjusted mud state parameters.

[0020] It should be further explained that the core of the system lies in constructing a closed-loop control system that integrates real-time sensing, dynamic modeling, intelligent decision-making, and proactive regulation. Specifically, the multi-source dynamic sensing unit for mud is implemented by deploying sensor arrays at key locations within the pile hole and along the grouting circulation pipeline. This array includes not only pressure sensors for monitoring slurry pressure and flow sensors for monitoring grouting rate, but more importantly, it integrates specialized sensors for online continuous monitoring of the physical properties of the mud, such as a mud viscometer based on vibration principles and a mud sand content monitor based on optical scattering principles. This allows for the real-time acquisition of multi-dimensional dynamic state parameter sequences, reflecting the essential fluid behavior of the mud, such as viscosity, density, and solid content.

[0021] The coupled behavior modeling and risk prediction unit can be implemented in hardware using an industrial computer or an embedded server. The industrial computer used in this unit must have a CPU with eight or more cores and a clock speed of at least 3.0 GHz, a GPU with at least 8 GB of video memory, at least 16 GB of RAM, and at least 512 GB of storage. The embedded server must have a dedicated edge computing processor with a clock speed of at least 2.0 GHz, at least 8 GB of RAM, and at least 256 GB of storage to ensure the real-time requirements of model calculation and data storage. Its built-in analysis model is a hybrid intelligent model that integrates the fluid dynamics control equations of mud in porous media as physical constraints with neural networks capable of learning temporal dependencies, such as long short-term memory networks. During operation, the unit couples the real-time inflow of multi-dimensional dynamic state parameter sequences with a pre-imported three-dimensional geological digital model containing the spatial distribution of karst caves to simulate the seepage and pressure transmission process of mud under complex geological conditions, thereby outputting a risk prediction map containing quantitative leakage probability values ​​and spatial coordinates of suspected leakage paths.

[0022] The multi-layered verification and early warning decision-making unit, as the "intelligent hub" of the system, first performs a first-layer self-consistency test on the initial risk signal output by the model, that is, simultaneously judging the persistence of the probability exceeding the limit and whether there is a coordinated mutation in the original sensor data. If it passes, the second layer of external evidence fusion is initiated, such as calling the phased scanning data of the ground-penetrating radar for spatial cross-verification. Based on this, the unit dynamically determines the final early warning level, such as observation, early warning, and action, through a confidence fusion algorithm based on Bayesian theory.

[0023] The core of the adaptive control unit is a knowledge base of control strategies containing a large number of cases. This knowledge base establishes a mapping relationship from "specific mud anomaly patterns" to "recommended construction parameter adjustment schemes." Among them, the recommended construction parameter adjustment schemes include adjusting the grouting pressure curve and changing the grout mix ratio. When a warning command is received, the unit automatically matches or generates specific control commands and sends them to the on-site grouting equipment for execution. Among them, the control commands include adjusting the frequency of the variable frequency pump and controlling the opening of the admixture addition valve. After execution, the system continuously monitors the mud parameter response for the next few minutes through a closed-loop verification module. If the risk indicators are confirmed to decrease, a verification success signal is fed back. The system learns and stores this successful case in the database for use in optimizing future decisions. This forms a complete intelligent closed loop from real-time status perception, fusion model prediction, multi-level credibility verification, to precise parameter intervention and effect feedback learning, realizing a fundamental transformation in the management of grout leakage risk from passive warning to proactive adaptive control.

[0024] The mud multi-source dynamic sensing unit includes: Pressure sensors are deployed around the pile and at key nodes of the pre-designed karst channels to monitor mud pressure. A flow sensor, installed in the grouting pipeline, is used to monitor the grouting flow rate; The mud property sensor group, set in the mud circulation loop, includes a viscosity sensor for online monitoring of mud viscosity and a sand content sensor for monitoring the content of solid particles in the mud.

[0025] It should be further explained that the specific implementation of the multi-source dynamic sensing unit for mud is as follows: the deployment of pressure sensors should be based on geological exploration data and construction design, and differentiated at the interface between soil and rock layers around the pile body, as well as at key upstream and downstream nodes of potential karst channels identified by the three-dimensional geological model, to form a spatial monitoring network capable of capturing pressure gradient anomalies.

[0026] The preferred flow sensor is an electromagnetic flowmeter or an ultrasonic flowmeter, which is directly installed in the main grouting pipeline to monitor the volumetric flow rate of the grout injected into the formation in real time. Its data is strictly synchronized with the pressure data in time. The mud property sensor group is a key component that distinguishes this solution from traditional solutions. Among them, the viscosity sensor uses a rotary or vibratory online viscometer, whose measuring probe is directly immersed in the mud circulation tank or bypass measuring pipe, continuously outputting a viscosity signal related to the frictional resistance within the mud. The sand content sensor uses an online monitoring instrument based on the principles of light scattering or ultrasonic attenuation. By analyzing the intensity of light or sound wave signals scattered by the mud flowing through the measuring unit in real time, it directly calculates the volume concentration of solid particles in the mud.

[0027] These sensors are connected to the data acquisition module via an industrial bus network. The data acquisition module synchronously acquires the raw signals from all sensors at a fixed sampling period, such as once per second, and performs preliminary analog-to-digital conversion and filtering. The pressure sensor in the mud multi-source dynamic sensing unit has a range of 0-10 MPa and an accuracy class of 0.2; the viscosity sensor has a range of 10-1000 mPa·s and an accuracy of ±2%; and the sand content sensor has a range of 0-50% and an accuracy of ±1%. The sampling frequency of all sensors is uniformly set to once per second. This sampling frequency can meet the dynamic change monitoring requirements of mud state parameters in karst pile foundation construction. Because the changes in mud state parameters during karst pile foundation construction have a certain lag, and the fluctuation period of on-site construction conditions is usually greater than 1 second, sampling once per second can completely capture the dynamic change trend of parameters, which meets the standard requirements of "continuous online monitoring" in the engineering field and can provide continuous and effective data support for subsequent risk prediction. Finally, a structured data frame containing timestamps, pressure values, flow values, viscosity values, and sand content values ​​is generated and uploaded to the subsequent modeling and analysis unit of the system in real time.

[0028] By synchronously collecting and fusing multiple parameters such as pressure, flow rate and mud physical properties, the system constructs a multi-dimensional state vector that reflects the relationship between the three factors of mud pressure, flow rate and physical properties, providing a richer and more direct data foundation that goes beyond single pressure or flow rate monitoring for accurately judging the risk of mud leakage.

[0029] The mud multi-source dynamic sensing unit also includes a borehole visual monitoring module, which is used to collect images of the mud surface at the pile borehole and calculate the rate of change of mud surface height through image recognition algorithms.

[0030] Further explanation is needed regarding the specific implementation of the borehole visual monitoring module: This module includes a fixed industrial camera with a protection level meeting outdoor industrial requirements. The camera is vertically suspended at a certain height directly above the borehole opening via a bracket, ensuring its field of view completely covers the borehole area. The camera continuously acquires color or grayscale video streams of the borehole opening at a set frame rate. At the image processing level, the system's built-in processor runs a specific image recognition algorithm. This algorithm first preprocesses each frame of the image, including grayscale conversion, noise reduction, and contrast enhancement. Then, it uses an edge detection algorithm to identify the distinct edge lines formed at the interface between the mud surface and air in the image. By calibrating the relative positional relationship between the camera and the borehole opening, the pixel coordinates of the liquid surface edge in the image are converted into actual physical height values.

[0031] To calculate the rate of change of mud level, the algorithm continuously tracks the difference in mud level between adjacent frames and, combined with the time interval of video frames, calculates the average change in mud level per unit time in real time, i.e., the rate of change. This visual monitoring data, as an independent monitoring dimension, is aligned and fused in time with sensor data such as pressure, flow rate, viscosity, and sand content. When mud leakage occurs, the volume of mud in the borehole may be lost rapidly, causing an abnormally increased rate of mud level drop. This visual signal can be cross-compared and verified with the injection flow rate displayed by the pipeline flow meter and the pressure sensor signal, thereby providing auxiliary judgment evidence from different physical principles for multi-layer verification and early warning decision-making units. This enhances the system's reliability and robustness in perceiving mud leakage events, especially sudden and rapid mud leakage.

[0032] The analysis model built into the Coupled Behavior Modeling and Risk Prediction Unit is a fusion model, which is constructed by embedding the mud fluid dynamics equation as a constraint into a time-series neural network. The Coupled Behavior Modeling and Risk Prediction Unit is configured to spatially correlate multiple dynamic state parameter sequences acquired in real time with the geological model, input them into the fusion model, and output a risk prediction map containing the probability of mud leakage and the spatial distribution of potential mud leakage areas.

[0033] It should be further explained that the specific implementation of the built-in fusion model in the coupled behavior modeling and risk prediction unit is as follows: The construction of the fusion model is first based on the principle of mass conservation and momentum conservation of mud flow in pore-fracture media, and a set of partial differential equations describing the relationship between pressure propagation and mud loss is established as a physical constraint framework; at the same time, a time series prediction model with a long short-term memory network as the basic architecture is constructed. The input layer of this network is designed to receive a multidimensional mud state parameter sequence arranged in time order.

[0034] The key to this fusion lies in using the physical residuals generated after discretizing the partial differential equations as regularization constraints during the model training phase. These constraints are incorporated into the total loss function of the neural network training, ensuring that while the network learns statistical patterns from historical data, its output must satisfy the fundamental governing equations of fluid mechanics as much as possible. The dataset used to train this model comes from historical pile foundation construction cases, containing complete time-series data of mud parameters, corresponding 3D geological model data, and annotations indicating whether grout leakage occurred.

[0035] In actual system operation, this unit receives structured data streams from the sensing unit in real time and organizes them into a sequence with fixed time windows. Simultaneously, it calls a pre-stored three-dimensional geological grid model representing the spatial distribution of strata and karst caves at the current construction pile location. The modeling unit correlates and maps the current mud state sequence with the grid attributes of the geological model, and then inputs this joint data into the trained fusion model. After calculation, the model outputs not only a probability value between zero and one, representing the possibility of slurry leakage in the near future, but also a three-dimensional probability distribution field corresponding to the spatial coordinates of the geological model. This distribution field uses different colors or values ​​to visually indicate the potential risk of slurry loss at each location in three-dimensional space, forming a risk prediction map.

[0036] This modeling approach, which combines deterministic physical laws with data-driven intelligent learning, enhances the predictive generalization ability and interpretability of results for conditions that have not undergone complex geological conditions.

[0037] The training dataset for the fusion model in the coupled behavior modeling and risk prediction unit comes from 100 complete engineering cases of pile foundation construction in karst areas over the past 5 years. Each case includes no less than 3,000 sets of time-series data of mud state parameters and corresponding geological models and leakage result annotations. The temporal neural network of the fusion model adopts a long short-term memory network architecture. The network has 5 hidden layers, with 128 neurons in each layer. The total loss function in the training phase is composed of a weighted average of the temporal prediction loss and the physical residual loss after discretization of the mud fluid dynamics equation. The weight coefficients are set to 0.6:0.4 according to the geological complexity of the engineering scenario. The model training iterations are 1,000 times, and the batch size is 64. After training, the accuracy of leakage risk prediction is no less than 92% as verified by the test set.

[0038] In the evidence fusion algorithm, the prior probability of Bayesian fusion is determined based on the historical occurrence rate of grout leakage events under the same geological conditions in the same region. If there is no historical data in the region, the default prior probability value is 0.15. When performing multi-source evidence fusion, the geological overlap evidence and the correlation evidence between adjacent boreholes are converted into likelihood values ​​in the 0-1 interval, and then combined with the prior probability to complete the calculation of the comprehensive early warning confidence.

[0039] The multi-layered verification and early warning decision unit is configured to execute a verification decision process that includes the following steps: S1: Receive grout leakage risk prediction information, perform self-check of the first layer model output, determine whether the predicted grout leakage probability continuously exceeds the first threshold to reach the preset number of cycles, and simultaneously determine whether there are abrupt changes in multiple dynamic state parameters that exceed the corresponding change threshold. S2: If both conditions in S1 are met simultaneously, a primary warning is triggered and the second layer of multi-source data cross-validation is initiated. The second layer of multi-source data cross-validation includes: spatially comparing the potential grout leakage area with real-time geological survey data, and analyzing the correlation of changes in mud state parameters of adjacent pile holes. S3: Based on the verification results of S2, calculate the comprehensive warning confidence level through evidence fusion algorithm, and determine the final warning level according to the range to which the comprehensive warning confidence level belongs. The warning level includes at least observation level, warning level and action level.

[0040] It should be further explained that the specific implementation of the verification decision process by the multi-layer verification and early warning decision unit is as follows: In the S1 model output self-check step, the unit receives the slurry leakage probability value from the prediction unit. This probability value is continuously compared with a pre-set probability threshold. When the probability value is found to be higher than this threshold for several consecutive sampling periods, it is considered to meet the probability continuous over-limit condition. At the same time, the unit performs online analysis on the raw data sequence of slurry pressure, flow rate, and viscosity parameters acquired in real time. By calculating the difference between adjacent time points or using the sliding window statistical method, it is determined whether the change of any parameter exceeds the change threshold set for it. For example, if the pressure decreases by more than a certain set value in a single sampling period, it is determined that there is a sudden change feature. Only when the two conditions of probability continuous over-limit and at least one key parameter sudden change are met synchronously in time will the primary early warning be triggered and enter S2.

[0041] In the S2 multi-source data cross-validation step, the system first accesses the stored or real-time received ground-penetrating radar scan data, which expresses recent changes in the subsurface medium in the form of point clouds or images. The unit performs spatial overlay analysis on the three-dimensional coordinates of the potential grout leakage area output by the prediction unit and the abnormal area reflected by the ground-penetrating radar data, and calculates the degree of overlap between the two as a verification evidence. At the same time, the system acquires the real-time mud pressure data of all adjacent pile holes under construction, calculates the correlation coefficient between the current pile hole pressure change and the pressure change of adjacent pile holes. If the correlation coefficient is high, it indicates that the change may be caused by regional stratum disturbance rather than isolated grout leakage. This analysis serves as another verification evidence.

[0042] In the S3 integrated early warning decision-making step, the unit adopts an evidence fusion framework based on Bayesian theory, which transforms multiple inputs, such as the self-check pass status in S1, the geological overlap evidence obtained in S2, and the correlation evidence between adjacent boreholes, into a unified likelihood. The final integrated early warning confidence is calculated by combining prior probabilities. This confidence is mapped to several pre-divided continuous numerical intervals, each interval corresponding to an early warning level. For example, a value in the middle interval corresponds to the "early warning level", while a value in a higher interval corresponds to the "action level" which requires immediate human intervention, thus completing the transformation from multi-source information to clear hierarchical instructions.

[0043] The multi-layer verification and early warning decision unit is also configured to: after determining the early warning level to be the early warning level, initiate the third-layer virtual scenario simulation; the virtual scenario simulation includes: simulating the virtual adjustment of at least one grouting construction parameter in the geological model, and simulating the change of mud state after adjustment based on the fusion model; if the simulation results show that the grout leakage risk index has decreased, then generating construction parameter adjustment suggestions corresponding to the virtual adjustment and attaching them to the early warning command.

[0044] It should be further explained that the specific implementation of the virtual scenario simulation steps is as follows: When the warning level is determined to be a warning level, the unit automatically activates the virtual simulation engine; the simulation engine first copies a complete set of mud state parameters and the corresponding geological model snapshot from the current real-time data to construct a virtual simulation environment that runs parallel to the real system. In this environment, the engine automatically generates a series of virtual adjustment schemes for grouting construction parameters based on a predefined rule base, such as simulating reducing the set pressure value of the main grouting pipeline by a fixed step, or simulating adding a certain amount of thickener to the mud circulation system to increase viscosity.

[0045] Subsequently, the simulation engine invokes the pre-trained fusion model, using the adjusted virtual parameters as new input conditions. Combined with the geological model, it rapidly simulates the state evolution of the virtual system over a future period with a shorter time step, and calculates the virtual grout leakage risk index corresponding to the end of the simulation. The engine compares and analyzes the risk index after the simulation with the initial risk index before the simulation. If the comparison results show that, under at least one virtual adjustment scheme, the risk index exhibits a clear downward trend and stabilizes at a low level, the simulation engine determines that the adjustment scheme is effective. It then selects the scheme with the most obvious risk reduction effect and encapsulates its specific parameter adjustment content, such as "reducing the grouting pressure setpoint by X MPa," into a structured construction parameter adjustment suggestion.

[0046] This suggestion, as an additional, optional intelligent decision support information, is immediately embedded into the warning instructions that will be sent to the adaptive control unit and on-site operators. This provides an optimized operation path based on forward-looking simulation verification before triggering actual equipment actions or manual intervention, thereby enhancing the scientific nature and proactivity of the system's decision-making.

[0047] The adaptive control unit includes a control strategy knowledge base and a closed-loop verification module; The control strategy knowledge base pre-stores the mapping relationship between abnormal patterns of mud state parameters under different geological scenarios and recommended grouting construction parameter adjustment strategies; The adaptive control unit is configured to match or generate construction parameter adjustment instructions from the control strategy knowledge base according to the level and content of the early warning instructions, and send them to the on-site grouting equipment. The closed-loop verification module is configured to continuously monitor the mud state parameters within a predetermined time period after the construction parameter adjustment command is executed. If the risk indicators are continuously decreasing and tending to stabilize, a verification pass signal is generated and fed back to the multi-layer verification and early warning decision unit to trigger the downgrade of the early warning level.

[0048] It should be further explained that the specific implementation of the adaptive control unit is as follows: The control strategy knowledge base is constructed by combining a relational database with a case reasoning framework. Its data tables store a large number of records abstracted from historical construction cases. Each record contains at least several key fields: the geological type code when the case occurred, the feature vector of the specific abnormal pattern presented by parameters such as mud pressure, flow rate, viscosity, and sand content when the warning is triggered, the set of specific control measures instructions taken at that time, such as the grouting pump frequency adjustment value, the admixture valve opening time, the change in the target pressure setting value, and the risk indicator change curve and the final verification conclusion recorded by the closed-loop verification module over a subsequent period after the control measures are implemented, including whether it is effective or ineffective.

[0049] The mapping relationship between the pre-stored abnormal patterns of mud state parameters and the adjustment strategies for grouting construction parameters in the control strategy knowledge base includes multiple specific cases. For example, when the mud viscosity is detected to drop sharply by 30% within 5 minutes and the sand content increases by 5% simultaneously, the control strategy is to add a viscosity enhancer accounting for 0.8%-1.2% of the total mud mass into the mud circulation loop, while simultaneously reducing the grouting pressure by 0.3-0.5 MPa. When the mud pressure at a key node around the pile is detected to drop by more than 1.2 MPa within 3 sampling cycles and the flow rate does not change significantly, the control strategy... Slightly increase the grouting flow rate by 10%-15% and adjust the water-cement ratio of the grout from 0.6 to 0.5; when the orifice visual monitoring module shows that the mud level change rate exceeds 0.5m / h and the pressure sensor data drops synchronously, the control strategy is to immediately start the grout replenishment process of the backup mud tank, set the replenishment rate to 1.2m³ / h, and at the same time increase the grouting pressure by 0.2-0.4MPa. The above control parameters have been verified by more than 50 sets of engineering tests, which can effectively realize the correction of mud state and control the risk of grout leakage.

[0050] When the unit receives an instruction from the early warning decision unit, its core matching engine first analyzes the current real-time mud parameters and extracts the current abnormal pattern feature vector. Then, it searches the knowledge base for several historical records with the same geological type code and the closest abnormal pattern feature vector. By analyzing the correlation between control measures and verification conclusions in these historical records, the matching engine generates one or more recommended construction parameter adjustment instructions and sends them to the programmable logic controller or dedicated actuator of the on-site grouting equipment.

[0051] Meanwhile, the closed-loop verification module is activated and starts a monitoring window of a preset duration. During this window, the module continuously obtains the latest mud state parameters from the sensing unit and calculates the moving average of key risk indicators. Among them, the key risk indicators include the real-time mud leakage probability output by the fusion model. The module has a preset judgment logic: if the average value of the risk indicators at the end of the monitoring window decreases by more than a set threshold compared with the baseline value before the control command is executed, and the indicator data fluctuates stably throughout the window period without a reverse upward trend, then the control is judged to have "passed verification".

[0052] The verified signal will be immediately fed back to the early warning decision unit, which can then downgrade the current early warning level, for example, from "early warning level" to "observation level". At the same time, the successful complete sequence of "current abnormal mode - execution instruction - verification passed" will be added to the control strategy knowledge base as a new case to optimize future matching recommendations.

[0053] The control strategy knowledge base has a self-learning function and is configured to: record the abnormal pattern of mud state parameters that triggers adjustment in each early warning event, the construction parameter adjustment command executed, and the verification result of the closed-loop verification module; when the same or similar abnormal pattern reappears, the adjustment strategy with a historically passed verification result will be recommended first.

[0054] It should be further explained that the specific implementation of the self-learning function of the control strategy knowledge base is as follows: This self-learning function is implemented through a continuously running background process that monitors every complete early warning and control event of the system. When a control action is completed and the closed-loop verification module generates a verification conclusion, the process automatically captures all relevant data of the current event. This data is encapsulated into a structured case object. This object includes the following fields: unique event identifier, timestamp, geological scene classification label, multi-parameter abnormal pattern vector extracted and normalized from the mud multi-source dynamic sensing unit when the early warning is triggered, the set of specific control instruction codes issued by the adaptive control unit, and the final verification status flag output by the closed-loop verification module.

[0055] The case was then added to the case table of the control strategy knowledge base. The knowledge base's retrieval and recommendation engine incorporates a similarity-based matching algorithm. When a new warning event occurs, the engine first calculates the distance between the current real-time anomaly pattern vector and the historical anomaly pattern vectors of all cases with the same geological scene in the knowledge base. The algorithm uses Euclidean distance or cosine similarity to measure the closeness between vectors.

[0056] The engine will filter out several historical cases with the smallest distance and check the verification status flags associated with these cases. If there are cases with a verification status of "passed", the engine will extract the set of control instructions corresponding to these cases and sort them according to their similarity to the current pattern. The control scheme with the highest similarity and that has passed verification will be given priority as the recommended strategy output. If the verification status of all similar cases is "failed", the engine may activate the backup strategy generation mechanism or prompt manual intervention to design a new scheme.

[0057] Through this mechanism, the system continuously accumulates effective handling experience tested in field practice during operation, and the accuracy and reliability of the knowledge base recommendations are continuously improved. This enables the system to provide empirically proven control suggestions more quickly and accurately when facing abnormal working conditions under repeated or similar geological conditions, forming an intelligent decision support system with the ability to accumulate experience and evolve.

[0058] It also includes a 3D visualization interactive terminal, which is connected to the coupled behavior modeling and risk prediction unit and the multi-layer verification and early warning decision-making unit. It is used to dynamically display the geological model, the real-time spatial distribution cloud map of mud state parameters, the risk prediction map, and the current early warning level and control status.

[0059] It should be further explained that the specific implementation of the 3D visualization interactive terminal is as follows: The terminal hardware is typically deployed in the construction site command center or supervision office, consisting of a workstation computer with graphics processing capabilities and a high-resolution monitor. The software runs a 3D rendering engine specifically developed for geological and construction data visualization. The terminal establishes data communication links with the system's modeling and prediction units and early warning decision-making units via wired or wireless networks, receiving structured data streams from these units in real time.

[0060] At the visualization level, the terminal software first loads and renders a pre-stored 3D geological model of the pile location, using different colors and textures to distinguish soil layers, rock layers, and karst cavities. The spatial distribution cloud map of real-time mud state parameters is generated through an interpolation algorithm: the system maps the readings of each pressure sensor distributed around the pile and inside the borehole to the corresponding 3D coordinate points on the geological model. For areas without sensors, spatial estimation algorithms such as Kriging interpolation are used to generate a continuous pressure field distribution, which is then overlaid on the 3D model as a semi-transparent color cloud map, with the color gradually changing from blue to red to correspond to the change in pressure from low to high.

[0061] The Kriging interpolation used to generate the spatial distribution cloud map of mud state parameters in the 3D visualization interactive terminal uses a spherical model for its variogram function, sets the interpolation search radius to 3 times the diameter of the pile hole, and sets the interpolation grid resolution to 0.5m×0.5m×0.5m. Through these parameter settings, accurate fitting and intuitive presentation of the spatial distribution of mud state parameters can be achieved.

[0062] The risk prediction map is displayed using the three-dimensional probability distribution field data output by the modeling unit, drawn in the terminal through another visual channel, with high-risk areas highlighted prominently. This other visual channel includes different levels of transparency or isosurfaces. The current warning level is displayed prominently on the interface as a combination of text labels and color blocks, while the control status is shown through dynamically updated lists or flowcharts, such as "Executing pressure reduction instructions" or "Control verification in progress."

[0063] The terminal supports user interaction, including rotating, scaling, and cutting the profile of the 3D model, as well as clicking on any location to query detailed geological attributes, real-time sensor readings, or predicted risk values, thus providing construction managers with an intuitive, comprehensive, and interactive global situational awareness and decision support interface.

[0064] The system's early warning management process forms a closed loop. The process is as follows: the multi-source dynamic sensing unit of mud acquires dynamic data of mud in real time, and the coupled behavior modeling and risk prediction unit analyzes the data to obtain the initial risk signal. Then, the multi-layer verification and early warning decision-making unit performs multiple filtering and confirmation before outputting graded instructions. Finally, the adaptive control unit performs targeted adjustments and verifies the effect. The adjusted state is collected again by the sensing unit and enters the next cycle, thereby realizing fully automated control from risk perception, intelligent decision-making to execution feedback.

[0065] It should be further explained that the specific implementation of the closed-loop early warning management process constructed by the system is as follows: The process starts from the mud multi-source dynamic sensing unit synchronously collecting on-site pressure, flow rate, viscosity and sand content data at a fixed sampling period, and processing the orifice visual image to obtain the liquid level change rate. These preprocessed and time-aligned multidimensional dynamic state parameters constitute a real-time data stream, which is continuously input to the coupled behavior modeling and risk prediction unit.

[0066] Upon receiving the data stream, this unit immediately performs spatial correlation and fusion calculations with a pre-stored 3D geological model. Through the built-in fusion model processing, it outputs a risk prediction map containing quantified probabilities and spatial distributions, thus completing the transformation from raw data to an initial risk signal. Next, the multi-layer verification and early warning decision-making unit receives this initial signal and executes its multi-layer verification logic: first, it performs a self-check of the model output to confirm the persistence of the risk signal and the co-mutation of the raw data; if successful, it further cross-verifies with real-time geological scan data and the status of adjacent pile holes, and finally calculates the comprehensive confidence level and determines the early warning level through an evidence fusion algorithm.

[0067] Based on this level, the adaptive control unit is activated. Its control strategy knowledge base matches or generates specific construction parameter adjustment instructions according to the current geological scene and mud anomaly patterns, and issues them to the execution mechanism of the on-site grouting equipment. After the instructions are executed, the system does not end. Instead, the closed-loop verification module within the adaptive control unit initiates a preset duration effect monitoring window. During this period, it continuously collects the adjusted mud state parameters and assesses the changing trends of risk indicators.

[0068] If monitoring results show that risk indicators are declining and stabilizing, a verification signal will be fed back to the early warning decision-making unit, triggering a dynamic downgrade of the early warning level. Simultaneously, the successful "condition-measure-effect" complete sequence will be stored in the knowledge base as a new case study. If the effect is not satisfactory, the system may maintain or raise the early warning level and try other strategies. This verified new status data, along with continuously collected real-time data, immediately serves as input for the next cycle, re-entering the perception, modeling, decision-making, and control process. In this way, data and instructions continuously flow and interact among the five stages of perception, analysis, decision-making, execution, and feedback, forming a continuous, cyclical automated control loop that enables the system to continuously track, evaluate, and adaptively intervene in the dynamic construction process.

[0069] Regarding the sampling frequency of the multi-source dynamic sensing unit for mud, from an engineering practice perspective, in pile foundation construction in karst areas, the changes in key properties of mud such as viscosity and sand content are a gradual process. The parameter precursor change cycle before sudden slurry leakage is usually in the range of several seconds to tens of seconds. A sampling frequency of once per second can completely cover this change cycle and can capture the sudden change characteristics of parameters in a timely manner. Compared with the common sampling frequency of similar monitoring systems in the industry: 0.5 times / second to 2 times / second, the sampling frequency of this system is in a reasonable and efficient range, fully meeting the engineering standard of "continuous online monitoring" and effectively solving the technical problem of insufficient real-time sensing capability of mud dynamic behavior.

[0070] The system's "fully automated control" features a clear mechanism for switching between automation and manual intervention. When the multi-level verification and early warning decision-making unit determines the early warning level to be observation or early warning, the system automatically executes the construction parameter adjustment instructions issued by the adaptive control unit, requiring no manual intervention throughout the process. When the early warning level reaches the action level, the system immediately triggers the switching mechanism, with a switching threshold of a comprehensive early warning confidence level ≥ 0.85. At this point, the system pauses the automatic control process and pushes audible and visual alarm signals and risk details to the on-site construction command center through a 3D visualization interactive terminal. Simultaneously, it pushes the construction parameter adjustment suggestions generated by the adaptive control unit. The system resumes execution after the on-site operator confirms the instructions or inputs new control instructions. During manual intervention, the system continuously collects mud state parameters and provides real-time feedback. If the risk indicators drop to the early warning level or below after manual intervention, the system automatically switches back to the fully automated control mode. If the risk indicators remain at the action level, the system updates the control suggestions every 5 minutes until the risk is eliminated. This seamless process ensures both the automation level of control and the necessity of manual decision-making under high-risk conditions, achieving an organic unity between automation and manual intervention.

[0071] By introducing real-time sensing of multi-dimensional dynamic properties such as mud viscosity and sand content, and combining this with geological models for coupled fluid behavior modeling, a risk identification mechanism that surpasses traditional single pressure and flow monitoring has been constructed. Its multi-layered verification logic, through cross-comparison and confidence fusion of model prediction results, real-time geological data, and adjacent well status, effectively filters out interference signals, improving the reliability and accuracy of early warning signals, and making early identification of slurry leakage risks more acute.

[0072] The system establishes a complete management loop, from intelligent early warning to adaptive parameter control, and further optimizes strategies through closed-loop verification and case learning. This enables continuous dynamic intervention and autonomous optimization of the construction process. This closed-loop mechanism not only provides precise guidance for handling risks when they occur, but also enhances the system's adaptability to complex working conditions by accumulating practical knowledge. As a result, it strengthens the overall control over construction quality and safety throughout the entire process, reducing reliance on human experience.

[0073] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0074] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A grout leakage early warning management system for karst cave pile foundation construction based on mud monitoring, characterized in that, include: The mud multi-source dynamic sensing unit is used to collect multiple dynamic state parameters of mud in the pile hole in real time. The coupled behavior modeling and risk prediction unit is connected to the mud multi-source dynamic sensing unit and is used to output mud leakage risk prediction information based on the multiple dynamic state parameters and the pre-stored geological model through the built-in analysis model. A multi-level verification and early warning decision unit is connected to the coupled behavior modeling and risk prediction unit. It is used to verify the leakage risk prediction information based on multi-level logic and generate graded early warning instructions based on the verification results. The adaptive control unit is connected to the multi-layer verification and early warning decision unit and the on-site grouting equipment, respectively. It is used to call the pre-stored control strategy library to output construction parameter adjustment instructions to the on-site grouting equipment according to the received early warning instructions, and to verify the effect based on the adjusted mud state parameters.

2. The grout leakage early warning management system for karst cave pile foundation construction based on mud monitoring as described in claim 1, characterized in that: The mud multi-source dynamic sensing unit includes: Pressure sensors are deployed around the pile and at key nodes of the pre-designed karst channels to monitor mud pressure. A flow sensor, installed in the grouting pipeline, is used to monitor the grouting flow rate; The mud property sensor group, set in the mud circulation loop, includes a viscosity sensor for online monitoring of mud viscosity and a sand content sensor for monitoring the content of solid particles in the mud.

3. The grout leakage early warning management system for karst cave pile foundation construction based on mud monitoring as described in claim 2, characterized in that: The mud multi-source dynamic sensing unit also includes a borehole visual monitoring module, which is used to collect mud surface images at the pile borehole opening and calculate the mud surface height change rate through an image recognition algorithm.

4. The grout leakage early warning management system for karst cave pile foundation construction based on mud monitoring as described in claim 1, characterized in that: The analysis model built into the coupled behavior modeling and risk prediction unit is a fusion model, which is constructed by embedding the mud fluid dynamics equation as a constraint into a time-series neural network. The coupled behavior modeling and risk prediction unit is configured to: spatially correlate the multiple dynamic state parameter sequences acquired in real time with the geological model, input the data into the fusion model, and output a risk prediction map containing the probability of mud leakage and the spatial distribution of potential mud leakage areas.

5. A grout leakage early warning management system for karst cave pile foundation construction based on mud monitoring as described in claim 1, characterized in that: The multi-layered verification and early warning decision unit is configured to execute a verification decision process that includes the following steps: S1: Receive the grout leakage risk prediction information, perform a self-check of the first layer model output, determine whether the predicted grout leakage probability continuously exceeds the first threshold to reach a preset number of cycles, and simultaneously determine whether there are any abrupt changes in the multiple dynamic state parameters that exceed the corresponding change threshold. S2: If both conditions in S1 are met simultaneously, a primary warning is triggered, and the second-level multi-source data cross-validation is initiated. The second layer of multi-source data cross-validation includes: spatially comparing potential grout leakage areas with real-time geological survey data, and analyzing the correlation of changes in mud state parameters of adjacent pile holes; S3: Based on the verification results of S2, calculate the comprehensive early warning confidence level through evidence fusion algorithm, and determine the final early warning level according to the range to which the comprehensive early warning confidence level belongs. The early warning level includes at least observation level, early warning level and action level.

6. A grout leakage early warning management system for karst cave pile foundation construction based on mud monitoring as described in claim 5, characterized in that: The multi-layer verification and early warning decision unit is further configured to: after determining the early warning level to be an early warning level, initiate the third-layer virtual scenario simulation; the virtual scenario simulation includes: simulating virtual adjustments to at least one grouting construction parameter in the geological model, and simulating changes in the mud state after adjustment based on the fusion model; if the simulation results show a decrease in the grout leakage risk index, then generating a construction parameter adjustment suggestion corresponding to the virtual adjustment, and attaching it to the early warning command.

7. A grout leakage early warning management system for karst cave pile foundation construction based on mud monitoring as described in claim 1, characterized in that: The adaptive control unit includes a control strategy knowledge base and a closed-loop verification module; The control strategy knowledge base pre-stores the mapping relationship between abnormal patterns of mud state parameters under different geological scenarios and recommended grouting construction parameter adjustment strategies. The adaptive control unit is configured to: match or generate construction parameter adjustment instructions from the control strategy knowledge base according to the level and content of the early warning instruction, and send them to the on-site grouting equipment; The closed-loop verification module is configured to: after the construction parameter adjustment command is executed, continuously monitor the mud state parameters within a subsequent predetermined time period; if the risk indicators are detected to be continuously decreasing and tending to stabilize, generate a verification pass signal and feed it back to the multi-layer verification and early warning decision unit to trigger a downgrade of the early warning level.

8. A grout leakage early warning management system for karst cave pile foundation construction based on mud monitoring, as described in claim 7, is characterized in that: The control strategy knowledge base has a self-learning function and is configured to: record the abnormal pattern of mud state parameters that triggers adjustment in each early warning event, the construction parameter adjustment command executed, and the verification result of the closed-loop verification module; when the same or similar abnormal pattern reappears, the adjustment strategy with a historically passed verification result is given priority recommendation.

9. A grout leakage early warning management system for karst cave pile foundation construction based on mud monitoring as described in claim 1, characterized in that: It also includes a three-dimensional visualization interactive terminal, which is connected to the coupled behavior modeling and risk prediction unit and the multi-layer verification and early warning decision-making unit, and is used to dynamically display the geological model, the real-time spatial distribution cloud map of mud state parameters, the risk prediction map, and the current early warning level and control status.

10. A grout leakage early warning management system for karst cave pile foundation construction based on mud monitoring according to any one of claims 1 to 9, characterized in that: The early warning management process constructed by the system forms a closed loop. The process is as follows: the multi-source dynamic sensing unit of mud acquires dynamic data of mud in real time, the coupled behavior modeling and risk prediction unit analyzes the data to obtain an initial risk signal, the multi-layer verification and early warning decision unit performs multiple filtering and confirmation and outputs graded instructions, and finally the adaptive control unit performs targeted adjustments and verifies the effect. The adjusted state is collected again by the sensing unit and enters the next cycle, thereby realizing the fully automated control from risk perception, intelligent decision-making to execution feedback.

Images