A dual-layer intelligent diagnostic method and system for the structural performance of urban underground pipelines

Abstract

This invention provides a two-layer intelligent diagnostic method and system for the structural performance of urban underground pipelines. By constructing a dynamic model for monitoring pipeline dynamic deflection, a solid physical and mechanical foundation is provided for diagnosis, effectively addressing the technical shortcomings of existing intelligent diagnostic models that lack physical rationality and are prone to deviating from the actual deformation patterns of pipelines. The dynamic model for monitoring urban underground pipeline dynamic deflection analyzes the variation law of pipeline deflection with pipeline structural performance, thereby accurately revealing the coupling mechanism between pipeline deflection and structural performance. Based on elastic dynamics theory, the model is solved through motion control equations in cylindrical coordinates, material constitutive relations, and frequency-wavenumber domain spectral analysis, accurately outputting dynamic pipeline deflection data. It clarifies the quantitative correlation between pipeline deflection data and pipeline structural performance parameters, providing clear mechanical theoretical support for subsequent diagnostic processes, significantly reducing the misjudgment rate of diagnostic results, and improving the reliability of the diagnosis.

Description

Technical Field

[0001] This invention relates to the technical field of pipeline performance testing, and in particular to a dual-layer intelligent diagnostic method and system for the structural performance of urban underground pipelines. Background Technology

[0002] As a core component of urban infrastructure, urban underground pipelines undertake key functions such as water supply and drainage, gas transmission, power and communication. The stability of their structural performance is directly related to the normal operation of the city, public safety, and the quality of life of the people.

[0003] The detection and diagnosis technologies for the structural performance of urban underground pipelines are mainly divided into two categories: traditional detection technologies and artificial intelligence-based detection and diagnosis technologies. Traditional detection technologies mainly include closed-circuit television (CCTV) inspection, ultrasonic testing, ground-penetrating radar (GPR) inspection, and manual visual inspection. These technologies rely heavily on manual operation, resulting in low detection efficiency, high labor intensity, limited detection range, and strong subjectivity. Traditional detection technologies can only locate pipelines, detect surface damage, or survey the surrounding environment. They cannot accurately obtain key structural performance parameters (such as elastic modulus and Poisson's ratio) of the pipeline body and surrounding soil, making it difficult to establish quantitative correlations between pipeline structural performance degradation and characteristic indicators such as deflection response. They also cannot conduct comprehensive and in-depth quality evaluation of pipeline structural performance, let alone achieve early warning and precise location of defects. With the application of artificial intelligence technology in the field of engineering inspection, data mining methods such as expert systems, Bayesian networks, and deep neural networks have been gradually introduced into underground pipeline fault diagnosis. However, in the diagnostic process, there is a lack of deep integration with the mechanical properties of the pipeline structure, and the failure to construct a dynamic model that conforms to the actual stress and deformation laws of the pipeline. This results in intelligent diagnostic models lacking physical theoretical support, easily leading to problems such as diagnostic results deviating from mechanical laws and misjudgments or omissions. Summary of the Invention

[0004] In view of the above problems, this invention is proposed to provide a two-layer intelligent diagnostic method and system for urban underground pipeline structure performance to overcome or at least partially solve the above problems. It can solve the problem that there is currently no technology that can deeply integrate this technical concept with elastic dynamics, multi-source data fusion technology and deep learning technology to build a complete two-layer intelligent diagnostic system for pipeline structure performance. It can also bring about the establishment of a knowledge-data collaborative driven two-layer intelligent diagnostic model, and achieve the effect of accurate judgment of pipeline structure performance through coarse-grained and fine-grained layered diagnosis.

[0005] Specifically, according to one aspect of the present invention, a two-layer intelligent diagnostic method for the structural performance of urban underground pipelines is characterized by comprising the following steps:

[0006] Step S1: Construct a dynamic model for monitoring pipeline deflection to establish a quantitative correspondence between pipeline deflection data and pipeline structural performance.

[0007] Step S2: Conduct a full-scale test on the pipeline, collect on-site monitoring data and full-scale test data, and obtain numerical simulation data of different pipeline structural performance based on the pipeline dynamic deflection monitoring dynamic model.

[0008] After converting the obtained field monitoring data, full-scale test data, and numerical simulation data into a unified format and time scale, data with strong correlation to the pipeline structure performance are selected to form a multi-source heterogeneous dataset.

[0009] Step S3: Merge and expand the data from the obtained multi-source heterogeneous dataset to obtain a diagnostic dataset;

[0010] Step S4: Perform coarse-grained diagnosis on the data in the obtained diagnostic dataset to obtain the pipelines with insufficient performance; perform fine-grained diagnosis on the pipelines with insufficient performance to obtain the diagnosis results of the pipelines.

[0011] Optionally, the following steps are also included:

[0012] Step S5: Update the diagnostic results obtained in step S4 into the comprehensive database, and call the diagnostic results in the comprehensive database through the console server.

[0013] Optionally, the construction of the dynamic model for monitoring pipeline deflection in step S1 to establish a quantitative correspondence between pipeline deflection data and pipeline structural performance includes:

[0014] Step S11: Construct a two-dimensional simplified axisymmetric analysis model of the pipe-soil system and establish a cylindrical coordinate system for the pipe-soil system;

[0015] Step S12: Based on the theory of elastic dynamics, the governing equations of motion for the pipe-soil system in cylindrical coordinates are as follows:

[0016] ;

[0017] in, For stress components, For radial and vertical displacements, The density of the material;

[0018] Step S13: Determine the constitutive relationship between the pipe body material and the surrounding soil material, wherein:

[0019] The pipeline body and the surrounding soil are both isotropic materials, and their constitutive relations are as follows:

[0020] ;

[0021] ;

[0022] in, Let Lamé constant be . With elastic modulus Compared to Poisson Related;

[0023] The pipeline body is made of isotropic material, and the surrounding soil layer is transversely isotropic material. Their constitutive relations are as follows:

[0024] ;

[0025] Among them, elastic constant From engineering parameters Sure.

[0026] Step S14: Solve the motion control equations using the efficient frequency domain-wavenumber domain spectral analysis method to obtain the pipe deflection data for each item;

[0027] Step S15: Based on the obtained pipe deflection data, establish the correspondence between the pipe deflection data and the pipe structural performance.

[0028] Optionally, in step S13, when both the pipe body and the surrounding soil are isotropic materials, the relationship between axisymmetric strain and displacement in the axisymmetric analysis model is as follows:

[0029] ;

[0030] When the pipeline body is an isotropic material and the surrounding soil layer is a transversely isotropic material, when When the constitutive relation of the material is described, it is an isotropic material relation.

[0031] Optionally, the step S14, which involves solving the motion control equations using an efficient frequency-wavenumber domain spectral analysis method to obtain various pipeline deflection data, includes:

[0032] Step S141: Perform a Fourier-Bessel transform on the motion control equations;

[0033] Step S142: Apply boundary and interface conditions; the boundary and interface conditions include at least: impact load boundary conditions, pipe-layer interface continuity or contact conditions, and outer boundary adopting absorbing boundary or infinite domain equivalent conditions.

[0034] Step S143: In the transformation domain, use the transfer matrix method or stiffness matrix method to accurately solve for the dynamic compliance matrix of the layered system. ;

[0035] Step S144: Reconstruct the displacement response in the physical space-time domain through double summation;

[0036] Step S145: Obtain the deflection time history curve and extract the basic feature vector reflecting the pipe structure performance from the obtained deflection time history curve. It must include at least: peak deflection Peak arrival time Bending energy The main frequency / band energy ratio and the half-peak width of the deflection basin are the pipeline deflection data mentioned in each item.

[0037] Optionally, in step S2, the method for selecting data with strong correlation to the pipeline structure performance to form a multi-source heterogeneous dataset is to use grey relational entropy analysis to quantify the correlation between the field monitoring data, the full-scale test data, and the numerical simulation data and the pipeline structure performance. The calculation formula is as follows:

[0038]

[0039] in, Let r represent the grey relational coefficient distribution mapping of the nth indicator to be screened, where r is the number of indicators to be screened. Let j be the sequence of the j-th indicator to be screened over time;

[0040] Based on the actual needs of the project, a preset gray entropy correlation threshold is set, and index data with correlation higher than the threshold are retained. The filtered data set forms the multi-source heterogeneous dataset.

[0041] Optionally, in step S3, the process of fusing and expanding the data in the obtained multi-source heterogeneous dataset to obtain a diagnostic dataset includes:

[0042] Step S31: Denoising and Completion Based on Enhanced Denoising Autoencoder: Train the enhanced denoising autoencoder (DAE), add noise to the input sample, and reconstruct it under the constraint of mask M to obtain the denoised and completed sample. ;

[0043] Step S32: Sample generation augmentation based on generative adversarial networks: Train the generative adversarial network (GAN) to make the generator output The discriminator is used to distinguish between real and generated samples, and the generator is used to output high-quality simulated deflection data.

[0044] Step S33: Inject physical constraints and remove invalid samples: Add a physical penalty term to the training loss. Remove generated samples that do not meet the constraints;

[0045] Step S34: Quality Control and Sample Set Union: Perform statistical consistency tests and anomaly detection on the generated samples to obtain the final generated sample set. and compared with the denoised and completed sample set obtained in step S31 Merge to form the diagnostic dataset.

[0046] Optionally, in step S4, the coarse-grained diagnosis of the data in the obtained diagnostic dataset to obtain the underperforming pipeline includes:

[0047] Step S401: Input the diagnostic dataset as the sample to be judged;

[0048] Step S402: Construct a multi-objective random forest model, taking the performance status of the pipeline structure as the classification objective, optimize the classification accuracy and feature importance ranking, and mine the deep core features of the samples;

[0049] Step S403: Calculate the multi-scale discrete entropy of data with different sampling periods, and select the period T with the smallest entropy value as the model update period;

[0050] Step S404: Principal component analysis is used to reduce the dimensionality of the deep features extracted by the multi-objective random forest model, retaining more than 95% of the original feature information;

[0051] Step S405: Input the dimensionality-reduced features into the support vector machine model, use the radial basis function as the kernel function, optimize the penalty factor and kernel parameters, and perform binary classification of pipeline "healthy state / insufficient performance state";

[0052] Step S406: Output coarse-grained diagnostic results. Samples that are determined to be in a state of insufficient performance will proceed to the fine-grained diagnostic process.

[0053] Optionally, in step S4, the underperforming pipe undergoes fine-grained diagnosis to obtain the diagnosis results for the pipe, including:

[0054] Step S407: Input the core feature vector of the pipeline sample that is determined to be underperforming by coarse-grained diagnosis, add multiple scene features, and form a composite feature set;

[0055] Step S408: The motion control equations, complex control equations of material constitutive relations, and correspondence rules between deflection characteristics and disease types of the dynamic model for monitoring pipeline deflection are embedded into the long short-term memory neural network loss function to construct a knowledge constraint layer.

[0056] Step S409: Add a multi-scale convolutional block structure to the input layer of the long short-term memory neural network model to extract the temporal features of the composite feature set in layers and learn the temporal correlation rules between features;

[0057] Step S410: Train the long short-term memory neural network model using pipeline defect type, severity, and key structural parameters as supervisory labels. During training, a physical penalty term is added to eliminate output results that violate mechanical laws, and the model weights are iteratively optimized.

[0058] Step S411: Input the composite feature set of the sample to be diagnosed, and output the specific disease type, severity level, and inverted structural performance parameters of the pipeline.

[0059] This invention also provides a dual-layer intelligent diagnostic system for the structural performance of urban underground pipelines, which runs any of the intelligent diagnostic methods described above, including: a pipeline deflection characteristic index monitoring subsystem, an urban underground pipeline comprehensive management database, a multi-source heterogeneous data fusion and dataset expansion subsystem, and a dual-layer intelligent diagnostic subsystem;

[0060] The pipeline deflection characteristic index monitoring subsystem includes a pipeline deflection measuring device, an automated drone, an IoT-based radar detection vehicle, a video surveillance robot, and a supporting underground pipeline network structure performance testing data acquisition workstation, used to collect multi-source heterogeneous raw data.

[0061] The urban underground pipeline integrated management database is used to provide data access control and data storage and retrieval call interfaces for the deep learning development board to provide a sample set of pipeline deflection characteristic indicators, and to store the entire process data of monitoring data, test data, and diagnostic results.

[0062] The multi-source heterogeneous data fusion and dataset augmentation subsystem includes an enhanced denoising autoencoder module, a convolution / transposed convolution module, a generator module, a discriminator module, and a loss function module, which are used to achieve data denoising and completion, generation and augmentation, and physical constraint verification.

[0063] The dual-layer intelligent diagnostic subsystem includes a pipeline dynamic deflection detection power module, a coarse-grained diagnostic module based on multi-objective random forest, and a fine-grained diagnostic module driven by knowledge and data collaboration, which are used to execute dual-layer diagnostic logic.

[0064] The present invention relates to a dual-layer intelligent diagnostic method and system for urban underground pipeline structural performance. By constructing a dynamic model for monitoring pipeline dynamic deflection, a solid physical and mechanical foundation for diagnosis is provided, effectively addressing the technical shortcomings of existing intelligent diagnostic models, such as lack of physical rationality and tendency to deviate from actual pipeline deformation patterns. The dynamic model for monitoring urban underground pipeline dynamic deflection analyzes the variation law of pipeline deflection with pipeline structural performance, thereby accurately revealing the coupling mechanism between pipeline deflection and structural performance. Based on elastic dynamics theory, the model solves the dynamic pipeline deflection data accurately through motion control equations in cylindrical coordinates, material constitutive relations, and frequency-wavenumber domain spectral analysis. This clarifies the quantitative correlation between pipeline deflection data and pipeline structural performance parameters, providing clear mechanical theoretical support for subsequent diagnostic processes, significantly reducing the misjudgment rate of diagnostic results, and improving diagnostic reliability. The system achieves a closed-loop process of "theoretical modeling—data acquisition—data processing—intelligent diagnosis," deeply integrating pipeline mechanics theory with data-driven algorithms, breaking the limitation of existing technologies where theoretical modeling and intelligent diagnosis are disconnected.

[0065] The above and other objects, advantages and features of the present invention will become more apparent to those skilled in the art from the following detailed description of specific embodiments of the invention in conjunction with the accompanying drawings. Attached Figure Description

[0066] The following sections will describe some specific embodiments of the invention in a detailed manner by way of example and not limitation, with reference to the accompanying drawings. The same reference numerals in the drawings denote the same or similar parts or portions. Those skilled in the art should understand that these drawings are not necessarily drawn to scale. In the drawings:

[0067] Figure 1 This is a flowchart illustrating a two-layer intelligent diagnostic method for the structural performance of urban underground pipelines according to an embodiment of the present invention.

[0068] Figure 2 This is a flowchart illustrating step S1 in a dual-layer intelligent diagnostic method for the structural performance of urban underground pipelines according to an embodiment of the present invention.

[0069] Figure 3 This is a flowchart illustrating step S14 of a dual-layer intelligent diagnostic method for the structural performance of urban underground pipelines according to an embodiment of the present invention.

[0070] Figure 4 This is a flowchart illustrating step S3 in a dual-layer intelligent diagnostic method for the structural performance of urban underground pipelines according to an embodiment of the present invention.

[0071] Figure 5 This is a flowchart illustrating step S4 in a dual-layer intelligent diagnostic method for the structural performance of urban underground pipelines according to an embodiment of the present invention.

[0072] Figure 6 This is a system architecture diagram of a dual-layer intelligent diagnostic system for the structural performance of urban underground pipelines according to an embodiment of the present invention;

[0073] Figure 7 This is a system hierarchy and data flow topology diagram of a two-layer intelligent diagnostic system for the structural performance of urban underground pipelines according to an embodiment of the present invention;

[0074] Figure 8 This is a diagram of pipe deflection test and measuring point layout for a dual-layer intelligent diagnostic method for the structural performance of urban underground pipelines according to an embodiment of the present invention.

[0075] Figure 9 This is a schematic diagram of a deflection basin for a dual-layer intelligent diagnostic method for the structural performance of urban underground pipelines according to an embodiment of the present invention. Detailed Implementation

[0076] Obviously, the accompanying drawings described below are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar scenarios based on these drawings without any inventive effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, any changes to design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of the content of this application.

[0077] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0078] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may represent singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to such processes, methods, products, or apparatus.

[0079] Figure 1 This is a flowchart illustrating a two-layer intelligent diagnostic method for the structural performance of urban underground pipelines according to an embodiment of the present invention. Figure 1-9 As shown in the figure, this embodiment of the invention provides a two-layer intelligent diagnostic method for the structural performance of urban underground pipelines, including the following steps:

[0080] Step S1: Construct a dynamic model for monitoring pipeline deflection to establish a quantitative correspondence between pipeline deflection data and pipeline structural performance.

[0081] Step S2: Conduct full-scale pipeline tests, collect on-site monitoring data and full-scale test data, and obtain numerical simulation data of different pipeline structural performance based on the pipeline dynamic deflection monitoring dynamic model.

[0082] After converting the obtained field monitoring data, full-scale test data, and numerical simulation data into a unified format and time scale, data with strong correlation to pipeline structural performance were selected to form a multi-source heterogeneous dataset.

[0083] Step S3: Merge and expand the data from the obtained multi-source heterogeneous dataset to obtain a diagnostic dataset;

[0084] Step S4: Perform coarse-grained diagnosis on the data in the obtained diagnostic dataset to identify underperforming pipelines; then perform fine-grained diagnosis on the underperforming pipelines to obtain the diagnostic results for those pipelines.

[0085] In this embodiment, a dynamic model for monitoring pipeline deflection is constructed to accurately establish a quantitative correspondence between pipeline deflection data and pipeline structural performance, providing clear mechanical references and improving the reliability of diagnosis. The model integrates full-scale pipeline test data, field monitoring data, and numerical simulation data obtained based on the dynamic model to form multi-source data support. Simultaneously, the data is fused and expanded, significantly improving the completeness and reliability of the diagnostic dataset and further enhancing diagnostic accuracy. Through a hierarchical diagnostic logic, a balance between diagnostic efficiency and accuracy is achieved, forming a synergistic effect of "rapid screening—precise positioning." This not only meets the system's complex data processing speed requirements but also ensures the accuracy of structural performance diagnostic results, solving the problem of quantifying, refining, and fully covering the structural performance diagnosis of urban underground pipelines.

[0086] In some embodiments of the present invention, the dual-layer intelligent diagnostic method for the structural performance of urban underground pipelines further includes the following steps:

[0087] Step S5: Update the diagnostic results obtained in step S4 into the comprehensive database, and call the diagnostic results in the comprehensive database through the console server.

[0088] In this embodiment, the obtained dual-layer intelligent diagnostic results of pipeline structural performance are updated and stored in the comprehensive management database of the console. The console server then displays a visual interface showing all the diagnostic results of pipeline structural performance. This further enhances the practicality and operability of the diagnostic results, effectively solving the technical shortcomings of existing technologies, such as scattered storage of diagnostic results, inconvenient viewing, and difficulty in intuitively grasping the health status of the entire pipeline network. It facilitates maintenance personnel to quickly view the diagnostic details of each pipeline segment, gain a comprehensive understanding of the overall performance of the underground pipeline network structure, significantly improve the timeliness and pertinence of maintenance decisions, and reduce maintenance management costs.

[0089] In some embodiments of the present invention, as shown in the appendix Figure 2 The construction of the dynamic model for monitoring pipeline deflection in step S1, to establish a quantitative correspondence between pipeline deflection data and pipeline structural performance, includes:

[0090] Step S11: Construct an axisymmetric analysis model of the pipeline-soil system and establish a cylindrical coordinate system for the pipeline-soil system. Specifically, establish an axisymmetric analysis model covering the pipeline body and the surrounding soil. The pipeline body material is considered an isotropic linear elastic body. The surrounding soil layer material, depending on the actual geological conditions, can be modeled as an isotropic or transversely isotropic linear elastic medium. The impact load is simplified as an axisymmetric transient point load acting on the center of the top of the pipeline. A coordinate system is established for subsequent mechanical analysis, as follows:

[0091] .

[0092] Step S12: Based on the theory of elastic dynamics, the governing equations of motion for the pipe-soil system in cylindrical coordinates are as follows:

[0093] ;

[0094] in, For stress components, For radial and vertical displacements, This represents the material density.

[0095] Step S13: Determine the constitutive relationship between the pipe body material and the surrounding soil material, wherein:

[0096] Both the pipeline body and the surrounding soil are isotropic materials, and their constitutive relations are as follows:

[0097] ;

[0098] ;

[0099] in, Let Lamé constant be . With elastic modulus Compared to Poisson Related;

[0100] The pipeline body is made of isotropic material, and the surrounding soil layer is made of transversely isotropic material. Their constitutive relations are as follows:

[0101] ;

[0102] Among them, elastic constant From engineering parameters Sure.

[0103] Step S14: Solve the motion control equations using the efficient frequency domain-wavenumber domain spectral analysis method to obtain various pipeline deflection data.

[0104] Step S15: Based on the obtained pipeline deflection data, establish the correspondence between the pipeline deflection data and the pipeline structural performance. Specifically, the dynamic deflection basin of the pipeline and the ground surface under transient impact loads can be output. This theoretical model, as a "forward model," provides a precise physical and mechanical basis for parameter inversion, feature extraction, and intelligent diagnosis in subsequent steps.

[0105] In this embodiment, the constructed dynamic model for monitoring pipeline deflection provides a solid physical and mechanical foundation for diagnosis, effectively addressing the technical shortcomings of existing intelligent diagnostic models, such as lack of physical rationality and tendency to deviate from the actual deformation patterns of pipelines. A dynamic model for monitoring the dynamic deflection of urban underground pipelines was constructed, and the variation law of pipeline deflection with pipeline structural performance was analyzed, thus accurately revealing the coupling mechanism between pipeline deflection and structural performance. Based on elastic dynamics theory, the model is solved through motion control equations in cylindrical coordinates, material constitutive relations, and frequency-wavenumber domain spectral analysis, accurately outputting dynamic pipeline deflection data. It clarifies the quantitative correlation between pipeline deflection data and pipeline structural performance parameters, providing clear mechanical theoretical support for subsequent diagnostic processes, significantly reducing the misjudgment rate of diagnostic results, and improving the reliability of diagnosis.

[0106] In some embodiments of the present invention, when both the pipe body and the surrounding soil are isotropic materials in step S13, the relationship between axisymmetric strain and displacement in the axisymmetric analysis model is as follows:

[0107] ;

[0108] When the pipeline body is an isotropic material and the surrounding soil layer is a transversely isotropic material, when At this time, the constitutive relation of the material is that of an isotropic material.

[0109] In this embodiment, determining the material relationships of the same line is to ensure the physical rationality of the dynamic model, solve the problem in the prior art that the model does not accurately match the material properties and the calculation results deviate from the actual deformation law of the pipeline, provide accurate mechanical parameter input for model solving, and improve the accuracy of the calculation results of dynamic deflection basin and core feature vector.

[0110] In some embodiments of the present invention, as shown in the appendix Figure 3 As shown, in step S14, the motion control equations are solved using an efficient frequency domain-wavenumber domain spectral analysis method to obtain various pipe deflection data, including:

[0111] Step S141: Perform a Fourier-Bessel transform on the motion control equations. Specifically, for the motion control equations in cylindrical coordinates (including radial, vertical displacement, and stress components), a joint Fourier-Bessel transform is used to transform the spatiotemporal domain... The displacement and stress functions within the domain are transformed into the frequency domain and wavenumber domain. The function within the function achieves dimensionality reduction. Specifically, it performs a Fourier transform on the vertical coordinate z, converting the vertical spatial variable into a wavenumber. (m is the wave number order); perform a Bessel transformation on the radial coordinate r to adapt to the axisymmetric properties of the cylindrical coordinate system; perform a Fourier transform on the time variable t to convert the time variable into angular frequency. (where n is the frequency order), and finally the original partial differential control equations in the spatiotemporal domain are transformed into linear algebraic equations in the frequency domain-wavenumber domain, eliminating the spatiotemporal coupling terms in the equations and simplifying the solution process.

[0112] Step S142: Apply boundary and interface conditions; boundary and interface conditions include at least: impact load boundary conditions, pipe-layer interface continuity or contact conditions, and absorbing boundary or infinite domain equivalent conditions for the outer boundary. Specifically, based on the actual stress and contact state of the pipe-soil system, three types of core constraint conditions are applied in the frequency domain-wavenumber domain to ensure that the solution results are consistent with reality: ① Impact load boundary conditions: Transform the transient point load into a load spectrum in the frequency domain-wavenumber domain and apply it to the corresponding position at the top center of the pipe; ② Pipe-soil interface constraint conditions: Apply continuous constraints (displacement continuity, stress continuity) or contact constraints (considering interface friction) according to the contact state between the pipe and the soil in actual engineering; ③ Outer boundary constraint conditions: Treat the outer boundary of the soil layer as an infinite domain and adopt absorbing boundary conditions to avoid interference from boundary reflection waves on the solution results, and equivalently simulate the mechanical response of the infinite underground soil. At the same time, combined with the material constitutive relation, the stress components are transformed into expressions for displacement components and substituted into the transformed linear algebraic equations to further simplify the solution system.

[0113] Step S143: Within the transformed domain, accurately solve the dynamic compliance matrix of the layered system using the transfer matrix method or the stiffness matrix method. Specifically, the transfer matrix method or the stiffness matrix method is used to efficiently solve the transformed and constrained linear algebraic equations to obtain the dynamic compliance matrix of the pipe-soil system in the frequency domain-wavenumber domain. This matrix quantifies the displacement response relationship of the pipeline at various locations under unit load in the frequency domain and wavenumber domain, covering the correspondence between radial and vertical displacement and wavenumber and angular frequency. During the solution process, matrix iterative optimization is used to improve the calculation accuracy, while avoiding the computational redundancy caused by infinite integration in traditional solution methods, thus greatly improving the solution efficiency.

[0114] Step S144: Reconstruct the displacement response in the physical spatiotemporal domain through double summation. Specifically, the obtained frequency-wavenumber domain dynamic compliance matrix is... By performing a double inverse Fourier-Bessel transform, the displacement response function in the frequency domain and wavenumber domain is inverted and reconstructed into the physical space-time domain. The pipeline displacement response function is obtained, yielding the radial and vertical displacement time history curves of the pipeline under transient impact loads. Finally, core feature vectors such as peak deflection, peak arrival time, and deflection energy are extracted from the displacement time history curves to output the pipeline's dynamic deflection basin. This completes the solution of the motion control equations, providing accurate mechanical calculation basis for subsequent data screening and intelligent diagnosis.

[0115] Step S145: Solve the deflection time history curve. Extract the basic feature vector reflecting the pipeline structural performance from the obtained deflection time history curve. It should include at least: peak deflection, peak arrival time, deflection energy, main frequency / band energy ratio, and half-peak width of the deflection basin. These are the various pipeline deflection data.

[0116] In this embodiment, the solution process, through a closed-loop operation of "transformation-constraint-solution-inverse transformation", balances solution efficiency and accuracy, effectively solving the technical problems of high difficulty and cumbersome calculation in solving the motion control equations of the pipeline-soil system in the spatiotemporal domain, and providing efficient and reliable core technical support for the theoretical modeling of the entire diagnostic method.

[0117] In some embodiments of the present invention, as shown in the appendix Figure 1 As shown, in step S2, the method for selecting data with strong correlation to pipeline structural performance to form a multi-source heterogeneous dataset is to use grey relational entropy analysis to quantify the correlation between field monitoring data, full-scale test data, and numerical simulation data and pipeline structural performance. The calculation formula is as follows:

[0118]

[0119] Where, represents the grey relational coefficient distribution mapping of the nth indicator to be screened, r is the number of indicators to be screened, and is the sequence of the jth indicator to be screened over time;

[0120] Based on the actual needs of the project, a preset gray entropy correlation threshold is set, and index data with correlation higher than the threshold are retained. The filtered data set forms a multi-source heterogeneous dataset.

[0121] Specifically, in step S2, firstly, when processing the three types of data, periodic monitoring of real underground pipelines in the city is carried out using equipment such as pipeline deflection measurement devices, IoT radar detection vehicles, and video surveillance robots. Deflection index data of the pipeline under natural operating conditions is collected, including dynamic parameters such as peak deflection, peak arrival time, and dominant frequency / band energy ratio, forming a real pipeline deflection index dataset as on-site pipeline monitoring data.

[0122] A test pipe section was constructed to match the specifications, materials, and burial environment of a real pipeline in a 1:1 scale. Sensors were deployed to collect deflection data of the test pipe section under artificially controlled impact loads and soil conditions. At the same time, combined with a multi-field coupled transient hydraulic model of pipeline structural performance, the instantaneous state parameters of pipeline confluence were quantitatively calculated using theoretical formulas to form a full-scale test index dataset as full-scale test data.

[0123] Based on the dynamic deflection monitoring model of the pipeline constructed in step S1, the stress state of the pipeline with different degrees of damage and different service lives is simulated on the computer. The structured data such as stress, strain, and displacement of the pipeline and soil are output as numerical simulation data, supplementing the two types of measured data.

[0124] Secondly, the differences in deflection indices between on-site monitoring data and full-scale test data were compared, and the reasons for these differences were analyzed. These reasons included differences in scenario conditions such as load intensity, soil density, and ambient temperature.

[0125] Adjust the scenario parameters of the full-scale test, such as adjusting the magnitude of the impact load and changing the type of test soil, so that the pipe bending state of the full-scale test is as close as possible to the actual state of the real pipe.

[0126] The format and time scale of the three types of data are unified, and unstructured waveform curves and text records are transformed into calculable numerical indicators; the time axes of data from different monitoring periods and different test batches are aligned to eliminate the problem of inconsistent data dimensions.

[0127] Finally, using "the degree of correlation between indicators and pipeline structural performance" as the core screening criterion, indicators that are sensitive to the health status of pipelines are retained, while redundant and low-correlation indicators are eliminated.

[0128] The grey relational entropy analysis method is used to quantify the degree of correlation between indicators. The core calculation formula is as follows:

[0129]

[0130] Where represents the grey relational coefficient distribution mapping of the nth indicator to be screened, r is the number of indicators to be screened, and is the sequence of the jth indicator to be screened over time.

[0131] Set a gray entropy correlation threshold based on the actual needs of the project, and retain indicators with a correlation higher than the threshold, such as peak deflection, energy ratio, peak arrival time and other core indicators.

[0132] The selected metrics are multi-source heterogeneous datasets that are strongly correlated with pipeline structural performance, and can be directly used for subsequent data fusion and intelligent diagnostic model training.

[0133] In this embodiment, a multi-source heterogeneous data-driven path of "full-scale testing—refined numerical simulation—regular on-site monitoring" is adopted, and gray relational entropy analysis is used to complete the index selection. This effectively solves the problems of single data source, inconsistent data quality, and redundant data interfering with diagnostic accuracy in existing technologies. By aligning multi-source data scenarios, the consistency of data from full-scale testing, numerical simulation, and on-site monitoring is ensured. By using gray relational entropy analysis to quantify the correlation between indicators and pipeline structural performance, highly correlated core indicators are selected, providing high-quality and highly targeted data input for subsequent intelligent diagnosis. At the same time, the comprehensiveness and effectiveness of the data are taken into account, laying a data foundation for high-precision diagnosis.

[0134] In some embodiments of the present invention, as shown in the appendix Figure 4 As shown, in step S3, the data from the obtained multi-source heterogeneous dataset are fused and expanded to obtain a diagnostic dataset including:

[0135] Step S31: Denoising and Completion Based on Enhanced Denoising Autoencoder: Train the enhanced denoising autoencoder (DAE), add noise to the input sample, and reconstruct it under the constraint of mask M to obtain the denoised and completed sample. Specifically, the input is a multi-source heterogeneous dataset obtained through filtering. This dataset includes pipeline field monitoring data, full-scale test data, and numerical simulation data on deflection characteristics, containing a small amount of noise, missing values, and abnormal fluctuations. Gaussian noise or impulse noise is randomly added to the input samples, and a mask matrix M is set to randomly mask some data features. An enhanced denoising autoencoder is trained, allowing the model to learn to accurately reconstruct noise-free and complete pipeline deflection data samples from noisy and partially masked inputs. Through iterative optimization of the model, noise and missing values ​​in the original data are repaired, resulting in a clean and complete denoised and imputed sample set. This addresses the issue of inconsistent data quality from multiple sources.

[0136] Step S32: Sample Generation Augmentation Based on Generative Adversarial Networks: Train a Generative Adversarial Network (GAN) to generate high-quality simulated deflection data. The generator outputs data, the discriminator distinguishes between real and generated samples, and the generator outputs real / generated samples. Specifically, construct a GAN architecture containing a generator G and a discriminator D. Train the generator: The generator receives random noise z and the pipe deflection feature vector V as input and outputs simulated deflection data samples. The goal is to generate simulated data with extremely high similarity to real samples. The discriminator is trained by receiving real samples (denoised and completed samples) and generated samples, distinguishing between them through a binary classification task, and outputting the probability that a sample is "real" or "generated". The generator and discriminator are trained alternately until a Nash equilibrium is reached, at which point the generator can output high-quality simulated deflection data that is difficult for the discriminator to distinguish.

[0137] Step S33: Inject physical constraints and remove invalid samples: Add a physical penalty term to the training loss. Samples that do not meet the constraints are discarded. Specifically, a physical penalty term is added to the loss function of the GAN model. ,in The constraints for pipeline elastic dynamics include the peak deflection range, stress-strain relationship, and displacement compatibility conditions. The simulated samples output by the generator are physically validated. If a sample does not conform to the mechanical deformation laws of the pipeline structure, its loss value is increased through a penalty term, forcing the generator to adjust its parameters. Simulated samples that do not meet the physical constraints are filtered out, ensuring that the remaining generated samples conform to the objective laws of actual pipeline deformation, avoiding "pseudo-data" from interfering with subsequent diagnostic models.

[0138] Step S34: Quality Control and Sample Set Union: Perform statistical consistency tests and anomaly detection on the generated samples to obtain the final generated sample set. and compared with the denoised and completed sample set obtained in step S31 The generated samples are merged to form a diagnostic dataset. Specifically, the generated samples, after being filtered by physical constraints, are validated using statistical indicators such as mean, variance, and distribution characteristics to ensure that the statistical characteristics of the generated samples are consistent with those of the real samples. An isolated forest or standard deviation method is used to remove a small number of outliers from the generated samples, further improving data quality. The generated sample set that passes quality control is then used. , and the obtained denoised and completed sample set The datasets are merged to form a final dataset that is sufficient in quantity and reliable in quality, which is the diagnostic dataset, used for training the next step of the two-layer intelligent diagnostic model.

[0139] In this embodiment, data fusion and augmentation are achieved by combining an enhanced denoising autoencoder (DAE) and a generative adversarial network (GAN), and physical constraints are injected, effectively addressing the technical challenges of noise, missing values, and insufficient sample quantity in multi-source data. The DAE achieves data denoising and completion, improving the quality of the original data; the GAN generates simulated samples that conform to the laws of physics and mechanics, compensating for the insufficient number of real samples; the introduction of a physical penalty term ensures the rationality of the generated samples, avoids interference from "pseudo-data" on diagnostic results, further improves data reliability, and provides a guarantee for the stable training of the subsequent two-layer intelligent diagnostic model.

[0140] In some embodiments of the present invention, as shown in the appendix Figure 5 As shown, in step S4, coarse-grained diagnosis is performed on the data in the obtained diagnostic dataset to identify underperforming pipelines, including:

[0141] Step S401: Input the diagnostic dataset as the sample to be judged. This dataset contains multi-source deflection data, including collected pipeline field monitoring data, full-scale test data, and numerical simulation data.

[0142] Step S402: Construct a multi-objective random forest model, using the performance status of the pipeline structure as the classification objective, optimizing classification accuracy and feature importance ranking, and mining deep core features of the samples. Specifically, a multi-objective random forest model is constructed, setting two optimization objectives: "classification accuracy of pipeline structure performance status" and "stability of feature importance ranking". The input dataset is manually labeled, with the labeling results divided into two categories: "healthy state" and "insufficient performance state", serving as supervision labels for model training. The multi-decision tree ensemble learning mechanism of MRF is used to perform layer-by-layer decomposition and correlation analysis of sample features. Core features with the greatest impact on performance status are selected, such as the peak deflection change rate and the anomaly of the main frequency energy ratio, while redundant features are automatically removed.

[0143] Step S403: Calculate the multi-scale discrete entropy of data from different sampling periods, and select the period T with the smallest entropy value as the model update period. Specifically, for monitoring data with different sampling periods, such as 1 day, 3 days, 7 days, and 15 days, calculate the multi-scale discrete entropy respectively. The discrete entropy value reflects the complexity of the data sequence; the smaller the entropy value, the stronger the data regularity and the less noise interference. Select the period T with the smallest discrete entropy value and set it as the update period of the structural performance detection model. This operation ensures that while taking into account the timeliness of diagnosis, the input data has stable regular characteristics, avoiding model fluctuations caused by frequent updates.

[0144] Step S404: Principal Component Analysis (PCA) is used to reduce the dimensionality of the deep features extracted by the multi-objective random forest (MRF) model, retaining more than 95% of the original feature information. Specifically, the deep core features extracted by MRF are dimensionality-reduced to address the "curse of dimensionality" problem caused by excessively high feature dimensionality, thereby reducing the computational complexity of subsequent models. The high-dimensional feature vectors are mapped to a low-dimensional space using the PCA algorithm, with a variance contribution rate threshold set at 95%. While retaining more than 95% of the original feature information, the high-dimensional features are compressed into a few principal component features.

[0145] Step S405: Input the dimensionality-reduced features into the Support Vector Machine (SVM) model, using the radial basis function (RBF) as the kernel function, optimize the penalty factor and kernel parameters, and perform binary classification of the pipeline as "healthy state / underperforming state". Specifically, input the principal component feature vectors after PCA dimensionality reduction into the SVM model, and select the radial basis function (RBF) as the kernel function. The kernel function can map linearly inseparable features in low-dimensional space to high-dimensional space to achieve linear separability. The penalty factor C and kernel function parameters of the SVM are optimized using a grid search method. This improves the classification accuracy and generalization ability of the model. The sample to be diagnosed is input into the trained SVM model, and the model outputs the class label of the sample, which is either "healthy" or "underperforming".

[0146] Step S406: Output coarse-grained diagnostic results. Samples identified as having insufficient performance will proceed to the fine-grained diagnostic process. Specifically, if a sample is identified as healthy, the diagnostic results will be directly stored in the comprehensive management database without proceeding to the next diagnostic step. If a sample is identified as having insufficient performance, its feature data and the abnormal features identified by MRF mining will be marked and fed into the second-layer fine-grained diagnostic module for more accurate disease localization and parameter inversion.

[0147] In some embodiments of the present invention, as shown in the appendix Figure 1 As shown, in step S4, the underperforming pipe undergoes fine-grained diagnosis, and the diagnosis results for the pipe include:

[0148] Step S407: Input the core feature vector of the pipeline samples identified as underperforming in the coarse-grained diagnostic analysis, and add multiple scenario features to form a composite feature set. Specifically, screen the underperforming pipeline samples output by the coarse-grained diagnostic analysis, and extract the core feature vector of these samples after MRF mining and PCA dimensionality reduction, including key indicators such as deflection peak anomaly rate, dominant frequency energy ratio offset, and peak arrival time delay. Supplement with multi-dimensional scenario features, such as pipeline burial years, soil type in the area, and temperature and humidity monitoring data for different seasons, to form a composite feature set for fine-grained diagnosis.

[0149] Step S408: Embed the motion control equations, complex control equations of material constitutive relations, and correspondence rules between deflection characteristics and disease types of the pipeline dynamic deflection monitoring dynamic model into the loss function of a long short-term memory neural network to construct a knowledge constraint layer. Specifically, the core equations of the pipeline dynamic deflection monitoring dynamic model, including the motion control equations of elastic dynamics and material constitutive relations, are transformed into mathematical constraints and embedded into the loss function of a long short-term memory neural network (LSTM). Incorporating domain expert experience, such as rules like "an increase in the half-peak width of the deflection basin corresponds to pipe structural cracking" and "a decrease in elastic modulus is positively correlated with the peak deflection value," a knowledge constraint layer is constructed to verify the rationality of the model output.

[0150] Step S409: Add a multi-scale convolutional block structure to the input layer of the Long Short-Term Memory (LSTM) neural network model to extract temporal features of the composite feature set hierarchically and learn the temporal correlation patterns between features. Specifically, before the input layer of the traditional LSTM model, a multi-scale convolutional block structure is added, with convolutional kernels of different sizes to capture short-term, medium-term, and long-term deflection temporal data features, corresponding to the performance degradation process of the pipeline at different stages. The composite feature set is input into the improved LSTM model, and the multi-scale convolutional blocks extract temporal features hierarchically. Then, the LSTM network learns the temporal correlation patterns between features, solving the problem that single-scale features cannot cover the full-cycle degradation information.

[0151] Step S410: Train a Long Short-Term Memory (LSTM) neural network model using pipeline defect type, severity, and key structural parameters as supervised labels. A physical penalty term is added during training to eliminate outputs that violate mechanical laws, and the model weights are iteratively optimized. Specifically, actual pipeline defect data is used as supervised labels, including defect type (cracks, corrosion, structural deformation, etc.), severity (mild / moderate / severe), and key structural parameters (elastic modulus, Poisson's ratio), to train the improved LSTM model. A physical penalty term is added during training. If the structural parameters output by the model violate the mechanical laws of the dynamic model, such as an elastic modulus lower than the material's theoretical lower limit, a penalty mechanism is triggered, adjusting the model weight parameters until the output conforms to physical constraints.

[0152] Step S411: Input the composite feature set of the sample to be diagnosed, and output the specific pipeline defect type, severity level, and inverted structural performance parameters. Specifically, after the model completes training, input the composite feature set of the sample to be diagnosed, and output three core diagnostic results: first, the specific pipeline defect type; second, the defect severity level; and third, the inverted key structural performance parameter values. The diagnostic results are correlated with the sample's original monitoring data and coarse-grained diagnostic conclusions, and synchronously stored in the integrated management database to provide accurate data support for subsequent pipeline operation and maintenance decisions.

[0153] In this embodiment, a two-layer intelligent diagnostic logic of "coarse-grained diagnosis - fine-grained diagnosis" is designed, which balances diagnostic efficiency and accuracy, effectively solving the technical problem that existing technologies cannot simultaneously meet the needs of large-scale pipeline network screening efficiency and accurate diagnosis. Coarse-grained diagnosis uses multi-objective random forest (MRF) feature mining, PCA dimensionality reduction, and SVM binary classification to quickly screen out pipelines with insufficient performance and eliminate healthy pipelines, significantly reducing the computational load of subsequent fine-grained diagnosis and improving the diagnostic efficiency of large-scale underground pipeline networks. Fine-grained diagnosis improves the LSTM model through knowledge embedding and multi-scale convolutional blocks to accurately identify the type, severity, and key structural performance parameters of pipeline defects, achieving accurate location and quantitative evaluation of pipeline defects and meeting the actual needs of precise operation and maintenance of urban underground pipelines.

[0154] This invention also provides a dual-layer intelligent diagnostic system for the structural performance of urban underground pipelines, used to run the intelligent diagnostic method in any of the above embodiments, as shown in the appendix. Figure 6-9 As shown, it includes a pipeline deflection characteristic index monitoring subsystem, an urban underground pipeline integrated management database, a multi-source heterogeneous data fusion and dataset expansion subsystem, and a two-layer intelligent diagnostic subsystem.

[0155] The pipeline deflection characteristic index monitoring subsystem includes a pipeline deflection measuring device, an automated drone, an IoT-based radar detection vehicle, a video surveillance robot, and a supporting underground pipeline network structure performance testing data acquisition workstation, used to collect multi-source heterogeneous raw data.

[0156] The urban underground pipeline integrated management database is used to provide data access control and data access call interface for the pipeline deflection characteristic index sample set of deep learning development boards, and to store the entire process data of monitoring data, test data, and diagnostic results.

[0157] The multi-source heterogeneous data fusion and dataset augmentation subsystem includes an enhanced denoising autoencoder module, a convolution / transposed convolution module, a generator module, a discriminator module, and a loss function module, which are used to achieve data denoising and completion, generation and augmentation, and physical constraint verification.

[0158] The dual-layer intelligent diagnostic subsystem includes a pipeline dynamic deflection detection power module, a coarse-grained diagnostic module based on multi-objective random forest, and a fine-grained diagnostic module driven by knowledge and data collaboration, which are used to execute dual-layer diagnostic logic.

[0159] From the perspective of a three-tier architecture, this embodiment is also composed of the following parts:

[0160] (1) Data access control layer, which consists of a comprehensive management database server, is connected to a geographic information system (GIS) server and a software standard data interface. The software standard data interface includes a pipeline geographic information query module and a pipeline structure performance diagnosis module. The interface function is supported by wired / wireless monitoring technology.

[0161] (2) Access control layer, carried by GIS server, provides GIS services, business query and operation services, computing services and database access relay access control services. Business query and operation services include C / S business system and B / S business system. Computing services support complex data analysis and processing.

[0162] (3) Application layer: It adopts a hybrid architecture mode of C / S business system and B / S business system to provide users with functions such as visualization of diagnostic results, data query, and operation and maintenance decision support.

[0163] This includes a hardware and software support platform: the hardware component includes a pipeline structure performance intelligent diagnostic terminal, a comprehensive management database server, an urban pipeline network GIS server, a data acquisition workstation, a monitoring and operation status visualization workstation, and detection instruments, meters, and sensors connected via the Internet of Things. The software component covers dynamic model calculation programs, data processing algorithms, deep learning models, and visualization software.

[0164] In this embodiment, a hardware-software integrated intelligent diagnostic system for urban underground pipelines was constructed, realizing a control platform that integrates pipeline deflection characteristic data acquisition, parameter calibration, and intelligent diagnostic algorithms. Based on a hybrid model combining C / S and B / S service modes, and using a geographic information system server and a comprehensive database server as the data processing platform, and employing a dual-layer intelligent diagnostic algorithm for dynamic deflection of urban underground pipelines driven by elastic dynamics, full-scale testing, numerical simulation, and "knowledge-data" collaboration, an intelligent diagnostic system for pipeline structural performance was designed and developed. This system can meet the complex data processing speed requirements of the system while ensuring the accuracy of the structural performance diagnostic results output, solving the problem that urban underground pipeline structural performance diagnosis cannot achieve quantification, precision, and full coverage.

[0165] This invention employs advanced knowledge embedding theory, finite element analysis and full-scale testing methods, deep learning technology, and complex system engineering implementation techniques to optimize and improve relevant aspects of urban underground pipeline structural performance diagnosis. It couples the elastic-dynamic relationship between pipeline structural performance and deflection, fully utilizing the technical route of "theoretical analysis—multi-dimensional data-driven screening—model construction—knowledge embedding." Based on software and hardware support, it provides comprehensive and professional intelligent diagnostic engineering implementation methods, ensuring that the various application subsystems of the dual-layer intelligent diagnostic method and system for urban underground pipeline structural performance operate collaboratively on a unified basic database. This provides strong technical support for the intelligent, accurate, and comprehensive diagnosis of urban underground pipeline structural performance, meeting the detection needs of most urban underground pipeline structural performance diagnosis.

[0166] Therefore, those skilled in the art should recognize that although numerous exemplary embodiments of the present invention have been shown and described in detail herein, many other variations or modifications conforming to the principles of the present invention can be directly determined or derived from the disclosure of the present invention without departing from the spirit and scope of the invention. Thus, the scope of the present invention should be understood and construed as covering all such other variations or modifications.

Claims

1. A dual-layer intelligent diagnostic method for the structural performance of urban underground pipelines, characterized in that, Includes the following steps: Step S1: Construct a dynamic model for monitoring pipeline deflection to establish a quantitative correspondence between pipeline deflection data and pipeline structural performance. Step S2: Conduct a full-scale pipeline test, collect on-site monitoring data and full-scale test data of the pipeline, and obtain numerical simulation data of different pipeline structural performance based on the pipeline dynamic deflection monitoring dynamic model. After converting the obtained field monitoring data, full-scale test data, and numerical simulation data into a unified format and time scale, data with strong correlation to the pipeline structure performance are selected to form a multi-source heterogeneous dataset. Step S3: Merge and expand the data from the obtained multi-source heterogeneous dataset to obtain a diagnostic dataset; Step S4: Perform coarse-grained diagnosis on the data in the obtained diagnostic dataset to obtain the pipelines with insufficient performance; perform fine-grained diagnosis on the pipelines with insufficient performance to obtain the diagnosis results of the pipelines.

2. The dual-layer intelligent diagnostic method for the structural performance of urban underground pipelines according to claim 1, characterized in that, It also includes the following steps: Step S5: Update the diagnostic results obtained in step S4 into the comprehensive database, and call the diagnostic results in the comprehensive database through the console server.

3. The dual-layer intelligent diagnostic method for the structural performance of urban underground pipelines according to claim 2, characterized in that, The construction of the dynamic model for monitoring pipeline deflection in step S1, to establish a quantitative correspondence between pipeline deflection data and pipeline structural performance, includes: Step S11: Construct a two-dimensional simplified axisymmetric analysis model of the pipe-soil system and establish a cylindrical coordinate system for the pipe-soil system; Step S12: Based on the theory of elastic dynamics, the governing equations of motion for the pipe-soil system in cylindrical coordinates are as follows: ； in, For stress components, For radial and vertical displacements, The density of the material; Step S13: Determine the constitutive relationship between the pipe body material and the surrounding soil material, wherein: The pipeline body and the surrounding soil are both isotropic materials, and their constitutive relations are as follows: ； ； in, Let Lamé constant be . With elastic modulus Compared to Poisson Related; The pipeline body is made of isotropic material, and the surrounding soil layer is transversely isotropic material. Their constitutive relations are as follows: ； Among them, elastic constant From engineering parameters Sure. Step S14: Solve the motion control equations using the efficient frequency domain-wavenumber domain spectral analysis method to obtain the pipe deflection data for each item; Step S15: Based on the obtained pipe deflection data, establish the correspondence between the pipe deflection data and the pipe structural performance.

4. The dual-layer intelligent diagnostic method for the structural performance of urban underground pipelines according to claim 3, characterized in that, In step S13, when both the pipe body and the surrounding soil are isotropic materials, the relationship between axisymmetric strain and displacement in the axisymmetric analysis model is as follows: ； When the pipeline body is an isotropic material and the surrounding soil layer is a transversely isotropic material, when When the constitutive relation of the material is mentioned, it is an isotropic material relation.

5. The dual-layer intelligent diagnostic method for the structural performance of urban underground pipelines according to claim 4, characterized in that, Step S14 involves solving the motion control equations using an efficient frequency-wavenumber spectral analysis method to obtain various pipe deflection data, including: Step S141: Perform a Fourier-Bessel transform on the motion control equations; Step S142: Apply boundary and interface conditions; the boundary and interface conditions include at least: impact load boundary conditions, pipe-layer interface continuity or contact conditions, and outer boundary adopting absorbing boundary or infinite domain equivalent conditions. Step S143: In the transformation domain, use the transfer matrix method or stiffness matrix method to accurately solve for the dynamic compliance matrix of the layered system. ; Step S144: Reconstruct the displacement response in the physical spacetime domain through double summation; Step S145: Obtain the deflection time history curve and extract the basic feature vector reflecting the pipe structure performance from the obtained deflection time history curve. It must include at least: peak deflection Peak arrival time Bending energy The main frequency / band energy ratio and the half-peak width of the deflection basin are the pipeline deflection data mentioned in each item.

6. The dual-layer intelligent diagnostic method for the structural performance of urban underground pipelines according to claim 5, characterized in that, In step S2, the method for selecting data with strong correlation to the pipeline structure performance to form a multi-source heterogeneous dataset is to use grey relational entropy analysis to quantify the correlation between the field monitoring data, the full-scale test data, and the numerical simulation data and the pipeline structure performance. The calculation formula is as follows: in, This represents the grey relational coefficient distribution mapping of the nth indicator to be screened, where r is the number of indicators to be screened. Let j be the sequence of the j-th indicator to be screened over time; Based on the actual needs of the project, a preset gray entropy correlation threshold is set, and index data with correlation higher than the threshold are retained. The filtered data set forms the multi-source heterogeneous dataset.

7. The dual-layer intelligent diagnostic method for the structural performance of urban underground pipelines according to claim 6, characterized in that, In step S3, the process of fusing and expanding the data from the obtained multi-source heterogeneous dataset to obtain a diagnostic dataset includes: Step S31: Denoising and Completion Based on Enhanced Denoising Autoencoder: Train the enhanced denoising autoencoder (DAE), add noise to the input sample, and reconstruct it under the constraint of mask M to obtain the denoised and completed sample. ; Step S32: Sample generation augmentation based on generative adversarial networks: Train the generative adversarial network (GAN) to make the generator output The discriminator is used to distinguish between real and generated samples, and the generator is used to output high-quality simulated deflection data. Step S33: Inject physical constraints and remove invalid samples: Add a physical penalty term to the training loss. Remove generated samples that do not meet the constraints; Step S34: Quality Control and Sample Set Union: Perform statistical consistency tests and anomaly detection on the generated samples to obtain the final generated sample set. and compared with the denoised and completed sample set obtained in step S31 The data are then merged to form the diagnostic dataset.

8. The dual-layer intelligent diagnostic method for the structural performance of urban underground pipelines according to claim 7, characterized in that, In step S4, the coarse-grained diagnosis of the data in the obtained diagnostic dataset to obtain the underperforming pipeline includes: Step S401: Input the diagnostic dataset as the sample to be judged; Step S402: Construct a multi-objective random forest model, taking the performance status of the pipeline structure as the classification objective, optimize the classification accuracy and feature importance ranking, and mine the deep core features of the samples; Step S403: Calculate the multi-scale discrete entropy of data with different sampling periods, and select the period T with the smallest entropy value as the model update period; Step S404: Principal component analysis is used to reduce the dimensionality of the deep features extracted by the multi-objective random forest model, retaining more than 95% of the original feature information; Step S405: Input the dimensionality-reduced features into the support vector machine model, use the radial basis function as the kernel function, optimize the penalty factor and kernel parameters, and perform binary classification of pipeline "healthy state / insufficient performance state"; Step S406: Output coarse-grained diagnostic results. Samples that are determined to be in a state of insufficient performance will proceed to the fine-grained diagnostic process.

9. The dual-layer intelligent diagnostic method for the structural performance of urban underground pipelines according to claim 8, characterized in that, In step S4, the underperforming pipe undergoes fine-grained diagnosis to obtain the diagnosis results, including: Step S407: Input the core feature vector of the pipeline sample that is determined to be underperforming by coarse-grained diagnosis, add multiple scene features, and form a composite feature set; Step S408: The motion control equations, complex control equations of material constitutive relations, and correspondence rules between deflection characteristics and disease types of the dynamic model for monitoring pipeline dynamic deflection are embedded into the long short-term memory neural network loss function to construct a knowledge constraint layer. Step S409: Add a multi-scale convolutional block structure to the input layer of the long short-term memory neural network model to extract the temporal features of the composite feature set in layers and learn the temporal correlation rules between features; Step S410: Train the long short-term memory neural network model using pipeline defect type, severity, and key structural parameters as supervisory labels. During training, a physical penalty term is added to eliminate output results that violate mechanical laws, and the model weights are iteratively optimized. Step S411: Input the composite feature set of the sample to be diagnosed, and output the specific disease type, severity level, and inverted structural performance parameters of the pipeline.

10. A dual-layer intelligent diagnostic system for urban underground pipeline structural performance, operating the intelligent diagnostic method described in any one of claims 1-9, comprising: The system includes a pipeline deflection characteristic monitoring subsystem, an urban underground pipeline integrated management database, a multi-source heterogeneous data fusion and dataset expansion subsystem, and a two-layer intelligent diagnostic subsystem. The pipeline deflection characteristic index monitoring subsystem includes a pipeline deflection measuring device, an automated drone, an IoT-based radar detection vehicle, a video surveillance robot, and a supporting underground pipeline network structure performance testing data acquisition workstation, used to collect multi-source heterogeneous raw data. The urban underground pipeline integrated management database is used to provide data access control and data storage and retrieval call interfaces for the deep learning development board to provide a sample set of pipeline deflection characteristic indicators, and to store the entire process data of monitoring data, test data, and diagnostic results. The multi-source heterogeneous data fusion and dataset augmentation subsystem includes an enhanced denoising autoencoder module, a convolution / transposed convolution module, a generator module, a discriminator module, and a loss function module, which are used to realize data denoising and completion, generation and augmentation, and physical constraint verification. The dual-layer intelligent diagnostic subsystem includes a pipeline dynamic deflection detection power module, a coarse-grained diagnostic module based on multi-objective random forest, and a fine-grained diagnostic module driven by knowledge and data collaboration, which are used to execute dual-layer diagnostic logic.

Images