Pipe gallery pipeline key connection point leakage monitoring system and method based on AI vision

Abstract

The invention relates to the field of computer vision and industrial safety monitoring, and discloses a pipe gallery pipeline key connection point leakage monitoring system and method based on AI vision, and the system comprises a data collection module, an intelligent analysis module, a 3D modeling module, a space-time verification module and an alarm prediction module. Real-time detection and three-dimensional accurate positioning of a leakage area are realized in combination with neural network optimization driven by physical simulation parameters; further through optical flow tracking and fluid mechanics verification, false alarm events of non-physical rules are screened out; and based on Bayesian decision and diffusion equation prediction, generating a leakage risk heat map and triggering graded alarm. The method solves the technical problems of low leakage detection precision, high false alarm rate and inaccurate positioning in a complex pipe gallery scene, has the advantages of high environmental adaptability, high anti-interference capability and prospective risk pre-judgment, and is suitable for intelligent safety monitoring of underground pipe galleries and petrochemical engineering pipelines.

Description

Technical Field

[0001] The present invention relates to the technical fields of computer vision and industrial safety monitoring, and specifically to a leakage monitoring system and method for key connection points of pipe gallery pipelines based on AI vision. Background Art

[0002] In the field of leakage monitoring of pipe gallery pipelines, traditional methods mainly rely on manual inspections and fixed-threshold sensor detections, which have problems such as poor environmental adaptability and high false alarm rates. Although vision-based detection technologies can identify some leakage characteristics, under complex lighting changes, pipeline surface reflections, and dynamic background interferences, the reliability of the detection results significantly decreases.

[0003] Although existing deep learning solutions have improved the detection accuracy in some scenarios, they are highly dependent on training data and have insufficient generalization ability in the face of small-sample scenarios with diverse pipeline materials and variable leakage patterns. In addition, multi-sensor fusion solutions are difficult to be deployed on a large scale in industrial scenarios due to high hardware costs and complex data alignment. At the same time, existing technologies generally lack in-depth verification of the physical laws of leakage, resulting in the inability to effectively distinguish real leakage from dynamic environmental interferences, which restricts the practicability and reliability of the detection system. Summary of the Invention

[0004] Aiming at the deficiencies of the existing technology, the present invention provides a leakage monitoring system and method for key connection points of pipe gallery pipelines based on AI vision, which solves the problems of high false alarm rate, low positioning accuracy, and lack of physical law verification existing in the existing leakage detection technologies in complex pipe gallery environments.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: A leakage monitoring system for key connection points of pipe gallery pipelines based on AI vision, including:

[0006] A data acquisition module, used for synchronously collecting visible light images, infrared thermal imaging data, and pressure sensing data to form a multi-modal data set;

[0007] An intelligent analysis module, which performs feature fusion analysis on the multi-modal data set based on a dynamic heterogeneous neural network and outputs a detection result including the coordinates of the leakage area and the leakage probability value;

[0008] A 3D modeling module, which receives the coordinates of the leakage area in the detection result, performs spatial mapping using the 3D model of the pipe gallery constructed by the neural radiance field, and corrects the temperature distribution field in combination with the real-time infrared thermal imaging data, and outputs a set of leakage points with three-dimensional coordinates;

[0009] A spatio-temporal verification module, which receives the set of leakage points and executes:

[0010] Tracking the movement trajectory of the leakage area based on the multi-modal optical flow method and calculating the diffusion rate in the time dimension;

[0011] Dynamically adjust the hydrodynamic diffusion equation in combination with the pipeline material parameters in the neural radiance field model to verify the diffusion pattern in the spatial dimension;

[0012] When the diffusion rate in the time dimension and the diffusion pattern in the spatial dimension meet the preset physical law matching degree, generate a leakage event with passing verification;

[0013] An alarm prediction module, which receives the leakage event with passing verification and the corresponding leakage probability value, and executes through a Bayesian decision-making model:

[0014] When the leakage probability value exceeds the dynamically optimized threshold and the verification matching degree reaches the confidence interval, trigger a hierarchical alarm;

[0015] Predict the leakage path based on the hydrodynamic diffusion equation and generate a warning message including a time-space diffusion heat map.

[0016] Preferably, the data acquisition module includes:

[0017] A visible light camera, an infrared thermal imager, a pressure sensor and a laser spectroscopy analyzer, where:

[0018] The laser spectroscopy analyzer is configured to detect the gas component concentration around the pipeline and generate gas abnormal distribution data;

[0019] The visible light camera and the infrared thermal imager achieve pixel-level spatial alignment through a coaxial pan-tilt head to form optical-thermal joint perception data;

[0020] The pressure sensor obtains the pipeline internal pressure fluctuation time series data;

[0021] The multi-modal data is jointly composed of spatially and temporally aligned optical images, thermodynamics images, pressure fluctuation sequences and gas concentration matrices.

[0022] Preferably, the intelligent analysis module includes:

[0023] A dynamic heterogeneous neural network unit, which consists of a first branch network and a second branch network:

[0024] The first branch network is constructed based on a lightweight convolutional network and is used to process visible light images and infrared thermal imaging data in real time;

[0025] The second branch network is trained based on the data generated by the meta-learning framework and physical simulation and is used for leakage detection in scenarios with a small training sample size;

[0026] A dynamic weight allocation unit dynamically adjusts the output weights of the two-branch network according to the physical simulation confidence of the second-branch network, the infrared anomaly intensity value calculated from the deviation between the thermal imaging data and the reference temperature field, and the fluctuation amplitude of the pressure sensing data.

[0027] Preferably, the data generated by the physical simulation includes:

[0028] Diffusion path simulation data of different pipeline materials in the leakage scenario;

[0029] Mapping relationship data between the thermodynamic parameters and infrared radiation characteristics of the pipeline material.

[0030] Preferably, the 3D modeling module includes:

[0031] A neural radiance field modeling unit for constructing a 3D neural radiance field model of the pipe gallery based on multi-view visible light and infrared data, where:

[0032] The neural radiance field modeling unit corrects the temperature distribution of the 3D model in combination with the thermal radiation characteristics of the pipeline material;

[0033] The temperature distribution is generated by dynamically adjusting the coefficient and fusing real-time infrared data and material thermodynamic parameters;

[0034] A leakage point positioning unit for mapping the leakage detection result output by the intelligent analysis module to the 3D model, tracking the movement trajectory of the leakage area by the optical flow method, and calculating the matching degree between the leakage point and the 3D model.

[0035] Preferably, the calculation process of the dynamic adjustment coefficient includes:

[0036] Construct a non-linear function f based on a multi-layer perceptron, whose input layer receives the normalized parameters of the real-time pipeline pressure P pipe and the environmental temperature T env ;

[0037] Optimize and train the function f through a meta-learning framework, and the training data set includes:

[0038] The thermal conductivity coefficient matrix of the pipeline material generated by physical simulation;

[0039] Temperature field distribution data under different pressure conditions;

[0040] Temperature-pressure correlation data of leakage events recorded in historical operation and maintenance;

[0041] In real-time calculation, preprocess the current pressure sensor reading and temperature sensor reading by standardization and input them into the trained function f;

[0042] The output layer generates a dynamically adjusted coefficient through the Sigmoid activation function, which is used for weighted fusion of real-time infrared data and material thermodynamics parameters.

[0043] Preferably, the spatio-temporal verification module includes:

[0044] A time dimension verification unit, which is used to track the movement trajectory of the leakage area by the optical flow method, calculate the leakage diffusion rate, and compare it with the preset pipeline material allowable diffusion rate threshold;

[0045] A space dimension verification unit, which verifies the diffusion mode of the leakage area based on the hydrodynamic diffusion equation, where the parameters of the diffusion equation are dynamically adjusted according to the pipeline material information and real-time pressure data in the neural radiance field model;

[0046] A consistency determination unit, which determines a false alarm and filters it when there is a logical conflict between the verification results in the time dimension and the space dimension.

[0047] Preferably, the optical flow method tracking adopts a multi-modal optical flow fusion algorithm, including:

[0048] Calculating the optical flow field based on gray-scale gradient for the visible light image;

[0049] Calculating the optical flow field based on temperature gradient for the infrared thermal imaging data;

[0050] Generating the movement trajectory of the leakage area by dynamically weighting and fusing the two optical flow fields.

[0051] Preferably, the alarm prediction module includes:

[0052] A Bayesian decision-making unit, which performs:

[0053] Dynamically updating the alarm prior probability based on the historical leakage event database, and generating a high-confidence alarm signal when the real-time detected leakage probability value exceeds the dynamically adjusted threshold;

[0054] Adapting to adjust the sensitivity interval of the dynamically adjusted threshold according to the current ambient temperature of the pipe gallery and the fluctuation range of the equipment operating pressure;

[0055] A diffusion prediction unit, which performs:

[0056] Combining the three-dimensional pipeline parameters provided by the neural radiance field model, generating a prediction map of the leakage diffusion path for a preset future period through hydrodynamic simulation;

[0057] Outputting a visualized heat map marking the diffusion direction, influence radius and risk level;

[0058] A dynamic threshold optimization unit, which performs:

[0059] Construct a multi-dimensional state space that includes environmental temperature and humidity, pressure fluctuation frequency, and historical false alarm records;

[0060] Iteratively optimize the dynamic adjustment threshold through a reinforcement learning algorithm to balance the conflicting metrics of real-time false alarm rate and missed alarm rate;

[0061] An alarm execution unit, in response to the high-confidence alarm signal, performs a hierarchical response action:

[0062] First-level response: Activate the on-site audible and visual alarm device in the pipe gallery and generate an emergency repair work order;

[0063] Second-level response: Push early warning information to the mobile terminal and mark the video stream of the key monitoring area.

[0064] The present invention also provides a method for monitoring leakage at key connection points of pipe gallery pipelines based on AI vision, including the following steps:

[0065] Synchronously collect visible light images, infrared thermal imaging data, and pressure sensing data in the pipe gallery area to generate a spatio-temporally aligned multi-modal data packet;

[0066] Input the multi-modal data packet into a dynamic heterogeneous neural network for feature fusion to generate a detection result that includes the coordinates and probability values of the leakage area in real time;

[0067] Map the detection result to a 3D pipe gallery model constructed by a neural radiance field, correct the temperature field distribution in combination with real-time infrared data, and output a three-dimensional set of leakage point coordinates;

[0068] Track the movement trajectory of the leakage area through a multi-modal optical flow method, and verify the spatio-temporal consistency of the leakage diffusion path based on the hydrodynamic diffusion equation to generate a verification result;

[0069] When the verification result meets the preset confidence condition, trigger a hierarchical alarm according to the leakage probability value, and predict the leakage diffusion trend based on the corrected 3D model.

[0070] The present invention provides a system and method for monitoring leakage at key connection points of pipe gallery pipelines based on AI vision.

[0071] It has the following beneficial effects:

[0072] 1. The present invention fuses multi-modal data through a dynamic heterogeneous neural network, significantly improving the detection accuracy while ensuring real-time performance. The lightweight network branch achieves millisecond-level response, and the high-precision branch is optimized based on physical simulation data, effectively solving the problem of missed detection in small-sample scenarios and being suitable for rapid leakage identification in complex industrial environments.

[0073] 2. The dynamic fusion mechanism of multi-modal data in the present invention can adapt to illumination changes, surface reflection, and temperature fluctuation interference. By dynamically adjusting the weights of infrared and visible light data, it ensures stable detection of leakage areas under extreme conditions such as low illumination and high reflection, improving the environmental robustness of the system.

[0074] 3. The dynamic fusion mechanism of multi-modal data in the present invention can adapt to illumination changes, surface reflection, and temperature fluctuation interference. By dynamically adjusting the weights of infrared and visible light data, it ensures stable detection of leakage areas under extreme conditions such as low illumination and high reflection, improving the environmental robustness of the system.

[0075] 4. The dual verification mechanism based on optical flow tracking and fluid mechanics in the present invention deeply combines the AI detection results with physical laws, effectively screening out false alarm events that do not conform to physical laws. This mechanism can distinguish real leaks from dynamic interferences (such as equipment movement and temporary foreign objects), significantly improving the credibility of detection results.

[0076] 5. Through the Bayesian decision-making model and dynamic threshold optimization, the present invention realizes intelligent adaptation of the alarm trigger logic. Combining with the prediction of fluid mechanics diffusion, the system not only gives real-time alarms but also anticipates the leakage diffusion trend, providing forward-looking decision-making support for the safe operation and maintenance of pipe corridors and improving the risk prevention and control ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 is a schematic diagram of the system architecture of the present invention;

[0078] Figure 2 is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0079] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0080] Please refer to the attached Figure 1 , the present invention provides a leakage monitoring system for key connection points of pipe corridor pipelines based on AI vision, aiming to achieve high-precision leakage monitoring of key connection points in the pipe corridor. This system can be widely applied to complex environments such as underground utility tunnels, oil pipelines, water supply pipelines, and chemical pipelines to achieve accurate leakage detection, three-dimensional positioning, trend prediction, and intelligent alarm, improving pipeline safety and maintenance efficiency.

[0081] As Figure 1 shown, the leakage monitoring system for key connection points of pipe corridor pipelines based on AI vision may include:

[0082] Data acquisition module: used to obtain visible light images, infrared thermal imaging data, pressure sensing data, and laser spectroscopy analysis data;

[0083] Intelligent analysis module: based on a dynamic heterogeneous neural network (DHNN), perform leakage detection on multimodal data and calculate the leakage probability;

[0084] 3D modeling module: use neural radiance fields (NeRF) technology to construct a 3D model of the pipe gallery environment, and combine the output results of the intelligent analysis module to locate the leakage points;

[0085] Spatio-temporal verification module: combine the optical flow tracking algorithm with fluid mechanics analysis to verify the consistency of the detected leakage area in the time and space dimensions;

[0086] Alarm prediction module: use the Bayesian decision algorithm to trigger an alarm and predict the leakage diffusion trend based on the fluid mechanics model.

[0087] The following is a detailed description of each module in the system of the present invention, comprehensively elaborating on the specific implementation principles, technical details, and processes of each module.

[0088] For the data acquisition module, it is used to synchronously collect visible light images, infrared thermal imaging data, pressure sensing data, and gas composition data of the pipe gallery area, and preprocess the multimodal data.

[0089] Exemplarily, the following sensors are arranged around the key connection points of the pipe gallery: visible light cameras (resolution ≥ 1920×1080, frame rate 30fps), infrared thermal imagers (wavelength range 8 - 14μm), pressure sensors (accuracy ±0.1MPa), and laser spectroscopy analyzers (detection accuracy 0.1ppm). Each sensor is connected through a hardware synchronization unit to ensure that the timestamp error is less than 1ms. Preferably, the visible light camera and the infrared thermal imager are installed coaxially to eliminate the viewing angle deviation.

[0090] First, perform contrast-limited adaptive histogram equalization (CLAHE) processing on the visible light images to eliminate image noise in the low-light environment of the pipe gallery. Specifically, the image is divided into 8×8 sub-regions, the histogram clipping threshold for each sub-region is 0.03, and the global image is reconstructed through bilinear interpolation. The signal-to-noise ratio (SNR) of the processed image is increased to more than 30dB, which can effectively extract the texture features of the leakage area.

[0091] Perform temperature normalization processing on the infrared thermal imaging data to eliminate the interference of ambient temperature fluctuations. The calculation formula is:

[0092]

[0093] where, TIR is the original temperature value, μ T is the historical average temperature, σ T is the standard deviation. Preferably, when |T norm | > 2.5, it is determined that the temperature is abnormal and the leakage candidate area is marked.

[0094] Calculate the real-time change rate of pipeline pressure. The formula is:

[0095]

[0096] where P t is the current pressure value, and Δt is the sampling interval (default 1 s). If ΔP exceeds the threshold θP (set to 20% of the historical fluctuation range), a pressure anomaly event is generated. Further, θP is dynamically adjusted according to the pipeline material. For example, θP for steel pipelines is 15% lower than that for polyethylene pipelines.

[0097] Align the data time axes of different sensors through the dynamic time warping (DTW) algorithm. Define the cost matrix D:

[0098] D(i,j) = |X i - Y j | + min(D(i - 1,j), D(i,j - 1), D(i - 1,j - 1))

[0099] where X i and Y j are the time series points of visible light and infrared data respectively. Preferably, a sliding window mechanism (window length 5 s) is used to achieve real-time alignment, and the calculation delay is less than 50 ms.

[0100] The gas concentration data collected by the laser spectroscopy analyzer is denoised by Kalman filtering and spatially matched with the infrared temperature anomaly area. Exemplarily, when the methane concentration and the corresponding area T norm > 2.0, it is determined as a gas leakage event.

[0101] The preprocessed multimodal data is transmitted to the intelligent analysis module through Gigabit Ethernet. The data packet format is defined as a quadruple:

[0102] DataPacket = <Image, T norm , ΔP, C gas >

[0103] where Image is the enhanced visible light / infrared fusion image, and C gas is the gas concentration vector.

[0104] Through multi-sensor synchronous acquisition and preprocessing, this embodiment can achieve low-latency and high-precision acquisition of the data of the pipe gallery environment.

[0105] For the intelligent analysis module, it is used to perform leakage detection and probability calculation on the multi-modal data obtained by the data acquisition module, and its core includes a dynamic heterogeneous neural network unit and a dynamic weight allocation unit.

[0106] Exemplarily, the first branch network adopts a lightweight MobileNetV4 architecture, inputs a visible light and infrared fusion image (size 256×256×3), and outputs a thermal map of the leakage area. Specifically, the network layer is configured as: 5 depthwise separable convolutional layers (kernel size 3×3), SE attention module (compression ratio 16), global average pooling layer, and Softmax classification layer. The inference latency of this branch on the NVIDIA Jetson AGX Xavier edge device is less than 30ms.

[0107] The second branch network is constructed based on the model-agnostic meta-learning (MAML) framework, and the input is a leakage data set generated by physical simulation. Preferably, the physical simulation data simulates the leakage diffusion process of different pipe materials (steel, polyethylene, PVC) through the finite element method, and generates multi-modal data including temperature field, pressure field, and fluid velocity field. The network structure is ResNet-34, and the initial parameters are updated through second-order gradient optimization during training:

[0108]

[0109] where α is the meta-learning rate (default 0.01), and L task is the task loss function.

[0110] The dynamic weight allocation unit calculates the branch weights according to the physical simulation result (S sim ), the infrared anomaly intensity (I IR ), and the pressure data (P pipr ):

[0111]

[0112] where φ(·) is a non-linear function optimized by meta-learning (implemented by a two-layer fully connected network), and ∈ is a small constant (1e-6) to prevent division by zero. Preferably, the infrared anomaly intensity value is obtained by calculating the standardized deviation of the temperature in the current detection area and the average temperature of the same location in history through statistical analysis. Preferably, when w1>0.7, the second branch network is activated for high-precision re-inspection.

[0113] The output feature map (size 16×16×512) of the first branch network and the feature map (size 16×16×512) of the second branch network are fused by dynamic weight weighting:

[0114] F fusion = w1·F branch1 +(1 - w1)·F bramch2

[0115] The fused feature map is output with a leakage probability value P after dimensionality reduction by a 1×1 convolutional layer leak ∈[0,1]. Further, when P leak > 0.85, a high-confidence leakage event is generated

[0116] Taking a steel pipe as an example, the physical simulation process includes:

[0117] 1. Establish a three-dimensional model of the pipe and set the material parameters (elastic modulus 200 GPa, Poisson's ratio 0.3);

[0118] 2. Simulate liquid leakage caused by a microcrack (size 0.1 mm) in COMSOL Multiphysics;

[0119] 3. Output simulation data including the temperature gradient field (ΔT = 5 - 15 °C) and the flow velocity field (0.1 - 1.2 m / s).

[0120] This data is used for few-shot training of the second branch network (only 50 groups of simulation data are required for each material type).

[0121] Exemplarily, three-dimensional leakage models of steel, polyethylene, and PVC pipes are established using finite element simulation software. First, set the crack size (in the range of 0.1 - 2.0 mm), fluid type (water / oil / gas), and pressure parameters (0.5 - 5.0 MPa). Calculate the diffusion path by solving the Navier-Stokes equation:

[0122]

[0123] where ρ is the fluid density, μ is the dynamic viscosity, v is the flow velocity field, and p is the pressure field. The output data includes the diffusion path coordinates (x t , y t , z t ) of the time series and the flow velocity v t , forming a spatio-temporal feature data set for training the second branch network

[0124] The thermal conductivity λ, specific heat capacity c p and surface emissivity ε of the pipe material are measured through experiments, and a quantitative relationship with the infrared radiation intensity is established:

[0125] E IR = εσT 4 +(1 - ε)E amb

[0126] where σ is the Stefan-Boltzmann constant, T is the surface temperature, and E amb is the environmental radiation intensity. Preferably, the mapping relationships of steel (ε = 0.85) and polyethylene (ε = 0.92) are embedded in the simulation model to generate infrared temperature field data matching the material.

[0127] Multi-physics field coupling simulation process:

[0128] 1. Geometric modeling: Construct a three-dimensional model of a pipeline with cracks in COMSOL (example crack length 1.5 mm);

[0129] 2. Parameter assignment: Input the thermodynamic parameters of the material (steel: λ = 45.8 W / (m·K), c p = 465 J / (kg·K));

[0130] 3. Boundary condition setting: Define the fluid inlet pressure (example value 2.5 MPa) and the environmental temperature (25 °C);

[0131] 4. Equation solving: Simultaneously solve the fluid dynamics equation and the heat conduction equation

[0132] 5. Data output: Generate a simulation data set including the diffusion path point cloud, the temperature gradient field and the infrared radiation intensity E IR of.

[0133] The diffusion path data generated by physical simulation in this embodiment can accurately reflect the leakage mode differences of pipelines with different materials.

[0134] For the 3D modeling module, it is used to construct the neural radiation field model of the pipe gallery environment and achieve precise leakage point positioning. Its core includes a neural radiation field modeling unit and a leakage point positioning unit.

[0135] Exemplarily, multi-view visible light cameras and infrared thermal imagers are used to collect the image data of the pipe gallery environment and input it into the neural radiation field (NeRF) model for three-dimensional reconstruction. The model generates the RGB color and temperature distribution at any view through the volume rendering equation:

[0136]

[0137] where σ is the density function, c is the color function, T IR is the temperature measured by the infrared sensor, M 材质 is the thermodynamic parameter of the material (steel: 0.85, polyethylene: 0.92), and γ is the dynamic adjustment coefficient.

[0138] The dynamic adjustment coefficient γ is generated by a multi-layer perceptron (MLP) optimized by meta-learning, and the real-time pressure P of the pipeline is inputpipe With the ambient temperature T env :

[0139] γ = f MLP (P pipe , T env ; θ)

[0140] where θ is a network parameter obtained by pre-training with a physical simulation dataset. Preferably, when P pipe > 2.0 MPa and T env > 40 °C, the value of γ is reduced to 0.3 - 0.5 to suppress infrared noise interference.

[0141] Specifically, the calculation process of the dynamic adjustment coefficient includes:

[0142] Construct a non-linear function f based on a multi-layer perceptron, whose input layer receives the normalized parameters of the real-time pressure P pipe of the pipeline and the ambient temperature T env ;

[0143] Optimize and train the function f through a meta-learning framework, and the training dataset includes:

[0144] The thermal conductivity coefficient matrix of the pipeline material generated by physical simulation;

[0145] The temperature field distribution data under different pressure conditions;

[0146] The temperature-pressure correlation data of leakage events recorded in historical operation and maintenance;

[0147] In real-time calculation, preprocess the current pressure sensor reading and temperature sensor reading by standardization, and input the trained function f;

[0148] The output layer generates a dynamic adjustment coefficient through the Sigmoid activation function, which is used to weight and fuse real-time infrared data and material thermodynamics parameters.

[0149] The leakage point positioning unit receives the leakage area coordinates (x i , y i ) output by the intelligent analysis module, and tracks its movement trajectory by the optical flow method. Define the matching degree index:

[0150]

[0151] where F NeRF is the three-dimensional temperature field rendered by the NeRF model, and F IR is the measured temperature value of the infrared sensor. When D match < 0.15, it is determined as a valid leakage point, otherwise it is marked as a false detection.

[0152] Through neural radiance field modeling and dynamic temperature correction, this embodiment can achieve high-precision reconstruction of the three-dimensional temperature field of the utility tunnel.

[0153] For the spatio-temporal verification module, it is used to verify the physical consistency of the leakage detection results in the time and space dimensions. Its core includes a time dimension verification unit, a space dimension verification unit, and a consistency determination unit.

[0154] Exemplarily, for a visible light image, calculate the optical flow field based on the gray gradient, and use the Horn-Schunck algorithm to solve the velocity vector (u v , v v ):

[0155] I x u v + I y v v + I t = 0

[0156] where I x , I y is the spatial gradient, and I t is the temporal gradient. Calculate the temperature gradient optical flow field (u v , v v ) for the infrared thermal imaging data by minimizing the temperature change error:

[0157]

[0158] Generate the comprehensive movement trajectory of the leakage area by dynamically weighting and fusing the two optical flow fields:

[0159] u = αu v + (1 - α)u t , v = αv v + (1 - α)v t

[0160] where α = σ(P pipe / P max ), σ(·) is the Sigmoid function, and P max is the maximum pressure-bearing value of the pipeline.

[0161] Calculate the diffusion rate of the leakage area according to the optical flow field:

[0162]

[0163] Compare v leak with the diffusion rate threshold v th allowed by the pipeline material. Exemplarily, for a steel pipeline, v th = 0.2 m / s, and for a polyethylene pipeline, v th = 0.5 m / s. When v leak> 1.5V th is marked as abnormal when

[0164] Based on the thermal conductivity coefficient λ of the pipeline material and the real-time pressure P output by the neural radiance field model pipe , dynamically adjust the parameters of the diffusion equation:

[0165]

[0166] where the diffusion coefficient D = k1λ + k2P pipe , k1 = 0.03, k2 = 0.12 are empirical constants. Solve the equation to obtain the theoretical diffusion region A theory , and calculate the overlap degree with the detection region A detect :

[0167]

[0168] When IoU < 0.6, it is determined that the space is inconsistent.

[0169] If the time dimension verification is abnormal (v leak > 1.5V th ) and the overlap degree in the space dimension is insufficient (IoU < 0.6), trigger the consistency determination unit. Exemplarily, when there is a logical conflict between the two (such as high-speed diffusion but high space matching degree), use the Bayesian decision model:

[0170]

[0171] where H0 is the false alarm hypothesis. When P(false alarm) > 0.8, filter the detection result.

[0172] This embodiment can achieve cross-sensor verification of the leakage movement trajectory through multi-modal optical flow fusion and hydrodynamic constraints.

[0173] For the alarm prediction module, it is used to comprehensively evaluate the leakage risk and generate an early warning signal. Its core includes a Bayesian decision unit, a diffusion prediction unit, and a dynamic threshold optimization unit.

[0174] Specifically, the alarm prediction module includes:

[0175] The Bayesian decision unit, which executes:

[0176] Dynamically update the prior probability of alarm based on the historical leakage event database, and generate a high-confidence alarm signal when the real-time detected leakage probability value exceeds the dynamically adjusted threshold;

[0177] According to the current ambient temperature of the pipe gallery and the fluctuation range of the equipment operating pressure, adaptively adjust the sensitivity interval of the dynamically adjusted threshold;

[0178] The diffusion prediction unit, which executes:

[0179] Combined with the three-dimensional pipeline parameters provided by the neural radiance field model, a leakage diffusion path prediction map for a future preset time period is generated through hydrodynamic simulation;

[0180] Output a visualized heat map marked with the diffusion direction, influence radius, and risk level;

[0181] The dynamic threshold optimization unit performs:

[0182] Construct a multi-dimensional state space including environmental temperature and humidity, pressure fluctuation frequency, and historical false alarm records;

[0183] Iteratively optimize the dynamic adjustment threshold through a reinforcement learning algorithm to balance the conflict indicators of real-time false alarm rate and missed alarm rate;

[0184] The alarm execution unit, in response to the high-confidence alarm signal, performs a hierarchical response action:

[0185] First-level response: Activate the on-site audible and visual alarm device in the pipe gallery and generate an emergency repair work order;

[0186] Second-level response: Push early warning information to the mobile terminal and mark the video stream of the key monitoring area.

[0187] First, construct a prior probability distribution based on historical leakage data and model it using the Beta distribution:

[0188] P(H1)~Beta(α,β)

[0189] where α is the number of historical leakage times and β is the number of normal events. When a leakage event is detected in real time, the posterior probability is updated through Bayes' theorem:

[0190]

[0191] where P(X|H1) is calculated from the leakage probability P output by the intelligent analysis module, and H0 is the hypothesis of no leakage. When P(H1|X)>0.9, an alarm is triggered. leak Calculate, and H0 is the hypothesis of no leakage. When P(H1|X)>0.9, an alarm is triggered.

[0192] Based on the pipeline material parameter λ and the real-time pressure P provided by the neural radiance field model pipe , solve the convection-diffusion equation to predict the diffusion range in the next 30 minutes:

[0193]

[0194] where the diffusion coefficient D = 0.05λ, and the velocity field v = k·P pipe ·n (k = 0.12, n is the pipeline normal vector). Discretely solve it using the finite difference method and output the heat map of the pollutant concentration distribution.

[0195] Dynamically adjust the alarm threshold θ using the Q-learning algorithm alarm , and the state space is defined as S = {T env , P pipe , IoU}, and the action space is the threshold adjustment amount Δθ ∈ {-0.1, 0, +0.1}. The reward function is designed as:

[0196] R = w1(1 - FPR) + w2(1 - FNR) - w3|Δθ|

[0197] where FPR is the false positive rate, FNR is the false negative rate, and the weight coefficients are w1 = 0.6, w2 = 0.3, and w3 = 0.1. Update the Q-table every 24 hours, with the learning rate η = 0.2 and the discount factor γ = 0.9.

[0198] This embodiment can achieve accurate leakage alarm through the combination of Bayesian decision-making and hydrodynamic prediction.

[0199] Please refer to the attached Figure 2 , and the present invention also provides a leakage monitoring method for key connection points of pipeline corridors based on AI vision, including the following steps:

[0200] S1. Cooperative acquisition of multi-modal data;

[0201] Synchronously collect environmental data through visible light cameras, infrared thermal imagers, and pressure sensors deployed at key connection points of the pipeline corridor. The visible light camera captures the texture details of the pipeline surface, the infrared thermal imager detects temperature abnormal areas, and the pressure sensor monitors the internal pressure fluctuations of the pipeline in real time. The multi-sensor data is synchronized through a hardware-level clock to ensure timestamp alignment and provide consistent input for subsequent analysis.

[0202] S2. Detection by a dynamic heterogeneous neural network;

[0203] Input the collected multi-modal data into a dynamic heterogeneous neural network:

[0204] The lightweight branch processes the visible light and infrared fusion image in real time to quickly locate the suspected leakage area;

[0205] The high-precision branch performs a fine analysis of the suspected area based on the leakage data generated by physical simulation (simulating the diffusion paths of pipelines with different materials);

[0206] Dynamic weight allocation adaptively adjusts the output weights of the two branches according to the infrared anomaly intensity, pressure fluctuation, and simulation results to balance real-time performance and detection accuracy.

[0207] S3. 3D leakage point positioning and modeling;

[0208] Construct a 3D model of the pipe gallery using Neural Radiance Field (NeRF) technology, and map the leakage coordinates output by the intelligent analysis module to the 3D space. By fusing real-time infrared temperature data with the thermodynamic parameters of the pipeline material (such as the thermal conductivity of a steel pipeline), correct the temperature distribution of the model to achieve millimeter-level precise positioning of the leakage point.

[0209] S4. Spatiotemporal consistency verification;

[0210] Combine dual verification of optical flow tracking and fluid mechanics analysis:

[0211] Optical flow tracking: Based on the fused optical flow field of visible light and infrared data, calculate the movement trajectory of the leakage area to verify the rationality of the diffusion rate in the time dimension;

[0212] Fluid mechanics verification: Dynamically adjust the parameters of the diffusion equation according to the pipeline material and pressure data, predict the leakage diffusion path, match the spatial overlap with the detection area, and screen out false alarms that do not follow physical laws.

[0213] S5. Trigger an alarm based on Bayesian decision-making and predict the leakage diffusion path;

[0214] Based on the Bayesian decision-making model, integrate historical data and real-time detection results, calculate the posterior probability of the leakage event, and dynamically optimize the alarm threshold to reduce false alarms. At the same time, predict the diffusion path in the future period through the fluid mechanics model, generate a heat map to mark high-risk areas, and guide the emergency response.

[0215] Through the collaborative innovation of multi-modal fusion detection, 3D physical modeling, interdisciplinary verification, and adaptive decision-making, this method constructs a complete chain of leakage monitoring system from data perception to intelligent decision-making, solves the deficiencies of traditional methods in terms of accuracy, real-time performance, and environmental adaptability, and provides reliable technical support for the safe operation and maintenance of the pipe gallery.

[0216] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. The leakage monitoring system for key connection points of pipe gallery pipelines based on AI vision is characterized in that Including: A data acquisition module, configured to synchronously acquire visible light images, infrared thermal imaging data, and pressure sensing data to form a multi-modal data set; An intelligent analysis module, which performs feature fusion analysis on the multi-modal data set based on a dynamic heterogeneous neural network, and outputs a detection result including the coordinates of the leakage area and the leakage probability value; A 3D modeling module, which receives the coordinates of the leakage area in the detection result, performs spatial mapping using the 3D model of the utility tunnel constructed by the neural radiance field, and corrects the temperature distribution field in combination with real-time infrared thermal imaging data, and outputs a set of leakage points with three-dimensional coordinates; A spatio-temporal verification module, which receives the set of leakage points and performs the following: Tracking the movement trajectory of the leakage area based on the multi-modal optical flow method, and calculating the diffusion rate in the time dimension; Combining the pipeline material parameters in the neural radiance field model to dynamically adjust the hydrodynamic diffusion equation to verify the diffusion mode in the spatial dimension; When the diffusion rate in the time dimension and the diffusion mode in the spatial dimension meet the preset physical law matching degree, generating a leakage event that passes the verification; An alarm prediction module, which receives the leakage event that passes the verification and the corresponding leakage probability value, and performs the following through the Bayesian decision-making model: When the leakage probability value exceeds the dynamically optimized threshold and the verification matching degree reaches the confidence interval, triggering a hierarchical alarm; Predicting the leakage path based on the hydrodynamic diffusion equation, and generating a warning message including a time-space diffusion heat map.

2. The leakage monitoring system for key connection points of pipe gallery pipelines based on AI vision according to claim 1, characterized in that, The data acquisition module includes: A visible light camera, an infrared thermal imager, a pressure sensor, and a laser spectroscopy analyzer, where: The laser spectroscopy analyzer is configured to detect the gas component concentration around the pipeline and generate gas abnormal distribution data; The visible light camera and the infrared thermal imager are pixel-level spatially aligned through a coaxial pan-tilt head to form optical-thermal joint perception data; The pressure sensor acquires the time series data of the internal pressure fluctuation of the pipeline; The multi-modal data is jointly composed of optically aligned optical images, thermodynamic images, pressure fluctuation sequences, and gas concentration matrices.

3. The leakage monitoring system for key connection points of pipe gallery pipelines based on AI vision according to claim 1, characterized in that, The intelligent analysis module includes: A dynamic heterogeneous neural network unit, which consists of a first branch network and a second branch network: The first branch network is constructed based on a lightweight convolutional network and is used to process visible light images and infrared thermal imaging data in real time; The second branch network is trained based on the meta-learning framework and the data generated by physical simulation, and is used for leakage detection in scenarios with a small training sample size; A dynamic weight allocation unit, which dynamically adjusts the output weights of the two branch networks according to the physical simulation confidence of the second branch network, the infrared anomaly intensity value calculated by the deviation between the thermal imaging data and the reference temperature field, and the fluctuation amplitude of the pressure sensing data.

4. The leakage monitoring system for key connection points of pipe gallery pipelines based on AI vision according to claim 3, characterized in that, The data generated by the physical simulation includes: Diffusion path simulation data of different pipeline materials in leakage scenarios; Mapping relationship data between the thermodynamic parameters and infrared radiation characteristics of pipeline materials.

5. The leakage monitoring system for key connection points of pipe gallery pipelines based on AI vision according to claim 1, wherein The 3D modeling module includes: A neural radiance field modeling unit, which is used to construct a 3D neural radiance field model of the utility tunnel based on multi-view visible light and infrared data, where: The neural radiance field modeling unit corrects the temperature distribution of the 3D model in combination with the thermal radiation characteristics of the pipeline material; The temperature distribution is generated by dynamically adjusting coefficients and fusing real-time infrared data with material thermodynamics parameters; A leakage point positioning unit is used to map the leakage detection results output by the intelligent analysis module to a 3D model, track the movement trajectory of the leakage area by the optical flow method, and calculate the matching degree between the leakage point and the 3D model.

6. The leakage monitoring system for key connection points of pipe gallery pipelines based on AI vision according to claim 5, characterized in that, The calculation process of the dynamically adjusted coefficient includes: Construct a nonlinear function f based on a multi-layer perceptron, whose input layer receives the real-time pipeline pressure P pipe With ambient temperature T env Normalization parameters of ; Optimizing and training the function f through a meta-learning framework, and the training data set includes: The matrix of the thermal conductivity coefficient of the pipeline material generated by physical simulation; Temperature field distribution data under different pressure conditions; Temperature-pressure correlation data of leakage events recorded in historical operation and maintenance; In real-time calculation, preprocess the current pressure sensor reading and temperature sensor reading in a standardized manner and input them into the trained function f; The output layer generates a dynamically adjusted coefficient through the Sigmoid activation function, which is used to weightedly fuse real-time infrared data and material thermodynamics parameters.

7. The leakage monitoring system for critical connection points of pipe gallery pipelines based on AI vision according to claim 1, characterized in that, The spatio-temporal verification module includes: A time dimension verification unit is used to track the movement trajectory of the leakage area by the optical flow method, calculate the leakage diffusion rate, and compare it with the preset threshold of the allowable diffusion rate of the pipeline material; A space dimension verification unit verifies the diffusion mode of the leakage area based on the fluid mechanics diffusion equation, where the parameters of the diffusion equation are dynamically adjusted according to the pipeline material information and real-time pressure data in the neural radiance field model; A consistency determination unit determines a false alarm and filters it when the verification results in the time dimension and the space dimension are logically conflicting.

8. The leakage monitoring system for key connection points of pipe gallery pipelines based on AI vision according to claim 7, characterized in that, The optical flow method tracking adopts a multi-modal optical flow fusion algorithm, including: Calculating the optical flow field based on the gray gradient for the visible light image; Calculating the optical flow field based on the temperature gradient for the infrared thermal imaging data; Generating the movement trajectory of the leakage area by fusing the two optical flow fields through dynamic weights.

9. The leakage monitoring system for key connection points of pipe gallery pipelines based on AI vision according to claim 1, wherein The alarm prediction module includes: A Bayesian decision-making unit performs: Dynamically updating the prior probability of the alarm based on the historical leakage event database, and generating a high-confidence alarm signal when the real-time detected leakage probability value exceeds the dynamically adjusted threshold; Adapting to adjust the sensitivity interval of the dynamically adjusted threshold according to the current ambient temperature of the pipe gallery and the fluctuation range of the equipment operating pressure; A diffusion prediction unit performs: Combining the three-dimensional pipeline parameters provided by the neural radiance field model to generate a prediction map of the leakage diffusion path for a preset future time period through fluid mechanics simulation; Outputting a visualized heat map marked with the diffusion direction, influence radius, and risk level; A dynamic threshold optimization unit performs: Constructing a multi-dimensional state space including environmental temperature and humidity, pressure fluctuation frequency, and historical false alarm records; Iteratively optimizing the dynamically adjusted threshold through a reinforcement learning algorithm to balance the conflicting indicators of the real-time false alarm rate and the missed alarm rate; An alarm execution unit, in response to the high-confidence alarm signal, performs a hierarchical response action: First-level response: Activate the on-site sound and light alarm device in the pipe gallery and generate an emergency maintenance work order; Second-level response: Push a warning message to the mobile terminal and mark the video stream of the key monitoring area.

10. A leakage monitoring method for key connection points of pipe gallery pipelines based on AI vision, which is applied to the leakage monitoring system for key connection points of pipe gallery pipelines based on AI vision according to any one of claims 1-9, characterized in that, Including the following steps: Synchronously collect visible light images, infrared thermal imaging data, and pressure sensing data in the pipe gallery area to generate a spatio-temporally aligned multi-modal data packet; Input the multi-modal data packet into a dynamic heterogeneous neural network for feature fusion, and generate detection results including the coordinates and probability values of the leakage area in real time; Map the detection results to the 3D utility tunnel model constructed by the neural radiance field, correct the temperature field distribution by combining real-time infrared data, and output a three-dimensional leakage point coordinate set; Track the movement trajectory of the leakage area by the multi-modal optical flow method, and verify the spatio-temporal consistency of the leakage diffusion path based on the hydrodynamic diffusion equation to generate a verification result; When the verification result meets the preset confidence condition, trigger a hierarchical alarm according to the leakage probability value, and predict the leakage diffusion trend based on the corrected 3D model.

Images