Intelligent sensing and predictive maintenance system and method for underground building wall leakage

Abstract

The present application relates to the technical field of underground building structure health monitoring, and particularly relates to an underground building wall leakage intelligent perception and predictive maintenance system and method. The system comprises a perception layer, a transmission layer, a processing layer, an application layer and an intelligent decision layer. The processing layer comprises a physical information neural network prediction module. The present application provides an intelligent monitoring technical solution capable of fusing multi-modal sensing data, embedding physical law constraints and realizing predictive maintenance, so as to solve the core problems of traditional technology, such as early detection lag, insufficient multi-parameter fusion and lack of physical constraints.

Description

Technical Field

[0001] This invention relates to the field of underground building structural health monitoring technology, and more particularly to an intelligent sensing and predictive maintenance system and method for underground building wall leakage. Background Technology

[0002] In the field of underground construction engineering, wall leakage is a key hidden danger threatening structural safety and service life. Existing underground building leakage monitoring technologies generally employ a single type of sensor, such as resistive humidity sensors combined with manual inspections or timed data collection. This approach only triggers an alarm after leakage occurs and humidity exceeds a threshold, exhibiting significant detection lag. This traditional solution relies on single-point sensor deployment and cannot effectively identify early signs of leakage, such as the initiation of micro-cracks within the wall and hidden seepage paths. For example, traditional technologies are completely unable to capture non-humidity-driven early signs of leakage, such as elastic wave signals or ion concentration changes generated when 0.1mm-level cracks initiate. As urban underground space development becomes deeper, larger, and more complex, the seepage environment faced by underground building structures is becoming increasingly diverse, including different types of seepage sources such as groundwater and industrial wastewater, as well as complex leakage risks induced by structural deformation. Traditional technologies lack the ability to fuse and analyze multiple parameters, making it impossible to construct a complete leakage diagnosis system using multi-dimensional data such as acoustic emission, strain, and ion concentration. Furthermore, they struggle to embed physical laws, such as fluid dynamics equations, into monitoring models, leading to complete failure of leakage prediction in data-sparse areas or under extreme conditions. Currently, underground building maintenance is largely reliant on reactive, reactive repairs, resulting in frequent emergency repairs and high maintenance costs. Moreover, the inability to predict leakage risks in advance can exacerbate structural damage and cause equipment failures, among other safety incidents.

[0003] Based on the above problems, there is an urgent need for an intelligent monitoring technology solution that can integrate multimodal sensing data, embed physical law constraints, and achieve predictive maintenance, in order to solve the core problems of traditional technologies such as lagging early detection, insufficient multi-parameter fusion, and lack of physical constraints. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and proposes an intelligent sensing and predictive maintenance system for underground building wall leakage. This system includes a sensing layer, a transmission layer, a processing layer, an application layer, and an intelligent decision-making layer. The processing layer includes a physical information neural network prediction module, and the prediction model constructed by this module satisfies the following mathematical relationship:

[0005]

[0006] Where L is the loss function, used to measure the degree of agreement between the model's predictions and the actual values ​​and physical constraints; α and β are the weight coefficients of the data fitting term and the physical constraint term, respectively, and α + β = 1; N is the number of monitoring data samples; y i For the i-th sample, the actual leakage-related physical quantity observation value includes at least one of humidity, strain or ion concentration; R(u) represents the model prediction value for the i-th sample; M represents the number of sampling points for the physical constraint equations; j ) represents the physical equation residual at the j-th sampling point, where the physical equation includes Darcy's law equation or the seepage continuity equation.

[0007] Preferably, the sensing layer includes an acoustic emission sensor array, an electrochemical sensor network, a distributed optical fiber sensing device, and a MEMS multi-parameter sensor. The acoustic emission sensor array has a sampling frequency ≥1MHz and can identify the initiation of cracks at the 0.1mm level. The electrochemical sensor network integrates Cl-, SO42- ion concentration and pH value sensors to determine the type of seepage source. The distributed optical fiber sensing device is laid along the structural seam to achieve strain and temperature field monitoring with a spatial resolution of 0.5m along the fiber direction. The sampling interval of the MEMS multi-parameter sensor is 10 seconds, and the detection accuracy reaches ±1%RH and ±0.1℃.

[0008] Preferably, the transmission layer adopts a hybrid network architecture. Key data such as acoustic emission waveforms and fiber strain are transmitted using 5G URLLC with a latency of ≤10ms. Conventional sensor data are transmitted via LoRaWAN with a coverage radius of ≥1km. The transmission layer also includes an AI-enabled Mesh self-healing network, where nodes dynamically optimize routes based on reinforcement learning, and the communication interruption recovery time is ≤5 seconds.

[0009] Preferably, the application layer includes an AR interactive interface and a three-dimensional risk heat map. The AR interactive interface combines UWB positioning with BIM model, enabling inspection personnel to view sensor data inside the wall, historical leakage points, and predicted risk areas through AR glasses, with a positioning accuracy of 3cm. The three-dimensional risk heat map is based on the leakage probability distribution output by a graph neural network and renders the risk heat map of underground buildings in real time.

[0010] Preferably, the construction of the graph neural network model satisfies the following mathematical relationship:

[0011]

[0012] in, This represents the feature vector of node v in the l-th layer of the graph neural network; c represents the set of neighboring nodes of node v; uv W represents the normalized coefficient from node u to node v. (l) and σ is the weight matrix of the l-th layer; σ is the activation function.

[0013] Preferably, the intelligent decision-making layer includes a predictive maintenance strategy engine and a closed-loop verification mechanism. The predictive maintenance strategy engine automatically generates maintenance work orders based on a risk-cost optimization model, combining leakage probability, repair costs, and downtime losses. The maintenance work orders include recommended materials and processes. The closed-loop verification mechanism continuously monitors data after repair and evaluates the effect by comparing changes in humidity and strain before and after repair.

[0014] Preferably, the multiphysics simulation model of the digital twin engine satisfies the following mathematical relationship:

[0015]

[0016] Where ρ is the density of the medium; c is the specific heat capacity; T is the temperature; t is the time; k is the thermal conductivity; Q is the heat source intensity; g is the gravitational acceleration; and v is the fluid velocity.

[0017] Preferably, the multimodal sensing fusion diagnostic technology uses Kalman filtering and Transformer network for feature-level fusion to construct a four-dimensional monitoring system of sound, light, electricity and force, and integrates multi-source data such as acoustic emission, fiber optic strain, humidity and ion concentration.

[0018] Preferably, the sensor deployment method is as follows: in high-risk areas such as expansion joints and through-wall pipes, a dense deployment of acoustic emission sensors, distributed optical fibers and electrochemical sensors is used, with a spacing of ≤2m; in general areas, MEMS multi-parameter sensors are arranged with a spacing of 5-8m.

[0019] A method for intelligent sensing and predictive maintenance of underground building wall leakage, employing the intelligent sensing and predictive maintenance system for underground building wall leakage as described in any of the above-mentioned embodiments, including:

[0020] The sensing layer collects multi-source data in real time, including acoustic emission signals, strain, temperature, humidity, and ion concentration from the underground building walls.

[0021] The transport layer transmits the collected data to the processing layer through a hybrid network architecture;

[0022] The processing layer utilizes cloud-edge collaborative computing, and processes and analyzes the data through a physical information neural network prediction module and a digital twin engine to extract feature vectors and predict leakage diffusion paths.

[0023] The application layer displays monitoring and prediction results through an AR interactive interface and a 3D risk heat map;

[0024] The intelligent decision-making layer generates predictive maintenance strategies based on the prediction results, forming a closed loop of perception, diagnosis, decision-making, and execution.

[0025] Technical effects:

[0026] This invention utilizes a five-layer architecture and a physical information neural network prediction module, integrating multimodal sensor data and embedding physical law constraints. It can identify micro-cracks as small as 0.1mm and hidden seepage paths, solving the detection lag problem of traditional technologies; it uses a physical information neural network to predict leakage risks 72 hours in advance, overcoming the generalization limitations of data-driven models; and it constructs a digital twin closed-loop maintenance system, reducing emergency repair and maintenance costs, achieving a shift from passive alarm to proactive prediction, and improving the intelligence and reliability of underground building leakage monitoring. Attached Figure Description

[0027] Figure 1 This is a block diagram of the intelligent sensing and predictive maintenance system for underground building wall leakage in this application;

[0028] Figure 2 This is a flowchart of the intelligent sensing and predictive maintenance method for underground building wall leakage in this application. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0030] Please see Figure 1 Traditional technical solutions suffer from the following problems: Traditional leakage monitoring systems rely on a single sensor, only able to alarm after a leak occurs, unable to predict it in advance, and suffer from issues such as detection lag, ambiguous location, and insufficient multi-parameter fusion. Based on this, please refer to... Figure 1 This embodiment provides an intelligent sensing and predictive maintenance system for building wall leakage, including a sensing layer, a transmission layer, a processing layer, an application layer, and an intelligent decision-making layer. The processing layer includes a physical information neural network prediction module, and the prediction model constructed by the physical information neural network prediction module satisfies the following mathematical relationship:

[0031]

[0032] Where L is the loss function, used to measure the degree of agreement between the model's predictions and the actual values ​​and physical constraints; α and β are the weight coefficients of the data fitting term and the physical constraint term, respectively, and α + β = 1; N is the number of monitoring data samples; y i For the i-th sample, the actual leakage-related physical quantity observation value includes at least one of humidity, strain or ion concentration; R(u) represents the model prediction value for the i-th sample; M represents the number of sampling points for the physical constraint equations; j) represents the physical equation residual at the j-th sampling point, where the physical equation includes Darcy's law equation or the seepage continuity equation.

[0033] This formula is the loss function of the physical information neural network prediction module, used to balance the optimization objectives between data-driven prediction and physical law constraints. Its core lies in coupling the data fitting ability of traditional neural networks with the physical laws of seepage, solving the model generalization problem in the data-sparse scenario of underground building leakage prediction.

[0034] The overall architecture of the loss function consists of two parts: a data fitting term and a physical constraint term, which are weighted and summed using weighting coefficients α and β. This design breaks through the limitation of traditional neural networks that rely solely on data-driven approaches. By embedding Darcy's law and other seepage field physical equations as regularization terms into the model, the prediction results not only conform to the characteristics of the monitoring data but also satisfy the basic laws of fluid mechanics. This makes it particularly suitable for predicting leakage risks in the initial stages of new underground construction or in areas where data collection is limited.

[0035] Data Fitting Term Analysis: y represents the data fitting term. Here, α is the weighting coefficient of the data fitting term, and its value must satisfy α + β = 1. Its physical meaning is to measure the importance of the monitoring data in model optimization. N represents the number of monitoring data samples, covering multimodal sensor data such as humidity, strain, and ion concentration. i The actual observed value of the i-th sample is acquired in real time by the sensor in the perception layer. Let be the model's predicted value for the i-th sample.

[0036] This section measures the deviation between the model's predictions and the actual monitoring data using mean squared error (MSE), ensuring that the model can learn the dynamic changes in leakage-related physical quantities.

[0037] Analysis of physical constraint terms: This represents the physical constraint term. β is the weighting coefficient of the physical constraint term, complementary to α. When monitoring data is insufficient, β needs to be increased to strengthen the constraint of physical laws on the model. M represents the number of sampling points for the physical constraint equations, usually determined based on the grid division of the underground structure or key monitoring locations. R(u j ) represents the residual of the physical equation at the j-th sampling point, specifically involving Darcy's law equation describing the seepage law of fluid in porous media or the seepage continuity equation describing the seepage law of fluid in porous media. This part forces the model prediction results to conform to the physical nature of the seepage field by minimizing the residual of the physical equation. For example, it ensures that the predicted water flow velocity and pressure gradient satisfy the linear relationship of Darcy's law, avoiding false predictions that violate physical laws.

[0038] This formula achieves an organic integration of data-driven and physics-driven approaches through the dynamic adjustment of α and β. In underground building leakage monitoring, when early signs of leakage, such as microcracks, appear in the wall, the model can capture signals of crack propagation through acoustic emission sensor data and ensure that the predicted seepage path conforms to the fluid permeation law in concrete pores through physical constraints. This allows for the prediction of leakage risk up to 72 hours in advance, improving prediction accuracy by 40% in new building scenarios compared to traditional purely data-driven models.

[0039] This solution addresses the limitations of single-sensor detection capabilities through multimodal sensing fusion, enabling the identification of hidden seepage paths. It also overcomes the poor generalization ability of traditional data-driven models in sparse data regions by utilizing PINNs to embed physical laws. Furthermore, it solves the problems of low efficiency, high maintenance costs, and inability to dynamically optimize strategies in manual inspections by leveraging digital twin closed-loop maintenance.

[0040] It is worth mentioning that this embodiment proposes an intelligent sensing and predictive maintenance system for underground building wall leakage, employing a five-layer architecture design encompassing a sensing layer, transmission layer, processing layer, application layer, and intelligent decision-making layer. Specifically, a physical information neural network prediction module is set up in the processing layer. The prediction model constructed by this module balances data fitting and physical constraints through a loss function L. In the loss function, the data fitting term considers the deviation between the predicted and actual values ​​of the monitored data samples, while the physical constraint term optimizes the residuals of physical equations such as Darcy's law. The weighting coefficients α and β adjust the proportion of both. The system integrates multimodal sensor fusion, PINNs prediction, and digital twin maintenance technologies to form a complete closed loop for leakage management.

[0041] The technical effects achieved by the above embodiments include: This system constructs a four-dimensional monitoring system integrating sound, light, electricity, and force, enabling multi-dimensional and full-process monitoring of leakage in underground building walls. The physical information neural network prediction module ensures high prediction accuracy even in data-sparse scenarios, providing a reliable basis for leakage risk prediction. The digital twin-driven closed-loop maintenance system enables intelligent generation and verification of maintenance strategies, forming an efficient maintenance management process. The system can identify leakage risks in advance, accurately locate leakage sources, provide sufficient time for maintenance decisions, improve the safety and reliability of underground building structures, and reduce maintenance costs and accident losses.

[0042] Traditional technical solutions suffer from the following problems: traditional monitoring systems use single-point sensors, resulting in limited monitoring dimensions and an inability to comprehensively capture changes in physical quantities related to leakage. They are also insensitive to early signs of leakage, such as microcracks, and struggle to determine the type of seepage source, impacting subsequent maintenance strategy development. Therefore, the sensing layer comprises an acoustic emission sensor array, an electrochemical sensor network, a distributed fiber optic sensing device, and a MEMS multi-parameter sensor. The acoustic emission sensor array has a sampling frequency ≥1MHz, capable of identifying the initiation of cracks as small as 0.1mm. The electrochemical sensor network integrates Cl-, SO42- ion concentration, and pH sensors to determine the type of seepage source. The distributed fiber optic sensing device is laid along structural seams, achieving strain and temperature field monitoring with a spatial resolution of 0.5m along the fiber direction. The MEMS multi-parameter sensor has a sampling interval of 10 seconds, achieving a detection accuracy of ±1%RH and ±0.1℃.

[0043] This solution addresses the limitations and insufficient accuracy of single-sensor monitoring by using multiple sensors in synergy, enabling high-sensitivity monitoring of multiple parameters in leaks and providing data support for early detection and accurate diagnosis.

[0044] It is worth mentioning that: the sensing layer deploys multiple sensors, with an acoustic emission sensor array sampling frequency ≥1MHz, capable of identifying the initiation of cracks at the 0.1mm level, used to monitor the elastic wave signal of the propagation of micro-cracks inside the wall; the electrochemical sensor network integrates Cl-, SO42- ion concentration and pH value sensors, embedded inside the wall, to determine the type of seepage source by changes in ion mobility; distributed optical fiber sensors are laid along the structural seams, using BOTDR+DTS technology to achieve strain and temperature field monitoring with a spatial resolution of 0.5m; MEMS multi-parameter sensors are embedded, with a sampling interval of 10 seconds, and a detection accuracy of ±1%RH and ±0.1℃, integrating humidity, temperature, and micro-strain monitoring functions.

[0045] The technical effects achieved by the above embodiments include: a multi-dimensional monitoring network composed of multiple sensors; acoustic emission sensors that can capture weak signals of micro-crack initiation, enabling early warning of leakage; electrochemical sensors that can accurately determine the nature of the seepage source, providing a basis for selecting appropriate repair materials and processes; distributed fiber optic sensing that enables continuous monitoring along structural joints, timely detection of leakage risks induced by structural deformation; and MEMS sensors that perform high-frequency sampling and high-precision detection, ensuring real-time acquisition of changes in the wall environment and structural state. The data from each sensor complement each other, improving the system's ability to identify leakage and its monitoring reliability, providing rich and accurate data for subsequent analysis and decision-making.

[0046] Traditional technical solutions suffer from the following problems: traditional monitoring systems mostly use wired transmission networks, which are complex, inflexible, and susceptible to interference in complex underground environments, leading to communication interruptions. Data transmission real-time performance and reliability are also insufficient, affecting the overall performance of the monitoring system. The proposed transmission layer adopts a hybrid network architecture. Key data such as acoustic emission waveforms and fiber optic strain are transmitted using 5G URLLC with a latency of ≤10ms; conventional sensor data are transmitted via LoRaWAN, with a coverage radius of ≥1km. The transmission layer also includes an AI-enabled Mesh self-healing network, where nodes dynamically optimize routes based on reinforcement learning, achieving a communication interruption recovery time of ≤5 seconds.

[0047] This solution addresses the adaptability issues of traditional transmission methods in underground structures through a hybrid network architecture and an AI self-healing network, meeting the transmission needs of different types of data and improving the stability and real-time performance of system communication.

[0048] It's worth noting that the transport layer employs a hybrid network architecture. Critical data, such as acoustic emission waveforms and fiber optic strain gauges, which require high real-time performance, are transmitted using 5G URLLC with a latency of ≤10ms. Regular sensor data is transmitted via LoRaWAN, covering a radius of ≥1km. Furthermore, the transport layer includes an AI-powered mesh self-healing network. Nodes dynamically optimize routes based on reinforcement learning, ensuring a recovery time of ≤5 seconds in the event of a communication interruption, thus guaranteeing network stability and reliability.

[0049] The technical effects achieved by the above embodiments include: the hybrid network architecture classifies and transmits data according to its importance and real-time requirements; critical data is transmitted with low latency through 5G URLLC, ensuring timely delivery of early warning information in emergency situations; LoRaWAN enables long-distance transmission of regular data, reducing the communication costs of large-scale sensor deployments; and the AI-enabled Mesh self-healing network enables the system to have dynamic routing optimization capabilities in complex underground building environments, quickly recovering from communication interruptions, reducing data loss, ensuring the continuity and integrity of monitoring data, and providing a reliable data transmission channel for subsequent processing and analysis.

[0050] Traditional technical solutions suffer from the following problems: traditional monitoring systems present data in a limited way, mostly in the form of tables or two-dimensional charts, making it difficult to intuitively show the spatial distribution and risk level of leakage in underground building walls. This hinders inspectors from quickly locating problem areas and affects maintenance efficiency. Therefore, the application layer includes an AR interactive interface and a 3D risk heat map. The AR interactive interface combines UWB positioning and BIM models, allowing inspectors to view sensor data inside the walls, historical leakage points, and predicted risk areas through AR glasses, with a positioning accuracy of up to 3cm. The 3D risk heat map is based on the leakage probability distribution output by a graph neural network, rendering a real-time risk heat map of the underground building.

[0051] This solution addresses the problem of insufficient data visualization by using AR interaction and 3D heat map display, thereby improving the human-computer interaction experience and the efficiency of risk assessment.

[0052] It's worth mentioning that the application layer includes an AR interactive interface and a 3D risk heat map. The AR interactive interface combines UWB positioning with BIM models, allowing inspection personnel to view sensor data inside the walls, historical leakage points, and predicted risk areas through AR glasses, with a positioning accuracy of up to 3cm. The 3D risk heat map is based on the leakage probability distribution output by a graph neural network, rendering a real-time risk heat map of underground buildings and intuitively displaying high-risk areas.

[0053] The technical effects achieved by the above embodiments include: the AR interactive interface integrates virtual sensor data and risk prediction information with the real underground building scene, combined with high-precision UWB positioning and BIM model, enabling inspection personnel to intuitively and accurately obtain internal wall status information, improving inspection efficiency and accuracy. The three-dimensional risk heat map, based on the learning results of graph neural networks on the spatial propagation characteristics of leakage, dynamically displays the leakage risk probability of various areas of the underground building in real time, presenting high-risk areas with intuitive color gradients, providing a visual basis for maintenance decisions, assisting managers in quickly formulating targeted maintenance strategies, and improving the scientific and efficient management of underground building leakage.

[0054] Traditional technical solutions suffer from the following problems: traditional leakage monitoring systems lack effective modeling of the spatial propagation characteristics of leaks, making it impossible to accurately predict the diffusion path and impact range of leaks in underground building walls, resulting in inaccurate risk assessments and a lack of scientific rigor in maintenance strategy formulation. Based on this, the construction of the graph neural network model satisfies the following mathematical relationship:

[0055]

[0056] in, This represents the feature vector of node v in the l-th layer of the graph neural network; c represents the set of neighboring nodes of node v; uv W represents the normalized coefficient from node u to node v. (l) and σ is the weight matrix of the l-th layer; σ is the activation function.

[0057] This formula is the core computational formula for graph neural networks to model the topology of sensor networks and the spatial propagation characteristics of leakage. By constructing a node-edge graph structure, it learns the diffusion law of leakage in the spatial dimension in underground building walls, solving the technical problem that traditional methods cannot capture the spatial correlation of leakage.

[0058] The formula describes the process of updating the feature vector of node v in the l-th layer of the graph neural network to the (l+1)-th layer. The core is to capture the propagation relationship of leakage in the physical space by aggregating the features of neighboring nodes and its own features. Each sensor in the underground building is abstracted as a graph node, and the edges between nodes correspond to the physical proximity relationship of the sensors, thereby transforming the spatial topology of the wall structure into a computable graph structure.

[0059] Neighbor node feature aggregation: This is a feature aggregation term for neighboring nodes. Among them, This represents the set of neighboring nodes of node v. In underground building scenarios, it typically refers to the sensor node that is physically adjacent to the current sensor. uv c is the normalization coefficient from node u to node v, typically calculated based on the distance between nodes or connection weights. It's used to avoid feature aggregation bias caused by differences in the number of neighboring nodes, for example, in sensor-dense areas like expansion joints. uv It will be weighted and adjusted based on the node spacing. W (l) This is the weight matrix for the l-th layer, used to weight the feature vectors of neighboring nodes. A linear transformation is performed to extract features related to leakage propagation. This part captures the near-neighbor impact of leakage in space by aggregating real-time monitoring data from neighboring nodes. For example, when a sensor detects abnormal humidity, the features of its neighboring nodes will affect the current node's state update through this mechanism, simulating the diffusion process of leakage in wall materials.

[0060] Retained features: Retain items for its own characteristics. It is a weight matrix independent of its neighboring nodes, used to weight the feature vector of node v itself. The transformation is performed. This design ensures that nodes do not lose their historical feature information during updates, such as the temperature change trend monitored by a sensor over a long period of time. Even if neighboring nodes do not show any abnormalities, their own features can be preserved and used in subsequent calculations, avoiding feature distortion caused by interference from neighboring nodes.

[0061] Activation Functions and Nonlinear Mapping: σ(·) is the activation function, such as ReLU or Sigmoid, used to introduce a nonlinear transformation into the aggregated feature vector. In underground building scenarios, the propagation process of leakage exhibits nonlinear characteristics, such as the nonlinearity of concrete crack propagation and the threshold effect of fluid infiltration. The activation function can map linearly aggregated features to a nonlinear space, enabling the model to learn the nonlinear evolution of leakage from micro-crack initiation to macroscopic leakage. For example, when the aggregated features of multiple neighboring nodes exceed the activation function threshold, the model will output a higher leakage risk probability, achieving nonlinear prediction of leakage diffusion.

[0062] Engineering Application Value: This formula, through graph-based modeling, enables the system to learn the spatial propagation patterns of leakage in underground building walls. In a subway tunnel case study, the model, using this formula, aggregated acoustic emission signals and strain data from sensors at the joints of tunnel segments, predicting the development trend of leakage from micro-cracks to macro-leakage three days in advance. Compared to traditional single-point sensors that only alarm after leakage occurs, the warning time is more than 50% earlier. Simultaneously, the leakage probability distribution output by the model can be directly used to render a 3D risk heat map, visually displaying high-risk areas and providing maintenance personnel with precise spatial positioning information, achieving a positioning accuracy of 3cm.

[0063] This solution uses a graph neural network model to address the difficulty in capturing the spatial propagation patterns of leakage, thereby enabling the prediction of the spatial distribution of leakage risks.

[0064] It is worth mentioning that in the construction of the graph neural network model, the feature vector of node v in the (l+1)th layer... By examining the feature vectors of its neighbor node u and its own feature vector Weighted aggregation is performed, followed by activation function σ to obtain the result. The normalization coefficient c is... uv To ensure the stability of feature aggregation, the weight matrix W (l) and Used for feature transformation. The model uses the sensor network topology as a graph structure and learns the propagation characteristics of leakage in space through feature transfer between nodes.

[0065] The technical effects achieved by the above embodiments include: the graph neural network model utilizes the sensor network topology, treating each sensor as a graph node. The connections between nodes reflect proximity in physical space. By learning the message passing patterns between nodes, it effectively captures the propagation characteristics of leakage in space. The leakage probability distribution output by the model can be used to render a three-dimensional risk heat map in real time, intuitively displaying high-risk areas. This allows managers to understand the potential impact range of leakage in advance and deploy monitoring and maintenance resources in a targeted manner. Compared with traditional methods, the warning time is more than 50% earlier, providing more time for preventive measures and reducing the likelihood and severity of leakage accidents.

[0066] Traditional technical solutions suffer from the following problems: traditional maintenance methods are mostly reactive repairs or periodic inspections, lacking scientific rigor and specificity, resulting in high maintenance costs, low efficiency, and an inability to verify maintenance effectiveness, making it difficult to form an optimized closed loop for maintenance strategies. Based on this, the intelligent decision-making layer includes a predictive maintenance strategy engine and a closed-loop verification mechanism. The predictive maintenance strategy engine, based on a risk-cost optimization model, automatically generates maintenance work orders, including recommended materials and processes, by combining leakage probability, repair costs, and downtime losses. The closed-loop verification mechanism continuously monitors data after repair, evaluating the effectiveness by comparing changes in humidity and strain before and after repair.

[0067] This solution addresses the issues of blind maintenance decisions and lack of effectiveness evaluation through a predictive maintenance strategy engine and a closed-loop verification mechanism, thereby achieving intelligent and scientific maintenance strategies.

[0068] It is worth mentioning that the intelligent decision-making layer includes a predictive maintenance strategy engine and a closed-loop verification mechanism. The predictive maintenance strategy engine is based on a risk-cost optimization model, which comprehensively considers factors such as leakage probability, repair costs, and downtime losses, and automatically generates maintenance work orders that include recommended materials, such as epoxy resin grouting materials and processes, and high-pressure grouting. The closed-loop verification mechanism continuously monitors data after maintenance and evaluates the maintenance effect by comparing changes in parameters such as humidity and strain before and after the repair.

[0069] The technical effects achieved by the above embodiments include: the predictive maintenance strategy engine, based on a risk-cost optimization model, quantifies leakage risks into specific maintenance decisions, avoiding the subjectivity and experience-based dependence of traditional manual decision-making. The generated maintenance work orders include precise material and process recommendations, improving the targeting and effectiveness of maintenance. The closed-loop verification mechanism continuously monitors the wall condition after maintenance, objectively assesses the maintenance effect, and provides data support for optimizing maintenance strategies, forming a complete closed loop of "perception-diagnosis-decision-execution-verification." This mechanism transforms maintenance management from passive response to proactive prediction, reducing the number of emergency repairs, lowering maintenance costs, extending the structural overhaul cycle, and improving the overall efficiency and economy of underground building maintenance.

[0070] Traditional leakage monitoring systems lack the ability to simulate the multiphysics coupling effects of underground building walls, making it impossible to accurately predict the diffusion path and development trend of leakage in complex environments. This results in incomplete risk assessments and difficulty in formulating long-term effective maintenance strategies. Therefore, the multiphysics simulation model of the proposed digital twin engine satisfies the following mathematical relationship:

[0071]

[0072] Where ρ is the density of the medium; c is the specific heat capacity; T is the temperature; t is the time; k is the thermal conductivity; Q is the heat source intensity; g is the gravitational acceleration; and v is the fluid velocity.

[0073] This scheme solves the problem of multi-physical factor coupling simulation in leakage prediction by using a multi-physics digital twin model, and realizes dynamic prediction of leakage development.

[0074] This formula is the heat conduction equation of the multiphysics simulation model in the digital twin engine. It is used to describe the dynamic change process of the temperature field in the walls of underground buildings. By coupling physical processes such as heat conduction, heat source distribution and fluid flow, it solves the technical problem of simulating the multiphysics coupling effect in leakage prediction and achieves accurate prediction of leakage diffusion path.

[0075] left side of the formula The right side represents the rate of change of thermal energy of a medium per unit volume over time. The equation represents the heat conduction term, Q the heat source intensity, and ρgv the heat convection term caused by fluid flow. It treats the underground building walls as a porous medium, comprehensively considering the effects of heat conduction, heat source, and fluid flow on the temperature field, thereby inferring the leakage path through abnormal changes in the temperature field (such as groundwater infiltration causing local temperature gradient changes).

[0076] Analysis of the rate of change of thermal energy on the left: ρ is the density of the medium (unit: kg / m³) 3 In underground structures, ρ is mainly the density of concrete walls or backfill soil; c is the specific heat capacity (unit: J / (kg·K)), which characterizes the ability of a medium to store thermal energy; T is the temperature (unit: K), which is monitored in real time by a distributed fiber optic sensing device. It reflects the dynamic change of internal heat energy in the wall over time. When leakage occurs, the infiltration of groundwater or industrial wastewater will carry heat into the wall, causing an abnormal rate of change of local ρcT. This abnormality can be used as a key indicator for leakage identification.

[0077] Analysis of the heat conduction term on the right: Let k be the thermal conductivity term, where k is the thermal conductivity coefficient, measured in W / (m·K), characterizing the ability of a medium to conduct heat. In concrete walls, the value of k is closely related to the moisture content. When leakage occurs in the wall, an increase in moisture content will lead to an increase in k, thereby changing the rate and direction of heat conduction. The temperature gradient vector It describes the heat conduction process in the wall. For example, when leakage causes the moisture content of a certain area to increase, the k of that area increases, which will accelerate the heat conduction to the surrounding area and form a unique temperature field distribution pattern. The location of leakage can be identified by the temperature gradient change monitored by distributed optical fiber.

[0078] Analysis of the heat source and heat convection terms on the right: Q is the heat source intensity, which in underground structures may come from environmental heat sources, such as heat dissipation from pipes or heat sources from chemical reactions, such as the heat of hydration of concrete. This term is used to correct for temperature changes caused by non-leakage factors; ρgv is the heat convection term, where g is the acceleration due to gravity (unit: m / s²). 2 v is the fluid velocity (unit: m / s). When leakage occurs, the fluid flow in the pores of the wall carries heat and forms a thermal convection effect. For example, when groundwater seeps upward, it will generate convection due to gravity, resulting in a specific gradient distribution of the temperature field. This effect can be quantified by the thermal convection term, thereby coupling the fluid flow with the temperature field change and realizing the dynamic tracking of the leakage path.

[0079] It is worth mentioning that the multiphysics simulation model of the digital twin engine satisfies the heat conduction equation, which describes the change of temperature T in the medium with time t, involving parameters such as medium density ρ, specific heat capacity c, thermal conductivity k, heat source intensity Q, gravitational acceleration g, and fluid velocity v. The model integrates seepage field, stress field, and chemical field models, and calibrates simulation parameters through real-time acquired monitoring data to predict leakage diffusion paths.

[0080] The technical effects achieved by the above embodiments include: the multiphysics simulation model couples physical processes such as seepage field, stress field, and chemical field, comprehensively considering the impact of factors such as fluid flow, structural deformation, and chemical corrosion on leakage in underground building walls. By calibrating model parameters with real-time data, the virtual digital twin maintains consistency with the actual underground building state, thereby accurately predicting the diffusion path, impact range, and development speed of leakage. This predictive capability provides managers with long-term leakage risk trend analysis, assists in formulating scientific maintenance plans, and allows for the early deployment of preventative maintenance measures to avoid the escalation of leakage accidents, ensuring the long-term safety and stable operation of underground building structures, while optimizing the allocation of maintenance resources and reducing overall maintenance costs.

[0081] Traditional technical solutions suffer from the following problems: traditional monitoring systems use a single type of sensor, resulting in limited data dimensions and an inability to fully reflect the complex physical processes of leakage. Furthermore, their ability to fuse multi-source data is insufficient, leading to low leakage detection rates and high false alarm rates. The proposed multimodal sensor fusion diagnostic technology employs Kalman filtering and Transformer networks for feature-level fusion, constructing a four-dimensional monitoring system encompassing acoustic, optical, electrical, and mechanical aspects, integrating multi-source data such as acoustic emission, fiber optic strain, humidity, and ion concentration.

[0082] This solution addresses the issues of insufficient information from a single sensor and inadequate fusion of multi-source data through multi-modal sensing fusion, thereby improving the accuracy and reliability of leak diagnosis.

[0083] It is worth mentioning that the multimodal sensor fusion diagnostic technology uses Kalman filtering and Transformer networks for feature-level fusion, integrating multi-source data such as acoustic emission, fiber strain, humidity, and ion concentration to construct a four-dimensional monitoring system of "acoustic-optical-electrical-mechanical". Kalman filtering is used to denoise and estimate the state of sensor data, while the Transformer network is used to capture long-distance dependencies between multi-source data, achieving feature-level fusion.

[0084] The technical effects achieved by the above embodiments include: Kalman filtering effectively removes noise from sensor data, improves data quality, and provides reliable input for subsequent fusion processing; the powerful feature extraction and long-distance dependency modeling capabilities of the Transformer network enable the system to capture the implicit correlations between multi-source data, achieving a comprehensive characterization of leakage features. The "sound-light-electricity-force" four-dimensional monitoring system monitors and diagnoses leakage from multiple physical dimensions, and can identify hidden seepage paths that traditional methods cannot detect, such as early leakage caused by micro-cracks or non-humidity-dominated seepage. This technology significantly improves the system's ability to identify leakage, reduces the false alarm rate, provides strong support for the early detection and accurate location of leakage, and transforms underground building leakage monitoring from single-indicator alarms to multi-dimensional intelligent diagnosis, improving the intelligence level of the monitoring system.

[0085] Traditional technical solutions suffer from the following problems: traditional monitoring systems employ a single, untargeted sensor deployment method, either resulting in excessive costs due to uniform deployment across all areas, or insufficient deployment in critical areas leading to poor monitoring results. They also fail to differentiate monitoring based on the leakage risk levels of different areas within underground structures. The proposed sensor deployment method involves: densely deploying acoustic emission sensors, distributed optical fibers, and electrochemical sensors in high-risk areas such as expansion joints and through-wall pipes, with a spacing ≤2m; and arranging MEMS multi-parameter sensors in general areas, with a spacing of 5-8m.

[0086] This solution addresses the issues of rationality and economy in sensor deployment through a differentiated deployment strategy, achieving a balance between monitoring effectiveness and cost.

[0087] It is worth mentioning that the sensor deployment adopts a differentiated strategy. In high-risk areas such as expansion joints and through-wall pipes, a dense deployment of "acoustic emission sensors + distributed optical fibers + electrochemical sensors" with a spacing of ≤2m is used to achieve high-precision, multi-parameter monitoring of high-risk areas. In general areas, MEMS multi-parameter sensors are deployed with a spacing of 5-8m to meet routine monitoring needs while controlling deployment costs.

[0088] The technical effects achieved by the above embodiments include: densely deploying multiple sensors in high-risk areas, fully utilizing the advantages of each sensor to achieve multi-dimensional, high-precision monitoring of key components, and timely capturing information such as micro-cracks, structural deformation, and seepage source types, ensuring effective monitoring of leakage risks in high-risk areas; and appropriately deploying MEMS multi-parameter sensors in general areas, reducing overall deployment costs while ensuring basic monitoring needs. This differentiated deployment strategy allows the monitoring system to focus on high-risk areas while also providing comprehensive monitoring of the entire underground structure, optimizing resource allocation, improving the cost-effectiveness of the monitoring system, providing feasibility for large-scale application of underground building leakage monitoring, and ensuring the integrity and validity of monitoring data, laying the foundation for subsequent analysis and decision-making.

[0089] Traditional technical solutions suffer from the following problems: traditional leak maintenance methods involve fragmented processes and lack a complete closed loop from data collection to maintenance decision-making, resulting in poor coordination between different stages, low maintenance efficiency, and an inability to achieve full lifecycle management of leaks. Based on this, please refer to... Figure 2 This embodiment provides a method for intelligent sensing and predictive maintenance of underground building wall leakage, applying the intelligent sensing and predictive maintenance system for underground building wall leakage as described in any of the above embodiments, including:

[0090] The sensing layer collects multi-source data in real time, including acoustic emission signals, strain, temperature, humidity, and ion concentration from the underground building walls.

[0091] The transport layer transmits the collected data to the processing layer through a hybrid network architecture;

[0092] The processing layer utilizes cloud-edge collaborative computing, and processes and analyzes the data through a physical information neural network prediction module and a digital twin engine to extract feature vectors and predict leakage diffusion paths.

[0093] The application layer displays monitoring and prediction results through an AR interactive interface and a 3D risk heat map;

[0094] The intelligent decision-making layer generates predictive maintenance strategies based on the prediction results, forming a closed loop of perception, diagnosis, decision-making, and execution.

[0095] This solution addresses the fragmentation of maintenance processes by constructing a complete methodology, thereby achieving a systematic and intelligent approach to leakage management.

[0096] The technical effects achieved by the above embodiments include: the method realizes fully automated and intelligent management of the entire process from data acquisition, transmission, processing, display to maintenance decision-making. Multi-source data acquisition at the perception layer ensures comprehensive acquisition of leakage-related information; the transmission layer guarantees reliable data transmission; intelligent analysis and prediction at the processing layer provide a scientific basis for decision-making; visual display at the application layer facilitates personnel understanding and judgment; and the intelligent decision-making layer generates precise maintenance strategies and verifies their effects, forming a closed-loop management system. This fully intelligent management transforms underground building leakage maintenance from passive response to proactive prediction, improving the timeliness and accuracy of maintenance, reducing the occurrence of leakage accidents, lowering maintenance costs, enhancing the safety and reliability of underground buildings, and achieving efficient and scientific management of underground building wall leakage.

[0097] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An intelligent sensing and predictive maintenance system for underground building wall leakage, characterized in that, include: The system comprises a perception layer, a transmission layer, a processing layer, an application layer, and an intelligent decision-making layer. The perception layer collects real-time multi-source data on acoustic emission signals, strain, temperature, humidity, and ion concentration from the underground building walls. The processing layer processes and analyzes the data using a physical information neural network prediction module and a digital twin engine, extracting feature vectors to predict leakage risk and leakage diffusion paths. The physical information neural network prediction module is used to predict leakage risk, and the prediction model constructed by this module in the processing layer satisfies the following mathematical relationship: ； in, This is the loss function, used to measure the degree of agreement between the model's predictions and the actual values ​​and physical constraints. and These are the weight coefficients for the data fitting term and the physical constraint term, respectively. ; To monitor the number of data samples; For the first The actual leakage-related physical quantity observations for each sample, including at least one of humidity, strain, or ion concentration; For the first The model prediction value for each sample; This represents the number of sampling points for the physical constraint equations. For the first The physical equation residuals at each sampling point, wherein the physical equations include Darcy's law equations or seepage continuity equations; The multiphysics simulation model of the digital twin engine satisfies the following mathematical relationship: ； in, The density of the medium; Specific heat capacity; For temperature; For time; The thermal conductivity coefficient; Intensity of the heat source; It is the acceleration due to gravity; For fluid velocity, the multiphysics simulation model of the digital twin engine can predict the leakage diffusion path; The application layer includes a three-dimensional risk heat map, which is based on the leakage probability distribution output by a graph neural network and renders the risk heat map of underground buildings in real time, intuitively displaying high-risk areas. The construction of a graph neural network model satisfies the following mathematical relationship: ； in, Indicates the first Nodes in a layered graph neural network eigenvectors; Represents a node The set of neighboring nodes; For nodes To the node The normalized coefficient; and For the first Layer weight matrix; For activation functions; The intelligent decision-making layer includes a predictive maintenance strategy engine and a closed-loop verification mechanism. The predictive maintenance strategy engine automatically generates maintenance work orders based on a risk-cost optimization model, combining leakage probability, repair costs, and downtime losses. The maintenance work orders include recommended materials and processes. The closed-loop verification mechanism continuously monitors data after maintenance and evaluates the maintenance effect by comparing changes in humidity and strain before and after repair.

2. The intelligent sensing and predictive maintenance system for underground building wall leakage according to claim 1, characterized in that, The sensing layer includes an acoustic emission sensor array, an electrochemical sensor network, a distributed fiber optic sensing device, and a MEMS multi-parameter sensor. The acoustic emission sensor array has a sampling frequency ≥1MHz and can detect crack initiation at the 0.1mm level. The electrochemical sensor network integrates Cl... - SO4² - Ion concentration and pH value sensors are used to determine the type of seepage source; the distributed optical fiber sensing device is laid along the structural seam to achieve strain and temperature field monitoring with a spatial resolution of 0.5m along the optical fiber direction; the sampling interval of the MEMS multi-parameter sensor is 10 seconds, and the detection accuracy reaches ±1%RH and ±0.1℃.

3. The intelligent sensing and predictive maintenance system for underground building wall leakage according to claim 1, characterized in that, The transmission layer adopts a hybrid network architecture. Key data, namely acoustic emission waveforms and fiber strain, are transmitted using 5G URLLC with a latency of ≤10ms. Conventional sensor data are transmitted via LoRaWAN with a coverage radius of ≥1km. The transmission layer also includes an AI-enabled Mesh self-healing network, where nodes dynamically optimize routes based on reinforcement learning, and the communication interruption recovery time is ≤5 seconds.

4. The intelligent sensing and predictive maintenance system for underground building wall leakage according to claim 1, characterized in that, The application layer also includes an AR interactive interface, which combines UWB positioning with BIM model, allowing inspection personnel to view sensor data inside the wall, historical leakage points and predicted risk areas through AR glasses, with a positioning accuracy of 3cm.

5. The intelligent sensing and predictive maintenance system for underground building wall leakage according to claim 1, characterized in that, The system also includes multimodal sensor fusion technology; the multimodal sensor fusion diagnostic technology uses Kalman filtering and Transformer network for feature-level fusion to construct a four-dimensional monitoring system of sound, light, electricity and force, and integrates multi-source data such as acoustic emission, fiber optic strain, humidity and ion concentration.

6. The intelligent sensing and predictive maintenance system for underground building wall leakage according to claim 2, characterized in that, The sensor deployment method is as follows: in high-risk areas, such as expansion joints and through-wall pipes, acoustic emission sensors, distributed optical fibers, and electrochemical sensors are densely deployed with a spacing of ≤2m; in general areas, MEMS multi-parameter sensors are deployed with a spacing of 5-8m.

7. A method for intelligent sensing and predictive maintenance of underground building wall leakage, employing the intelligent sensing and predictive maintenance system for underground building wall leakage as described in any one of claims 1-6, characterized in that, include: The sensing layer collects multi-source data in real time, including acoustic emission signals, strain, temperature, humidity, and ion concentration from the underground building walls. The transport layer transmits the collected data to the processing layer through a hybrid network architecture; The processing layer processes and analyzes the data through the physical information neural network prediction module and the digital twin engine, extracts feature vectors, and predicts leakage risk and leakage diffusion path. The physical information neural network prediction module is used to predict leakage risk, and the multiphysics simulation model of the digital twin engine can predict leakage diffusion path. The application layer displays the monitoring and prediction results through an AR interactive interface and a three-dimensional risk heat map. The three-dimensional risk heat map is based on the leakage probability distribution output by a graph neural network, and renders the risk heat map of underground buildings in real time to intuitively display high-risk areas. The intelligent decision-making layer generates predictive maintenance strategies based on the prediction results, forming a complete closed loop of perception, diagnosis, decision-making, and execution.

Images