Tunnel lining vault anti-disengaging monitoring method, system, equipment and medium

Abstract

The invention relates to the technical field of tunnel engineering construction, and discloses a tunnel lining vault anti-disengaging monitoring method, system and device and a medium, and the method comprises the following steps: obtaining an original design model and point cloud data of a lining vault, constructing a target model, and marking pre-embedded sensor points and candidate mobile sampling points; dynamically collecting a pressure signal and a trigger signal of a target node in a concrete pouring process, and processing the pressure signal and the trigger signal to generate time sequence data; performing time sequence anomaly detection through a pre-trained bidirectional long-short-term memory network, and outputting void early warning; deducing void probability distribution by using a graph neural network based on time sequence data and early warning; dynamically updating a sampling strategy according to the uncertainty of the distribution map, and iteratively optimizing a monitoring result; and finally, generating a slurry supplementing instruction for the area with the probability value exceeding the slurry supplementing threshold value. According to the embodiment of the invention, the global coverage of the pre-embedded sensor is combined with the dynamic supplementation of the candidate points, and the problem of a space blind area in a traditional method is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel engineering construction, and in particular to a tunnel lining vault anti-cavitation monitoring method, system, equipment and medium. Background Art

[0002] In tunnel construction, lining vault voids are a key safety hazard. During concrete pouring, cavities can easily form in the vault due to factors such as impeded slurry flow and solidification shrinkage. This can lead to a decrease in the lining's bearing capacity, accelerated steel corrosion, and even safety hazards such as collapse. Therefore, effective monitoring of vault voids is crucial for ensuring tunnel construction quality and operational safety.

[0003] Existing technology utilizes a three-dimensional pressure sensor placed within the vault's exhaust pipe and an electromagnetic wave sensor clipped to the grouting hole in the formwork. Through a buoyancy-lever mechanism, these sensors convert pressure changes in the discharged grout during concrete pouring, or trigger signals indicating overflow, into electrical signals, enabling online monitoring and alarming of the vault's voiding status. After pouring is complete, the sensors continue to monitor the pressure distribution at the interface between the lining and the rock mass, assisting in assessing the structural density and bond quality later in the process.

[0004] However, the aforementioned sensor system is only discretely deployed at the pre-buried exhaust pipe and the overflow hole. The monitoring range is limited to these fixed locations and cannot cover the entire curved surface of the vault. If a void or loose concrete occurs between two sensor points or outside the transmission mechanism's radius, the local void may not be discovered and remedied in a timely manner, creating the risk of missed detection in monitoring "blind spots." Summary of the Invention

[0005] In view of this, the present application provides a tunnel lining vault anti-cavitation monitoring method, system, equipment and medium to solve the above problems.

[0006] In a first aspect, a tunnel lining vault anti-cavitation monitoring method is provided, which is applied to a monitoring platform and includes: Obtain the original design model of the lining vault, obtain point cloud data of the inner surface of the lining vault through an acquisition device, construct a target model based on the point cloud data and the original design model, mark the embedded sensor point and the first candidate mobile sampling point on the target model, and use any point among the embedded sensor point and the first candidate mobile sampling point as the target node; Dynamic sampling is performed based on the target model. Dynamic sampling includes collecting signal data of the target node during the concrete pouring process. The signal data includes pressure signal data at the embedded exhaust pipe and trigger signal data of the template overflow hole. Processing the signal data to generate first time series data; Input the first time series data into a pre-trained bidirectional long short-term memory network model to perform time series anomaly detection. When an anomaly is detected, an air gap warning is output. The air gap warning includes the warning time and the node identifier of the corresponding target node. Based on the first time series data and the void warning, the lining vault is inferred through the graph neural network to obtain the first void probability distribution map; Determine the uncertainty of the first void probability distribution graph, and dynamically update the sampling strategy based on the uncertainty; After the dynamic sampling strategy is updated and executed, the second time series data is obtained and the first void probability distribution map is updated. When the uncertainty of all target nodes is lower than the preset threshold, the dynamic sampling is terminated and the second void probability distribution map is output. For the node area in the second void probability distribution map whose probability value exceeds the preset grouting threshold, a grouting operation instruction is generated and sent to the on-site construction terminal.

[0007] The above technical solution, by fusing point cloud data with the original design model to construct a target model, and combining pre-buried sensor points with candidate mobile sampling points, implements dynamic signal acquisition, time series anomaly detection and graph neural network inference, which can accurately locate the position and range of the lining vault void in real time; further combined with uncertainty assessment and dynamic sampling strategy update, it realizes adaptive optimization of monitoring accuracy and efficiency, and finally generates grouting operation instructions that can be directly issued to the construction terminal, solving the spatial blind spot problem of traditional methods and greatly improving the intelligence level of tunnel vault anti-voiding monitoring and the accuracy of construction guidance.

[0008] Optionally, point cloud data of the inner surface of the lining vault is acquired through an acquisition device. Based on the point cloud data and the original design model, a target model is constructed, including: Use laser scanning or drone photogrammetry to collect point clouds of the inner surface of the lining vault and pre-process the collected point cloud data; The pre-processed point cloud data is reconstructed into a mesh surface using a triangulation algorithm; Perform rigid body alignment and fusion of the mesh surface and the original design model to generate a target model with coordinate mapping relationship.

[0009] The above technical solution utilizes laser scanning or drone photogrammetry to acquire point cloud data of the inner surface of the lining vault. This, combined with a triangulation algorithm, reconstructs the mesh surface and integrates it with the original design model. This allows for the rapid and highly accurate construction of a target model reflecting the vault's actual geometry. This process removes noise points through point cloud preprocessing and ensures coordinate consistency between the design model and the actual structure through rigid body alignment. This solves the problems of low efficiency and large errors associated with traditional manual measurement, provides a precise three-dimensional spatial reference for subsequent sensor placement and void analysis, and ensures the geometric accuracy of monitoring node location annotation and signal analysis, enabling accurate mapping of monitoring data to the actual structural locations.

[0010] Optionally, processing the signal data to generate the first time series data includes: The pressure signal data is calibrated according to the preset sampling frequency; Perform pulse counting and duration analysis on the trigger signal data to extract the time window of each trigger event. The time window starts at the beginning of the rising edge of the trigger signal and ends at the end of the falling edge of the trigger signal. For the calibrated pressure signal data, in each time window, the high-frequency noise in the current time window is removed by the wavelet threshold denoising algorithm, and the data in the current time window is reconstructed to obtain the reconstructed signal data; The reconstructed signal data corresponding to each time window are spliced ​​to obtain an overall signal set, and the overall signal set is normalized to generate first time series data.

[0011] The above technical solution designs a multi-stage data processing process for pressure signals and trigger signals: the pressure signal is calibrated according to a preset frequency to ensure a consistent data time base; key event periods during the concrete pouring process (such as the onset and end of overflow) are accurately located through pulse counting and time window analysis of the trigger signal; a wavelet threshold denoising algorithm is used within the time window to remove high-frequency noise and reconstruct the signal, effectively retaining the low-frequency characteristic signal reflecting the void state; and finally, normalization is used to eliminate data dimensional differences. This processing method significantly improves the signal-to-noise ratio and feature integrity of the original signal, providing high-quality input data for subsequent time series model analysis, avoiding anomaly detection errors caused by noise interference, and enhancing the monitoring system's anti-interference ability.

[0012] Optionally, the first time series data is input into a pre-trained bidirectional long short-term memory network model to perform time series anomaly detection. When an anomaly is detected, outputting a gap warning includes: Adding position code to the first time series data; Inputting the first time series data into the bidirectional long short-term memory network model, and extracting time series features based on position coding through the encoder of the bidirectional long short-term memory network model; Each time series feature is scored through the decoder of the bidirectional long short-term memory network model, and a shortage warning is output when the score is higher than the preset abnormal score threshold.

[0013] The above technical solution, by adding positional encoding to time series data, converts the sequential features of the time series into spatialized inputs that can be recognized by the model. Combined with the encoder-decoder structure of the bidirectional long short-term memory network model, it can simultaneously capture the past and future contextual features of the signal and effectively extract long-distance dependent time series change patterns. By scoring each time series feature and setting an anomaly threshold, quantitative detection of abnormal fluctuations in degassing-related signals is achieved. Compared with the traditional unidirectional time series model, the bidirectional long short-term memory network model can more comprehensively analyze the signal evolution trend, improve the sensitivity and accuracy of anomaly detection, ensure timely triggering of warnings in the early stages of degassing risks, and gain a time window for construction intervention.

[0014] Optionally, based on the first time series data and the void warning, the lining vault is inferred through a graph neural network to obtain a first void probability distribution map including: Call the three-dimensional coordinates of each target node in the target model, calculate the Euclidean distance between each target node, and form a weighted adjacency matrix; The first time series data and void warning corresponding to each target node are spliced ​​in node order to generate a feature vector matrix; the weighted adjacency matrix and the feature vector matrix are input into a multi-layer graph attention network, and the feature vectors of adjacent target nodes are aggregated through adaptive attention weights to output the first void probability distribution map covering the entire vault.

[0015] The above technical solution constructs a weighted adjacency matrix based on the three-dimensional coordinates of each node in the target model, quantifying the spatial correlation between monitoring nodes; splicing time series data and void warning information into a feature vector matrix, inputting it into a multi-layer graph attention network for feature aggregation, and dynamically allocating the influence of adjacent nodes through adaptive attention weights. This method breaks through the limitations of traditional single-point monitoring and uses graph structure to model the spatial correlation relationship of vault nodes. It can infer the overall void probability distribution from local abnormal signals and generate a visual risk map covering the entire vault. Compared with independent analysis of single-point data, the graph neural network method can more accurately identify the spatial propagation characteristics of the void area, provide construction personnel with an intuitive risk distribution reference, and facilitate the formulation of a global quality control strategy.

[0016] Optionally, determining the uncertainty of the first empty probability distribution graph and dynamically updating the sampling strategy based on the uncertainty includes: Calculate the uncertainty of each target node based on the first empty probability distribution graph; Sort the target nodes from high to low according to their uncertainty, and select the target nodes whose uncertainty is greater than the preset threshold as the next round of sampling objects; Perform sensor wake-up or mobile deployment operations on the target node where the sampling object is located to conduct a new round of signal collection.

[0017] The above technical solution calculates the uncertainty of each node in the void probability distribution diagram and dynamically sorts them, focusing monitoring resources on high-uncertainty areas and performing targeted sampling by waking up or moving sensors. This dynamic sampling strategy avoids the blindness and redundancy of traditional fixed sampling, significantly reduces the amount of data collected while ensuring monitoring accuracy, and reduces sensor energy consumption and data processing costs. By iteratively updating the sampling strategy, the uncertainty of each node is gradually reduced, and the monitoring process progresses from coarse-grained detection to refined positioning, forming a closed-loop feedback mechanism of "detection-inference-optimization", ultimately achieving high-precision positioning of the void area, providing a reliable basis for subsequent grouting operations.

[0018] Optionally, for a node area in the second void probability distribution map where the probability value exceeds a preset grouting threshold, generating a grouting operation instruction and sending it to an on-site construction terminal includes: For each target node whose probability value exceeds a preset grouting threshold in the second void probability distribution map, adjacent over-threshold nodes are merged according to a spatial clustering algorithm to determine a continuous void area; For continuous void areas, a grouting operation plan is generated based on the area, depth estimation, and construction accessibility, and the grouting operation plan is formatted into grouting operation instructions that can be recognized by the construction terminal; The grouting operation instructions are sent to the on-site construction terminal and the execution status is fed back.

[0019] The above technical solution uses a spatial clustering algorithm to merge adjacent over-threshold nodes in high-probability void node areas, accurately identifying continuous void areas and avoiding scattered grouting instructions caused by single-point anomalies. It also generates a grouting plan based on regional area, depth estimation, and construction accessibility to ensure the engineering feasibility and cost-effectiveness of the plan. The plan is formatted as instructions recognizable by the construction terminal and the execution status is fed back, achieving seamless integration between the monitoring system and the construction site. This technology solves the subjectivity and inefficiency of traditional manual judgment of grouting areas. By generating standardized operation instructions through automated algorithms, it improves the pertinence and efficiency of grouting operations, ensures the integrity and durability of the tunnel lining structure, and technically reduces the risk of later structural damage caused by void residues.

[0020] In a second aspect of the present application, a tunnel lining vault anti-cavitation monitoring system is provided, which includes a model building module, a dynamic signal acquisition module, a time series data generation module, a time series anomaly detection module, a cavity probability inference module, a sampling strategy optimization module, and a grouting instruction generation module, wherein: a model building module configured to obtain an original design model of the lining vault, obtain point cloud data of the inner surface of the lining vault through an acquisition device, build a target model based on the point cloud data and the original design model, mark the embedded sensor point and the first candidate mobile sampling point on the target model, and use any point among the embedded sensor point and the first candidate mobile sampling point as a target node; A dynamic signal acquisition module is configured to perform dynamic sampling based on the target model. Dynamic sampling includes collecting signal data of target nodes during concrete pouring. The signal data includes pressure signal data at the embedded exhaust pipe and trigger signal data of the template overflow hole. a time series data generating module configured to process the signal data and generate first time series data; a time series anomaly detection module configured to input the first time series data into a pre-trained bidirectional long short-term memory network model to perform time series anomaly detection, and output an air gap warning when an anomaly is detected, the air gap warning including a warning time and a node identifier of a corresponding target node; a void probability inference module configured to infer the lining vault through a graph neural network based on the first time series data and the void warning to obtain a first void probability distribution map; a sampling strategy optimization module configured to determine an uncertainty of the first void probability distribution graph and dynamically update the sampling strategy based on the uncertainty; The grouting instruction generation module is configured to obtain the second time series data and update the first void probability distribution map after the dynamic sampling strategy is updated and executed. When the uncertainty of all target nodes is lower than the preset threshold, the dynamic sampling is terminated and the second void probability distribution map is output. For the node area in the second void probability distribution map whose probability value exceeds the preset grouting threshold, a grouting operation instruction is generated and sent to the on-site construction terminal.

[0021] In the third aspect of the present application, an electronic device is provided, including a processor, a memory, a user interface and a network interface, the memory is used to store instructions, the user interface and the network interface are both used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device performs any of the methods described above.

[0022] In a fourth aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores instructions. When the instructions are executed, any one of the methods described above is executed.

[0023] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: By integrating point cloud data with the original design model to construct a target model, and combining pre-buried sensor points with candidate mobile sampling points, dynamic signal acquisition, time series anomaly detection and graph neural network inference are implemented, which can accurately locate the position and range of the lining vault void in real time; further combining uncertainty assessment and dynamic sampling strategy updates, adaptive optimization of monitoring accuracy and efficiency is achieved, and finally grouting operation instructions are generated that can be directly issued to the construction terminal, solving the spatial blind spot problem of traditional methods and greatly improving the intelligence level of tunnel vault anti-voiding monitoring and the accuracy of construction guidance. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is an exemplary system architecture diagram of a tunnel lining vault anti-cavitation monitoring method or a tunnel lining vault anti-cavitation monitoring system using the present application; Figure 2 This is a flow chart of a tunnel lining vault anti-cavitation monitoring method according to an embodiment of the present application; Figure 3 This is a schematic diagram of a module of a tunnel lining vault anti-cavitation monitoring system according to an embodiment of the present application; Figure 4 It is a structural diagram of an electronic device disclosed in an application embodiment.

[0025] Explanation of the accompanying drawings: 100, system architecture; 101, first terminal device; 102, second terminal device; 103, third terminal device; 104, network; 105, server; 301, model building module; 302, dynamic signal acquisition module; 303, time series data generation module; 304, time series anomaly detection module; 305, void probability inference module; 306, sampling strategy optimization module; 307, slurry filling instruction generation module; 401, processor; 402, communication bus; 403, user interface; 404, network interface; 405, memory. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments.

[0027] In the description of the embodiments of this application, words such as "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "for example" or "for instance" in the embodiments of this application should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "for example" or "for instance" is intended to present the relevant concepts in a concrete manner.

[0028] In the description of the embodiments of the present application, the term "multiple" means two or more. For example, multiple systems refer to two or more systems, and multiple screen terminals refer to two or more screen terminals. In addition, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized.

[0029] This embodiment discloses a tunnel lining vault anti-cavitation monitoring method or a tunnel lining vault anti-cavitation monitoring system. Figure 1 A schematic diagram of an exemplary system architecture of an embodiment of a tunnel lining vault anti-cavitation monitoring method or a tunnel lining vault anti-cavitation monitoring system to which the present application may be applied is shown.

[0030] like Figure 1 As shown, system architecture 100 may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. Network 104 is a medium for providing communication links between terminal devices 101, 102, 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables.

[0031] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as model training applications, video recognition applications, web browser applications, social platform software, etc.

[0032] Terminal devices 101, 102, 103 can be hardware or software. When terminal devices 101, 102, 103 are hardware, they can be various electronic devices with display screens, including but not limited to smart phones, tablet computers, e-book readers, MP3 (Moving Picture Experts Group Audio Layer III, Moving Picture Experts Compression Standard Audio Layer 3) players, MP4 (Moving Picture Experts Group Audio Layer IV, Moving Picture Experts Compression Standard Audio Layer 4) players, laptop computers and desktop computers, etc. When terminal devices 101, 102, 103 are software, they can be installed in the electronic devices listed above. It can be implemented as multiple software or software modules (for example, multiple software or software modules used to provide distributed services), or it can be implemented as a single software or software module. No specific limitation is made here.

[0033] When terminals 101, 102, and 103 are hardware, they may also be equipped with a video capture device. The video capture device may be any device capable of capturing video, such as a camera, a sensor, and the like. Users can use the video capture device on terminals 101, 102, and 103 to capture video.

[0034] The server 105 may be a server that provides various services, such as a background server that processes data displayed on the terminal devices 101, 102, and 103. The background server may analyze and process the received data, and may feed back the processing results (such as recognition results) to the terminal device.

[0035] It should be noted that the server can be either hardware or software. When the server is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. When the server is software, it can be implemented as multiple software or software modules (e.g., multiple software or software modules used to provide distributed services), or as a single software or software module. No specific limitations are given here.

[0036] It should be understood that Figure 1 The number of terminal devices, networks, and servers in the above description is merely illustrative. Any number of terminal devices, networks, and servers may be used as needed. In particular, if target data does not need to be acquired remotely, the above system architecture may not include a network, but may instead include only terminal devices or servers.

[0037] Figure 2 : is a flow chart of a tunnel lining vault anti-cavitation monitoring method in an embodiment of the present application, such as Figure 2 As shown, this embodiment includes: Step S201: Obtain the original design model of the lining vault, obtain point cloud data of the inner surface of the lining vault through an acquisition device, construct a target model based on the point cloud data and the original design model, mark the embedded sensor point and the first candidate mobile sampling point on the target model, and use any point among the embedded sensor point and the first candidate mobile sampling point as the target node.

[0038] For example, the original design model of the lining vault is obtained, for example, from a tunnel design institute, using the original CAD (Computer Aided Design) 3D design file of the tunnel lining vault, which contains key information such as the circumferential lining segment assembly plan, vault cross-section lines, and segment interface locations. The original CAD 3D design file is converted to STL (stereolithography) format and imported into the monitoring platform, retaining attributes such as segment units, seam lines, and thickness tolerances in the model metadata.

[0039] Furthermore, acquisition equipment (such as a portable laser radar carried on a tripod) is deployed on site along the centerline of the arch to collect circumferential and axial point clouds. The point cloud and STL model are then incorporated into the registration module. First, a rough alignment is performed based on the plane features of the pipe segment interface. Then, the ICP (Iterative Closest Point) algorithm is applied iteratively until the residual error is ≤5mm, resulting in a fusion model with a coordinate mapping relationship. Pre-buried pressure / trigger sensor points are evenly arranged on the target model, and multiple mobile sampling candidate points are generated between them. The coordinates and numbers of all nodes are stored in a database. When the system is initialized, some nodes are selected in circumferential order as the first round of sampling targets. Subsequently, the target nodes can be adaptively adjusted based on the acquisition results and uncertainty.

[0040] In one possible implementation, point cloud data of the inner surface of the lining vault is acquired by an acquisition device, and a target model is constructed based on the point cloud data and the original design model. Specifically, the method includes: acquiring point cloud data of the inner surface of the lining vault by laser scanning or drone photogrammetry, and preprocessing the acquired point cloud data; reconstructing the preprocessed point cloud data into a mesh surface by using a triangulation algorithm; and rigidly aligning and fusing the mesh surface with the original design model to generate a target model with a coordinate mapping relationship.

[0041] Specifically, point cloud data of the inner surface of the lining vault is acquired through acquisition equipment, such as a portable lidar mounted on a measuring tripod. The instrument maintains a scanning distance of approximately 1.2m from the vault's inner surface and moves sequentially in horizontal and vertical steps of 0.1m, ensuring a point cloud overlap rate of at least 60% at each location. Simultaneously, a drone-mounted structured light camera is used to perform supplemental scanning at the junction of the vault's invert and side walls to obtain detailed data on complex curved areas. Coordinate reference calibration is performed using a multi-faceted reflective target before each measurement. After acquisition, the raw point cloud is coarsely filtered using a ground control station to remove external obstructions such as surface water and construction scaffolding.

[0042] Furthermore, the target model is constructed. For example, the preprocessed point cloud data and the converted STL design model are simultaneously loaded into the registration module of the monitoring platform. The initial alignment method based on plane features is applied to match the flat feature surface at the interface of the pipe segment with the corresponding plane in the point cloud data to complete the rough alignment. The ICP algorithm is further used for iterative registration until the average residual error between the model and the point cloud data is less than 5mm, generating a fused target model with coordinate mapping. The fused point cloud is triangulated and reconstructed using the Delaunay triangulation algorithm to generate a closed mesh surface, and the normal vector of each triangle unit on the surface is calculated to support subsequent sampling and positioning. Median filtering is used to smooth areas with high local noise to ensure that the mesh surface has no obvious high-frequency bumps.

[0043] Furthermore, according to engineering specifications, 32 pre-buried pressure sensor points (8×4) are evenly arranged on the inner surface of the arch along the axial and circumferential directions, ensuring that the minimum spacing is not less than 1m; at the same time, the three-dimensional coordinates, purpose (pressure / trigger) and number information of each point are recorded in the node database. Between the pre-buried points, a total of 48 first candidate mobile sampling points are generated on the grid surface every 0.5m for subsequent mobile sampling by robotic arms or portable sensors. These points are temporarily deployed, and the coordinates are also stored in the platform. During the system initialization phase, according to the uniform coverage strategy, 16 nodes are selected from the above 80 nodes (32 pre-buried points + 48 candidate points) in circumferential order as the first round of sampling targets. In each subsequent round, the selected objects can be dynamically adjusted from all 80 points based on the previous acquisition results and uncertainty indicators.

[0044] Step S202 : Dynamic sampling is performed based on the target model. The dynamic sampling includes collecting signal data of the target node during the concrete pouring process. The signal data includes pressure signal data at the embedded exhaust pipe and trigger signal data of the template overflow hole.

[0045] For example, the pre-buried exhaust pipe pressure sensor and the template overflow hole trigger switch arranged in the arch are activated according to the constructed target model. The pressure signal is filtered in real time at a high rate of 200Hz (1-50Hz bandpass) by the edge controller and marked with a node ID and timestamp. The trigger signal is generated by a photoelectric or micro switch when the slurry overflows. The edge controller performs hardware debouncing on the rising and falling edges and records the absolute time. Then, the processed pressure waveform and the trigger event list are reported to the monitoring platform together with the current perfusion status.

[0046] Step S203: Process the signal data to generate first time series data.

[0047] For example, the raw pressure current signal (4-20 mA) uploaded by the edge controller is first linearly converted into a pressure value (unit: kPa) according to the sensor calibration curve, and the timestamps of each channel are synchronized based on the perfusion start signal. At the same time, the pulse of the photoelectric switch is hardware-debounced and the time of each closing (rising edge) and opening (falling edge) is extracted to form an event list. Then, the pressure waveform is decomposed and reconstructed using the Daubechies-4 wavelet three-level decomposition and soft thresholding within the time window from 0.1 s before the start to 0.1 s after the end of each trigger event. If adjacent windows overlap, they are merged. Finally, all denoised waveform segments are seamlessly spliced ​​in chronological order, and missing samples caused by delays or blind spots are filled by linear interpolation. Then, the minimum and maximum values ​​of the overall waveform are calculated, and the pressure signal and the corresponding event duration are normalized to the [0, 1] interval and spliced ​​into a unified 2×N (N is a positive integer) matrix along the channel dimension. Finally, an absolute timestamp and node ID are attached to each sampling point, thereby generating a bidirectional long short-term memory network (LSTM) for subsequent processing. The first time series data used for Short-Term Memory (STM) time series anomaly detection.

[0048] In one possible implementation, the signal data is processed to generate first time series data, specifically including: calibrating the pressure signal data according to a preset sampling frequency; performing pulse counting and duration analysis on the trigger signal data to extract the time window of each trigger event, where the time window starts at the start moment of the rising edge of the trigger signal and ends at the end moment of the falling edge of the trigger signal; for the calibrated pressure signal data, within each time window, high-frequency noise within the current time window is removed by a wavelet threshold denoising algorithm, and the data within the current time window is reconstructed to obtain reconstructed signal data; the reconstructed signal data corresponding to each time window are spliced ​​to obtain an overall signal set, and the overall signal set is normalized to generate the first time series data.

[0049] Specifically, the on-site edge controller collects 4–20mA current signals from the exhaust pipes of each node at a preset frequency of 200Hz. On the platform side, the current values ​​are linearly mapped to pressure values ​​(unit: kPa) according to the sensor's factory calibration curve. The timestamps of all data channels are uniformly calibrated using the concrete pouring pump start signal as the time "zero point" to ensure synchronization of the pressure waveform timing between different nodes. Hardware debounce (jitter threshold 5ms) is performed on the pulse signals of the photoelectric switches equipped with each template overflow hole. The absolute time of each rising edge (initial slurry overflow) and falling edge (cessation of slurry overflow) is counted. The interval corresponding to each pair of rising and falling edges is considered a "trigger event time window," starting at the start time of the rising edge and ending at the end time of the falling edge. For example, if a slurry overflow lasts 0.8s, the event window is the 0.8s interval.

[0050] Furthermore, the calibrated pressure waveform within each trigger event window is decomposed into three levels using the Daubechies-4 wavelet. A threshold is automatically set based on the standard deviation of the noise in the current window, and soft threshold filtering is performed on the high-frequency coefficients of each layer. The thresholded coefficients are reconstructed into clean pressure signal segments using an inverse wavelet transform, fully removing high-frequency noise generated by construction vibration and electromagnetic interference. The reconstructed signal segments corresponding to all event windows are seamlessly spliced ​​together in chronological order of their start and end times. If data gaps of less than 0.02 seconds exist between adjacent windows (due to communication delays or trigger blind spots), linear interpolation is used to fill in the missing samples at the splicing points to ensure continuity across the entire pressure signal segment.

[0051] Furthermore, the minimum pressure value P of the entire signal after splicing is calculated min and the maximum pressure value P max ; Normalize the pressure value P(t) at any time t in the signal: Synchronize the duration of each trigger event Δt (seconds) to the maximum duration Δt of the entire segment max Normalize and generate an event intensity sequence of the same length as the pressure signal. It is spliced ​​with the normalized event intensity sequence along the channel dimension to obtain a 2×N matrix, where N is the total number of sampling points; the absolute timestamp and corresponding node ID are recorded for each sampling point, and the first time series data is finally output.

[0052] In step S204, the first time series data is input into a pre-trained bidirectional long short-term memory network model to perform time series anomaly detection. When an anomaly is detected, an air gap warning is output. The air gap warning includes the warning time and the node identifier of the corresponding target node.

[0053] Exemplarily, the first time series data matrix of each target node is read, and combined with the axial and circumferential coordinates of the node in the vault model, it is mapped into a position encoding vector of the same length as the time series using the sine and cosine functions, and spliced ​​into a 4×N input matrix with the normalized pressure signal and event intensity channel; then, the system sends the matrix into the pre-trained bidirectional long short-term memory network encoder in batches according to the nodes, traverses the sequence in both directions to extract forward and reverse hidden features, and the decoder performs a nonlinear transformation on the feature vector of each time step and outputs a scalar anomaly score. Once the score at a certain moment is higher than the preset threshold, the ID and absolute timestamp of this node are immediately recorded to generate an air gap warning.

[0054] In one possible implementation, the first time series data is input into a pre-trained bidirectional long short-term memory network model to perform time series anomaly detection, and an out-of-order warning is output when an anomaly is detected. Specifically, the method includes: adding position coding to the first time series data; inputting the first time series data into the bidirectional long short-term memory network model, and extracting time series features based on the position coding through the encoder of the bidirectional long short-term memory network model; and scoring each time series feature through the decoder of the bidirectional long short-term memory network model, and outputting an out-of-order warning when the score is higher than a preset anomaly score threshold.

[0055] Specifically, the first time series data matrix (shape 2×N) of each target node is read from the database, and the axial (along the tunnel axis coordinates) and circumferential (segment circumferential coordinates) positions of the node in the vault target model are queried; a one-dimensional position vector P1 is generated by linear interpolation along the time axis according to the axial coordinate, and a one-dimensional position vector P2 is generated by the circumferential coordinates. The sine / cosine function is used to map P1 and P2 to encoding vectors of the same length as the time series, respectively, and they are spliced ​​together with the pressure and event intensity channels to form a 4×N input matrix with position encoding.

[0056] Furthermore, each 4×N matrix with positional encoding is fed into the pre-trained bidirectional LSTM encoder in batches according to nodes; the encoder traverses the sequence in parallel in the forward and reverse time directions, extracting temporal features in the past and future windows, such as the pattern before and after a sharp drop in pressure, the low-pressure platform accompanied by continuous long-term triggering events, etc., and finally outputs the bidirectional concatenated hidden feature vector h at each time step. t The hidden feature sequence {h1, h2, ..., h n} is passed into the bidirectional LSTM decoder moment by moment; the decoder processes the input feature h at each time step. t Perform nonlinear transformation and generate a scalar anomaly score s t , reflecting the degree of deviation of the signal from the normal perfusion pattern at that moment. Based on the historical monitoring data, a scoring threshold S0 is determined in the verification stage, for example, S0 = 0.85; when the score s of a node at the t*th moment is t*≥S0, immediately record the unique ID and absolute timestamp t* of the node and generate a void warning.

[0057] Furthermore, the platform regularly (e.g., every 5 minutes) or after the first warning is generated, sorts the void warnings of all nodes during the current grouting process in chronological order to form a node number-warning time list; this list is pushed to the construction terminal and operation and maintenance center in the form of message notifications, so that the corresponding nodes can be quickly located on site and excavation, void detection, or grouting operations can be carried out.

[0058] In step S205 , based on the first time series data and the void warning, the lining vault is inferred through a graph neural network to obtain a first void probability distribution map.

[0059] Exemplarily, the three-dimensional coordinates of all target nodes are extracted from the target model, and the Euclidean distance between the nodes is calculated. A weighted adjacency matrix (edge ​​weight is the inverse of the distance) is constructed based on the set adjacency threshold. At the same time, time pooling is performed on the first time series data of each node to extract statistical features such as mean, variance, and peak value, and the node's air gap warning information (whether it is abnormal, number of abnormalities, etc.) is spliced ​​into a complete feature vector. The adjacency matrix and the node feature matrix are then input into the pre-trained GAT (Graph Attention Network). The network adaptively aggregates the features of neighboring nodes and outputs a air gap probability value (0-1) for each node at the top layer. After being mapped by the Sigmoid function, it represents the risk level of the node being air gapped. The platform projects the probability values ​​of all nodes back into the vault target model and interpolates to generate a first air gap probability distribution map covering the entire vault area, which is visualized in the three-dimensional model as a heat map. At the same time, the distribution map and node data are stored in the database for subsequent dynamic sampling strategy updates and risk closed-loop analysis.

[0060] In one possible implementation, based on the first time series data and the void warning, the lining vault is inferred through a graph neural network to obtain a first void probability distribution map, specifically including: calling the three-dimensional coordinates of each target node in the target model, calculating the Euclidean distance between each target node, and forming a weighted adjacency matrix; splicing the first time series data and void warning corresponding to each target node in node order to generate a feature vector matrix; inputting the weighted adjacency matrix and the feature vector matrix into a multi-layer graph attention network, aggregating the feature vectors of adjacent target nodes through adaptive attention weights, and outputting a first void probability distribution map covering the entire vault.

[0061] Specifically, the three-dimensional coordinates (X, Y, Z) of all 16 target nodes are read from the target model database; the Euclidean distance d between any two nodes is calculated. ij , and A ij As edge weight, A 16×16 weighted adjacency matrix A is formed. For each target node, statistics such as mean, standard deviation, maximum drop rate, and trigger frequency are calculated on the time dimension of its first time series data to obtain several time series features. Whether the node has ever issued an air gap warning and the number of warnings are then added as additional features. The data are concatenated in node number order to obtain a 16×F feature vector matrix X (F is the sum of the feature dimensions).

[0062] Furthermore, the weighted adjacency matrix A and the eigenvector matrix F are input into GAT together; each layer of GAT uses the adaptive attention mechanism to pay attention to the neighbor nodes j of each node i according to A ij Assign weights to the node feature similarity and aggregate their features. In the last layer of GAT, a real value Z is output for each node. i , Z i Mapped to p by Sigmoid function i ∈[0,1], that is, the probability of node i being empty. The probability of all nodes being empty p i The three-dimensional position of the target model is mapped back, and a continuous probability distribution surface is drawn on the triangulated mesh based on four-node interpolation. The platform is presented in a three-dimensional view with a gradient color band, from cold colors (low probability) to warm colors (high probability), intuitively showing the risk of the entire vault falling out.

[0063] Step S206: determining the uncertainty of the first empty probability distribution graph, and performing a dynamic sampling strategy update based on the uncertainty.

[0064] For example, first read the probability p of all target nodes being empty i , and calculate the uncertainty u i , filter out all u i Nodes exceeding the threshold U0 select the 1-2 nearest and reachable points in their respective candidate mobile sampling point lists; then, a wake-up command is sent via the industrial wireless network to the embedded pressure sensors or mobile arms at these sampling points, entering the 200Hz high-frequency mode and re-collecting the pressure and overflow trigger signals during the next concrete pouring process.

[0065] In one possible implementation, the uncertainty of the first air-out probability distribution graph is determined, and a dynamic sampling strategy is updated based on the uncertainty, specifically including: calculating the uncertainty of each target node for the first air-out probability distribution graph; sorting the target nodes from high to low according to the uncertainty, and selecting the target nodes whose uncertainty is greater than a preset threshold as the next round of sampling objects; performing a sensor wake-up or mobile deployment operation on the target node where the sampling object is located to perform a new round of signal acquisition.

[0066] Specifically, the probability of missing all 16 target nodes p is loaded from the database.i (range 0-1), calculate the uncertainty u for each node i =p i ×(1-p i ), the value is in p i =0.5 is the largest, indicating that the model is most ambiguous in judging the node. i Sort from high to low and compare with the preset uncertainty threshold U0 (for example, 0.2); select all u from the sorting results i >U0 of the node set u, for example, the u of nodes 2, 4, and 6 are 0.24, 0.23, and 0.22 respectively, all exceeding the threshold, then they are listed as the next round of sampling objects. For each selected node i, select 1-2 points that are closest to node i and can be reached on site from the first mobile candidate sampling points. For example, there are two points A and B near node 4 (0.5m and 0.6m away from node 4 respectively), and A, which is closer, is preferred. The list of sampling objects for the next round is sent to the edge controller via the industrial wireless network. After receiving the instruction, the edge controller sends a wake-up command to the embedded pressure sensors at these sampling points, or drives the mobile arm to move the portable sensor to the specified coordinates. During the next concrete pouring process, the pressure and overflow trigger signals of these supplementary sampling points are collected and pre-processed in real time. After the collection is completed, they are reported together with the node ID and timestamp and marked as the kth round of supplementary sampling data.

[0067] Step S207: After the dynamic sampling strategy is updated and executed, the second time series data is obtained and the first void probability distribution map is updated. When the uncertainty of all target nodes is lower than the preset threshold, the dynamic sampling is terminated and the second void probability distribution map is output. For the node area in the second void probability distribution map whose probability value exceeds the preset grouting threshold, a grouting operation instruction is generated and sent to the on-site construction terminal.

[0068] For example, after completing the dynamic supplementary sampling, the new round of signal data is merged with the original time series to generate the second time series data of each target node, and the GAT inference process is repeated to obtain the updated second empty probability distribution diagram; then the uncertainty of each node is calculated again. If all u i If all the values ​​are lower than the uncertainty threshold U0, the sampling accuracy is determined to have reached the standard and the dynamic sampling is terminated. At this time, all the p i For nodes ≥ the preset grouting threshold p0 (e.g., 0.7), the spatial clustering algorithm is used to merge adjacent over-threshold nodes into continuous areas. A detailed grouting operation plan (including grouting pressure, grouting volume, and operation sequence) is generated based on the area, estimated depth, and construction accessibility. The formatted grouting operation instructions are sent to the on-site construction terminal through the industrial wireless network.

[0069] In one possible implementation, for the node area whose probability value in the second void probability distribution map exceeds the preset grouting threshold, a grouting operation instruction is generated and sent to the on-site construction terminal, specifically including: for each target node whose probability value exceeds the preset grouting threshold in the second void probability distribution map, adjacent over-threshold nodes are merged according to the spatial clustering algorithm to determine a continuous void area; for the continuous void area, a grouting operation plan is generated based on the area, depth estimation and construction accessibility of the area, and the grouting operation plan is formatted into a grouting operation instruction that can be recognized by the construction terminal; the grouting operation instruction is sent to the on-site construction terminal and the execution status is fed back.

[0070] Specifically, all target nodes with pi≥p0 (such as 0.7) are screened out from the second void probability distribution map, such as nodes 4, 5, and 6; a clustering algorithm based on a distance threshold (such as 1.0m) (such as DBSCAN (Density-Based Spatial Clustering of Applications with Noise)) is applied to the three-dimensional coordinates of these nodes, and nodes with a distance of ≤1.0m are automatically merged to obtain several continuous void areas (such as the area of ​​segments 5-7 on the third link). The spatial projection area (such as 0.8m) of each continuous area is calculated. 2 ) and estimate the void depth (e.g., an average depth of 5 cm) based on the normal distance from the area surface to the inner surface of the lining in the target model; and evaluate the accessibility and optimal grouting entrance of each area based on the on-site construction conditions (segment joint location, adjacent supports, and construction channel width).

[0071] Furthermore, the coordinates of the grouting points and the grouting sequence are determined for each area. For example, grouting is performed from the top of the area first, and then the grouting is distributed to both sides in sequence. The grouting material ratio (such as a water-binder ratio of 0.45 for cement slurry), single-point grouting pressure (such as 1.2 MPa), grouting volume (such as 10 L per point) and the interval time between each point (such as 5 minutes per point) are determined. The information of each grouting point (node ​​cluster ID, three-dimensional coordinates, grouting pressure, grouting volume, grouting sequence and time interval) is encapsulated into a JSON (JavaScript Object Notation) instruction package that can be recognized by the construction terminal. The instruction package is sent to the on-site grouting control terminal through the industrial wireless network or the OPC communication protocol. After completing the grouting of each node, the construction terminal uploads the execution status in real time (such as grouting start, grouting completion, pressure abnormality); the operation progress is updated according to the feedback, and an alarm is automatically triggered when abnormal feedback occurs, prompting the operation and maintenance personnel to handle it on site.

[0072] Figure 3 Schematic diagram of a module of a tunnel lining arch anti-cavitation monitoring system according to an embodiment of the present application. Figure 3As shown, the system includes: a model building module 301, a dynamic signal acquisition module 302, a time series data generation module 303, a time series anomaly detection module 304, a void probability inference module 305, a sampling strategy optimization module 306 and a grouting instruction generation module 307, wherein: The model construction module 301 is configured to obtain an original design model of the lining vault, obtain point cloud data of the inner surface of the lining vault through an acquisition device, construct a target model based on the point cloud data and the original design model, mark the embedded sensor point and the first candidate mobile sampling point on the target model, and use any point among the embedded sensor point and the first candidate mobile sampling point as a target node; Dynamic signal acquisition module 302 is configured to perform dynamic sampling based on the target model. Dynamic sampling includes collecting signal data of target nodes during concrete pouring. The signal data includes pressure signal data at the embedded exhaust pipe and trigger signal data of the template overflow hole. A time series data generating module 303 is configured to process the signal data and generate first time series data; The time series anomaly detection module 304 is configured to input the first time series data into a pre-trained bidirectional long short-term memory network model to perform time series anomaly detection, and output an air gap warning when an anomaly is detected. The air gap warning includes the warning time and the node identifier of the corresponding target node; A void probability inference module 305 is configured to infer the lining vault through a graph neural network based on the first time series data and the void warning to obtain a first void probability distribution map; a sampling strategy optimization module 306 configured to determine the uncertainty of the first empty probability distribution graph and dynamically update the sampling strategy based on the uncertainty; The grouting instruction generation module 307 is configured to obtain the second time series data and update the first void probability distribution map after the dynamic sampling strategy is updated and executed. When the uncertainty of all target nodes is lower than the preset threshold, the dynamic sampling is terminated and the second void probability distribution map is output. For the node area in the second void probability distribution map whose probability value exceeds the preset grouting threshold, a grouting operation instruction is generated and sent to the on-site construction terminal.

[0073] Optionally, the model building module 301 is further configured to: Use laser scanning or drone photogrammetry to collect point clouds of the inner surface of the lining vault and pre-process the collected point cloud data; The pre-processed point cloud data is reconstructed into a mesh surface using a triangulation algorithm; Perform rigid body alignment and fusion of the mesh surface and the original design model to generate a target model with coordinate mapping relationship.

[0074] Optionally, the time series data generating module 303 is further configured to: The pressure signal data is calibrated according to the preset sampling frequency; Perform pulse counting and duration analysis on the trigger signal data to extract the time window of each trigger event. The time window starts at the beginning of the rising edge of the trigger signal and ends at the end of the falling edge of the trigger signal. For the calibrated pressure signal data, in each time window, the high-frequency noise in the current time window is removed by the wavelet threshold denoising algorithm, and the data in the current time window is reconstructed to obtain the reconstructed signal data; The reconstructed signal data corresponding to each time window are spliced ​​to obtain an overall signal set, and the overall signal set is normalized to generate first time series data.

[0075] Optionally, the timing anomaly detection module 304 is further configured to: Adding position code to the first time series data; Inputting the first time series data into the bidirectional long short-term memory network model, and extracting time series features based on position coding through the encoder of the bidirectional long short-term memory network model; Each time series feature is scored through the decoder of the bidirectional long short-term memory network model, and a shortage warning is output when the score is higher than the preset abnormal score threshold.

[0076] Optionally, the miss probability inference module 305 is further configured to: Call the three-dimensional coordinates of each target node in the target model, calculate the Euclidean distance between each target node, and form a weighted adjacency matrix; The first time series data and void warning corresponding to each target node are spliced ​​in node order to generate a feature vector matrix; the weighted adjacency matrix and the feature vector matrix are input into a multi-layer graph attention network, and the feature vectors of adjacent target nodes are aggregated through adaptive attention weights to output the first void probability distribution map covering the entire vault.

[0077] Optionally, the sampling strategy optimization module 306 is further configured to: Calculate the uncertainty of each target node based on the first empty probability distribution graph; Sort the target nodes from high to low according to their uncertainty, and select the target nodes whose uncertainty is greater than the preset threshold as the next round of sampling objects; Perform sensor wake-up or mobile deployment operations on the target node where the sampling object is located to conduct a new round of signal collection.

[0078] Optionally, the grouting instruction generating module 307 is further configured to: For each target node whose probability value exceeds a preset grouting threshold in the second void probability distribution map, adjacent over-threshold nodes are merged according to a spatial clustering algorithm to determine a continuous void area; For continuous void areas, a grouting operation plan is generated based on the area, depth estimation, and construction accessibility, and the grouting operation plan is formatted into grouting operation instructions that can be recognized by the construction terminal; The grouting operation instructions are sent to the on-site construction terminal and the execution status is fed back.

[0079] It should be noted that the above embodiments provide systems that implement their functions using only the division of the above functional modules as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0080] This embodiment also discloses an electronic device, referring to Figure 4 The electronic device may include: at least one processor 401 , at least one communication bus 402 , a user interface 403 , a network interface 404 , and at least one memory 405 .

[0081] The communication bus 402 is used to implement the connection and communication between these components.

[0082] The user interface 403 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 403 may also include a standard wired interface and a wireless interface.

[0083] The network interface 404 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0084] The processor 401 may include one or more processing cores. The processor 401 utilizes various interfaces and lines to connect various parts of the entire server, and executes various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 405, as well as calling data stored in the memory 405. Optionally, the processor 401 may be implemented in the form of at least one hardware component selected from digital signal processing (DSP), field-programmable gate array (FPGA), and programmable logic array (PLA). The processor 401 may integrate one or a combination of a central processing unit (CPU), a graphics processing unit (GPU), and a modem. The CPU primarily processes the operating system, user interface, and application programs; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; and the modem is used to handle wireless communications. It is understood that the modem may not be integrated into the processor 401 and may be implemented separately on a single chip.

[0085] Among them, the memory 405 may include a random access memory (RAM) or a read-only memory (Read-Only Memory). Optionally, the memory 405 includes a non-transitory computer-readable storage medium. The memory 405 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 405 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data involved in the above-mentioned various method embodiments, etc. The memory 405 may also be optionally at least one storage device located away from the aforementioned processor 401. As Figure 4 As shown, the memory 405 as a computer storage medium may include an operating system, a network communication module, a user interface module and an application program of a tunnel lining arch anti-cavitation monitoring method.

[0086] exist Figure 4In the electronic device shown, the user interface 403 is mainly used to provide an input interface for the user and obtain data input by the user; and the processor 401 can be used to call an application program for a tunnel lining arch anti-de-airing monitoring method stored in the memory 405. When executed by one or more processors 401, the electronic device executes one or more methods as described in the above embodiments.

[0087] It should be noted that for the aforementioned method embodiments, for simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required for this application.

[0088] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0089] In the several embodiments provided in this application, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are merely schematic, such as the division of units, which is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interface, and the indirect coupling or communication connection of devices or units can be electrical or other forms.

[0090] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0091] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0092] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a memory 405 and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned memory 405 includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a magnetic disk, or an optical disk.

[0093] The above is only an exemplary embodiment of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the disclosure of the specification, those skilled in the art will easily think of other embodiments of the present disclosure. This application is intended to cover any variations, uses or adaptive changes of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary technical means in the technical field that are not recorded in the present disclosure. The description and examples are to be regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A tunnel lining vault anti-cavitation monitoring method, characterized in that: Applied to a monitoring platform, the method includes: Obtaining an original design model of the lining vault, acquiring point cloud data of the inner surface of the lining vault using an acquisition device, constructing a target model based on the point cloud data and the original design model, marking a pre-embedded sensor point and a first candidate mobile sampling point on the target model, and using any point among the pre-embedded sensor point and the first candidate mobile sampling point as a target node; Performing dynamic sampling based on the target model, the dynamic sampling comprising collecting signal data of the target node during concrete pouring, the signal data comprising pressure signal data at the embedded exhaust pipe and trigger signal data of the template overflow hole; Processing the signal data to generate first time series data; Inputting the first time series data into a pre-trained bidirectional long short-term memory network model to perform time series anomaly detection, and outputting an air gap warning when an anomaly is detected, the air gap warning including the warning time and the node identifier corresponding to the target node; Based on the first time series data and the void warning, inferring the lining vault through a graph neural network to obtain a first void probability distribution map; Determining the uncertainty of the first empty probability distribution graph, and dynamically updating the sampling strategy based on the uncertainty; After the dynamic sampling strategy is updated and executed, the second time series data is obtained and the first void probability distribution map is updated. When the uncertainty of all the target nodes is lower than the preset threshold, the dynamic sampling is terminated and the second void probability distribution map is output. For the node area in the second void probability distribution map whose probability value exceeds the preset grouting threshold, a grouting operation instruction is generated and sent to the on-site construction terminal.

2. The method according to claim 1, characterized in that The step of acquiring point cloud data of the inner surface of the lining vault by an acquisition device and constructing a target model based on the point cloud data and the original design model includes: Collecting point cloud data of the inner surface of the lining vault by laser scanning or drone photogrammetry, and preprocessing the collected point cloud data; Reconstructing the pre-processed point cloud data into a mesh surface using a triangulation algorithm; The mesh surface is rigidly aligned and fused with the original design model to generate the target model with a coordinate mapping relationship.

3. The method according to claim 1, characterized in that The processing of the signal data to generate first time series data includes: Calibrate the pressure signal data according to a preset sampling frequency; Perform pulse counting and duration analysis on the trigger signal data to extract a time window for each trigger event, where the time window starts at the start time of the rising edge of the trigger signal and ends at the end time of the falling edge of the trigger signal; For the calibrated pressure signal data, in each time window, high-frequency noise in the current time window is removed by a wavelet threshold denoising algorithm, and the data in the current time window is reconstructed to obtain reconstructed signal data; The reconstructed signal data corresponding to each of the time windows are spliced ​​to obtain an overall signal set, and the overall signal set is normalized to generate the first time series data.

4. The method according to claim 1, wherein Inputting the first time series data into a pre-trained bidirectional long short-term memory network model to perform time series anomaly detection, and outputting a gap warning when an anomaly is detected includes: Adding a position code to the first time series data; Inputting the first time series data into the bidirectional long short-term memory network model, and extracting time series features based on the position code through the encoder of the bidirectional long short-term memory network model; Each of the time series features is scored by the decoder of the bidirectional long short-term memory network model, and the out-of-stock warning is output when the score is higher than a preset abnormal score threshold.

5. The method according to claim 1, wherein The inference of the lining vault by using a graph neural network based on the first time series data and the void warning to obtain a first void probability distribution map includes: Calling the three-dimensional coordinates of each target node in the target model, calculating the Euclidean distance between each target node, and forming a weighted adjacency matrix; The first time series data and the air gap warning corresponding to each target node are concatenated in node order to generate a feature vector matrix; The weighted adjacency matrix and the eigenvector matrix are input into a multi-layer graph attention network, and the eigenvectors of adjacent target nodes are aggregated through adaptive attention weights to output the first void probability distribution map covering the entire vault.

6. The method according to claim 1, characterized in that Determining the uncertainty of the first void probability distribution graph and dynamically updating the sampling strategy based on the uncertainty includes: Calculating the uncertainty of each target node based on the first empty probability distribution graph; Sort the target nodes from high to low according to the uncertainty, and select the target nodes whose uncertainty is greater than the preset threshold as the next round of sampling objects; A sensor wake-up or mobile deployment operation is performed on the target node where the sampling object is located to perform a new round of signal collection.

7. The method according to claim 1, characterized in that The step of generating a grouting operation instruction for a node area having a probability value exceeding a preset grouting threshold in the second void probability distribution graph and sending the instruction to the on-site construction terminal includes: For each target node in the second void probability distribution graph whose probability value exceeds a preset grouting threshold, merging adjacent exceeding-threshold nodes according to a spatial clustering algorithm to determine a continuous void area; For the continuous void area, a grouting operation plan is generated based on the area, depth estimation and construction accessibility of the area, and the grouting operation plan is formatted into a grouting operation instruction that can be recognized by the construction terminal; The grouting operation instruction is sent to the construction terminal on site and the execution status is fed back.

8. A tunnel lining vault anti-cavitation monitoring system, characterized in that: The system includes a model building module, a dynamic signal acquisition module, a time series data generation module, a time series anomaly detection module, a void probability inference module, a sampling strategy optimization module, and a grouting instruction generation module, among which: The model building module is configured to obtain an original design model of the lining vault, obtain point cloud data of the inner surface of the lining vault through an acquisition device, build a target model based on the point cloud data and the original design model, mark the embedded sensor point and the first candidate mobile sampling point on the target model, and use any point among the embedded sensor point and the first candidate mobile sampling point as a target node; The dynamic signal acquisition module is configured to perform dynamic sampling based on the target model, wherein the dynamic sampling includes collecting signal data of the target node during the concrete pouring process, wherein the signal data includes pressure signal data at the embedded exhaust pipe and trigger signal data of the template overflow hole; The time series data generating module is configured to process the signal data to generate first time series data; The time series anomaly detection module is configured to input the first time series data into a pre-trained bidirectional long short-term memory network model to perform time series anomaly detection, and output an air gap warning when an anomaly is detected, wherein the air gap warning includes a warning time and a node identifier corresponding to the target node; The void probability inference module is configured to infer the lining vault through a graph neural network based on the first time series data and the void warning to obtain a first void probability distribution map; The sampling strategy optimization module is configured to determine the uncertainty of the first void probability distribution graph and dynamically update the sampling strategy based on the uncertainty; The grouting instruction generation module is configured to obtain second time series data and update the first void probability distribution map after the dynamic sampling strategy is updated and executed. When the uncertainty of all the target nodes is lower than the preset threshold, the dynamic sampling is terminated and the second void probability distribution map is output. For the node area in the second void probability distribution map whose probability value exceeds the preset grouting threshold, a grouting operation instruction is generated and sent to the on-site construction terminal.

9. An electronic device, characterized in that: The electronic device comprises a processor, a memory, a user interface and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are both used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores instructions, and when the instructions are executed, the method according to any one of claims 1 to 7 is executed.

Images