A heat pipe leakage detection method based on DFEM and PINNs

Abstract

The application discloses a kind of heat pipe leakage detection methods based on DFEM and PINNs, applied to heat pipe leakage detection technical field. Including the following steps: through multi-physical field simulation, the dynamic change of temperature, pressure and stress field in the process of leakage is simulated, and a three-dimensional heat pipe leakage finite element model is constructed;Using PINNs and multi-task learning strategy, the temperature field prediction and leakage detection task are optimized collaboratively, and the position, scale and development trend of the leakage source are identified in complex environment;Through the fusion of real-time distributed optical fiber sensing data, the rapid identification and dynamic prediction ability of the system to the leakage source is improved.Compared with the traditional method, the application has significant advantages in accuracy, efficiency and real-time response capability, especially in high temperature, high pressure and long distance pipeline application scenarios, can effectively improve the accuracy and calculation efficiency of leakage detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of heat pipe leakage detection, and more particularly to a heat pipe leakage detection method based on DFEM and PINNs. BACKGROUND

[0002] As an important component of energy transmission systems, heat pipes are widely used in heating, industrial heating, and chemical transportation. However, as the service life of the pipes increases, factors such as pipe aging, corrosion, and external forces may cause the pipes to leak, which can have a serious impact on energy utilization, pipe safety, and environmental protection. Pipe leakage not only leads to energy waste, but also can cause environmental pollution, fires, or explosions, which seriously threaten public safety.

[0003] Existing heat pipe leakage detection techniques mainly include monitoring methods based on local sensors (temperature, pressure sensors) and traditional finite element methods (FEM). The monitoring methods based on local sensors mainly rely on temperature, pressure, and acoustic sensors, but these methods often face problems such as high signal noise, inaccurate positioning, and difficulty in effectively evaluating the specific location, scale, and development trend of the leakage. Especially in long-distance pipes and complex environments such as high temperature and high pressure, the limitations of traditional methods are more obvious. Traditional finite element simulation methods often face high computational complexity when dealing with complex geometric structures and multi-physical field coupling problems, and lack adaptability and real-time performance when facing dynamic changes in leakage sources. At the same time, data-driven methods have difficulty in accurately reflecting the actual leakage situation in the environment with limited data quality and quantity, especially in complex working conditions and high-noise data, the prediction accuracy of the model is low. How to combine finite element simulation with data-driven technology and use real data (such as real-time sensor data) to optimize the model has become the key to improving detection accuracy and efficiency. Therefore, it is a problem that needs to be solved by those skilled in the art to propose a heat pipe leakage detection method based on the fusion of DFEM and PINNs. SUMMARY

[0004] Therefore, the present application provides a heat pipe leakage detection method based on DFEM and PINNs, which simulates the changes of temperature, pressure and stress field during the leakage process of heat pipe by DFEM technology, and realizes the organic combination of multi-physical field information and data-driven model by PINNs, effectively improving the accuracy, efficiency and real-time performance of leakage detection, and is suitable for large-scale pipe network monitoring tasks in complex working conditions.

[0005] To achieve the above purpose, the present application provides the following technical scheme:

[0006] A heat pipe leakage detection method based on DFEM and PINNs, comprising the following steps:

[0007] S1, establish a three-dimensional finite element model simulating the heat pipe according to the structural parameters, material properties and boundary conditions of the heat pipe, and simulate the dynamic evolution process of temperature, pressure and stress field of the pipe under leakage state by using deep finite element method;

[0008] S2, set a point leakage source in the three-dimensional finite element model, record the spatio-temporal change data induced by leakage through multi-physical field simulation, and generate physical data samples for training of physical information neural network;

[0009] S3, embed the heat conduction equation, fluid dynamics equation and elastic mechanics equation as physical constraints into the physical information neural network, and construct a leakage detection framework integrating physical model and data-driven capability;

[0010] S4, construct a joint loss function containing physical residual term and data error term, which is used to measure the degree of coincidence of the output of the physical information neural network with the residual of the physical equation and the fitting degree of the simulation and measured data;

[0011] S5, introduce a multi-task learning strategy into the physical information neural network, and use a parallel sub-task mode to cooperatively train the temperature field prediction, pressure field prediction and leakage detection tasks;

[0012] S6, real-time collect the temperature, stress and vibration signals inside and around the heat pipe by distributed optical fiber sensing technology, and input the processed signals into the trained physical information neural network for comparison and analysis with the simulation results.

[0013] Optionally, S1 further comprises:

[0014] Adopt adaptive mesh technology to dynamically adjust the grid density according to the change of physical field gradient near the leakage source in the three-dimensional finite element model, and improve the local simulation accuracy.

[0015] Optionally, S2 specifically comprises:

[0016] S21, set the leakage source in the three-dimensional finite element model of the heat pipe as a point leakage source, simulate the influence of the leakage source on the temperature field, pressure field and stress field inside and outside the heat pipe by setting the position, size, leakage rate, type, temperature, physical properties of the fluid and boundary conditions of the heat pipe;

[0017] S22, use the point leakage source model to represent the point leakage source by simulating the local temperature and pressure change, and set the temperature boundary condition and pressure boundary condition;

[0018] S23, simulate and generate various physical field data related to leakage, including the distribution of temperature field, pressure field and stress field, by using deep finite element method.

[0019] S24, extract the physical field data change in the simulation process and analyze the location, scale and development trend of the leakage source.

[0020] Optionally, S3 is specifically:

[0021] The physical equation is embedded into the neural network, the physical information neural network adopts a multi-layer perception structure, the input is space and time coordinates x, y, t, pipe geometric parameters and initial boundary conditions, and the output is temperature T(x, y, t), pressure p(x, y, t), stress u(x, y, t), density p, thermal conductivity k, specific heat capacity c p , viscosity of the fluid m, stress in the pipe s, and strain e;

[0022] The output of the network is differentiated by the automatic differentiation technique to construct the residual term of the physical equation:

[0023]

[0024] In the formula, R T (x, y, t) represents the residual term of the physical equation, p is the density of the fluid in the pipe, T is the temperature of the pipe, represents the gradient, k is the thermal conductivity of the pipe, c p is the specific heat capacity of the fluid, and Q is the heat source term in the pipe heat conduction equation.

[0025] Optionally, the expression of the joint loss function in S4 is:

[0026] L total = l1L data + l2L phys

[0027]

[0028] In the formula, L total is the joint loss, L phys is the physical loss term, L data is the data loss term, l1 and l2 are weight coefficients for balancing the data loss and the physical loss; T j and P j are the temperature and pressure in the simulation result in the data loss term, and are the actual measured data, and M is the number of data points; T i and P i are the temperature and pressure in the simulation result in the physical loss term, m is the viscosity of the fluid, and N is the number of sampling points of the physical field.

[0029] Optionally, S5 is specifically:

[0030] By adopting a multi-task learning strategy, the leakage detection task, the temperature field prediction task and the pressure field prediction task share the intermediate layer features of the physical information neural network, the physical information neural network can efficiently transmit valuable information between tasks, thereby improving the learning effect of each task, and the specific tasks include:

[0031] The leakage detection task: by analyzing the changes of temperature, pressure and stress field in the heat pipe, the position, scale and development trend of the leakage source are detected in real time;

[0032] The temperature field prediction task: the change of the temperature field inside and outside the pipe is predicted by the physical information neural network;

[0033] The pressure field prediction task: the pressure fluctuation caused by the leakage source is predicted by the physical information neural network;

[0034] In the multi-task learning strategy, all tasks share the optimized loss function, and the loss function includes a physical loss term and a data loss term, which ensures that the model not only conforms to the physical law, but also fits the actual data.

[0035] Optionally, S6 is specifically:

[0036] Based on the distributed optical fiber sensing technology, the temperature, stress and vibration signals inside and outside the pipe are collected in real time. In order to realize data fusion, the optical fiber sensing data and simulation data need to be preprocessed and standardized. The standardized data eliminates the dimensional differences between different data sources, so that different data can be compared and fused. Through data fusion technology, the optical fiber sensing data and simulation data are standardized to eliminate the difference between data sources, and then a comprehensive feature vector is generated. The physical information neural network is used to automatically learn and respond to the prediction of the fused data, to identify the position, scale and development trend of the leakage source in real time, thereby improving the response capability and accuracy of the leakage detection system.

[0037] According to the above technical solution, compared with the prior art, the present application provides a heat pipe leakage detection method based on DFEM and PINNs, which has the following beneficial effects:

[0038] (1) The present application can accurately simulate the dynamic changes of multi-physical fields (such as temperature field, pressure field, stress field, etc.) in the process of heat pipe leakage by combining DFEM and PINNs. Compared with the traditional single physical field modeling method, the present application can comprehensively reflect the multi-aspect influence of leakage on the pipe system, including temperature, pressure and stress, etc. by considering the geometric structure, material properties, fluid dynamics characteristics and external environmental factors of the pipe. Through the synergistic effect of optimized physical field simulation and data-driven analysis, the present application improves the accuracy of leakage position, scale and development trend judgment, and provides more reliable theoretical support for pipe leakage detection;

[0039] (2) The application improves the detection accuracy of the leakage source by introducing a fusion scheme of actual sensor data and simulation data, combining real-time monitoring technology and a deep learning model; through the real-time acquisition of optical fiber sensing data, the position and scale of the leakage source can be quickly identified, and the accuracy and real-time performance of the system are enhanced through comparison and analysis with the simulation results; the detection method has strong adaptability and can handle data changes in complex environments, providing strong technical support for real-time monitoring and maintenance of heat pipes. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of the provided drawings.

[0041] Figure 1 The flow chart of the heat pipe leakage detection method of the present application;

[0042] Figure 2 The flow chart of the physical information neural network training and optimization of the present application;

[0043] Figure 3 The schematic diagram of the simulation result of the underground heat pipe simulation point leakage of the present application. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0045] The embodiments of the present application disclose a heat pipe leakage detection method based on DFEM and PINNs, as shown in Figure 1 The method comprises the following steps:

[0046] S1, a three-dimensional finite element model of a simulated heat pipe is established according to the structural parameters, material properties and boundary conditions of the heat pipe, and a deep finite element method is used to simulate the dynamic evolution process of the temperature, pressure and stress field of the pipe under the leakage state;

[0047] S2, a point leakage source is set in the three-dimensional finite element model, and through multi-physical field simulation, the time and space change data induced by the leakage are recorded to generate physical data samples for training of the physical information neural network;

[0048] S3. The heat conduction equation, fluid dynamics equation, and elasticity equation are embedded as physical constraints into the physical information neural network to construct a leak detection framework that integrates physical models and data-driven capabilities.

[0049] S4. Construct a joint loss function that includes physical residuals and data error terms, which are used to measure the degree of conformity of the physical information neural network output to the physical equation residuals and the degree of fit to simulation and measured data, respectively.

[0050] S5. Introduce a multi-task learning strategy into the physical information neural network and use parallel sub-tasks to collaboratively train the temperature field prediction, pressure field prediction and leakage detection tasks.

[0051] S6. Real-time acquisition of temperature, stress, and vibration signals inside and around the thermal pipeline using distributed fiber optic sensing technology, followed by standardization and fusion processing, is input into a trained physical information neural network and compared with simulation results for analysis.

[0052] Furthermore, S1 also includes:

[0053] Adaptive meshing technology is used to dynamically adjust the mesh density based on the changes in the physical field gradient near the leakage source in the three-dimensional finite element model, thereby improving the accuracy of local simulation.

[0054] In this embodiment of the invention, the three-dimensional finite element model includes key geometric and physical information such as the pipe's inner diameter, wall thickness, burial depth, and material properties. For example, assuming the pipe burial depth is 3m, the inner diameter D = 800mm, the transmission medium is hot water, and the pipe material is steel, the modeling process needs to set parameters such as the thermal conductivity, elastic modulus, and Poisson's ratio of the steel, and consider the initial operating conditions such as the fluid's temperature, pressure, and flow velocity to accurately describe the system's thermo-mechanical coupling behavior. The establishment of the three-dimensional finite element model lays the foundation for subsequent multiphysics simulations.

[0055] The coupling model of multiphysics is as follows:

[0056] Heat conduction equation: In a thermal pipe, the temperature field change is affected by heat conduction. The following heat conduction equation is used in the modeling of heat conduction in the pipe:

[0057]

[0058] In the formula, T is the temperature of the pipe, k is the thermal conductivity of the pipe, with a value of 1.2 W / (mk), and Q is the heat source term caused by the leakage source;

[0059] Hydrodynamic equation: the pressure change caused by the fluid flow inside the heat pipe is described by the hydrodynamic equation, and the water is an incompressible fluid, and the Navier-Stokes equation can be used to describe the fluid disturbance and pressure fluctuation inside the pipe caused by leakage:

[0060]

[0061] In the formula, p is the pressure of the fluid, μ is the viscosity of the fluid, v is the velocity of the fluid, and ρ is the fluid density;

[0062] Stress field equation: the stress field model is used to describe the stress distribution inside the pipe caused by leakage, and the stress inside the pipe is closely related to its deformation, which can be described by the following elastic mechanics equation:

[0063] σ=E·∈

[0064] In the formula, σ is the stress inside the pipe, E is the elastic modulus of the pipe material, and ∈ is the strain.

[0065] Further, S2 is specifically:

[0066] S21, setting the leakage source in the three-dimensional finite element model of the heat pipe as a point leakage source, simulating the influence of the leakage source on the temperature field, pressure field and stress field inside and outside the heat pipe by setting the position, size, leakage rate, type, temperature, physical properties of the fluid and boundary conditions of the heat pipe;

[0067] S22, using a point leakage source model to represent a point leakage source by simulating local temperature and pressure changes, setting temperature boundary conditions and pressure boundary conditions;

[0068] S23, using deep finite element method for simulation, generating various physical field data related to leakage, including temperature field, pressure field and stress field distribution;

[0069] These data will be used as input in the subsequent leakage detection model for training and verification by deep neural network. Using DFEM for simulation, the simulation results are shown in Figure 3 The generated data can accurately reflect the multi-physical field changes caused by the leakage source, especially the temperature fluctuations, flow rate changes and pressure fluctuations near the leakage source and other complex phenomena. Through these high-precision simulation data, the deep neural network can better fit the leakage characteristics in the actual environment;

[0070] S24, extracting the physical field data changes in the simulation process and analyzing the position, size and development trend of the leakage source.

[0071] In the embodiments of the present application, the location of the leak source is determined by the geometry of the pipeline, and a specific point in the pipeline is usually taken as the leak source, assuming that the diameter of the point leak source is 24 mm. The leak rate and flow rate and other parameters are set by the operating conditions of the pipeline and the specific circumstances of the leak point. By simulating the changes in temperature, pressure and flow field caused by the leak source, further accurate simulation data for leak detection is provided. The flow rate of the leak source can be represented by the following formula:

[0072] m = p v A

[0073] In the formula, m is the flow rate, p is the fluid density, A is the cross-sectional area of the leak source, and v is the fluid velocity. By adjusting the flow rate, location and size of the leak source, the effects of different leak conditions on the pipeline system can be effectively simulated, providing reliable input for subsequent leak detection.

[0074] In the embodiments of the present application, it is assumed that the leak occurs at a certain point of the pipeline, the diameter of the leak source is set to 24 mm, and the flow rate and leak rate are known parameters. The changes caused by the point leak source will cause local temperature fluctuations, pressure changes and flow field disturbances inside the pipeline. In the three-dimensional finite element model of the heat pipeline, the point leak source can be represented by simulating the local temperature and pressure changes:

[0075]

[0076] Δσ = E · Δ∈

[0077]

[0078] In the formula, C p is the specific heat capacity of the fluid inside the pipeline, which is 871, Δ∈ is the strain change, Δσ is the stress change, Δv is the velocity change, ΔT is the temperature change, T in , T out are the temperatures at both ends of the pipeline, and Q L represents the heat loss or heat flow caused by the leak.

[0079] Temperature boundary condition: The temperatures at both ends of the pipeline are usually known, and these temperature changes will affect the temperature field inside the pipeline through the heat conduction equation. Near the leak source location, the temperature inside the pipeline will fluctuate dramatically. Assuming that the temperatures at both ends of the pipeline are T in and T out , the temperature boundary condition can be represented as:

[0080] T in = 353.15 K, T out = 300 K

[0081] Pressure boundary condition: The pressure of the fluid in the pipeline is usually determined by an external source. When a leak occurs, the pressure of the fluid will change due to the presence of the leak source. The presence of the leak source will cause changes in the internal pressure of the pipeline, especially fluctuations in the pressure near the leak source. The changes in the pressure field are simulated by the fluid dynamics Navier-Stokes equation: Assuming that the fluid pressures at both ends of the pipeline are P in and P out , the pressure boundary condition can be expressed as:

[0082] P(0)=P in =405300Pa,P(L)=P out =403300Pa.

[0083] Further, S3 is specifically:

[0084] The training and optimization process of the physical information neural network is shown in Figure 2 , the physical equation (such as the heat conduction equation, Navier-Stokes equation and fluid dynamics equation) is embedded into the neural network, the physical information neural network adopts a multi-layer perceptron structure, the input is the spatial and temporal coordinates x, y, t, the pipeline geometric parameters and the initial boundary conditions, and the output is the temperature T(x, y, t), the pressure p(x, y, t), the stress u(x, y, t), the density ρ, the thermal conductivity k, the specific heat capacity c p , the viscosity μ of the fluid, the stress σ in the pipeline and the strain ∈.

[0085] The output of the network is differentiated by automatic differentiation technology to construct the residual term of the physical equation:

[0086]

[0087] In the formula, R T (x, y, t) represents the residual term of the physical equation, ρ is the density of the fluid in the pipeline, T is the temperature of the pipeline, represents the gradient, k is the thermal conductivity of the pipeline, c p is the specific heat capacity of the fluid, and Q is the heat source term in the heat conduction equation of the pipeline.

[0088] The traditional finite element method (FEM) can well simulate the changes of temperature field and pressure field in the heat and power pipeline, however, using FEM alone in simulating the coupling process of complex multi-physical field has high calculation complexity and limited precision. Therefore, the embodiment of the present application combines DFEM and PINNs to improve the simulation precision and reduce the calculation complexity.

[0089] To simplify the calculation of higher-order derivatives and improve training stability, this invention introduces an energy method to improve the training process of PINNs. In the problem of thermal pipeline leak detection, the total energy of the physical system consists of changes in physical fields such as temperature and stress. The energy method optimizes the loss function of the neural network by minimizing the total energy, avoiding the calculation of higher-order derivatives. Specifically, the energy method loss function is in the form of:

[0090]

[0091] In the formula, N(θ) is the output of the neural network (e.g., temperature field, flow velocity field, etc.), K is the global stiffness matrix, P is the global load vector, and θ is the input model parameter of the neural network. By learning the physical model, the specific form and value of the stiffness matrix K and the load vector P are automatically inferred from the data.

[0092] Furthermore, to address issues such as incomplete data and high noise in leak detection, a composite loss function combining the residuals of the physical control equations and the errors of finite element / measured data is constructed to optimize the output of the PINNs network. The expression for the joint loss function in S4 is as follows:

[0093] L total =λ1L data +λ2L phys

[0094]

[0095] In the formula, L total For joint losses, L phys For physical loss, L data For the data loss term, λ1 and λ2 are weighting coefficients used to balance data loss and physical loss; T j and P j These are the temperature and pressure from the simulation results, respectively, in the data loss term. and For actual measurement data, M is the number of data points; T i and P i These represent the temperature and pressure from the simulation results in the physical loss term, respectively; μ is the fluid viscosity; and N is the number of sampling points for the physical field. The data loss term ensures that the neural network output matches the DFEM or measured data, while the physical loss term ensures that the network output satisfies the residual constraints of the governing equations.

[0096] Furthermore, S5 specifically refers to:

[0097] By employing a multi-task learning strategy, the leak detection task, temperature field prediction task, and pressure field prediction task share the intermediate layer features of the physical information neural network. This allows the physical information neural network to efficiently transfer valuable information between tasks, thereby improving the learning performance of each task. Specific tasks include:

[0098] Leakage detection task: By analyzing the changes in temperature, pressure and stress field within the thermal pipeline, the location, scale and development trend of the leak source are detected in real time; the temperature, pressure and stress fluctuations caused by the leak source serve as the detection basis, and the physical information neural network accurately locates the leak source and assesses its scale through analysis of the changes, supporting dynamic monitoring and efficient prediction of the leak source under different operating conditions.

[0099] Temperature field prediction task: Predict changes in the temperature field inside and outside the pipeline through a physical information neural network, especially temperature fluctuations in the leak source area. Combining historical temperature data and physical models, the neural network predicts temperature changes in real time, helps identify leak events, and provides additional temperature information. Especially in high-temperature environments, temperature fluctuations caused by leak sources are crucial to the stability of the pipeline system.

[0100] Pressure field prediction task: Predicting pressure fluctuations caused by leaks using a physical information neural network; leaks are often accompanied by pressure fluctuations within pipelines, especially under high-pressure environments, where the impact of leaks on the pressure field is significant. The neural network, by combining fluid dynamics equations and sensor data, accurately predicts the dynamic changes in the pressure field, further improving the detection accuracy of leaks.

[0101] In a multi-task learning strategy, all tasks share an optimized loss function, which includes a physical loss term and a data loss term, ensuring that the model not only conforms to physical laws but also fits the actual data.

[0102] By optimizing the comprehensive loss function, the model can efficiently learn the characteristics of each physical field and has a higher adaptability when dealing with complex leakage scenarios.

[0103] Furthermore, S6 specifically refers to:

[0104] Based on distributed fiber optic sensing technology, temperature, stress, and vibration signals inside and outside pipelines are collected in real time. To achieve data fusion, the fiber optic sensing data and simulation data first need to be preprocessed and standardized. The standardized data eliminates the dimensional differences between different data sources, enabling comparison and fusion between different data. Through data fusion technology, the fiber optic sensing data and simulation data are standardized to eliminate differences in data sources, thereby generating a comprehensive feature vector. The physical information neural network is then used to automatically learn and predict the response of the fused data, identifying the location, scale, and development trend of the leak source in real time, thus significantly improving the response capability and accuracy of the leak detection system.

[0105] In this embodiment of the invention, the standardization processing formula is as follows:

[0106]

[0107] In the formula, X represents the original data. min and X max These are the minimum and maximum values ​​of the data, respectively. The standardized data X norm It can eliminate the differences in units between different data sources, making it possible to compare and integrate different data.

[0108] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0109] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting leakage of a heat pipe based on DFEM and PINNs, characterized in that, The method comprises the following steps: S1, a three-dimensional finite element model of the heat pipeline is established according to the structural parameters, material properties and boundary conditions of the heat pipeline, and a deep finite element method is used to simulate the dynamic evolution process of the temperature, pressure and stress field of the pipeline under the leakage state; S2, a point leakage source is set in the three-dimensional finite element model, and the time and space change data induced by the leakage are recorded through multi-physical field simulation to generate physical data samples for training of the physical information neural network; S3, the heat conduction equation, the fluid dynamics equation and the elastic mechanics equation are embedded into the physical information neural network as physical constraints to construct a leakage detection framework integrating the physical model and the data driving capability; S4, a joint loss function including a physical residual term and a data error term is constructed to measure the coincidence degree of the physical information neural network output to the physical equation residual and the fitting degree to the simulation and measured data; S5, a multi-task learning strategy is introduced into the physical information neural network, and the temperature field prediction, pressure field prediction and leakage detection tasks are collaboratively trained in a parallel sub-task mode; S6, the temperature, stress and vibration signals inside and around the heat pipeline are collected in real time through a distributed optical fiber sensing technology, and are input into the trained physical information neural network after standardization and fusion processing, and are compared and analyzed with the simulation results.

2. The method of claim 1, wherein, In S1, the adaptive grid technology is used to dynamically adjust the grid density according to the change of the physical field gradient near the leakage source in the three-dimensional finite element model, and the local simulation accuracy is improved. In S2, the leakage source in the three-dimensional finite element model of the heat pipeline is set as a point leakage source, the influence of the leakage source on the temperature field, pressure field and stress field inside and outside the heat pipeline is simulated by setting the position, size, leakage rate, type, temperature, physical properties of the fluid and boundary conditions of the heat pipeline, a point leakage source model is used to simulate the local temperature and pressure change, the temperature boundary condition and the pressure boundary condition are set, a deep finite element method is used for simulation, various physical field data related to the leakage are generated, including the distribution of the temperature field, the pressure field and the stress field, and the position, size and development trend of the leakage source are extracted and analyzed during the simulation process.

3. The method of claim 1, wherein, In S3, the network output is differentiated through automatic differentiation technology, and the residual term of the physical equation is constructed. In S4, the expression of the joint loss function is: In S5, the multi-task learning strategy is used to share the intermediate layer features of the physical information neural network among the leakage detection task, the temperature field prediction task and the pressure field prediction task, the physical information neural network can efficiently transmit valuable information among the tasks, thereby improving the learning effect of each task, and the specific tasks include: The leakage detection task: the position, size and development trend of the leakage source are detected in real time by analyzing the changes of the temperature, pressure and stress field in the heat pipeline; The temperature field prediction task: the change of the temperature field inside and outside the pipeline is predicted through the physical information neural network; 4. The method of claim 1, wherein, The pressure field prediction task: the pressure fluctuation caused by the leakage source is predicted through the physical information neural network. The physical equation is embedded into the neural network, the physical information neural network adopts a multi-layer perception structure, the input is space and time coordinates , pipe geometric parameters and initial boundary conditions, and the output is temperature , pressure , stress , density , thermal conductivity , specific heat capacity , viscosity of fluid , stress in the pipe , strain ; ​ wherein represents a physical equation residual term, is the pipe fluid density, is the pipe temperature, represents a gradient, is the pipe thermal conductivity, is the fluid specific heat capacity, is the heat source term in the pipe heat conduction equation.

5. The method of claim 1, wherein, ​ wherein, is the joint loss, is the physical loss term, is the data loss term, and are the weight coefficients for balancing the data loss and the physical loss; and are the temperature and pressure in the simulation results in the data loss term, respectively, and are the actual measured data, M is the number of data points; and are the temperature and pressure in the simulation results in the physical loss term, respectively, is the viscosity of the fluid, N is the number of sampling points of the physical field, denotes the gradient, is the thermal conductivity of the pipe, is the specific heat capacity of the fluid, is the pipe fluid density.

6. The method of claim 1, wherein, ​ ​ ​ ​ ​ In the multi-task learning strategy, all tasks share the optimized loss function, which contains physical loss and data loss, to ensure that the model not only meets the physical law, but also fits the actual data.

7. The method of claim 1, wherein, S6 is specifically: Based on the distributed optical fiber sensing technology, the temperature, stress and vibration signals inside and outside the pipeline are collected in real time. In order to realize data fusion, the optical fiber sensing data and simulation data need to be preprocessed and standardized. The standardized data eliminates the dimensional differences between different data sources, so that different data can be compared and fused. Through data fusion technology, the optical fiber sensing data and simulation data are standardized, the data source differences are eliminated, and then the comprehensive feature vector is generated. The physical information neural network is used to automatically learn and respond to the fused data, to identify the location, scale and development trend of the leakage source in real time, so as to improve the response ability and accuracy of the leakage detection system.

Images