Geomembrane defect detection system under soil and stone medium coverage condition and method thereof

Abstract

The invention relates to the technical field of engineering, in particular to a geomembrane defect detection system and method under the earth and stone medium coverage condition, and the geomembrane defect detection system comprises a multi-mode adaptive sensing module, an edge calculation module, a data fusion module, a digital twinning module and a decision and application module; compared with the defects that single resistivity or geological radar detection is adopted in the prior art, interference of medium conductivity and metal impurities is likely to happen, and the deep resolution is insufficient, the scheme adopts multi-mode sensor network cooperative work, and a dynamic density layout and time-phased acquisition strategy is combined; through multi-mode complementation of an electrode array, an ultra-wideband radar, a distributed optical fiber and an MEMS vibration sensor, a dominant sensing mode is automatically switched in a conductive clay area, temperature gradient interference is avoided through night acquisition, the penetration depth is increased to 15 meters through multi-band radar fusion, and metal interference is inhibited; the detection precision and the anti-interference capability in a complex medium environment are remarkably improved, and the problems of high false alarm rate and high omission ratio of a traditional method are effectively solved.

Description

Technical Field

[0001] The present invention relates to the field of engineering technology, and in particular to a geomembrane defect detection system and method under soil and rock medium covering conditions. Background Art

[0002] In the field of civil engineering anti-seepage, geomembranes (such as HDPE and LDPE) are widely used in landfills, tailings ponds, water conservancy projects, and other scenarios due to their excellent impermeability and corrosion resistance. However, geomembrane defect detection presents significant challenges in working conditions where the soil-rock overburden layer is thick (typically 5-15 meters) and the medium composition is complex (such as a mixture of clay, gravel, and metal impurities).

[0003] While traditional detection methods such as electrical impedance tomography (ERT) and geological radar (GPR) can partially identify leaks, they have significant limitations. ERT is susceptible to background interference in conductive clay or high-humidity areas, resulting in resistivity errors exceeding 30%, leading to a high false alarm rate. GPR's penetration depth drops sharply from a theoretical 10 meters to less than 3 meters in areas rich in metal impurities due to severe electromagnetic wave scattering, and its resolution for millimeter-scale pores is insufficient. Furthermore, existing systems often use a fixed electrode or sensor density, which is difficult to adapt to soil heterogeneity. For example, over-deployment in uniform sand and gravel areas wastes resources, while insufficient density leads to missed detections in joints or areas with sudden slope changes. In one case, uneven sensor coverage at a tailings pond missed three cracks, ultimately leading to a leak. While manual inspections can address some of these shortcomings, they are inefficient and lack real-time monitoring capabilities, especially in areas with deep cover or dangerous slopes, which significantly increases labor costs and safety risks.

[0004] Existing technologies often rely on cloud computing for data processing, but field projects often face network instability or bandwidth limitations, resulting in delays of up to several hours in detection results and failing to meet the real-time response requirements for leak incidents. For example, after a heavy rainstorm, a landfill failed to locate a leak in time due to data transmission delays, causing contaminants to spread to the surrounding soil. While recent research has attempted to incorporate distributed fiber optic sensing or drone-based infrared thermal imaging, multimodal data fusion capabilities are insufficient. For example, the spatiotemporal alignment error between fiber optic vibration signals and resistivity data exceeds 0.5 meters, and there is a lack of correlation analysis for small defects (less than 1 cm). Consequently, the actual detection rate is only 70%-80%. Furthermore, existing systems for assessing geomembrane aging remain at the static empirical model stage, failing to consider the real-time effects of environmental stresses such as temperature, humidity, and dynamic loads. This results in lifespan prediction errors exceeding 30%. For example, a water conservancy project failed to anticipate material softening caused by high temperatures, leading to large-scale cracking of the geomembrane after only three years of service, shortening the predicted lifespan by 50%.

[0005] Therefore, there is an urgent need for a full-process solution that integrates multimodal sensing, edge intelligent compression, graph neural network fusion, and digital twin life prediction to break through the multiple bottlenecks of accuracy, efficiency, and reliability. Summary of the Invention

[0006] In order to overcome the problems raised in the above background technology, the present invention proposes a geomembrane defect detection system and method under soil and rock medium covering conditions.

[0007] The technical solution of the present invention is: a geomembrane defect detection system under soil and rock medium covering conditions, comprising: Multimodal adaptive sensing module, used to collect electrical, acoustic, and electromagnetic wave signals in soil and rock media in real time through a multimodal sensor network; Edge computing module, used to compress data in real time and trigger high-precision algorithms through edge computing technology; A data fusion module for correlating multimodal data using graph neural networks; Digital twin module, used to simulate the aging process of geomembranes, dynamically optimize alarm thresholds and life prediction; The decision-making and application module is used to drive sound and light alarms, coordinate positioning, and automated repair equipment based on data analysis results, achieving closed-loop control of the entire process from defect detection to repair execution.

[0008] Preferably, the multimodal adaptive sensing module specifically includes: A11: Electrode array module, used to locate leaks by dynamically deploying flexible electrodes at a high density and detecting resistivity changes; A12: Electromagnetic wave detection module, used to penetrate deep coverage using ultra-wideband radar and identify areas with abnormal dielectric constants; A13: Vibration module, used to capture vibration signals through optical fiber and MEMS sensors to distinguish noise from real defects; A14: Optical auxiliary module, used to combine mobile phone infrared and laser scanning to screen shallow temperature anomalies and surface deformation.

[0009] A15: Salt solution tracer unit, used to verify the leak location by conductivity changes after injecting salt water.

[0010] Preferably, the electrode array module is laid out using a dynamic density strategy, and the specific steps include: S11: Preliminary scanning: First, metal electrodes are laid out with an initial electrode spacing of 5m, and the geological radar is started simultaneously to perform a large-scale scan. The electrical impedance and electromagnetic wave reflection data are obtained using the laid metal electrodes and the geological radar to generate a preliminary resistivity distribution map and a reflection anomaly heat map; S12: Identify abnormal areas. Mark the ERT low-resistance area and the GPR reflection amplitude mutation area as suspected defects. Use the DBSCAN algorithm to cluster the abnormal points, eliminate isolated noise, and determine the scope of the area that needs to be encrypted.

[0011] S13: Dynamic encrypted deployment, inserting mobile electrodes at 0.5m intervals within the boundaries of the clustered abnormal area; S14: Real-time data fusion, combined with the newly added electrode data, updates the ERT imaging, and iterates until the detection resolution meets the requirements; S15: Resource recovery and reuse. When the encrypted area is detected and confirmed to be free of defects, the mobile electrodes are recovered and reused in other areas.

[0012] Preferably, when the electrode array module is in operation, data is collected between 20:00 and 6:00 every day, and a three-dimensional resistivity distribution map is output, and low-resistance abnormal areas are marked.

[0013] Preferably, the edge computing module compresses data in real time and triggers a high-precision algorithm through edge computing technology, specifically including: S21: Data compression, compressing the raw sensor data through wavelet transform, reducing the volume of the raw sensor data to 10% of the original volume; S22: Anomaly detection, setting a dynamic threshold based on the initial data collected by the multimodal adaptive sensing module, and detecting the data of the original sensor according to the set dynamic threshold; S23: Model inference, using the lightweight CNN built into the edge computing module to analyze abnormal data and analyze the specific defect type and preliminary coordinates.

[0014] Preferably, when the edge computing module compresses the original sensor data through wavelet transform and compresses the volume of the original sensor data to 10% of the original volume, the method specifically includes: S31: Perform multi-level decomposition of the ERT voltage signal using the Daubechies wavelet basis to separate the low-frequency approximate component and the high-frequency detail component; S32: Perform hard threshold filtering on high-frequency detail components, retain low-frequency energy, and discard low-amplitude noise coefficients; S33: Based on the retained approximation and threshold processed detail components, the compressed signal is restored by inverse wavelet transform; S34: Perform FFT calculation on the vibration sensor waveform to generate spectrum energy distribution; S35: intercepts the effective frequency band of 0-500Hz and discards high-frequency noise; S36: Normalize the amplitude to an 8-bit integer, ignore the phase information, and store the non-zero coefficients.

[0015] Preferably, when using the built-in lightweight CNN of the edge computing module to analyze abnormal data and analyze the specific defect type and preliminary coordinates, the structure of the lightweight CNN includes: A21: Input layer, including the resistivity image channel and the vibration spectrum feature channel. The resistivity image channel is used to input a single-channel grayscale resistivity image of size 64×64×1. The vibration spectrum feature is used to input a 64-dimensional vector of normalized energy values ​​of vibration data after FFT processing. A22: Image processing branch, including convolution layer 1, convolution layer 2, and convolution layer 3. Convolution layer 1 is a 3×3 convolution and 8 filters, convolution layer 2 is a 3×3 convolution and 16 filters, and convolution layer 3 is a 3×3 convolution and 32 filters. A23: concatenation layer, used to concatenate the 2048-dimensional vector of the flattened image with the 64-dimensional vector of the vibration spectrum to obtain a 2112-dimensional feature; A24: Fully connected layer, including fully connected layer 1 and fully connected layer 2. Fully connected layer 1 is used to reduce the 2112-dimensional features to 124 dimensions, and fully connected layer 2 is used to reduce the 124-dimensional features to 3 dimensions.

[0016] Preferably, when the data fusion module uses a graph neural network to associate multimodal data, it specifically includes: S41: Spatiotemporal alignment, which synchronizes all sensor data in time through GPS clock tags and spatially registers the sensor data; S42: Graph neural network analysis: Graph neural network is used to analyze the spatiotemporally aligned data to obtain the predicted points and defect probabilities of geomembrane defects. S43: Generate a three-dimensional defect map, divide the space into 0.1m×0.1m×0.1m voxels, assign a defect probability value to each voxel, and visualize the defect probability value through color.

[0017] Preferably, when using a graph neural network to analyze the data after spatiotemporal alignment to obtain the predicted points and defect probabilities of geomembrane defects, the structure of the graph neural network is positioned as follows: each sensor is a node containing a feature vector, and edges are established when the physical distance between nodes is less than one meter. The graph convolution layer is used to aggregate information from neighboring nodes, and the attention mechanism is introduced to dynamically allocate association weights between nodes. Finally, the defect probability of each node is output through the Sigmoid function.

[0018] As a preferred option, the digital twin module simulates the aging process of geomembranes and dynamically optimizes alarm thresholds and life prediction, specifically including: S51: Material attenuation equation modeling. Based on the characteristics of HDPE materials, a tensile strength attenuation equation is established to quantify the effects of time, temperature, and load on aging. The principle formula of the tensile strength attenuation equation is: ; in, is the initial tensile strength, is the time attenuation coefficient, is the first stress sensitivity coefficient, and , is the second stress sensitivity coefficient, and , is the cumulative load stress; S52: Real-time data input, using fiber optic temperature measurement, MEMS pressure sensor, and humidity sensor to collect environmental parameters including temperature, stress, and humidity in real time; S53: Dynamic threshold adjustment, combining rule engine and random forest model prediction to optimize alarm triggering conditions; S54: Remaining life calculation, using finite element simulation to simulate the aging rate of stress concentration areas and generate a remaining life thermodynamic map; S55: Risk classification, which divides risk levels according to defect probability and remaining life to guide repair priorities; S56: Maintenance suggestion generation, automatically prompting geomembrane replacement or reinforcement plans for areas where the lifespan is less than the set threshold and the defect probability is greater than 70%; S57: Threshold adaptive adjustment, dynamically adjusting sensor sensitivity according to environmental changes; S58: Visual output, showing life prediction, risk distribution and maintenance work orders through WebGL interactive interface or CAD compatible format.

[0019] The method for detecting defects of geomembrane under soil and rock medium covering conditions includes the following steps: S61: Multimodal data acquisition, dynamic placement of electrode arrays and ultra-wideband radar scanning, real-time acquisition of resistivity, electromagnetic waves, vibration, and optical signals, marking abnormal areas; S62: Edge intelligent compression and detection, wavelet transform compresses ERT data, FFT truncates vibration spectrum, and lightweight CNN analyzes anomaly types and generates preliminary coordinates; S63: Multi-source data fusion, spatiotemporal alignment of sensor data, and graph neural network integration of electrical, acoustic, and temperature features to generate a 3D defect probability map; S64: Digital twin and life prediction, based on material aging models and real-time environmental data, dynamically adjusts alarm thresholds, predicts remaining life and divides risk levels.

[0020] Beneficial effects of the present invention: 1. Compared to existing technologies that use single resistivity or geological radar detection, which are susceptible to interference from dielectric conductivity, metal impurities, and insufficient deep-layer resolution, this solution uses a multimodal sensor network working in conjunction with dynamic density deployment and time-segmented acquisition strategies. Through the multimodal complementarity of electrode arrays, ultra-wideband radar, distributed optical fiber, and MEMS vibration sensors, it automatically switches to the dominant sensing mode in conductive clay areas, collects data at night to avoid temperature gradient interference, and multi-band radar fusion increases penetration depth to 15 meters and suppresses metal interference. This solution significantly improves detection accuracy and anti-interference capabilities in complex media environments, achieving deep defect positioning errors of less than 0.3 meters and a millimeter-level hole recognition rate of 95%, effectively addressing the high false alarm rate and missed detection rate of traditional methods. 2. Compared to existing technologies that rely on cloud computing, resulting in high latency and bandwidth pressure, this solution uses edge computing modules to achieve real-time data compression and localized analysis. It uses wavelet transform and FFT spectrum truncation technology to compress raw data to 10%, and combines this with a lightweight CNN model (with 1MB of parameters) to achieve millisecond-level defect classification and coordinate positioning. This solution shortens response time to less than 0.1 seconds, reduces data transmission energy consumption by 90%, supports real-time detection and decision-making in offline environments, and significantly improves adaptability and system reliability in complex field conditions. 3. Compared to existing technologies that independently process multi-sensor data and suffer from large spatiotemporal alignment errors, this solution builds a sensor node topology map based on a graph neural network (GNN). This solution dynamically correlates multimodal features such as resistivity, vibration, and temperature through a spatiotemporal alignment algorithm and an attention mechanism. This solution achieves deep fusion of multi-source data, increasing the detection rate of defects under 1 cm to 95%, achieving a 3D imaging positioning accuracy of ±0.1 meter, and reducing the false alarm rate by 50%. Furthermore, it supports seamless integration with engineering modeling software (BIM / CAD), providing visual decision support for precise construction and repair. 4. Compared with the shortcomings of static threshold alarms and manual repair based on existing technologies, this solution integrates material aging models and environmental data through a digital twin platform, dynamically optimizes alarm thresholds and predicts remaining life (error <10%), while simultaneously linking robots and drones to achieve closed-loop control of the entire process. Based on the HDPE tensile strength attenuation equation and real-time temperature, humidity, and load data, the system adaptively adjusts detection sensitivity and repair strategies, increasing repair efficiency to 2 locations per hour, reducing material waste by 60%, and shortening maintenance response time in high-risk areas to less than 30 minutes, significantly improving project safety and overall lifecycle management efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Shown is a schematic diagram of the geomembrane defect detection system under soil and rock medium covering conditions of the present invention; Figure 2What is shown is a schematic diagram of the workflow of the geomembrane defect detection method under soil and rock medium covering conditions of the present invention. DETAILED DESCRIPTION

[0022] The present invention will be further described below with reference to the accompanying drawings and examples.

[0023] See also Figure 1-Figure 2 The present invention provides an embodiment: a geomembrane defect detection system under soil and rock medium covering conditions, comprising: Multimodal adaptive sensing module, used to collect electrical, acoustic, and electromagnetic wave signals in soil and rock media in real time through a multimodal sensor network; Edge computing module, used to compress data in real time and trigger high-precision algorithms through edge computing technology; A data fusion module for correlating multimodal data using graph neural networks; Digital twin module, used to simulate the aging process of geomembranes, dynamically optimize alarm thresholds and life prediction; The decision-making and application module is used to drive sound and light alarms, coordinate positioning, and automated repair equipment based on data analysis results, achieving closed-loop control of the entire process from defect detection to repair execution.

[0024] As described above, the multimodal adaptive sensing module in the present invention can collect multiple signals in real time and comprehensively obtain information in soil and rock media; the edge computing module compresses data in real time and triggers high-precision algorithms to improve data processing efficiency and accuracy; the data fusion module uses graph neural networks to associate multimodal data and deeply mine the value of data; the digital twin module simulates the aging process of geomembranes, dynamically optimizes alarm thresholds and life predictions, and improves the accuracy and foresight of detection; the decision-making and application module realizes closed-loop control of the entire process from defect detection to repair execution, and can timely drive sound and light alarms, coordinate positioning and start automated repair equipment to ensure the safe and stable operation of geomembranes and reduce maintenance costs and risks.

[0025] Furthermore, the multimodal adaptive sensing module specifically includes: A11: Electrode array module, used to locate leaks by dynamically deploying flexible electrodes at a high density and detecting resistivity changes; A12: Electromagnetic wave detection module, used to penetrate deep coverage using ultra-wideband radar and identify areas with abnormal dielectric constants; A13: Vibration module, used to capture vibration signals through optical fiber and MEMS sensors to distinguish noise from real defects; A14: Optical auxiliary module, used to combine mobile phone infrared and laser scanning to screen shallow temperature anomalies and surface deformation.

[0026] A15: Salt solution tracer unit, used to verify the leak location by conductivity changes after injecting salt water.

[0027] As described above, the electrode array module in this invention uses a dynamic density layout of flexible electrodes to detect resistivity changes to locate leaks, accurately pinpointing the leak location. The electromagnetic wave detection module uses ultra-wideband radar to penetrate deep layers and identify areas with abnormal dielectric constants, effectively detecting potential problems in deep soil and rock media. The vibration module uses optical fibers and MEMS sensors to capture vibration signals and distinguish between noise and real defects, improving the accuracy of defect identification. The optical assistance module combines mobile phone infrared and laser scanning to screen for shallow temperature anomalies and surface deformation, enabling rapid detection of shallow abnormalities. The salt solution tracing unit verifies the leak location by detecting changes in conductivity after salt water injection, providing a reliable supplementary means for leak location confirmation. These multiple modules work together to achieve comprehensive, multi-level, and precise detection of geomembrane defects.

[0028] Furthermore, the electrode array module is laid out using a dynamic density strategy. The specific steps include: S11: Preliminary scanning: First, metal electrodes are laid out with an initial electrode spacing of 5m, and the geological radar is started simultaneously to perform a large-scale scan. The electrical impedance and electromagnetic wave reflection data are obtained using the laid metal electrodes and the geological radar to generate a preliminary resistivity distribution map and a reflection anomaly heat map; S12: Identify abnormal areas. Mark the ERT low-resistance area and the GPR reflection amplitude mutation area as suspected defects. Use the DBSCAN algorithm to cluster the abnormal points, eliminate isolated noise, and determine the scope of the area that needs to be encrypted.

[0029] S13: Dynamic encrypted deployment, inserting mobile electrodes at 0.5m intervals within the boundaries of the clustered abnormal area; S14: Real-time data fusion, combined with the newly added electrode data, updates the ERT imaging, and iterates until the detection resolution meets the requirements; S15: Resource recovery and reuse. When the encrypted area is detected and confirmed to be free of defects, the mobile electrodes are recovered and reused in other areas.

[0030] As described above, the present invention arranges metal electrodes at 5m intervals through preliminary scanning and cooperates with geological radar to obtain data to generate relevant images, providing a basis for subsequent analysis; abnormal area identification marks suspected defects and uses algorithm clustering to eliminate noise, accurately determines the area to be encrypted, and avoids blind encryption; dynamic encryption is deployed to insert mobile electrodes at 0.5m intervals at the boundary of the abnormal area, which significantly improves the detection accuracy of the defective area; real-time data fusion is combined with the newly added electrode data to update the imaging and iterate, ensuring that the detection resolution meets the requirements and the reliability of the detection results; the resource recovery and reuse mechanism recycles the mobile electrodes for other areas when there are no defects in the encrypted area, effectively improving the utilization rate of electrode resources and reducing the detection cost.

[0031] Furthermore, when the electrode array module is working, it collects data from 20:00 to 6:00 every day, outputs a three-dimensional resistivity distribution map, and marks low-resistance abnormal areas.

[0032] As described above, the technical solution of the electrode array module in the present invention working during a specific time period (20:00-6:00) and outputting a three-dimensional resistivity distribution map and marking low-resistance abnormal areas has achieved remarkable results: selecting the night time period for data collection can effectively avoid complex electromagnetic interference during the day (such as industrial electricity, high-frequency radiation of electronic equipment) and human activity noise (construction machinery vibration, interference from people walking), greatly improving signal purity and data accuracy; the output three-dimensional resistivity distribution map presents the spatial distribution characteristics of the resistivity of the soil and rock medium in a three-dimensional and intuitive form, helping detection personnel to quickly understand the electrical property changes of the medium around the geomembrane; accurately marking low-resistance abnormal areas provides clear guidance for the rapid locking of potential defects such as leakage and damage, significantly shortening the defect investigation time, improving detection efficiency, ensuring the long-term safe operation of the geomembrane, and reducing the risk of engineering accidents and maintenance costs caused by defects.

[0033] Furthermore, when the edge computing module compresses data in real time and triggers high-precision algorithms through edge computing technology, it specifically includes: S21: Data compression, compressing the raw sensor data through wavelet transform, reducing the volume of the raw sensor data to 10% of the original volume; S22: Anomaly detection, setting a dynamic threshold based on the initial data collected by the multimodal adaptive sensing module, and detecting the data of the original sensor according to the set dynamic threshold; S23: Model inference, using the lightweight CNN built into the edge computing module to analyze abnormal data and analyze the specific defect type and preliminary coordinates.

[0034] As described above, in the data compression stage, the present invention compresses the volume of raw sensor data to 10% of its original size through wavelet transform, greatly reducing the amount of data transmission, lowering network bandwidth occupancy, improving data transmission efficiency, and saving storage resources for subsequent processing; in the anomaly detection stage, a dynamic threshold is set based on the initial data collected by the multimodal adaptive sensing module and the raw sensor data is detected accordingly, which can more accurately capture abnormal changes in the data and improve the sensitivity and accuracy of anomaly detection; in the model inference stage, the built-in lightweight CNN is used to analyze the abnormal data, which can quickly and accurately determine the specific defect type and preliminary coordinates, providing strong support for subsequent precise positioning and repair, and realizing efficient and accurate data processing and defect analysis as a whole.

[0035] Furthermore, when the edge computing module compresses the raw sensor data through wavelet transform and reduces the volume of the raw sensor data to 10% of its original size, it specifically includes: S31: Perform multi-level decomposition of the ERT voltage signal using the Daubechies wavelet basis to separate the low-frequency approximate component and the high-frequency detail component; S32: Perform hard threshold filtering on high-frequency detail components, retain low-frequency energy, and discard low-amplitude noise coefficients; S33: Based on the retained approximation and threshold processed detail components, the compressed signal is restored by inverse wavelet transform; S34: Perform FFT calculation on the vibration sensor waveform to generate spectrum energy distribution; S35: intercepts the effective frequency band of 0-500Hz and discards high-frequency noise; S36: Normalize the amplitude to an 8-bit integer, ignore the phase information, and store the non-zero coefficients.

[0036] As described above, the present invention decomposes the ERT voltage signal through Daubechies wavelet basis multi-level decomposition and processes the high-frequency detail components, thereby effectively retaining the key information of low-frequency energy and discarding the low-amplitude noise coefficient, thereby reducing redundant data; the signal is restored based on the retained approximation and the processed detail components by inverse wavelet transform, ensuring that the main features of the signal are retained; the vibration sensor waveform is calculated by FFT, the effective frequency band is intercepted, and the high-frequency noise is discarded, thereby further purifying the data; the amplitude is normalized to an 8-bit integer, the phase information is ignored, and the non-zero coefficients are stored. Under the premise of ensuring data availability, the data volume is greatly compressed to 10% of the original size, which significantly reduces the data storage and transmission costs, improves the data processing efficiency, and provides strong support for subsequent real-time analysis and decision-making.

[0037] Furthermore, when using the lightweight CNN built into the edge computing module to analyze abnormal data and analyze the specific defect type and preliminary coordinates, the structure of the lightweight CNN includes: A21: Input layer, including the resistivity image channel and the vibration spectrum feature channel. The resistivity image channel is used to input a single-channel grayscale resistivity image of size 64×64×1. The vibration spectrum feature is used to input a 64-dimensional vector of normalized energy values ​​of vibration data after FFT processing. A22: Image processing branch, including convolution layer 1, convolution layer 2, and convolution layer 3. Convolution layer 1 is a 3×3 convolution and 8 filters, convolution layer 2 is a 3×3 convolution and 16 filters, and convolution layer 3 is a 3×3 convolution and 32 filters. A23: concatenation layer, used to concatenate the 2048-dimensional vector of the flattened image with the 64-dimensional vector of the vibration spectrum to obtain a 2112-dimensional feature; A24: Fully connected layer, including fully connected layer 1 and fully connected layer 2. Fully connected layer 1 is used to reduce the 2112-dimensional features to 124 dimensions, and fully connected layer 2 is used to reduce the 124-dimensional features to 3 dimensions.

[0038] As described above, the present invention sets the resistivity image channel and the vibration spectrum feature channel through the input layer, and inputs the resistivity single-channel grayscale image and the 64-dimensional vector of the normalized energy value respectively, integrates multi-source data information, and provides rich features for defect analysis; the image processing branch gradually extracts the deep-level features of the resistivity image through three convolution layers with different numbers of filters, thereby enhancing the ability to capture image features; the splicing layer splices the image flattened vector with the vibration spectrum vector, integrating the image and spectrum features, so that the model can comprehensively consider different types of data information; the fully connected layer gradually compresses the feature dimension through two dimensionality reductions, and finally obtains a 3D feature, which not only reduces the amount of model calculation, but also accurately extracts key features for defect type and preliminary coordinate analysis, while ensuring the accuracy of the analysis, and improves the computational efficiency of the model, making it more suitable for edge computing scenarios.

[0039] Furthermore, the data fusion module uses graph neural networks to associate multimodal data, specifically including: S41: Spatiotemporal alignment, which synchronizes all sensor data in time through GPS clock tags and spatially registers the sensor data; S42: Graph neural network analysis: Graph neural network is used to analyze the spatiotemporally aligned data to obtain the predicted points and defect probabilities of geomembrane defects. S43: Generate a three-dimensional defect map, divide the space into 0.1m×0.1m×0.1m voxels, assign a defect probability value to each voxel, and visualize the defect probability value through color.

[0040] As described above, the data fusion module in the present invention has obvious advantages in technical solutions of associating multimodal data using graph neural networks: the spatiotemporal alignment link uses GPS clock tags to achieve time synchronization and spatial registration of all sensor data, eliminating the deviations of different sensors in the time and space dimensions, and providing a unified and accurate data basis for subsequent analysis; the graph neural network analysis link uses its powerful relational modeling capabilities to conduct in-depth mining of the spatiotemporal aligned data, accurately obtain the predicted points and defect probabilities of geomembrane defects, and improve the accuracy and reliability of defect prediction; the three-dimensional defect map generation link divides the space into fine voxels and assigns defect probability values, and through color visualization output, intuitively presents the spatial distribution of geomembrane defects, which facilitates inspection personnel to quickly understand the defect location and severity, provides clear guidance for subsequent repair work, and effectively improves the efficiency and quality of geomembrane defect detection and treatment.

[0041] Furthermore, when using graph neural networks to analyze the data after spatiotemporal alignment to obtain the predicted points and defect probabilities of geomembrane defects, the structure of the graph neural network is positioned as follows: each sensor is a node with a feature vector, and edges are established when the physical distance between nodes is less than one meter. The graph convolution layer is used to aggregate the information of neighboring nodes, and the attention mechanism is introduced to dynamically allocate the association weights between nodes. Finally, the defect probability of each node is output through the Sigmoid function.

[0042] As described above, the present invention takes each sensor as a node and assigns a feature vector to it, constructs a clear data representation, and can fully explore the characteristics of the sensor itself; edges are established between nodes based on a physical distance of less than one meter, and the connection relationship between nodes is reasonably constructed, which conforms to the actual physical scene; the graph convolution layer is used to aggregate the information of neighboring nodes, effectively integrating the data in the local area and enhancing the model's learning ability for local features; the attention mechanism is introduced to dynamically allocate the association weights between nodes, so that the model can automatically adjust the weights according to the actual degree of association between different nodes, thereby improving the model's ability to model complex relationships; finally, the defect probability of each node is output through the Sigmoid function, which intuitively reflects the possibility of defects at the location of each sensor, provides a reliable basis for accurately locating the geomembrane defect prediction point, and helps to improve the accuracy and efficiency of defect detection.

[0043] Furthermore, the digital twin module simulates the aging process of geomembranes and dynamically optimizes alarm thresholds and life predictions, specifically including: S51: Material attenuation equation modeling. Based on the characteristics of HDPE materials, a tensile strength attenuation equation is established to quantify the effects of time, temperature, and load on aging. The principle formula of the tensile strength attenuation equation is: ; in, is the initial tensile strength, is the time attenuation coefficient, is the first stress sensitivity coefficient, and , is the second stress sensitivity coefficient, and , is the cumulative load stress; S52: Real-time data input, using fiber optic temperature measurement, MEMS pressure sensor, and humidity sensor to collect environmental parameters including temperature, stress, and humidity in real time; S53: Dynamic threshold adjustment, combining rule engine and random forest model prediction to optimize alarm triggering conditions; S54: Remaining life calculation, using finite element simulation to simulate the aging rate of stress concentration areas and generate a remaining life thermodynamic map; S55: Risk classification, which divides risk levels according to defect probability and remaining life to guide repair priorities; S56: Maintenance suggestion generation, automatically prompting geomembrane replacement or reinforcement plans for areas where the lifespan is less than the set threshold and the defect probability is greater than 70%; S57: Threshold adaptive adjustment, dynamically adjusting sensor sensitivity according to environmental changes; S58: Visual output, showing life prediction, risk distribution and maintenance work orders through WebGL interactive interface or CAD compatible format.

[0044] As described above, the material attenuation equation modeling in the present invention quantifies the effects of time, temperature, and load on aging based on the characteristics of HDPE materials, providing a scientific basis for accurately simulating the aging process of geomembranes; real-time data input collects environmental parameters through multiple sensors, ensuring the timeliness and accuracy of the data, and providing reliable data support for subsequent analysis; dynamic threshold adjustment combines the rule engine and random forest model to predict and optimize the alarm trigger conditions, improving the accuracy and adaptability of the alarm; the remaining life calculation uses finite element simulation to generate a heat map, which intuitively presents the aging status and remaining life of each area of ​​the geomembrane; risk grading is divided into levels according to defect probability and remaining life, which helps to rationally plan repair work and improve resource utilization efficiency; maintenance suggestion generation automatically prompts replacement or reinforcement plans for specific areas, providing specific guidance for actual maintenance work; threshold adaptive adjustment dynamically adjusts sensor sensitivity according to environmental changes, enhancing the stability and reliability of the system; visual output displays key information through multiple interface formats, making it easier for managers to intuitively understand the status of the geomembrane, providing strong support for decision-making, and overall improving the intelligent level of geomembrane aging monitoring, life prediction, and maintenance management.

[0045] The method for detecting defects of geomembrane under soil and rock medium covering conditions includes the following steps: S61: Multimodal data acquisition, dynamic placement of electrode arrays and ultra-wideband radar scanning, real-time acquisition of resistivity, electromagnetic waves, vibration, and optical signals, marking abnormal areas; S62: Edge intelligent compression and detection, wavelet transform compresses ERT data, FFT truncates vibration spectrum, and lightweight CNN analyzes anomaly types and generates preliminary coordinates; S63: Multi-source data fusion, spatiotemporal alignment of sensor data, and graph neural network integration of electrical, acoustic, and temperature features to generate a 3D defect probability map; S64: Digital twin and life prediction, based on material aging models and real-time environmental data, dynamically adjusts alarm thresholds, predicts remaining life and divides risk levels.

[0046] As described above, the multimodal data acquisition in the present invention collects multiple signals in real time and marks abnormal areas by dynamically deploying multiple devices, comprehensively obtaining information about the geomembrane and its surrounding environment, and providing a rich data basis for accurate detection; the edge intelligent compression and detection link uses wavelet transform, FFT truncation and lightweight CNN to effectively compress data, analyze abnormal types and generate preliminary coordinates, while reducing data transmission and storage costs, quickly locking the defect location and improving detection efficiency; multi-source data fusion generates a three-dimensional defect probability map through spatiotemporal alignment and graph neural network fusion of multi-source features, intuitively presenting the defect distribution and enhancing the accuracy and reliability of defect location; digital twins and life prediction dynamically adjust the alarm threshold, predict the remaining life and divide the risk level based on material aging models and real-time environmental data, realizing dynamic monitoring and scientific evaluation of the geomembrane status, providing a strong basis for timely repair and maintenance, helping to extend the service life of the geomembrane and ensure the safe and stable operation of the project.

[0047] Example 1: Landfill geomembrane defect detection and repair Technical solution application overview: Under the condition of landfill cover layer (8m thickness), multimodal adaptive sensing module is deployed: Dynamic electrode array deployment: Flexible electrodes were deployed at 5-meter intervals. Data was collected at night (8:00 PM - 6:00 AM). Combined with ultra-wideband radar scanning, a three-dimensional resistivity distribution map was generated, marking low-resistance anomaly areas (resistivity <80 Ω·m). Leakage vibration signals were captured synchronously using distributed optical fibers and MEMS sensors to eliminate mechanical noise interference.

[0048] Data compression and edge analytics: The ERT data was compressed to 10% using Daubechies wavelet transform, and the 0-500 Hz frequency band of the vibration signal was intercepted. A lightweight CNN model (input 64×64 resistivity image + 64-dimensional vibration spectrum) was used to classify two hole defects (confidence level >90%).

[0049] Multi-source data fusion: After spatiotemporal alignment, a graph neural network (building edges when node distance is less than 1m and integrating resistivity, vibration, and temperature features) generates a three-dimensional defect map and locates three joint leakage points (with a probability of >95%).

[0050] Digital Twin Optimization: Based on the HDPE aging model and real-time load / humidity data, the remaining life in high-risk areas is predicted to be less than 1 year, and the resistivity threshold in the rainy season is dynamically adjusted to 70Ω·m.

[0051] Closed-loop repair execution: The UWB-guided robot drilled and grouted defects >5cm (epoxy resin cured in 30 minutes), and the drone sprayed polyurea material (3mm anti-seepage layer) on small holes. After the repair, the leakage volume decreased by 98%.

[0052] Technical effects: Detection accuracy: ±0.1m positioning, 1cm hole detection rate >95%. Efficiency improvement: Deployment time reduced by 50%, repair efficiency reaches 3 sites / hour. Cost optimization: Electrode usage reduced by 40%, and operation and maintenance costs reduced by 35%.

[0053] Example 2: Health monitoring and maintenance of geomembrane in mine tailings ponds Technical solution application overview: In the tailings pond scenario covered by highly conductive clay, the system performs the following operations: Multimodal sensing: The electrical impedance method is prohibited, and acoustic sensing (fiber optic vibration + acoustic emission) and UWB radar (low frequency 0.1GHz penetrating 10m covering layer) are preferred, combined with salt solution tracing to verify the leakage point.

[0054] Edge computing optimization: The vibration signal is FFT-compressed and transmitted to the edge node, where a lightweight CNN model identifies microcracks (with an 85% confidence level), triggering high-density ERT encryption detection (with a spacing of 0.5 m).

[0055] Data fusion and simulation: GNN analysis correlates vibration energy with dielectric constant anomalies. The digital twin predicts a remaining life of six months in high-stress areas and dynamically lowers the vibration threshold to 0.3 m / s².

[0056] Automated maintenance: The repair robot injects grout to reinforce the cracked area based on the three-dimensional defect map (accuracy ±0.1m), and the drone simultaneously sprays anti-corrosion coating to extend the service life of the geomembrane by more than 2 years.

[0057] Technical effects: Adaptability to complex media: Detection accuracy in conductive clay areas exceeds 90%, with a 50% reduction in false alarms. Extended service life: Digital twin model error is less than 8%, and maintenance response time in high-risk areas is less than 4 hours. Enhanced safety: Unmanned repairs are possible in hazardous environments (deeply covered areas and steep slopes), reducing manual intervention by 80%.

[0058] The embodiments of the present invention are described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Various changes can be made within the scope of knowledge of those skilled in the art without departing from the spirit of the present invention.

Claims

1. Geomembrane defect detection system under soil and rock medium covering conditions; its characteristics are: include: Multimodal adaptive sensing module, used to collect electrical, acoustic, and electromagnetic wave signals in soil and rock media in real time through a multimodal sensor network; Edge computing module, used to compress data in real time and trigger high-precision algorithms through edge computing technology; A data fusion module for correlating multimodal data using graph neural networks; Digital twin module, used to simulate the aging process of geomembranes, dynamically optimize alarm thresholds and life prediction; The decision-making and application module is used to drive sound and light alarms, coordinate positioning, and automated repair equipment based on data analysis results, achieving closed-loop control of the entire process from defect detection to repair execution.

2. The geomembrane defect detection system under soil and rock medium covering conditions according to claim 1 is characterized in that: The multimodal adaptive sensing module specifically includes: A11: Electrode array module, used to locate leaks by dynamically deploying flexible electrodes at a high density and detecting resistivity changes. The module collects data between 8:00 PM and 6:00 AM daily, outputs a three-dimensional resistivity distribution map, and marks areas of abnormally low resistance. A12: Electromagnetic wave detection module, used to penetrate deep coverage using ultra-wideband radar and identify areas with abnormal dielectric constants; A13: Vibration module, used to capture vibration signals through optical fiber and MEMS sensors to distinguish noise from real defects; A14: Optical auxiliary module, used to combine mobile phone infrared and laser scanning to screen shallow temperature anomalies and surface deformation. A15: Salt solution tracer unit, used to verify the leak location by conductivity changes after injecting salt water.

3. The geomembrane defect detection system under soil and rock medium covering conditions according to claim 2 is characterized in that: The electrode array module is laid out using a dynamic density strategy. The specific steps include: S11: Preliminary scanning: First, metal electrodes are laid out with an initial electrode spacing of 5m, and the geological radar is started simultaneously to perform a large-scale scan. The electrical impedance and electromagnetic wave reflection data are obtained using the laid metal electrodes and the geological radar to generate a preliminary resistivity distribution map and a reflection anomaly heat map; S12: Identify abnormal areas. Mark the ERT low-resistance area and the GPR reflection amplitude mutation area as suspected defects. Use the DBSCAN algorithm to cluster the abnormal points, eliminate isolated noise, and determine the scope of the area that needs to be encrypted. S13: Dynamic encrypted deployment, inserting mobile electrodes at 0.5m intervals within the boundaries of the clustered abnormal area; S14: Real-time data fusion, combined with the newly added electrode data, updates the ERT imaging, and iterates until the detection resolution meets the requirements; S15: Resource recovery and reuse. When the encrypted area is detected and confirmed to be free of defects, the mobile electrodes are recovered and reused in other areas.

4. The geomembrane defect detection system under soil and rock medium covering conditions according to claim 3 is characterized by: When the edge computing module compresses data in real time and triggers high-precision algorithms through edge computing technology, it specifically includes: S21: Data compression, compressing the raw sensor data through wavelet transform, reducing the volume of the raw sensor data to 10% of the original volume; S22: Anomaly detection, setting a dynamic threshold based on the initial data collected by the multimodal adaptive sensing module, and detecting the data of the original sensor according to the set dynamic threshold; S23: Model inference, using the lightweight CNN built into the edge computing module to analyze abnormal data and analyze the specific defect type and preliminary coordinates.

5. The geomembrane defect detection system under soil and rock medium covering conditions according to claim 4 is characterized in that: The edge computing module compresses the raw sensor data through wavelet transform, reducing the volume of the raw sensor data to 10% of its original size. Specifically, it includes: S31: Perform multi-level decomposition of the ERT voltage signal using the Daubechies wavelet basis to separate the low-frequency approximate component and the high-frequency detail component; S32: Perform hard threshold filtering on high-frequency detail components, retain low-frequency energy, and discard low-amplitude noise coefficients; S33: Based on the retained approximation and threshold processed detail components, the compressed signal is restored by inverse wavelet transform; S34: Perform FFT calculation on the vibration sensor waveform to generate spectrum energy distribution; S35: intercepts the effective frequency band of 0-500Hz and discards high-frequency noise; S36: Normalize the amplitude to an 8-bit integer, ignore the phase information, and store the non-zero coefficients.

6. The geomembrane defect detection system under soil and rock medium covering conditions according to claim 5 is characterized in that: When using the lightweight CNN built into the edge computing module to analyze abnormal data and analyze the specific defect type and preliminary coordinates, the structure of the lightweight CNN includes: A21: Input layer, including the resistivity image channel and the vibration spectrum feature channel. The resistivity image channel is used to input a single-channel grayscale resistivity image of size 64×64×1. The vibration spectrum feature is used to input a 64-dimensional vector of normalized energy values ​​of vibration data after FFT processing. A22: Image processing branch, including convolution layer 1, convolution layer 2, and convolution layer 3. Convolution layer 1 is a 3×3 convolution and 8 filters, convolution layer 2 is a 3×3 convolution and 16 filters, and convolution layer 3 is a 3×3 convolution and 32 filters. A23: concatenation layer, used to concatenate the 2048-dimensional vector of the flattened image with the 64-dimensional vector of the vibration spectrum to obtain a 2112-dimensional feature; A24: Fully connected layer, including fully connected layer 1 and fully connected layer 2. Fully connected layer 1 is used to reduce the 2112-dimensional features to 124 dimensions, and fully connected layer 2 is used to reduce the 124-dimensional features to 3 dimensions.

7. The geomembrane defect detection system under soil and rock medium covering conditions according to claim 6 is characterized in that: The data fusion module uses graph neural networks to associate multimodal data, specifically including: S41: Spatiotemporal alignment, which synchronizes all sensor data in time through GPS clock tags and spatially registers the sensor data; S42: Graph neural network analysis: Graph neural network is used to analyze the spatiotemporally aligned data to obtain the predicted points and defect probabilities of geomembrane defects. S43: Generate a three-dimensional defect map, divide the space into 0.1m×0.1m×0.1m voxels, assign a defect probability value to each voxel, and visualize the defect probability value through color.

8. The geomembrane defect detection system under soil and rock medium covering conditions according to claim 7 is characterized in that: When using graph neural networks to analyze data that has been aligned in time and space to obtain the predicted points and defect probabilities of geomembrane defects, the structure of the graph neural network is positioned as follows: each sensor is a node with a feature vector, and edges are established when the physical distance between nodes is less than one meter. The graph convolution layer is used to aggregate information from neighboring nodes, and the attention mechanism is introduced to dynamically allocate association weights between nodes. Finally, the defect probability of each node is output through the Sigmoid function.

9. The geomembrane defect detection system under soil and rock medium covering conditions according to claim 8 is characterized in that: The digital twin module simulates the aging process of geomembranes and dynamically optimizes alarm thresholds and life predictions, specifically including: S51: Material attenuation equation modeling, establishing a tensile strength attenuation equation based on HDPE material properties, and quantifying the effects of time, temperature, and load on aging; S52: Real-time data input, using fiber optic temperature measurement, MEMS pressure sensor, and humidity sensor to collect environmental parameters including temperature, stress, and humidity in real time; S53: Dynamic threshold adjustment, combining rule engine and random forest model prediction to optimize alarm triggering conditions; S54: Remaining life calculation, using finite element simulation to simulate the aging rate of stress concentration areas and generate a remaining life thermodynamic map; S55: Risk classification, which divides risk levels according to defect probability and remaining life to guide repair priorities; S56: Maintenance suggestion generation, automatically prompting geomembrane replacement or reinforcement plans for areas where the lifespan is less than the set threshold and the defect probability is greater than 70%; S57: Threshold adaptive adjustment, dynamically adjusting sensor sensitivity according to environmental changes; S58: Visual output, showing life prediction, risk distribution and maintenance work orders through WebGL interactive interface or CAD compatible format.

10. The method for detecting defects of geomembranes under soil and rock medium covering conditions according to claim 9, characterized in that: The following steps are involved: S61: Multimodal data acquisition, dynamic placement of electrode arrays and ultra-wideband radar scanning, real-time acquisition of resistivity, electromagnetic waves, vibration, and optical signals, marking abnormal areas; S62: Edge intelligent compression and detection, wavelet transform compresses ERT data, FFT truncates vibration spectrum, and lightweight CNN analyzes anomaly types and generates preliminary coordinates; S63: Multi-source data fusion, spatiotemporal alignment of sensor data, and graph neural network integration of electrical, acoustic, and temperature features to generate a 3D defect probability map; S64: Digital twin and life prediction, based on material aging models and real-time environmental data, dynamically adjusts alarm thresholds, predicts remaining life and divides risk levels.

Images