Electric field fusion detection method and system for underground diaphragm wall leakage and medium

Abstract

The invention discloses an electric field fusion detection and intelligent diagnosis method and system for leakage of a two-in-one diaphragm wall, and belongs to the technical field of underground structure anti-seepage detection. The method comprises the steps that a distributed electrode array is vertically arranged along the diaphragm wall; executing dual-mode acquisition, acquiring three-dimensional resistivity data reflecting electrical structures of the diaphragm wall and a surrounding medium through an active mode, and recording a natural potential sequence generated by seepage through a passive mode; performing space-time correlation fusion on the resistivity anomaly of the active mode and the natural potential anomaly of the passive mode, and extracting a fusion feature vector; and inputting the data into a leakage diagnosis artificial intelligence model, and outputting space coordinates, scale estimation and leakage rate grading of leakage defects. The model is constructed by adopting simulation data pre-training and actual measurement data fine tuning. The method effectively eliminates interference of uneven reinforcement cages and stratums, realizes accurate positioning and quantitative evaluation of leakage hidden dangers, and has the advantages of no damage, intelligence, high precision and the like.

Description

Technical Field

[0001] This invention belongs to the field of quality inspection technology for urban rail transit and underground engineering, and particularly relates to an electric field fusion detection method, system and medium for leakage of underground continuous wall. Background Technology

[0002] As urban underground space development becomes deeper and more complex, diaphragm walls, combining the functions of foundation pit support and the exterior wall of the main structure, have been widely used in subway stations and deep foundation pit projects. As the first line of defense against water damage in underground structures, the construction quality of diaphragm walls directly affects the safety and service life of buildings. However, due to the complexity of geological conditions, limitations of the construction environment, and fluctuations in joint treatment processes, diaphragm walls are prone to leakage at wall joints, areas of loose concrete, or in the presence of mud layers. Traditional leakage detection methods mainly rely on manual inspections, water injection tests, or single geophysical exploration methods, which have played a certain role in ensuring the safe operation of underground structures.

[0003] However, existing diaphragm wall leakage detection technologies face numerous challenges in practical applications. First, traditional single geophysical methods (such as high-density resistivity methods or acoustic transmission methods) are highly susceptible to shielding effects from the dense steel reinforcement cage within the diaphragm wall and strong electrical interference, resulting in low spatial resolution and difficulty in accurately locating minute leakage channels within the complex coupled field of "reinforced concrete-soil." Second, current technologies primarily focus on static detection of physical structural defects, lacking real-time perception of the dynamic seepage process of groundwater. This makes it difficult to effectively distinguish between static structural inhomogeneities and dynamic, real leakage, leading to severe ambiguity and false alarm rates in the detection results. Furthermore, current data processing relies heavily on human experience, lacking quantitative assessment methods for the scale and rate of leakage under complex conditions, thus failing to meet the urgent needs of modern smart construction sites for automated, intelligent, and preventative monitoring. Summary of the Invention

[0004] To address the problems in existing technologies regarding the detection of leakage in diaphragm walls, such as low resolution due to interference from reinforcing cages and complex geological formations, difficulty in distinguishing between static defects and dynamic seepage, and lack of intelligent quantitative assessment methods, this invention provides an electric field fusion detection method, system, and medium for diaphragm wall leakage.

[0005] This invention is implemented as follows: an electric field fusion detection method for leakage in underground continuous walls, characterized by comprising the following steps:

[0006] S1. Distributed electrode arrays are deployed in the underground continuous wall and its surrounding area according to the preset plan, and a mapping relationship between the spatial position of the electrodes and the three-dimensional model of the underground continuous wall is established. S2. Perform dual-mode electric field data acquisition, wherein: in active mode, control the selected electrode pair to inject current into the underground continuous wall and the surrounding medium, measure the potential response of the remaining electrodes, and obtain a dataset reflecting the three-dimensional electrical structure of the underground continuous wall and the surrounding medium; in passive mode, monitor the natural potential difference between the electrodes, record the natural potential sequence that changes over time, and capture the natural electric field signal generated by seepage. S3. Spatiotemporally correlate and fuse the abnormal electrical parameters identified by the active mode with the natural potential anomalies captured by the passive mode, and extract the fused feature vector. S4. Input the fused feature vector into the pre-trained leakage diagnosis artificial intelligence model, and the model outputs the diagnosis result of the leakage defect.

[0007] In the above technical solution, preferably, in step S1, the layout scheme of the distributed electrode array includes: arranging several longitudinal survey lines vertically along the wall of the underground continuous wall; increasing the electrode layout density at the joint position of the wall segment, the wall segment corresponding to the change of geological conditions, and the connection between the wall and the base plate.

[0008] In the above technical solution, preferably, step S3 includes the following specific processes: performing inversion processing on the data of the active mode to generate a three-dimensional resistivity distribution map of the underground continuous wall area and identifying low-resistivity anomaly areas; processing the data of the passive mode to extract the spatiotemporal distribution characteristics of natural potential anomalies; calculating the spatial overlap between the low-resistivity anomaly area and the natural potential anomaly area in a unified three-dimensional space-time coordinate system, and analyzing the evolution trend of natural potential signals in the overlapping area to construct the fused feature vector.

[0009] In the above technical solution, preferably, the diagnostic results in step S4 include: spatial coordinates of the leakage defect, quantitative estimation of the defect size, leakage rate classification, and confidence index to distinguish between real leakage and non-leakage interference factors.

[0010] In the above technical solution, preferably, the leakage diagnosis artificial intelligence model is constructed in the following way: a parameterized comprehensive geoelectric model of the underground continuous wall is established, and typical defects with different parameters are set to simulate the seepage field under different hydraulic gradients; through forward modeling, a simulation training dataset corresponding to different defect-seepage conditions is generated, the simulation training dataset includes active mode response data and passive mode natural potential data, and is labeled with defect parameters; a deep learning network model is constructed, and pre-trained using the simulation training dataset to establish a mapping relationship between the fused electric field features and defect parameters; a measured verification dataset is collected, and the pre-trained deep learning network model is fine-tuned to obtain a fixed diagnostic model.

[0011] In the above technical solution, preferably, the deep learning network model uses a three-dimensional convolutional neural network to process the data of the active mode, and uses a one-dimensional convolutional neural network or a recurrent neural network to process the time series data of the passive mode, and fuses the features of the two.

[0012] This invention proposes an electric field fusion detection and intelligent diagnosis system for underground continuous wall leakage. Applying the aforementioned method, the system is characterized by comprising: a distributed electrode array, composed of multiple electrodes, for performing the deployment and potential sensing described in step S1; a multi-functional acquisition node, connected to the distributed electrode array, controlled to switch between current emission mode and potential signal reception mode, for performing the active source excitation and passive signal monitoring described in step S2; and a central processing and control unit, communicatively connected to the multi-functional acquisition node, for controlling the system timing and running the leakage diagnosis artificial intelligence model to execute steps S3 and S4.

[0013] In the above technical solution, preferably, the distributed electrode array includes electrodes arranged on the outer soil side and the inner foundation pit side of the underground continuous wall, forming a clamping observation structure.

[0014] The present invention proposes an electronic device, characterized in that it includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0015] The present invention proposes a computer-readable storage medium, characterized in that it stores a computer program thereon, which, when executed by a processor, implements the steps of the above-described method.

[0016] Compared with the prior art, the present invention has the following significant advantages and beneficial effects: This invention significantly improves detection accuracy and anti-interference capability through the deep integration of active mode imaging and passive mode monitoring. Utilizing the three-dimensional resistivity distribution acquired in active mode, the static physical structure of the wall and its surrounding medium can be clearly characterized, identifying potential defects such as non-compact concrete, mud inclusions, or voids. Simultaneously, by combining passive mode monitoring of the natural potential signal generated by the seepage kinetic effect, the dynamic characteristics of groundwater flow can be directly captured. This "static-dynamic combined" detection mechanism enables the system to correlate structural defects with fluid activity, effectively eliminating electrical interference caused by dense steel reinforcement cages within the wall, material inhomogeneity, or differences in the geological background. This fundamentally solves the technical problems of inaccurate positioning and high false alarm rates in traditional geophysical exploration methods.

[0017] This invention possesses comprehensive detection capabilities and multi-dimensional decision support information. The system can not only accurately locate the three-dimensional spatial position of leak points, but also semi-quantitatively to quantitatively assess the scale of defects (such as equivalent cavity size) and the intensity of leakage activity (such as leakage rate grading) through an artificial intelligence diagnostic model, and provide the confidence level of the diagnostic results. This provides refined data support for subsequent engineering treatment, avoiding the blind application of reinforcement measures such as grouting and sealing. In model construction, a method of "numerical simulation with massive data pre-training + measured data fine-tuning" is adopted. This method leverages the advantage of accurate simulated data labels while also taking into account the actual engineering environment, successfully overcoming the bottleneck of insufficient labeled data faced by artificial intelligence in the field of engineering detection, and ensuring the reliability and generalization ability of the diagnostic results.

[0018] This invention achieves non-destructive, efficient, and preventative testing with strong engineering applicability. The entire testing process is based on geophysical non-destructive testing, causing no damage to the diaphragm wall structure. Furthermore, the deployment and data collection scheme is flexible and adaptable to complex foundation pit construction environments. Standardized operating procedures reduce reliance on individual experience. This solution is not only suitable for quality acceptance after wall completion but can also be applied throughout the entire foundation pit excavation process and the service life of the main structure, enabling early detection and warning of potential leakage, thus possessing significant practical value in ensuring the safe operation of underground engineering projects. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the system structure and layout in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the principle of the natural potential method for detection in an embodiment of the present invention; Figure 3 This is a flowchart of model training and diagnosis in an embodiment of the present invention; Figure 4 This is a logic block diagram of data fusion and recognition in an embodiment of the present invention. Detailed Implementation

[0020] 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 of the invention and are not intended to limit the scope of protection of the invention.

[0021] The "two-wall-in-one" underground continuous wall described in this invention refers to a vertical support system that serves as a maintenance structure during the foundation pit engineering stage and as a permanent exterior wall of the building during the main structure stage. Due to its dense distribution of reinforcing bars, complex joint forms, and crossing of multiple strata, it has extremely high requirements for the accuracy of leakage detection and its anti-interference ability.

[0022] Example 1 This embodiment discloses an electric field fusion detection method for leakage in diaphragm walls. By extracting and intelligently identifying the electrical characteristics of the diaphragm wall and its surrounding medium from multiple dimensions, it achieves accurate identification of leakage defects. The specific implementation logic of this method is as follows: First, perform step S1: deploy a distributed electrode array in the underground continuous wall and its surrounding area according to the preset plan, and establish a mapping relationship between the spatial position of the electrodes and the three-dimensional model of the underground continuous wall.

[0023] When deploying the distributed electrode array, considering the unique structural characteristics of the diaphragm wall, this embodiment abandons the traditional horizontal, evenly spaced deployment method and instead adopts a targeted deployment strategy. Specifically, several longitudinal survey lines are arranged vertically along the diaphragm wall to form a detection network covering the entire depth. To improve the ability to detect key leakage risk points, the electrode density is artificially increased in wall sections with joints (such as I-beam or steel section joints), sections with drastic geological changes (such as the interface between sand and gravel layers and rock strata), and areas with complex stresses where the wall connects to the foundation pit floor. For example, the electrode spacing can be set to 1.0 to 1.5 meters in conventional sections, while in the aforementioned high-risk areas, the electrode spacing can be reduced to 0.3 to 0.5 meters to obtain higher resolution potential gradient data.

[0024] In this embodiment, the layout and fixing of the electrode array are fully integrated with the construction technology of the diaphragm wall to ensure the coupling stability between the sensing unit and the measured medium. Specifically, for monitoring inside the diaphragm wall, the electrodes are fixed in a pre-embedded manner. High-strength insulating clamps are used to fix the electrodes in groups to the inner side of the longitudinal reinforcing bars of the steel cage, and an insulating partition is installed between the electrodes and the reinforcing bars to prevent direct electrical short circuits from the steel cage to the electrodes. For monitoring outside the wall or subsequent supplementary monitoring, specially designed flexible electrodes are attached to the wall surface using micro-drilling or jet grouting. To ensure excellent electrical contact performance between the electrodes and the diaphragm wall concrete or surrounding soil, a solid conductive coupling agent composed of bentonite, graphite powder, and a high water-retaining polymer is pre-coated onto the electrode sensing surface. The coupling agent can absorb water and expand during the process of entering the tank and pouring concrete, forming a dense, low-resistance and flexible electrical contact layer at the interface between the electrode and the concrete. This effectively reduces the contact resistance (usually to below 500Ω), thereby eliminating signal distortion caused by interface voids or drying shrinkage cracks, and laying the physical foundation for the acquisition of high-precision electric field data.

[0025] After deployment, the absolute three-dimensional coordinates of each electrode are obtained using high-precision measuring instruments (such as total stations or RTK), and then mapped onto the three-dimensional building information model (BIM) or digital twin model of the diaphragm wall. This mapping process establishes a one-to-one correspondence between electric field data acquisition points and physical spatial structures, laying the foundation for accurately restoring resistivity anomalies to their actual spatial locations within the wall.

[0026] Next, step S2 is executed: dual-mode electric field data acquisition is performed. The core of this step lies in obtaining complementary information reflecting the static structure of the medium and the dynamic activity of the fluid through the organic combination of active excitation and passive monitoring modes.

[0027] In active mode, selected electrode pairs are controlled to inject current into the diaphragm wall and surrounding medium, while the potential response of the remaining electrodes is measured to obtain a dataset reflecting the three-dimensional electrical structure of the diaphragm wall and surrounding medium. Specifically, the central processing and control unit schedules multi-functional acquisition nodes, selecting a pair of electrodes in the electrode array as current emitters (A, B), and injecting AC or DC signals of a preset frequency and intensity into the formation through a constant current source or regulated power supply. Simultaneously, the remaining electrodes (M, N) in receiving mode synchronously record the induced potential at various points in the formation. By systematically switching and scanning the emitting electrode pairs, a high-density resistivity measurement dataset covering the entire target body is formed.

[0028] Because of the significant electrical differences between the diaphragm wall itself, the internal reinforcing cage, the water-rich area affected by leakage, and the normal strata, the dataset acquired under active mode, after inversion processing, can reveal the three-dimensional spatial distribution of resistivity inside and around the diaphragm wall. For example, when there are mud inclusions, pores, or a surge in local water content due to leakage inside the wall, the area will exhibit obvious low resistivity anomalies. The advantage of active mode is that it is less affected by environmental fluctuations and can provide a stable reference background for the physical structural defects of the wall.

[0029] In this embodiment, the current excitation process in active mode fully considers the balance between signal-to-noise ratio and electrode polarization effect under the strong electrical interference environment of the diaphragm wall. Specifically, the current intensity injected by the system control multi-functional acquisition node into the diaphragm wall and surrounding medium is set to the range of 10mA to 500mA. For sections with better geological conditions and higher background resistivity, a smaller current (e.g., 10mA-50mA) is used to protect the electrodes and reduce electrochemical polarization. For deep strata or areas with high conductivity and dense reinforcement, the current is increased to 200mA-500mA to enhance the penetration depth of the potential response signal. At the same time, in order to effectively suppress the polarization effect on the electrode surface and eliminate DC background noise in the underground environment, the system uses a low-frequency alternating current as the excitation source, and its alternating frequency range is strictly controlled between 0.1Hz and 10Hz. By performing multiple cyclic acquisitions and superimposing and averaging within this frequency range, it is possible not only to avoid interference from the power frequency (50Hz) and its higher harmonics, but also to ensure that the current field achieves a quasi-static stable distribution in the concrete wall and surrounding soil, thereby obtaining a high-fidelity apparent resistivity observation dataset.

[0030] In passive mode, the natural potential difference between the electrodes is monitored, and the natural potential sequence changing over time is recorded to capture the natural electric field signal generated by seepage. Please refer to [link to relevant documentation]. Figure 2 The core physics of this model lies in the electrokinetic effect. When groundwater is driven by a pressure gradient and flows through joints, cracks, or loose channels formed by construction defects in a diaphragm wall, the electric double layer between the fluid and the solid wall undergoes relative displacement, thereby generating a jet potential at both ends of the flow path and forming a natural electric field. In this embodiment, all multi-functional acquisition nodes are switched to a high input impedance receiving state (e.g., input impedance not less than 100MΩ) to record the natural potential fluctuations of each monitoring electrode relative to the reference electrode in real time.

[0031] Through continuous monitoring, the system can capture the dynamic characteristics of seepage intensity as the excavation progress of the foundation pit, fluctuations in internal and external water levels, or rainfall events evolve. This non-human-interventional signal acquisition can directly reflect whether there is active hydraulic connection in the current wall. To improve the reliability of the signal, the system performs detrending processing and noise filtering on the acquired raw spontaneous potential signal to eliminate stray current interference caused by electrode polarization, power grid frequency, and the operation of large construction machinery.

[0032] In this embodiment, the potential acquisition in passive mode strictly adheres to the reference benchmark requirements of high-precision electrical resistivity tomography, eliminating environmental noise and system errors through the scientific deployment of reference electrodes. Specifically, the system sets up at least one reference electrode, positioned in an "electrically quiet zone" far from the excavation area, the operating area of ​​large construction machinery, and high-voltage transmission lines. The electrode depth is typically located in stable strata below the groundwater level to ensure spatial consistency and temporal stability of the reference potential. To completely eliminate zero-point drift caused by the electrode's own chemical polarization effect, a specially designed non-polarized electrode is used, preferably a saturated copper sulfate electrode or a silver-silver chloride electrode. This non-polarized electrode achieves ion exchange with the soil through a porous ceramic material, providing extremely low and highly stable polarization potential compensation. By performing differential calculations between the potential signals of each monitoring electrode and the reference benchmark in real time, the system can accurately extract microvolt-level potential fluctuations caused by seepage electrostatic effects, significantly improving the system's sensitivity in capturing early, weak leakage signals.

[0033] Step S3: Spatiotemporally correlate and fuse the anomalous electrical parameters identified by active pattern recognition with the natural potential anomalies captured by passive pattern recognition to extract the fused feature vector. This process is crucial for eliminating detection ambiguity and achieving precise localization. Please refer to [link to relevant documentation]. Figure 4 First, a high-precision 3D inversion calculation is performed on the apparent resistivity dataset acquired in the active mode. During the inversion process, the geometric structure and electrical parameters of the diaphragm wall and its internal reinforcing cage are introduced as prior constraint information. A regularized inversion algorithm is used to generate a 3D resistivity distribution map of the diaphragm wall region. In this map, the system automatically identifies and extracts low-resistivity anomaly areas with resistivity significantly lower than the background value by setting a dynamic threshold. These areas typically indicate physical structural defects within the wall, such as non-dense concrete, mud inclusions, voids, or groundwater filling.

[0034] Simultaneously, gradient field transformation is performed on the data acquired in passive mode to extract the spatiotemporal distribution characteristics of spontaneous potential anomalies. In this embodiment, spontaneous potential abrupt change points are identified by calculating the second derivative of the potential field in space; these points typically correspond to the inlet or outlet locations of seepage. Subsequently, the spatial overlap between the low-resistivity anomaly region and the spontaneous potential anomaly region is calculated in a unified three-dimensional space-time coordinate system. The calculation of spatial overlap involves determining the distance between the geometric centroids of the two anomaly regions and quantitatively analyzing the overlap ratio of the anomaly body envelope volume.

[0035] By analyzing the trend of the spontaneous potential signal evolution over time within the overlapping area, the activity level of the defect is further confirmed. For example, if the variance of spontaneous potential fluctuations in a certain low-resistivity area increases with rainfall, then this area is assigned a very high leakage weight. Based on the above correlation analysis, a fusion feature vector is extracted and constructed. This feature vector is a tensor containing multi-dimensional information, specifically including: the three-dimensional spatial coordinates of the center of the suspected leakage area, the equivalent geometric scale of the anomaly, the attenuation ratio of resistivity at the center, the peak value of the spontaneous potential gradient, the time stability parameters of the potential sequence, and environmental correlation factors (such as the real-time monitoring of the water level difference between the inside and outside of the wall). This fusion feature vector integrates the dual information of "static structural defects" and "dynamic fluid activity," providing a high-information-density discrimination basis for subsequent artificial intelligence models for leakage diagnosis.

[0036] In this embodiment, to address the characteristics of inconsistent physical dimensions and extremely large numerical ranges in the fused feature vectors, a combined normalization method based on extremum mapping and nonlinear logarithms is adopted to ensure rapid convergence and global optimization stability of the artificial intelligence model during training. Specifically, for resistivity values, considering that their distribution typically exhibits a dynamic range of several orders of magnitude, a logarithmic transformation is first applied to compress the dynamic range, followed by a Min-Max mapping function to transform it to the [0,1] interval. For natural potential and its gradient values, zero-mean normalization is directly used to handle their original fluctuations, and they are linearly scaled to the [0,1] interval according to a preset monitoring extremum limit. Through the above normalization process, the influence of physical dimensions between different sensor signals is eliminated, enabling the model to balance the contribution weights of structural electrical anomalies and permeation electrodynamic responses when calculating the loss function, effectively avoiding gradient explosion or neuron inactivation problems caused by differences in feature values.

[0037] Step S4: Input the fused feature vector into the pre-trained leakage diagnosis AI model, and the model outputs the diagnosis result of the leakage defect. In this embodiment, the construction process of the leakage diagnosis AI model fully considers the scarcity and complexity of underground engineering data, and adopts the technical path of "numerical simulation big data-driven pre-training combined with measured small sample data fine-tuning" to solidify the final diagnosis model. For the model construction and training process, please refer to [link to relevant documentation]. Figure 3 .

[0038] First, a parameterized comprehensive geoelectric model of the diaphragm wall is established. In the simulation environment, based on common diaphragm wall structural parameters in actual engineering (such as wall thickness, depth, and reinforcement ratio of the steel cage) and surrounding strata parameters (such as resistivity and permeability coefficient of different soil layers), a field model containing various typical defects is set up. These typical defects include, but are not limited to: mud inclusions at wall joints, localized honeycomb and pitting of concrete, through cracks, and over-excavation and backfilling at the wall base. Simultaneously, seepage fields under different intensities are simulated by varying the hydraulic gradient. Forward modeling is performed using finite element or finite difference algorithms to generate simulation training datasets corresponding to different defect-seepage conditions. This dataset not only includes simulated active mode response data (apparent resistivity spatial distribution) and passive mode data (spontaneous sequence of spontaneous potential), but each sample also carries precise label information such as defect type, spatial location, geometric dimensions, and leakage rate.

[0039] In this embodiment, the training dataset for the leakage diagnosis AI model is constructed based on rigorous physical field coupling forward simulation to ensure that the model has a strong physical mechanism support. Specifically, the forward simulation process employs refined flow-electric coupling equations. By establishing the governing equations for the Darcy permeability field and the electrical conduction field, it simulates the electrokinetic effects generated when groundwater flows through wall defects under pressure gradient drive and the resulting changes in the potential field distribution. During the simulation, the contributions of different porosities, permeabilities, fluid conductivity, and fracture geometric parameters to the electric field response are comprehensively considered. To ensure the model has strong generalization ability and robustness, parametric scanning technology is used to generate training samples covering hundreds of geological conditions and defect combinations, with the total sample size exceeding one hundred thousand. This massive, high-precision labeled data not only covers the electric field response characteristics under extreme conditions but also, by superimposing random environmental noise following a Gaussian distribution, greatly enriches the model's noise resistance to complex interference backgrounds in actual engineering, thus providing a sufficient and solid training foundation for the feature extraction and discrimination logic of subsequent deep learning networks.

[0040] Subsequently, a deep learning network model is constructed. In this embodiment, the deep learning network model adopts a multi-path parallel feature processing architecture. For the 3D inversion volume features generated by the active mode, the model uses a 3D convolutional neural network (3DCNN) for spatial feature extraction to identify the geometric shape and topological structure of the resistivity anomaly; for the time series features generated by the passive mode, the model uses a 1D convolutional neural network (1DCNN) or a recurrent neural network (RNN, such as a long short-term memory network LSTM) for temporal evolution feature extraction. The two sets of features are concatenated through a feature fusion layer and input into a fully connected layer or attention mechanism module to achieve deep integration of "structure-fluid" information. The network is pre-trained using the aforementioned simulation training dataset to establish a complex nonlinear mapping relationship between the fused electric field features and defect parameters, enabling the model to have a preliminary leakage detection capability.

[0041] To better reflect real-world engineering environments, a field validation dataset was collected for fine-tuning. This dataset, derived from known engineering cases or controlled field experiments, includes realistic background noise and complex underground interference. Through transfer learning, while preserving the low-level feature extraction capabilities of the pre-trained model, high-level decision parameters were adjusted to solidify the final diagnostic model. This approach effectively addresses the challenges of insufficient labeled data and poor generalization faced by artificial intelligence in the field of diaphragm wall detection.

[0042] In actual diagnostics, when the fused feature vector extracted in step S3 is input into the model, the model outputs diagnostic results with multi-dimensional quantitative features. Specifically, the diagnostic results include the three-dimensional spatial coordinates of the leakage defect (achieving centimeter-level location), a quantitative estimate of the defect size (such as equivalent cavity volume or crack width), and a leakage rate classification (such as minor leakage, moderate leakage, and severe leakage). Crucially, the model also outputs a confidence index to quantify the reliability of the diagnostic conclusion. Through in-depth mining of the fused features, the model can keenly capture the subtle differences between static anomalies caused by highly conductive materials such as steel cages and the dynamic electric field induced by seepage, thereby effectively distinguishing real leakage defects from non-leakage factors such as uneven wall materials and environmental electromagnetic interference, significantly reducing false alarm and false negative rates.

[0043] In this embodiment, the output of leakage diagnosis results is presented through a visual graphical interface and quantitative evaluation indicators. Specifically, the example interface of the system output results includes a three-dimensional interactive transparent model of the ground-connected wall, where leakage defects are rendered as clouds with different color scales. The hue represents the leakage rate classification (e.g., red represents high-velocity leakage, and yellow represents high-water-content areas induced by seepage), and the size of the cloud intuitively reflects the geometric scale estimate of the defect. At the quantitative diagnosis level, the system provides a confidence index for each identified abnormal area. The calculation logic of this index is based on the output probability distribution of the Softmax layer at the end of the deep learning network, combined with the correlation bias of multi-dimensional features. In specific calculations, the model not only extracts the original probability value of the corresponding "leakage" item in the Softmax output category, but also uses the geometric consistency of the spatiotemporal distribution of active mode resistivity anomalies and passive mode potential anomalies as a weighting factor to correct the original probability. For example, the closer the spatial distance between the anomaly centers extracted by the two types of modes, the higher their confidence weight. This multi-criteria fusion confidence calculation method can quantitatively inform technicians of the reliability of diagnostic conclusions, providing scientific numerical criteria for engineering decisions such as whether to immediately carry out grouting and sealing on site.

[0044] Example 2 Based on the aforementioned method principle, this embodiment discloses an electric field fusion detection and intelligent diagnosis system for diaphragm wall leakage, such as... Figure 1 This system achieves integrated operations from low-level perception to high-level decision-making. It mainly consists of a distributed electrode array, multi-functional acquisition nodes, and a central processing and control unit.

[0045] A distributed electrode array, serving as a front-end sensing component, is deployed in the diaphragm wall and its surrounding area according to a pre-defined scheme. In a preferred embodiment, the distributed electrode array employs a clamp-type observation structure, meaning that the electrodes are not only deployed on the inner side of the diaphragm wall (the pit side) but also on the outer side (the soil side) through pre-embedding or drilling. This structure allows the controlled current to penetrate the entire wall cross-section, thereby obtaining three-dimensional electrical scanning data at various depths within the wall, significantly improving the resolution of hidden defects in the middle of the wall.

[0046] The multi-functional acquisition node is the core execution unit of the system, electrically connected to the electrodes in the distributed electrode array. Each node integrates a high-precision current excitation module, a weak signal detection module, and a high-speed switching array. Under the command and control of the central processing and control unit, the acquisition node can flexibly switch between current emission mode and potential signal reception mode within milliseconds. In active mode, a designated node injects high-voltage DC or low-frequency AC current into a designated electrode; in passive mode, the node automatically switches to a high-input-impedance potential measurement state. The node also has an internal synchronization clock module to ensure that the entire system has a strict time reference when acquiring natural potential time series, thereby achieving synchronous capture of seepage dynamic signals.

[0047] In this embodiment, the communication architecture and packaging process of the multi-functional acquisition node are specially designed to adapt to the extremely complex construction environment of underground engineering. Specifically, the multi-functional acquisition node is wiredly cascaded with the central processing and control unit via an industrial-grade RS485 bus or CAN bus, ensuring that data transmission still has extremely high anti-interference capability and real-time performance even in the strong electromagnetic shielding environment generated by the dense steel cage in deep foundation pits. Simultaneously, for complex working conditions with limited local wiring, the node can be equipped with an industrial-grade wireless mesh network module, achieving self-organized signal transmission through multi-hop relay technology, ensuring the connectivity of a large-scale monitoring network. Regarding hardware protection, considering the long-term high humidity, high water pressure, and cement slurry corrosion environment around the diaphragm wall, all acquisition nodes use high-strength stainless steel or special engineering plastic shells, combined with a fully sealed potting process, achieving an overall protection level of IP68. This stringent packaging process ensures that even when the node is immersed for a long time or in a high-pressure seepage zone, its internal core acquisition circuit and communication module maintain stable performance, thus providing reliable hardware support for all-weather, full-cycle diaphragm wall health monitoring.

[0048] The central processing and control unit connects to multiple multi-functional acquisition nodes via communication links, undertaking the tasks of system timing control, data aggregation, and diagnostic decision-making. This unit has a built-in high-performance processor for running the aforementioned artificial intelligence model for leakage diagnosis. During operation, the central unit first issues a grid configuration command, driving each node to complete dual-mode data acquisition according to a preset sequence. The massive amount of raw data acquired is aggregated to the central unit via a bus for real-time preprocessing, 3D inversion, and feature fusion. Finally, by retrieving the fixed diagnostic model, a 3D leakage cloud map of the underground continuous wall is directly output to the display terminal.

[0049] Example 3: This embodiment also provides an electronic device, including a memory and a processor. The memory stores a computer program for executing the above-described method for detecting leakage in underground continuous walls. When the processor executes the program, it can automatically complete the entire closed-loop operation from acquisition node control and multi-source electric field data fusion to artificial intelligence diagnosis. Furthermore, this embodiment also includes a computer-readable storage medium, on which instructions stored, when executed by a computer, can implement the steps described in Embodiment 1. This combination of hardware and software allows the present invention to function as a standalone intelligent detection instrument or be integrated into an automated foundation pit monitoring platform.

[0050] The technical solution provided by this invention effectively solves the problem of strong interference from a single geophysical exploration method caused by the complex steel mesh inside a diaphragm wall through a combination of active resistivity imaging and passive natural potential monitoring, elevating leakage detection from traditional "morphological recognition" to the level of "functional activity discrimination." Based on an artificial intelligence diagnostic model that integrates numerical simulation and measured data, it can automatically extract weak defect features in complex electric fields, achieving standardization and intelligence in the diagnostic process and effectively avoiding subjective errors in manual interpretation. The distributed, switchable hardware architecture adopted by this system has strong adaptability to engineering sites, and can be used not only for short-term quality acceptance after wall completion but also for long-term dynamic monitoring throughout the entire foundation pit excavation process, providing key technical support for preventative maintenance.

[0051] In addition, by fusing data from the artificially excited electric field in active mode and the natural electric field in passive mode, this solution unexpectedly achieves "online real-time self-calibration" of the polarization potential of the distributed electrode array itself. That is, it uses the known response function in active mode to reverse the level drift in passive mode. This enables the system to maintain monitoring accuracy stability for several years in underground highly corrosive environments where manual maintenance is not possible, breaking through the technical bottleneck of traditional electrochemical sensors that are difficult to serve for a long time.

[0052] The aforementioned AI model for leakage diagnosis, in the process of extracting and fusing feature vectors, not only identifies current leakage defects but also demonstrates sensitivity to the "electrical fingerprint" of subtle stress concentrations and evolution within the diaphragm wall. That is, before the hydraulic gradient causes the physical leakage channel to fully open, the model can capture abnormal piezoelectric / thermoelectric composite responses caused by changes in the internal stress distribution of the concrete, thus achieving "advanced warning" of diaphragm wall structural failure and shifting the detection point from "post-event discovery" to "early warning." Furthermore, the clamp-mounted distributed electrode array, under constant current excitation in active mode, unexpectedly forms a virtual "differential shielding field" within the wall cross-section, automatically canceling common-mode interference generated by stray currents from surrounding urban rail transit. Its nonlinear improvement in signal-to-noise ratio enables the system to maintain extremely high discrimination accuracy in the highly complex environment of ultra-deep foundation pits. This natural suppression capability against urban background electromagnetic noise far exceeds the expected effect of traditional signal processing methods. The organic combination of the above-mentioned technical features enables the present invention to achieve a leap from structural detection to functional diagnosis, while also possessing strong adaptability to complex underground environments and forward-looking health prediction capabilities.

[0053] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting leakage in underground diaphragm walls using electric field fusion, characterized in that, Includes the following steps: 、 S1. Distributed electrode arrays are deployed in the underground continuous wall and its surrounding area according to the preset plan, and a mapping relationship between the spatial position of the electrodes and the three-dimensional model of the underground continuous wall is established. S2. Perform dual-mode electric field data acquisition, wherein: in active mode, control the selected electrode pair to inject current into the underground continuous wall and the surrounding medium, measure the potential response of the remaining electrodes, and obtain a dataset reflecting the three-dimensional electrical structure of the underground continuous wall and the surrounding medium; in passive mode, monitor the natural potential difference between the electrodes, record the natural potential sequence that changes over time, and capture the natural electric field signal generated by seepage. S3. Spatiotemporally correlate and fuse the abnormal electrical parameters identified by the active mode with the natural potential anomalies captured by the passive mode, and extract the fused feature vector. S4. Input the fused feature vector into the pre-trained leakage diagnosis artificial intelligence model, and the model outputs the diagnosis result of the leakage defect.

2. The electric field fusion detection method for leakage in underground diaphragm walls according to claim 1, characterized in that, In step S1, the layout scheme of the distributed electrode array includes: arranging several longitudinal survey lines vertically along the underground continuous wall; increasing the electrode layout density at wall joint locations, wall sections corresponding to changes in geological conditions, and at the connection between the wall and the base plate.

3. The electric field fusion detection method for leakage in underground diaphragm walls according to claim 1, characterized in that, The specific process of step S3 includes: performing inversion processing on the data of the active mode to generate a three-dimensional resistivity distribution map of the underground continuous wall area and identifying low-resistivity anomaly areas; processing the data of the passive mode to extract the spatiotemporal distribution characteristics of natural potential anomalies; calculating the spatial overlap between the low-resistivity anomaly area and the natural potential anomaly area in a unified three-dimensional space-time coordinate system, and analyzing the evolution trend of natural potential signals in the overlapping area to construct the fused feature vector.

4. The electric field fusion detection method for leakage in underground diaphragm walls according to claim 1, characterized in that, The diagnostic results described in step S4 include: spatial coordinates of the leakage defect, quantitative estimation of the defect size, leakage rate classification, and confidence index for distinguishing between real leakage and non-leakage interference factors.

5. The electric field fusion detection method for leakage in underground diaphragm walls according to claim 1, characterized in that, The leakage diagnosis artificial intelligence model is constructed in the following way: a parameterized comprehensive geoelectric model of the underground continuous wall is established, and typical defects with different parameters are set to simulate the seepage field under different hydraulic gradients; through forward modeling, simulation training datasets corresponding to different defect-seepage conditions are generated. The simulation training datasets include active mode response data and passive mode natural potential data, and are labeled with defect parameters. A deep learning network model is constructed and pre-trained using the simulated training dataset to establish a mapping relationship between fused electric field features and defect parameters. Collect experimental verification datasets, fine-tune the pre-trained deep learning network model, and obtain a fixed diagnostic model.

6. The electric field fusion detection method for leakage in underground diaphragm walls according to claim 5, characterized in that, The deep learning network model uses a three-dimensional convolutional neural network to process the active mode data, and a one-dimensional convolutional neural network or a recurrent neural network to process the passive mode time series data, and then fuses the features of the two.

7. An electric field fusion detection and intelligent diagnosis system for leakage in underground diaphragm walls, employing the method described in any one of claims 1-6, characterized in that, include: A distributed electrode array, consisting of multiple electrodes, is used to perform the deployment and potential sensing described in step S1. A multi-functional acquisition node is connected to the distributed electrode array and is controlled to switch between current emission mode and potential signal reception mode, used to perform the active source excitation and passive signal monitoring described in step S2. The central processing and control unit is communicatively connected to the multi-functional acquisition node, used to control the system timing, and run the leakage diagnosis artificial intelligence model to execute steps S3 and S4.

8. The electric field fusion detection and intelligent diagnosis system for leakage in underground diaphragm walls according to claim 7, characterized in that, The distributed electrode array includes electrodes deployed on the outer side of the underground continuous wall and the inner side of the foundation pit, forming a clamping observation structure.

9. An electronic device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, It stores a computer program thereon, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

Images