A tunnel service performance multi-dimensional information perception method based on multi-source heterogeneous data fusion
By fusing multi-source heterogeneous data and using a physical neural network model, the problems of accuracy and multiple solutions in inversion analysis during tunnel structure monitoring were solved, enabling high-precision identification and in-depth diagnosis of tunnel defects.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HEFEI UNIV OF TECH
- Filing Date
- 2026-05-18
- Publication Date
- 2026-08-04
AI Technical Summary
Existing technologies for tunnel structure monitoring suffer from limitations of single detection techniques and insufficient fusion of multi-source data, resulting in low accuracy and multiple solutions in inversion analysis, making it difficult to accurately identify complex situations involving various defects.
A multi-source heterogeneous data fusion method is adopted. Data is collected by deploying multiple types of sensing devices, preprocessed to construct a standardized dataset, and iteratively optimized using a physical neural network model. Logical judgment is then performed by combining the coupling relationship of multi-dimensional physical parameters to achieve in-depth diagnosis of tunnel defects.
It improves the accuracy and robustness of tunnel structure inversion analysis, and can accurately identify defects such as lining voids, crack types and carbonization corrosion, providing a scientific basis for operation and maintenance decisions.
Smart Images
Figure CN122508335A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of tunnel engineering monitoring technology, and in particular relates to a method for sensing multi-dimensional information on tunnel service performance based on multi-source heterogeneous data fusion. Background Technology
[0002] As a typical underground concealed engineering project, the service performance of tunnel structures is affected by a variety of factors, including the stress state of the surrounding rock, voids behind the lining, crack propagation in the lining, and carbonation and deterioration of the concrete. These factors interact and directly affect the long-term safety and durability of the tunnel. To obtain this multi-dimensional information, various non-destructive testing and structural monitoring technologies are widely used in engineering, such as ground-penetrating radar, infrared thermal imaging, ultrasonic testing, electromagnetic induction, acoustic emission technology, distributed fiber optic sensing, and video image monitoring. These technologies can, to a certain extent, enable the perception and inversion analysis of the tunnel structural state, providing data support for tunnel operation and maintenance.
[0003] However, existing technologies still have significant shortcomings in practical applications. On the one hand, single detection technologies often have inherent limitations. For example, ground-penetrating radar is susceptible to interference from internal steel reinforcement in the lining, and ultrasonic detection has low accuracy and the results are not intuitive, resulting in limited accuracy and scope of inversion analysis and difficulty in dealing with the "multiple solutions" problem commonly encountered in inversion analysis. On the other hand, existing inversion analysis methods are mostly based on the assumption of a single defect, ignoring the complex situation of multiple defects coexisting in actual engineering. This leads to insufficient consideration of the mutual influence between detection signals, resulting in a large deviation between the inversion results and the actual structural state. Therefore, there is an urgent need to develop a tunnel multi-dimensional information sensing method that can integrate multi-source heterogeneous data and synergistically consider the coupling effects of multiple physical fields to improve the accuracy and robustness of inversion analysis. Summary of the Invention
[0004] To address the aforementioned technical problems, this invention proposes a method for perceiving multi-dimensional information on tunnel service performance based on multi-source heterogeneous data fusion, thereby resolving the issues present in the existing technologies.
[0005] To achieve the above objectives, this invention provides a method for sensing multi-dimensional information on tunnel service performance based on multi-source heterogeneous data fusion, comprising: Multi-source heterogeneous monitoring data are collected based on various types of sensing devices deployed in the tunnel construction area; The multi-source heterogeneous monitoring data are preprocessed to obtain a standardized dataset; Based on the standardized dataset, a monitoring data loss function is constructed, and a total loss function for the physical neural network model containing the monitoring data loss function is also constructed. The spatiotemporal coordinates of the tunnel monitoring area are input into the neural network model. Based on the total loss function of the physical neural network model, iterative optimization is performed. While updating the network weights, the physical parameters to be inverted are corrected, and the multidimensional physical parameters obtained by inversion are output. Logical judgment is performed based on the coupling relationship between the multidimensional physical parameters to obtain the tunnel defect diagnosis results.
[0006] Optionally, the process of preprocessing the multi-source heterogeneous monitoring data to obtain a standardized dataset includes: Anomalies in the multi-source heterogeneous monitoring data are identified and removed using integrity detection to obtain corrected data. The corrected data is denoised using a denoising strategy that matches the data types in the multi-source heterogeneous monitoring data to obtain denoised data. The denoised data is time-aligned based on timestamp matching to obtain time-aligned data. Based on the deployment location of the sensing devices, the time-aligned data is spatially aligned to obtain spatiotemporally aligned data. The spatiotemporally aligned data is normalized to obtain the standardized dataset.
[0007] Optionally, the expression for the monitoring data loss function constructed based on the standardized dataset is as follows: ; In the formula, This is a data loss function for monitoring. The number of samples for monitoring data, The neural network predicted the value. To ensure the accuracy of the monitoring data, The spatial location vector of the monitoring point. The time point for monitoring data.
[0008] Optionally, the process of constructing the total loss function of the physical neural network model includes: A solid mechanics loss function is constructed based on the momentum balance equation of solid mechanics, a thermodynamic loss function is constructed based on the heat conduction equation, an electromagnetic loss function is constructed based on Maxwell's equations, and a wave mechanics loss function is constructed based on the scalar wave equation. The physical driving loss function is obtained by combining the solid mechanics loss function, thermodynamic loss function, electromagnetic loss function and wave mechanics loss function. Construct a boundary loss function based on Dirichlet and Neumann boundary conditions; A continuity conditional loss function is constructed based on the continuity of the field and flux at the contact surface; The physical driving loss function, boundary loss function, continuity condition loss function, and monitoring data loss function are combined to obtain the total loss function of the physical neural network model.
[0009] Optionally, the expression for the total loss function of the physical neural network model is: ; In the formula, For the overall loss function, Here, κ represents the parameters of the physical field neural network model, and κ represents the multidimensional information to be inverted. The physical-driven loss function, p The physical field equations representing the fusion process are as follows: Construct a boundary loss function for the conditions. To monitor the data loss function, This is the continuous conditional loss function.
[0010] Optionally, the process of inputting the spatiotemporal coordinates of the tunnel monitoring area into a neural network model, iteratively optimizing the model based on its total loss function, updating the network weights while correcting the physical parameters to be inverted, and outputting the inverted multidimensional physical parameters includes: The spatiotemporal coordinates of the tunnel monitoring area are used as input layer data and fed into a fully connected neural network. Nonlinear calculations are performed through the weights and activation functions inside the network to output the predicted value of the multidimensional physical field. The predicted value of the multidimensional physical field is differentiated using automatic differentiation technology to obtain the partial derivative terms; Substitute the partial derivative terms and the physical parameters to be inverted into the solid mechanics equilibrium equation, heat conduction equation, Maxwell's equations and wave equation, calculate the residual values caused by the inability to close each physical equation, and obtain the physical driving loss function. The mean square error is calculated based on the sensor observation data and the predicted physical values to obtain the monitoring data loss function; The total loss function is obtained by combining the physical drive loss function and the monitoring data loss function. An optimizer is used to iteratively optimize based on minimizing the total loss function. While updating the neural network weights through backpropagation, the physical parameters to be inverted are automatically corrected according to the gradient descent direction until the total loss function converges, at which point the multidimensional physical parameters obtained by inversion are output.
[0011] Optionally, the process of obtaining tunnel defect diagnosis results by performing logical determination based on the coupling relationship between the multidimensional physical parameters includes: Decision 1 is made based on the elastic modulus obtained from the inversion; if the elastic modulus decreases, then decision 2 is made. The determination 2 is based on thermal conductivity. If the thermal conductivity decreases synchronously, the diagnosis result of the back of the lining being detached is output. If the thermal conductivity does not decrease synchronously, the determination 2a is based on the anisotropic characteristics of the elastic modulus. Based on the anisotropic characteristics of the elastic modulus, a determination is made 2a. If anisotropy exists, a diagnosis result of structural damage is output; if anisotropy does not exist, a diagnosis result of initial damage is output.
[0012] Optionally, the process of obtaining tunnel defect diagnosis results by performing logical determination based on the coupling relationship between the multidimensional physical parameters further includes: Decision 1 is made based on the elastic modulus obtained from the inversion; if the elastic modulus does not decrease, then proceed to decision 3. Based on the wave velocity, the determination 3 is performed. If the wave velocity does not scatter, the structural health diagnosis result is output. If the wave velocity does scatter, the determination 3a is performed based on the conductivity. Based on the conductivity, a judgment 3a is made. If there is a synchronous decrease in conductivity, a diagnosis result of durability defects is output. If there is no synchronous decrease in conductivity, a diagnosis result of porosity change is output.
[0013] The present invention also provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described thereon.
[0014] The present invention also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.
[0015] Compared with the prior art, the present invention has the following advantages and technical effects: This invention overcomes the limitations and low accuracy of single-detection technologies by collecting multi-source heterogeneous monitoring data through various types of sensing devices. By preprocessing the multi-source heterogeneous data to form a standardized dataset, it eliminates data scale differences and spatiotemporal asynchrony issues between different monitoring methods, providing a unified data foundation for subsequent fusion analysis.
[0016] This invention constructs a total loss function for a physical neural network model that includes a loss function for monitoring data. It embeds the governing equations of multiple physics fields, such as solid mechanics, thermodynamics, electromagnetism, and wave mechanics, as prior knowledge into the neural network, achieving a deep integration of data-driven and physical sensing. This method can fully utilize the complementary information between multi-source monitoring data, effectively solving the technical problems of low inversion accuracy and multiple solutions caused by the assumption of a single disease in traditional inversion analysis.
[0017] This invention simultaneously corrects the physical parameters to be inverted during the iterative optimization process of the neural network, achieving high-precision inversion of multi-dimensional service performance parameters such as the stress state of the surrounding rock, lining voids, crack propagation, and concrete carbonation. The multi-dimensional physical parameters obtained from the inversion can comprehensively reflect the true service state of the tunnel structure, overcoming the limitation of traditional methods that can only detect a single type of defect.
[0018] This invention uses logical judgment based on the coupling relationship between multi-dimensional physical parameters. Through joint analysis of parameters such as elastic modulus, thermal conductivity, wave velocity, and electrical conductivity, it can accurately identify different types of defects such as lining voids, crack types, and carbonization corrosion, achieving in-depth diagnosis of tunnel structural defects. This method significantly improves the accuracy and reliability of defect identification through mutual verification and cross-validation of multiple physical parameters, providing a scientific basis for tunnel operation and maintenance decisions. Attached Figure Description
[0019] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a general architecture diagram of an embodiment of the present invention; Figure 2 This is a diagram of the neural network model architecture for physical parameter inversion analysis according to an embodiment of the present invention. Figure 3 This is a flowchart illustrating the disease depth diagnosis process according to an embodiment of the present invention. Detailed Implementation
[0020] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.
[0021] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.
[0022] Example 1 The main objective of this invention is to first use multiple monitoring instruments to distinguish and detect multi-source heterogeneous data, and then utilize the advantages of multi-source heterogeneous data fusion to achieve inversion analysis of multi-dimensional information of tunnel structure, thereby improving the accuracy of the inversion analysis results.
[0023] like Figure 1 As shown, this embodiment provides a method for perceiving multi-dimensional information on tunnel service performance based on multi-source heterogeneous data fusion, including the following steps: S1. Acquisition of multi-source heterogeneous data.
[0024] This includes collecting multi-source heterogeneous monitoring data of the tunnel and its surrounding environment using various sensing methods during tunnel construction or operation, specifically including the following steps: S1.1, Deploy various types of sensing devices in the tunnel construction area and its influence range to collect multi-source heterogeneous data reflecting the tunnel structural status, stratum response and construction conditions; The sensing device includes at least: Mechanical sensing devices are used to acquire information on displacement, strain, stress, or acceleration of structures or surrounding rock. These devices include: laser point clouds, strain gauges, vibrating wire strain gauges, fiber optic strain sensors, distributed fiber optic strain sensors, stress gauges, and earth pressure cells.
[0025] Thermal sensing devices are used to acquire temperature field information of tunnel structures or surrounding media. Thermal sensing devices include: temperature sensors, fiber Bragg grating temperature sensors, and infrared thermal imagers.
[0026] Electromagnetic sensing devices are used to acquire electromagnetic response information of underground media. The main type of electromagnetic sensing device is ground-penetrating radar.
[0027] Wave-type sensing devices are used to acquire propagation response information of sound waves, elastic waves, or electromagnetic waves. Wave-type sensing devices include: acoustic emission sensors, seismographs, and microseismic monitoring sensors.
[0028] S1.2 Continuous or intermittent monitoring of the tunnel construction process or service status is carried out through sensing devices to obtain raw monitoring data, and the raw monitoring data is classified and stored according to data source, physical attributes and acquisition frequency; among them, different types of monitoring data are heterogeneous in terms of time resolution, spatial distribution and data dimensions.
[0029] S1.3, uniformly identify the acquired multi-source heterogeneous monitoring data, assign a unique data label to each type of data, and the data label shall include at least the data type identifier, collection time information and spatial location information.
[0030] S2, Data Preprocessing.
[0031] The multi-source heterogeneous monitoring data obtained in step S1 are preprocessed to eliminate noise, unify scale, and achieve spatiotemporal alignment of the data to obtain a standardized dataset. This process includes the following sub-steps: S2.1 Perform integrity checks on the original monitoring data, identify missing data, abnormal data and invalid data, and remove or correct abnormal data according to preset rules to obtain corrected data; Abnormal data includes, but is not limited to, data points that are outside the physically reasonable range, data with sudden noise, or data errors caused by communication anomalies.
[0032] S2.2, Denoising the corrected data after integrity detection is performed to obtain denoised data. Denoising processing includes, but is not limited to, filtering, smoothing or statistical methods to reduce the impact of environmental noise and equipment noise on the monitoring data. Different types of monitoring data adopt denoising strategies that match their physical characteristics.
[0033] S2.3, Time alignment processing is performed on the denoised data. By matching timestamps or interpolating, monitoring data with different sampling frequencies are mapped to a unified time base to obtain time-aligned data. S2.4, Spatial alignment processing is performed on the time-aligned data. Based on the deployment location or spatial mapping relationship of the sensing devices, the monitoring data is unified to the preset spatial coordinate system to obtain the time-space aligned data. S2.5 Normalize or dimensionless the spatiotemporally aligned data to eliminate the influence of dimensional differences between different physical quantities on subsequent inversion analysis. S2.6 Organize the preprocessed multi-source heterogeneous data according to the preset data structure to form a standardized dataset for inversion analysis.
[0034] S3, a physical neural network model for the fusion of multi-source heterogeneous information.
[0035] A monitoring data loss function is constructed based on the standardized dataset, and a total loss function for the physical neural network model incorporating this monitoring data loss function is also constructed. This embodiment employs multi-source heterogeneous information for multi-dimensional information perception of tunnel structures. This embodiment constructs an inversion analysis method based on a physical neural network model. Physical neural networks are currently widely used, and the key to the physical neural network model in this embodiment lies in constructing a suitable loss function. This loss function integrates the fundamental equations of solid mechanics, thermodynamics, electromagnetism, and wave mechanics to achieve inversion analysis of multi-source heterogeneous data. The total loss function of the physical neural network model in this embodiment is... As shown in the following formula (1) in: θ These are the parameters of the physical field neural network model. κ The multidimensional information to be inverted includes in-situ stress and characteristic parameters of telecommunications defects in tunnel structures. To control the loss function of the return trip, where p The physical field equations representing the fusion (solid mechanics) thermodynamics Electromagnetism wave equation ). This is a loss function constructed based on boundary conditions. The loss function, based on multi-source heterogeneous detection data, is determined during iterative training. κ The key, This is the contact condition loss function.
[0036] The calculation method for the multinomial loss function is as follows: S3.1 The fundamental equation of solid mechanics is the momentum balance equation, which describes the balance between internal stress and body force in a material. It is the core governing equation for describing the mechanical field in continuum mechanics.
[0037] (2) in, For stress tensor; Physical force (such as gravity, inertial force); Density; Let be the displacement vector. The constitutive equation used to construct the mapping relationship between physical quantities is: (3) (4) in, For strain tensor; The elastic-rigid tensor (the fourth elastic modulus in the linear elastic case, which can be represented by Young's modulus E and Poisson's ratio ν for homogeneous isotropic materials). The residuals of the PDE equations constructed based on the fundamental equations of solid mechanics. The formula is: (5) In the formula, For divergence operators, For Cauchy stress tensor, For physical strength, For density distribution, It is acceleration.
[0038] Solid mechanical loss function for: (6) In the formula, The number of samples collected.
[0039] S3.2 The fundamental equation of thermodynamics is the heat conduction equation, which describes the diffusion of heat within a medium in time and space and is the main governing equation characterizing the evolution of the thermal field.
[0040] (7) in,T For temperature; ρ Density; z Specific heat capacity; k Thermal conductivity; Q This is an internal heat source item.
[0041] PDE equation residuals constructed based on fundamental thermodynamic equations The formula is: (8) In the formula, Let be a function of specific heat capacity as a function of time. x represents the spatial coordinate variable, and t represents the time coordinate variable. To be changed k ( x Spatial distribution function of the thermal conductivity of a material. This represents the spatial distribution of temperature over time. Let be the distribution function of the internal heat source term in time and space.
[0042] Therefore, the thermodynamic loss function for: (9) S3.3 The fundamental equations of electromagnetism are described by Maxwell's equations, among which the key equation most often used for time-varying field sensing is Faraday's law of electromagnetic induction, which reflects that a magnetic field that changes with time will generate an electric field.
[0043] (10) (11) in, Electric field strength; ∇×(⋅) represents the magnetic flux density; ∇×(⋅) represents the curl operator; denoted as the rate of change of the magnetic field over time. The magnetic field strength; Current density; It is the electric displacement; This is the displacement current term. The constitutive relation is: (12) in, is the dielectric constant (or dielectric tensor); μ is the magnetic permeability; For electrical conductivity (or conductivity tensor), the residuals of the PDE equations constructed based on the fundamental equations of electromagnetism are given. and The formula is: (13) Therefore, the electromagnetic loss function for: (14) S3.4 The fundamental equation of wave mechanics is the scalar wave equation, which describes the propagation characteristics of waves such as sound waves, elastic waves, and electromagnetic waves in a medium. It is the basic equation for wave field analysis.
[0044] (15) in, p Sound pressure (or scalar wave quantity); c Wave speed; ρ Let be the density of the medium. The acoustic constitutive equation is: (16) PDE equation residuals constructed based on the fundamental equations of wave mechanics The formula is: (17) In the formula, For scalar wave fields, For the wave source term of the excitation source, The phase velocity is used to represent the speed at which a wave propagates in a medium. This is the acceleration term of the wave field.
[0045] Therefore, the wave loss function for: (18) S3.5 Boundary conditions can generally be divided into Dirichlet boundary conditions and Neumann boundary conditions. For solid mechanics, thermodynamics, electromagnetism, and wave mechanics, the boundary condition loss function is uniformly expressed as: (19) in, These are the main variables of a physical field, such as displacement, temperature, electric field, and sound pressure. Generally, these are fluxes, such as stress, heat flux, and electromagnetic flux. Furthermore, The colon superscript indicates the result of neural network inference, while Superscripts indicate the true values of boundary conditions.
[0046] S3.6 Similarly, the continuity conditional loss function at the contact surface is divided into field continuity and flux continuity, and the formula is: (20) In this context, the superscripts (1) and (2) represent the field quantity and flux on both sides of the contact surface, respectively. This is the normal vector of the contact surface.
[0047] S3.7, the monitoring data loss function is expressed as: (twenty one) In the formula, This is a data loss function for monitoring. The number of samples for monitoring data, The neural network predicted the value. To ensure the accuracy of the monitoring data, The spatial location vector of the monitoring point. The time point for monitoring data.
[0048] S4. Multidimensional physical parameter inversion analysis.
[0049] This step, based on PINN, constructs an intelligent analysis model combining data-driven and physical perception. By minimizing the total loss function of the physical neural network model, it achieves accurate inversion of the physical parameters of the tunnel surrounding rock and structure. While updating the network weights, it corrects the physical parameters to be inverted, outputting the inverted multidimensional physical parameters. The neural network architecture is as follows: Figure 2 As shown. The specific implementation process is as follows: S4.1, Forward Calculation: First, construct a data-driven proxy model based on a deep neural network (DNN). This involves mapping the spatiotemporal coordinates of the tunnel monitoring area. Data is fed into the fully connected neural network as input, and then processed by the network's internal weights. The fully connected neural network performs nonlinear operations using activation functions and outputs predicted values of multidimensional physical fields. Predicted value Includes displacement ,temperature ,electric field and sound pressure Four categories of physical quantities. Then, sensor observation data collected on-site were introduced. As a supervisory label, the mean squared error between the predicted and observed values is calculated to obtain the loss function for the data-driven part. This is used to constrain the output of the neural network to approximate the real physical field.
[0050] S4.2, Loss Function Construction: Next, a physics perception and inversion module is constructed, embedding partial differential equations as prior knowledge into the network. The predicted physics field output from step S4.1 is differentiated using automatic differentiation (Auto-Diff) technology to obtain the partial derivative terms (e.g., ...). (etc.). Simultaneously, define the set of physical parameters to be inverted. ,gather Including elastic modulus Thermal conductivity Electrical conductivity and medium density The partial derivatives are compared with the physical parameters to be inverted. By substituting the equations of equilibrium in solid mechanics, the heat conduction equation, Maxwell's equations, and the wave equation, the residual values resulting from the inability to close the physical equations are calculated, thereby constructing the loss function of the physical driving part. .
[0051] S4.3, Inversion Analysis: Finally, construct the total loss function. The system employs an optimizer (such as Adam or L-BFGS) to perform iterative optimization. During the process of minimizing the total loss function, the system implements a dual update mechanism: firstly, it updates the weights of the neural network through backpropagation. To improve the accuracy of physical field prediction; on the other hand, to automatically correct the physical parameters to be inverted based on the gradient descent direction. When the total loss function converges to a preset threshold, the iteration stops, and the inversion analysis continues. The value represents the actual physical and mechanical parameters of the tunnel structure and surrounding rock.
[0052] The specific physical parameters representing multidimensional disease information in different physical fields are shown in Table 1.
[0053] Table 1 S5. Diagnosis of tunnel defects based on multidimensional physical parameter coupling.
[0054] This embodiment requires using the coupling relationship between the inverted multidimensional physical parameters for logical judgment, identifying the disease type, and alternately verifying the disease information with multidimensional physical information to provide remediation suggestions. The overall process is as follows: Figure 3 As shown.
[0055] like Figure 3 As shown, first, decision 1 is executed to check the elastic modulus. Has a significant decrease occurred? If the elastic modulus... A significant decrease indicates that the structural stiffness has been compromised, and the process then proceeds to judgment 2, further verifying the thermal conductivity. Whether the temperature decreases synchronously. The physical basis for this judgment is that air has a significant insulating effect; if the thermal conductivity... A decrease in thermal conductivity indicates the presence of voids within the structure, and the system will output diagnostic result A, classifying it as a lining back void or cavity defect. For this type of defect, the system will then perform a quantitative assessment, calculating the geometric dimensions of the void, such as its area and depth. Conversely, if the thermal conductivity... If the decrease is not synchronized (indicating that the medium is still continuous), proceed to judgment 2a and check the elastic modulus. Does it exhibit significant anisotropy? If significant directional characteristics exist, output diagnostic result B1, classifying it as a directional macroscopic crack, indicating relatively severe structural damage; if no significant directional characteristics exist, output diagnostic result B2, classifying it as a distributed microcrack, representing the early stage of structural damage. If the elastic modulus is found in judgment 1... If there is no significant decrease (i.e., the structural macroscopic stiffness is normal), the process then proceeds to judgment 3, checking the wave velocity of the sound wave or elastic wave. Is there significant attenuation or scattering? If the wave velocity... An anomaly was detected, indicating a change in the internal density or mass of the structure. The process then proceeds to judgment 3a, which triggers a check on conductivity. Whether they decrease synchronously. If conductivity A significant decrease occurs simultaneously (usually reflecting changes in pore water alkalinity or variations in deep-seated materials), and the system outputs diagnostic result C1, indicating severe carbonization or dissolution and other durability defects; if the conductivity... If there is no significant synchronous decrease, the diagnostic result C2 is output, indicating general material degradation or porosity change. If, under normal stiffness conditions, the wave velocity... No abnormalities were found, and the system output diagnostic result D, determining that the current tunnel structure is healthy and has no significant abnormal characteristics. After completing the diagnosis and indicator calculation for all the above-mentioned dimensions of defects, all data streams are converged to Decision 4, where the system comprehensively assesses whether the severity of the current defect has reached the pre-set safety warning threshold. If the defect is determined to be severe (reaching the warning threshold), the system will trigger Action Recommendation 1, automatically issuing a safety alarm and pushing an instruction to the management platform to immediately conduct on-site inspection or special repair. If the defect is determined to be minor or the structure is healthy (not reaching the warning threshold), the system will execute Action Recommendation 2, archiving and recording the multi-dimensional physical parameters and diagnostic conclusions, and recommending continuous monitoring of its evolution trend in subsequent routine operations and maintenance. At this point, the single closed-loop diagnostic process is complete.
[0056] This embodiment constructs a physical neural network model that integrates multiple physics fields, enabling inversion analysis using multi-source heterogeneous data fusion. By combining the advantages of various detection methods, different data corroborate and complement each other, improving the accuracy of inversion analysis.
[0057] The present invention also provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described thereon.
[0058] The present invention also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.
[0059] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for sensing multi-dimensional information on tunnel service performance based on multi-source heterogeneous data fusion, characterized in that, Includes the following steps: Multi-source heterogeneous monitoring data are collected based on various types of sensing devices deployed in the tunnel construction area; The multi-source heterogeneous monitoring data are preprocessed to obtain a standardized dataset; Based on the standardized dataset, a monitoring data loss function is constructed, and a total loss function for the physical neural network model containing the monitoring data loss function is also constructed. The spatiotemporal coordinates of the tunnel monitoring area are input into the neural network model. Based on the total loss function of the physical neural network model, iterative optimization is performed. While updating the network weights, the physical parameters to be inverted are corrected, and the multidimensional physical parameters obtained by inversion are output. Logical judgment is performed based on the coupling relationship between the multidimensional physical parameters to obtain the tunnel defect diagnosis results.
2. The method for sensing multi-dimensional information on tunnel service performance based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The process of preprocessing the multi-source heterogeneous monitoring data to obtain a standardized dataset includes: Anomalies in the multi-source heterogeneous monitoring data are identified and removed using integrity detection to obtain corrected data. The corrected data is denoised using a denoising strategy that matches the data types in the multi-source heterogeneous monitoring data to obtain denoised data. The denoised data is time-aligned based on timestamp matching to obtain time-aligned data. Based on the deployment location of the sensing devices, the time-aligned data is spatially aligned to obtain spatiotemporally aligned data. The spatiotemporally aligned data is normalized to obtain the standardized dataset.
3. The method for sensing multi-dimensional information on tunnel service performance based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The expression for the monitoring data loss function constructed based on the standardized dataset is as follows: ； In the formula, To monitor the data loss function, The number of samples for monitoring data, The neural network predicted the value. To ensure the accuracy of the monitoring data, The spatial location vector of the monitoring point. The time point for monitoring data.
4. The method for sensing multi-dimensional information on tunnel service performance based on multi-source heterogeneous data fusion according to claim 3, characterized in that, The process of constructing the total loss function of the physical neural network model includes: A solid mechanics loss function is constructed based on the momentum balance equation of solid mechanics, a thermodynamic loss function is constructed based on the heat conduction equation, an electromagnetic loss function is constructed based on Maxwell's equations, and a wave mechanics loss function is constructed based on the scalar wave equation. The physical driving loss function is obtained by combining the solid mechanics loss function, thermodynamic loss function, electromagnetic loss function and wave mechanics loss function. Construct a boundary loss function based on Dirichlet and Neumann boundary conditions; A continuity conditional loss function is constructed based on the continuity of the field and flux at the contact surface; The physical driving loss function, boundary loss function, continuity condition loss function, and monitoring data loss function are combined to obtain the total loss function of the physical neural network model.
5. The method for sensing multi-dimensional information on tunnel service performance based on multi-source heterogeneous data fusion according to claim 4, characterized in that, The expression for the total loss function of the physical neural network model is: ； In the formula, For the overall loss function, Here, κ represents the parameters of the physical field neural network model, and κ represents the multidimensional information to be inverted. The physical-driven loss function, p The physical field equations representing the fusion process are as follows: Construct a boundary loss function for the conditions. To monitor the data loss function, This is the continuous conditional loss function.
6. The method for sensing multi-dimensional information on tunnel service performance based on multi-source heterogeneous data fusion according to claim 1, characterized in that, The process of inputting the spatiotemporal coordinates of the tunnel monitoring area into a neural network model, iteratively optimizing the model based on its total loss function, updating the network weights while correcting the physical parameters to be inverted, and outputting the inverted multidimensional physical parameters includes: The spatiotemporal coordinates of the tunnel monitoring area are used as input layer data and fed into a fully connected neural network. Nonlinear calculations are performed through the weights and activation functions inside the network to output the predicted value of the multidimensional physical field. The predicted value of the multidimensional physical field is differentiated using automatic differentiation technology to obtain the partial derivative terms; Substitute the partial derivative terms and the physical parameters to be inverted into the solid mechanics equilibrium equation, heat conduction equation, Maxwell's equations and wave equation, calculate the residual values caused by the inability to close each physical equation, and obtain the physical driving loss function. The mean square error is calculated based on the sensor observation data and the predicted physical values to obtain the monitoring data loss function; The total loss function is obtained by combining the physical drive loss function and the monitoring data loss function. An optimizer is used to iteratively optimize based on minimizing the total loss function. While updating the neural network weights through backpropagation, the physical parameters to be inverted are automatically corrected according to the gradient descent direction until the total loss function converges, at which point the multidimensional physical parameters obtained by inversion are output.
7. The method for sensing multi-dimensional information on tunnel service performance based on multi-source heterogeneous data fusion according to claim 6, characterized in that, The process of obtaining tunnel defect diagnosis results by performing logical judgment based on the coupling relationship between the multidimensional physical parameters includes: Decision 1 is made based on the elastic modulus obtained from the inversion; if the elastic modulus decreases, then decision 2 is made. The determination 2 is based on thermal conductivity. If the thermal conductivity decreases synchronously, the diagnosis result of the back of the lining being detached is output. If the thermal conductivity does not decrease synchronously, the determination 2a is based on the anisotropic characteristics of the elastic modulus. Based on the anisotropic characteristics of the elastic modulus, a determination is made 2a. If anisotropy exists, a diagnosis result of structural damage is output; if anisotropy does not exist, a diagnosis result of initial damage is output.
8. The method for sensing multi-dimensional information on tunnel service performance based on multi-source heterogeneous data fusion according to claim 7, characterized in that, The process of obtaining tunnel defect diagnosis results by making logical judgments based on the coupling relationship between the multidimensional physical parameters also includes: Decision 1 is made based on the elastic modulus obtained from the inversion; if the elastic modulus does not decrease, then proceed to decision 3. Based on the wave velocity, the determination 3 is performed. If the wave velocity does not scatter, the structural health diagnosis result is output. If the wave velocity does scatter, the determination 3a is performed based on the conductivity. Based on the conductivity, a judgment 3a is made. If there is a synchronous decrease in conductivity, a diagnosis result of durability defects is output. If there is no synchronous decrease in conductivity, a diagnosis result of porosity change is output.
9. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in claim 1.
10. A storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in claim 1.