An underground engineering concealed structure defect intelligent detection method and system

Abstract

The application relates to the technical field of structural health detection, and discloses an underground engineering concealed structure defect intelligent detection method and system, which comprises the following steps: collecting electromagnetic reflection waves, elastic stress waves and apparent resistivity data of an underground concealed structure and recording the spatial pose coordinates and time stamps of each data; adopting a dynamic time warping algorithm for time-space registration to construct a multi-modal data tensor; inputting the tensor into a physical information neural network for defect feature inversion, utilizing a loss function containing a Maxwell equation and a wave equation residual loss term to output defect parameters containing position, size and medium physical properties; converting the parameters into a three-dimensional point cloud and mapping the three-dimensional point cloud to a building information model to construct a digital twin, and simultaneously extracting the three-dimensional point cloud to a finite element analysis solver as a boundary weakening and material degradation condition to calculate a bearing capacity evaluation value and generate a safety warning signal. The application realizes high-precision inversion of defects and quantitative evaluation of structural mechanics.

Description

Technical Field

[0001] This invention relates to the field of structural health detection technology, and in particular to an intelligent detection method and system for hidden structural defects in underground engineering. Background Technology

[0002] With the acceleration of urbanization, the construction scale of underground engineering projects (such as tunnels and underground utility tunnels) is constantly expanding. During long-term service, concealed structures are susceptible to defects such as voids, looseness, and water seepage due to geological conditions and external loads, which seriously threaten the safe operation of the project.

[0003] Currently, the detection of hidden structural defects in underground engineering mainly relies on single detection technologies (such as ground-penetrating radar). However, single-physical-field detection is easily interfered with in complex geological environments, exhibiting a "different objects, same spectrum" phenomenon, which can easily lead to misjudgments and missed detections. Meanwhile, existing data processing methods largely depend on purely data-driven deep learning models, lacking underlying constraints based on physical laws, resulting in insufficient inversion accuracy and generalization ability. Furthermore, existing detection results mostly remain at the surface identification of defect location and geometric dimensions, failing to integrate the medium's physical properties of microscopic defects with the mechanical bearing capacity of the macroscopic structure. This prevents the achievement of a technological closed loop from "detecting defects" to "quantitatively assessing structural safety," making it difficult to meet the increasingly stringent requirements for intelligent health monitoring of underground structures. Summary of the Invention

[0004] This invention provides an intelligent detection method and system for hidden structural defects in underground engineering. It integrates multi-physics spatiotemporal aligned perception, physical information neural network inversion based on Maxwell's equation and wave equation constraints, and finite element mechanical calculation driven by digital twins to achieve high-precision identification of hidden structural defects in underground engineering and quantitative safety early warning of overall bearing capacity.

[0005] This invention provides an intelligent detection method for hidden structural defects in underground engineering, comprising:

[0006] S1. Collect electromagnetic reflection wave data, elastic stress wave data and apparent resistivity data of the concealed structure of underground engineering, and record the spatial pose coordinates and timestamps when collecting each data.

[0007] S2. Based on the spatial pose coordinates and timestamps of the various data, the electromagnetic reflection wave data, elastic stress wave data and apparent resistivity data are spatiotemporally registered using a dynamic time warping algorithm to construct a multimodal data tensor.

[0008] S3. Input the multimodal data tensor into a preset physical information neural network to perform defect feature inversion and output the defect parameters of the underground engineering concealed structure. The defect parameters include defect location coordinates, geometric dimensions and medium physical properties. The loss function of the preset physical information neural network includes a partial differential equation residual loss term constructed from Maxwell's equations describing the propagation law of electromagnetic waves and the wave equation describing the propagation law of elastic waves.

[0009] S4. The defect location coordinates, geometric dimensions, and medium physical properties are converted into three-dimensional point cloud data, and the three-dimensional point cloud data is mapped to a preset building information model of the underground project to construct a digital twin of the underground structure containing defect attributes; wherein, the defect attributes are composed of the defect location coordinates, geometric dimensions, and medium physical properties;

[0010] S5. Extract the defect location coordinates, geometric dimensions, and medium physical properties in the digital twin of the underground structure as boundary weakening and material deterioration conditions into the finite element analysis solver, calculate the bearing capacity assessment value of the hidden structure of the underground project at the defect location coordinates, and compare it with the preset safety threshold to generate a structural safety early warning signal.

[0011] Furthermore, S1 specifically includes:

[0012] S101. Deploy a multi-source sensor array and a combined positioning and timing system on the mounting platform, and use a hardware-level clock synchronization mechanism as the master control clock to synchronously send trigger commands to the multi-source sensor array; wherein, the multi-source sensor array includes a ground-penetrating radar antenna array, an acoustic sensor, and a high-density electrical resistivity tomography electrode array.

[0013] S102. Under the joint control of the trigger command, the multi-source sensor array is used to perform physical field excitation and reception on the concealed structure of the underground project, and electromagnetic reflection wave data, elastic stress wave data and apparent resistivity data are acquired respectively; wherein, the ground penetrating radar antenna array acquires the electromagnetic reflection wave data, the acoustic sensor acquires the elastic stress wave data, and the high-density electrical resistivity meter electrode array acquires the apparent resistivity data.

[0014] S103. The combined positioning and timing system is used to calculate the spatial pose coordinates of the platform in real time, and the spatial pose coordinates and master control timestamp acquired in the same trigger cycle are used as data tag headers and bound to the corresponding electromagnetic reflection wave data, elastic stress wave data and apparent resistivity data to output a single acquisition data frame.

[0015] Furthermore, S2 specifically includes:

[0016] S201. Based on the spatial pose coordinates, construct a three-dimensional spatial grid with a unified resolution. According to the pre-calibrated extrinsic parameter matrix, transform the acquisition origin of each data into the three-dimensional spatial grid. Use the Kriging spatial interpolation algorithm or the cubic spline interpolation algorithm to resample and project the discrete electromagnetic reflection wave data, elastic stress wave data and apparent resistivity data onto the same spatial grid nodes to complete the lateral spatial coordinate alignment.

[0017] S202. Using the spatially aligned electromagnetic reflection wave data as a reference sequence, the cumulative cost matrix of the elastic stress wave data and apparent resistivity data with the reference sequence is calculated using a dynamic time warping algorithm, so as to perform nonlinear time stretching and compression on the elastic stress wave data and apparent resistivity data to complete longitudinal waveform feature registration.

[0018] S203. The spatiotemporally registered electromagnetic reflection wave data, elastic stress wave data, and apparent resistivity data are respectively subjected to dimensional normalization processing and stacked according to the channel dimension to construct the multimodal data tensor.

[0019] Furthermore, S202 specifically includes:

[0020] The time series of the electromagnetic reflection wave data is extracted as the reference sequence, and the time series of the elastic stress wave data is extracted as the sequence to be normalized.

[0021] Calculate the Euclidean distance between corresponding data points of the reference sequence and the sequence to be normalized to construct a distance matrix;

[0022] Based on the distance matrix, the cumulative cost matrix is ​​calculated using the state transition equation, and the regularized path with the minimum total cumulative distance is found by backtracking through a dynamic programming algorithm.

[0023] Based on the regularization path, the sequence to be regularized is locally extended or compressed nonlinearly so that its peak and trough features are aligned with the reference sequence on the equivalent physical interface; and the same dynamic time regularization algorithm is used to regularize the one-dimensional depth profile data of the apparent resistivity data to the reference sequence.

[0024] Furthermore, S3 specifically includes:

[0025] S301. Construct a physical information neural network that includes a data input layer, multiple hidden layers and a physical parameter output layer. Input the multimodal data tensor into the physical information neural network to extract physical coupling features and output the predicted proxy physical field value and the defect parameters to be inverted.

[0026] S302. Construct a total loss function based on the predicted proxy physical field values, the actual measured physical field values, and the partial differential equations of physical laws; wherein, the total loss function is a weighted combination of the data fitting loss term and the residual loss term of the partial differential equations;

[0027] S303. An optimization algorithm is used to perform backpropagation updates of network weights to minimize the total loss function. When the total loss function converges, the defect parameters are extracted from the physical parameter output layer, and the defect parameters are logically compared and verified with a preset physical constant threshold range. After passing the logical comparison and verification, the parameters are stored in the database.

[0028] Furthermore, in S302, the total loss function The calculation formula is:

[0029]

[0030] in, and For adaptive dynamic weighting coefficients, For data fitting loss term, This is the residual loss term in the partial differential equation;

[0031] The data fitting loss term The calculation formula is:

[0032]

[0033] The residual loss term of the partial differential equation Electromagnetic wave residual term and elastic wave residuals Composition; wherein, the electromagnetic wave residual term The calculation formula is:

[0034]

[0035] The elastic wave residual term The calculation formula is:

[0036]

[0037] In the above formula, This represents the total number of registration data points actually sampled by the sensor. and These represent the electric field intensity predicted by the network and the actual measured electric field intensity at the i-th data point, respectively. and These are the elastic displacement predicted by the network and the actual measured elastic displacement at the i-th data point, respectively. The number of points to be randomly selected; The Laplace operator is used; t is the time variable. This represents the second-order partial derivative with respect to time; The permeability of free space, It is the vacuum permittivity; Let be the relative permittivity of the network output at the j-th collocation point; Let the density of the medium output by the network be the density at the j-th collocation point; Let be the elastic modulus of the medium output by the network at the j-th collocation point.

[0038] Furthermore, S4 specifically includes:

[0039] S401. Based on the defect location coordinates and the geometric dimensions, reconstruct the three-dimensional geometric bounding box and perform meshing discretization to generate a basic spatial point cloud, and bind the medium physical properties as a physical scalar field to each point of the basic spatial point cloud to generate the three-dimensional point cloud data.

[0040] S402. Obtain the building information model of the pre-built underground project, calculate the coordinate transformation matrix between the coordinate system of the sensing device and the reference coordinate system of the building information model, and use the coordinate transformation matrix to perform a spatial affine transformation on the three-dimensional point cloud data to obtain the target point cloud data registered in the reference coordinate system.

[0041] S403. Create a custom attribute extension set corresponding to the target point cloud data in the building information model, write the target point cloud data and the defect attribute into the custom attribute extension set, and establish a spatial topological dependency relationship with the original hidden structural components through spatial intersection determination, thereby constructing the digital twin of the underground structure containing the defect attribute.

[0042] Furthermore, S5 specifically includes:

[0043] S501. Import the digital twin of the underground structure containing defect attributes into the preset finite element analysis solver, and perform three-dimensional solid mesh generation on the hidden structure based on the defect location coordinates and geometric dimensions.

[0044] S502. The physical properties of the medium are converted into material degradation constitutive parameters of the finite element element, and the initial stiffness matrix of the finite element element in the defect region is degraded and reconstructed based on the material degradation constitutive parameters to obtain the damage stiffness matrix.

[0045] S503. Apply the actual boundary load under the operating state of the underground project to the outer boundary of the finite element model, call the finite element analysis solver to solve the displacement field and stress field of the overall structure, and calculate the bearing capacity assessment value of the hidden structure at the coordinates of the defect location.

[0046] S504. The load-bearing capacity assessment value is compared with the preset safety threshold. If the load-bearing capacity assessment value does not meet the preset safety threshold, a structural safety early warning signal containing the early warning level and defect location is generated.

[0047] Furthermore, S502 specifically includes:

[0048] Define a damage reduction factor generated by the dynamic mapping of the physical properties of the medium. Based on the damage reduction factor, reconstruct the initial stiffness matrix of the finite element elements within the defect region using the following formula:

[0049]

[0050] in, Here is the damage stiffness matrix. This is the damage reduction factor. This is the initial stiffness matrix.

[0051] This invention also provides an intelligent detection system for hidden structural defects in underground engineering projects. Based on the intelligent detection method for hidden structural defects in underground engineering projects described above, the system includes:

[0052] The data acquisition module is used to collect electromagnetic reflection wave data, elastic stress wave data and apparent resistivity data of the concealed structure of underground engineering, and to record the spatial pose coordinates and timestamps when collecting each data item.

[0053] The spatiotemporal registration module is used to perform spatiotemporal registration of the electromagnetic reflection wave data, elastic stress wave data and apparent resistivity data based on the spatial pose coordinates and timestamps of the various data, using a dynamic time warping algorithm, in order to construct a multimodal data tensor.

[0054] The feature inversion module is used to input the multimodal data tensor into a preset physical information neural network to perform defect feature inversion and output the defect parameters of the underground engineering concealed structure. The defect parameters include defect location coordinates, geometric dimensions and medium physical properties. The loss function of the preset physical information neural network includes a partial differential equation residual loss term constructed from Maxwell's equations describing the propagation of electromagnetic waves and the wave equation describing the propagation of elastic waves.

[0055] A digital twin construction module is used to convert the defect location coordinates, geometric dimensions, and medium physical properties into three-dimensional point cloud data, and map the three-dimensional point cloud data into a preset building information model of the underground engineering to construct a digital twin of the underground structure containing defect attributes; wherein, the defect attributes are composed of the defect location coordinates, geometric dimensions, and medium physical properties;

[0056] The assessment and early warning module is used to extract the location coordinates, geometric dimensions, and physical properties of the medium of the defect in the digital twin of the underground structure as boundary weakening and material deterioration conditions and input them into the finite element analysis solver. It calculates the bearing capacity assessment value of the hidden structure of the underground project at the defect location coordinates and compares it with a preset safety threshold to generate a structural safety early warning signal.

[0057] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0058] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0059] The beneficial effects of this invention are as follows:

[0060] This invention overcomes the limitations of single sensors being susceptible to interference by synchronously acquiring multi-physics field detection data and using a dynamic time warping algorithm for spatiotemporal registration to construct a multimodal data tensor, significantly improving the comprehensiveness of heterogeneous data perception in complex media. It introduces a physical information neural network, constructing Maxwell's equations describing electromagnetic wave propagation and the wave equations describing elastic wave propagation as partial differential equation residual loss terms, which are then integrated into the model inversion, improving the accuracy and robustness of the inversion of defect physical properties under complex geological conditions. The inverted 3D point cloud is mapped into a building information model to construct a digital twin, and physical parameters are transformed into boundary weakening and material degradation conditions, extracted into the finite element analysis solver, achieving accurate calculation and intelligent early warning of the local bearing capacity of concealed structures. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of a method flow according to an embodiment of the present invention.

[0062] Figure 2 This is a schematic diagram of the device structure according to an embodiment of the present invention.

[0063] Figure 3 This is a schematic diagram of the internal structure of a computer device according to an embodiment of the present invention.

[0064] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0065] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0066] like Figure 1As shown, this invention provides an intelligent detection method for hidden structural defects in underground engineering, comprising:

[0067] S1. Collect electromagnetic reflection wave data, elastic stress wave data and apparent resistivity data of the concealed structure of underground engineering, and record the spatial pose coordinates and timestamps when collecting each data.

[0068] In a preferred embodiment of the present invention, step S1 specifically includes the following sub-steps:

[0069] S101. Deploy a multi-source heterogeneous sensor array and a spatiotemporal positioning and timing system, and perform joint triggering acquisition based on a hardware-level clock synchronization mechanism.

[0070] Specifically, a multi-source detection hardware system is constructed based on an intelligent inspection robot or a mounting platform. The sensor array includes: a ground-penetrating radar (GPR) antenna array for acquiring dielectric constant differences; a piezoelectric ceramic acoustic sensor or distributed acoustic sensing (DAS) fiber optic probe for acquiring acoustic impedance differences; and a capacitively coupled high-density electrical resistivity tomography (ERT) electrode array for acquiring moisture content characteristics. Simultaneously, a combined positioning and timing system comprising a lidar (LiDAR), an inertial measurement unit (IMU), and a total station tracking prism is rigidly mounted on the mounting platform.

[0071] Those skilled in the art should understand that rigidly fixing each sensor is to establish a unified robot body coordinate system, thereby eliminating calibration extrinsic parameter errors between multiple sensors.

[0072] To eliminate time differences in data acquisition, a hardware-level clock synchronization protocol (such as the IEEE 1588 PTP precise time protocol or the BeiDou / GPS PPS signal) is used as the master clock. The main control computer synchronously sends trigger commands to the ground penetrating radar, acoustic sensors, and high-density electrical resistivity tomography (EDT) instrument. The timestamp resolution reaches the microsecond (μs) level, ensuring that different physical field sensors simultaneously start sampling at the same physical plane while the platform is moving.

[0073] S102. Acquiring Multiphysics Field Detection Data of Hidden Structures: Under the joint triggering mechanism, each sensor array synchronously excites and receives physical fields from hidden structures in underground engineering (such as tunnel concrete lining and the surrounding rock behind it). The specific process is as follows:

[0074] Collecting electromagnetic reflection wave data: The ground-penetrating radar emits high-frequency nanosecond-level electromagnetic pulse waves towards the concealed structure, receives and records the electromagnetic reflection waves generated when encountering interfaces with abrupt changes in dielectric constant, such as voids or reinforcing bars, forming the first time-series signal sequence.

[0075] Acquiring elastic stress wave data: A mechanical exciter or ultrasonic transmitter probe is used to apply transient excitation to the wall surface. An acoustic sensor array receives the reflected and refracted elastic stress waves propagating inside the concrete and surrounding rock, forming a second time-series signal sequence that is sensitive to the mechanical strength of the medium.

[0076] Acquiring apparent resistivity data: The high-density electrical resistivity meter injects low-frequency alternating current into the structure according to the preset electrode spacing and measures the potential difference between the receiving electrodes. Based on the electrode arrangement coefficient, it calculates the apparent resistivity sequence, which is extremely sensitive to moisture, and forms the third time-series signal sequence.

[0077] S103, Spatial pose calculation of data and binding of multidimensional spatiotemporal labels.

[0078] While collecting the above three types of data, the integrated positioning and timing system, based on the Simultaneous Localization and Mapping (SLAM) algorithm, calculates the spatial pose coordinates of the platform in the global engineering coordinate system in real time. The spatial pose coordinates have six degrees of freedom (6-DOF), represented as P=(x,y,z,α,β,γ), where x,y,z are three-dimensional translation coordinates, and α,β,γ are roll, pitch, and yaw angles.

[0079] The spatial pose coordinates P and master control timestamp T acquired within the same trigger cycle are used as data headers and bound to the corresponding electromagnetic reflection wave data, elastic stress wave data, and apparent resistivity data, respectively. The structure of the output single-acquisition data frame can be represented as follows:

[0080]

[0081] in, Represents electromagnetic reflected wave data. Represents elastic stress wave data, This represents apparent resistivity data. Through strong binding of multidimensional spatiotemporal labels, an alignment benchmark is provided for dynamic time warping (DTW) of data at different sampling frequencies.

[0082] S2. Based on the spatial pose coordinates and timestamps of the various data, the electromagnetic reflection wave data, elastic stress wave data and apparent resistivity data are spatiotemporally registered using a dynamic time warping algorithm to construct a multimodal data tensor.

[0083] In a preferred embodiment of the present invention, step S2 specifically includes the following sub-steps:

[0084] S201. Global spatial meshing interpolation alignment based on spatial pose coordinates and timestamps.

[0085] First, extract the master control timestamp and spatial pose coordinates (3D translation and attitude angles) from a single data acquisition frame. Then, construct a uniform resolution 3D spatial mesh (VoxelGrid) within the pre-defined global physical coordinate system of the underground engineering project.

[0086] Due to the different physical dimensions and installation positions of various sensors, the origin of each data acquisition point is transformed into a unified three-dimensional spatial grid based on the pre-calibrated extrinsic parameter matrix of the rigid structure of the mounting platform. To address the issue of inconsistent spatial sampling intervals among different sensors, Kriging spatial interpolation or cubic spline interpolation is employed to resample and project the discrete electromagnetic reflection wave data, elastic stress wave data, and apparent resistivity data onto the same spatial grid nodes, eliminating resolution differences in spatial dimensions and achieving alignment of lateral spatial coordinates.

[0087] S202. Vertical depth / temporal feature registration based on dynamic time warping (DTW) algorithm.

[0088] After completing the lateral spatial alignment, to address the issue of misalignment in the time domain (or equivalent depth domain) of reflected echoes from the same defect due to different wave velocities, a Dynamic Time Warping (DTW) algorithm is employed to perform nonlinear stretching and compression alignment of the waveform sequence. The specific process is as follows:

[0089] Time series of electromagnetic reflected waves Using the time series of elastic stress waves as the reference sequence The sequence to be normalized;

[0090] Calculate the Euclidean distance between corresponding data points in sequences X and Y, and construct a distance matrix D of size N×M, where the matrix elements are... Based on the distance matrix D, the cumulative cost matrix C is defined, and its state transition equation is:

[0091]

[0092] By using dynamic programming, we backtrack from C(N,M) to C(1,1) to find a regular path W (Warping Path) that minimizes the total cumulative distance.

[0093] Based on the normalized path W, the time series of elastic stress waves is locally nonlinearly extended or compressed to ensure that its peak / trough characteristics are strictly aligned with the waveform of the electromagnetic reflected wave at the equivalent physical interface (such as a cavity boundary). Subsequently, the same DTW algorithm is used to normalize the one-dimensional depth profile data of apparent resistivity to the reference sequence.

[0094] S203, Channel stacking and physical dimension normalization of multimodal data tensors.

[0095] After spatiotemporal registration of S201 and S202, the three types of detection data achieved strict alignment in both horizontal spatial nodes and vertical waveform characteristics. To eliminate the significant differences in the dimensions of different physical fields (e.g., resistivity amplitude may reach thousands of ohm-meters, while electromagnetic reflection wave amplitude is in the microvolt range), Z-score standardization was performed on the three types of data to make them conform to a standard normal distribution with a mean of 0 and a variance of 1.

[0096] Finally, using the horizontal spatial span as the width, the regularized vertical sampling points as the height, and the three physical field data types as the channels, a concatenation of channels is performed to construct a multimodal data tensor with dimensions H×W×C (where the number of channels C=3). This tensor is similar to a three-channel RGB color image, but each pixel contains electromagnetic, acoustic, and electrical features perfectly aligned to that spatial location, serving as the standard input format for physical information neural networks.

[0097] S3. Input the multimodal data tensor into a preset physical information neural network to perform defect feature inversion and output the defect parameters of the underground engineering concealed structure. The defect parameters include defect location coordinates, geometric dimensions and medium physical properties. The loss function of the preset physical information neural network includes a partial differential equation residual loss term constructed from Maxwell's equations describing the propagation law of electromagnetic waves and the wave equation describing the propagation law of elastic waves.

[0098] In a preferred embodiment of the present invention, step S3 specifically includes the following sub-steps:

[0099] S301. Construct a pre-defined multi-branch physical information neural network model.

[0100] A physical information neural network is constructed, comprising a data input layer, multiple hidden layers, and a physical parameter output layer. The multimodal data tensor output in step S2 is input into this network. Deep convolution operations in the hidden layers are used to extract deep physical coupling features of electromagnetic reflected waves, elastic stress waves, and apparent resistivity at the same spatiotemporal node. The input variable of the network is a given spatiotemporal coordinate point (x, z, t), and the output variable comprises two parts: the first part is the surrogate physical field prediction value (including the predicted electric field intensity at the spatial point). and predicting elastic displacement The second part consists of the parameters of the hidden structural defects to be inverted, specifically including the coordinates of the defect locations. Geometric dimensional parameters (such as equivalent radius r or length and width limits) and physical properties of the medium (including relative permittivity) and medium density ).

[0101] S302. Constructing the total loss function based on multi-physics coupling.

[0102] To ensure that the output of the physical information neural network strictly follows the laws of electromagnetism and acoustic / dynamic physics, a total loss function is defined. It consists of the data fitting loss term. and the residual loss term of partial differential equations Composed of weighted combinations:

[0103]

[0104] in, and These are adaptive dynamic weighting coefficients.

[0105] Specifically, the data fitting loss term The formula used to measure the mean square error between the network-predicted field values ​​and the actual sensor measurements is:

[0106]

[0107] The residual loss term of the partial differential equation Further decomposed into electromagnetic wave residual terms and elastic wave residuals The electromagnetic wave residual term is constructed from the time-domain wave equation simplified from Maxwell's equations in a two-dimensional passive space, and its formula is:

[0108]

[0109] The elastic wave residual term is constructed from the wave equation describing the propagation law of elastic stress waves, and its formula is:

[0110]

[0111] In the above formula, This represents the total number of registration data points actually sampled by the sensor. and These represent the electric field intensity predicted by the network and the electric field intensity actually measured by the ground penetrating radar at the i-th data point, respectively. and These represent the elastic displacement predicted by the network and the elastic displacement actually measured by the acoustic sensor at the i-th data point, respectively. The number of collocation points randomly sampled within the internal computational domain and its boundaries; The Laplace operator is used; the network calculates the second derivatives of the electric field strength and displacement with respect to spatial coordinates using automatic differentiation techniques; t is the time variable. This represents the second-order partial derivative with respect to time; The permeability of free space, is the vacuum permittivity, and both are known physical constants; Let the relative permittivity of the medium output by the network be the value at the j-th collocation point. Let the density of the medium output by the network be the density at the j-th collocation point; Let J be the elastic modulus of the medium output by the network at the j-th collocation point (which can be expressed as bulk modulus or Young's modulus for different geological bodies).

[0112] S303. Perform closed-loop iterative optimization to output defect parameters and perform physical logic verification on the defect parameters.

[0113] A second-order optimization algorithm is used to perform backpropagation updates of the network weights in order to minimize the total loss function. During the iteration process, the network calculates the second derivatives of the electric field strength and displacement with respect to spatial coordinates using automatic differentiation techniques, and corrects the descent direction of the network gradient using the residual loss term of the partial differential equation. The iteration stops when the total loss function converges to below a preset error threshold. At this point, the converged defect location coordinates, geometric dimensions, and physical properties of the medium are extracted from the physical parameter output layer.

[0114] The extracted defect parameters are logically compared and verified with a preset physical constant threshold range. Specifically, the dielectric constant output is verified. Whether it falls within a reasonable range of known underground media (such as air, groundwater, or concrete). If the defect parameters are verified through physical logic, the inversion result is deemed valid, and the defect parameters are stored in the database.

[0115] S4. The defect location coordinates, geometric dimensions, and medium physical properties are converted into three-dimensional point cloud data, and the three-dimensional point cloud data is mapped to a preset building information model of the underground project to construct a digital twin of the underground structure containing defect attributes; wherein, the defect attributes are composed of the defect location coordinates, geometric dimensions, and medium physical properties.

[0116] In a preferred embodiment of the present invention, step S4 specifically includes the following sub-steps:

[0117] S401. Generation of 3D point cloud data of defects based on multi-dimensional feature binding.

[0118] First, extract the defect location coordinates (e.g., spatial center point position) and geometric dimensions (e.g., length, width, height, or equivalent radius) output in step S3, and reconstruct the three-dimensional geometric bounding box of the defect in local space. Then, according to a preset spatial resolution, perform meshing discretization on the internal space of the three-dimensional geometric bounding box to generate a basic spatial point cloud composed of countless discrete points. Finally, use the medium physical properties (including relative permittivity, medium density, and elastic modulus, etc.) output in step S3 as a physical scalar field and bind them one by one to each point of the basic spatial point cloud. In the multi-dimensional three-dimensional point cloud data generated in this way, any discrete point... The data structure can be represented as a multidimensional tuple:

[0119]

[0120] in, Let these be the spatial coordinates of the point. Let be the relative permittivity at that point. For density, This is the elastic modulus. This step completes the 3D instantiation from macroscopic parameters to a microscopic physical point cloud.

[0121] S402. Affine transformation registration between the sensor coordinate system and the BIM reference coordinate system.

[0122] The Building Information Model (BIM) of the pre-built underground project is acquired. Since the 3D point cloud data is generated based on the global engineering coordinate system of the sensing device, while the pre-built BIM typically uses its own design reference coordinate system, there is a spatial offset between the two. The preset spatial calibration relationship between the positioning origin of the platform and the BIM design reference origin is extracted, and the coordinate transformation matrix is ​​calculated. This coordinate transformation matrix includes a 3D rotation matrix R and a translation vector T.

[0123] Using the coordinate transformation matrix, for each discrete point in the 3D point cloud data Perform affine transformation in three-dimensional space:

[0124]

[0125] After coordinate transformation, the target point cloud data that is absolutely registered in the BIM reference coordinate system is obtained. This eliminates the spatial position error between virtual and real spaces.

[0126] S403, bottom-level topology binding and construction of digital twins of underground structures.

[0127] After spatial registration, the point cloud is not simply rendered visually in BIM software; instead, semantic binding of the underlying data is performed. Based on the Industry Foundation Classes (IFC) data exchange standard, a custom Property Set (Pset) is created for the defect in the BIM database. The target point cloud data, along with its carried defect location coordinates, geometric dimensions, and physical properties, are fully encapsulated and written into the custom Property Set. Simultaneously, a spatial intersection determination algorithm is used to establish the spatial topological dependency between the defect Property Set and the intruded original concealed structural components (such as tunnel segments and secondary lining concrete) in the BIM.

[0128] After the aforementioned underlying data writing and topological relationship reconstruction, a digital twin of the underground structure containing defect attributes is generated, which not only has a three-dimensional visual appearance but also has real defect physical parameters attached to the internal component attribute nodes, providing geometric and material input conditions for mechanical finite element mesh generation.

[0129] S5. Extract the defect location coordinates, geometric dimensions, and medium physical properties in the digital twin of the underground structure as boundary weakening and material deterioration conditions into the finite element analysis solver, calculate the bearing capacity assessment value of the hidden structure of the underground project at the defect location coordinates, and compare it with the preset safety threshold to generate a structural safety early warning signal.

[0130] In a preferred embodiment of the present invention, step S5 specifically includes the following sub-steps:

[0131] S501, Mesh Discretization and Adaptive Encryption of Multiphysics Digital Twins.

[0132] The digital twin of the underground structure containing defect attributes, constructed in step S4, is imported into a preset finite element analysis solver via an application programming interface (API). The solver first extracts the defect location coordinates and geometric dimensions from the twin and performs three-dimensional solid mesh generation on the hidden structure. To balance computational efficiency and accuracy, an adaptive mesh generation technique is adopted, i.e., a coarse mesh is generated in the normal structural region, while a high-density, finer mesh element is generated in the stress concentration sensitive region extending outward from the defect location coordinates (e.g., a region three times the equivalent radius).

[0133] S502. Material degradation and stiffness reconstruction based on the physical properties of the medium.

[0134] Extract the physical properties of the medium (such as relative permittivity) bound in the digital twin. Medium density And the elastic modulus M), which is converted into the material degradation constitutive parameters of the finite element element.

[0135] Specifically, the damage reduction factor is defined. ( This coefficient is dynamically generated by mapping the physical properties of the medium. For example, when determining the density of the medium in a certain region... Extremely low and relative permittivity When considering (air void characteristics), the reduction factor of the unit in this region. When a region is identified as a water-rich, loosely compacted area, an empirical formula is applied. The reduction factor. Then, the initial stiffness matrix of the finite element elements within the defect region is calculated using the following formula. Perform degradation reconstruction to obtain the damage stiffness matrix. :

[0136]

[0137] This step allows the originally abstract physical detection parameters to substantially participate in the underlying calculation of the structural mechanical response, thus constituting the conditions for material degradation.

[0138] S503, Applying engineering boundary loads and calculating bearing capacity assessment values.

[0139] Apply real boundary loads under the operating conditions of underground engineering to the outer boundary of the finite element model. The boundary loads include, but are not limited to, the water and soil pressure of the surrounding rock, the self-weight of the structure, and the dynamic loads on the ground surface.

[0140] The finite element analysis solver (such as an implicit statics algorithm) is invoked to solve for the displacement and stress fields of the overall structure. Based on this, the load-bearing capacity assessment value of the hidden structure at the coordinates of the defect location is calculated. In this embodiment, the load-bearing capacity assessment value is preferably the local stress intensity ratio. It is defined as: the maximum value of the first principal stress at the stress concentration point at the defect edge. Standard value of design compressive strength of concrete materials The ratio, that is:

[0141]

[0142] Those skilled in the art will understand that the load-bearing capacity assessment value can also be expressed equivalently using the overall structural safety factor or yield strain limit.

[0143] S504, threshold comparison logic and generation of graded early warning signals.

[0144] Obtain the preset safety threshold (e.g., critical stress intensity ratio) that conforms to the design specifications for underground engineering. The calculated bearing capacity assessment value is compared with the preset safety threshold in real time. If... If the structure is deemed safe, only the defect record is archived; otherwise... If the local load-bearing capacity is insufficient, and there is a risk of brittle failure or cracking, a structural safety warning signal will be automatically triggered.

[0145] according to If the error exceeds the warning level, a graded early warning report is generated. The early warning report not only includes the alarm level (such as red, orange, yellow), but also outputs the corresponding defect location coordinates, geometric dimension profile, and suggested grouting reinforcement area through the feedback channel, thereby guiding precise maintenance on the engineering site.

[0146] like Figure 2 As shown, the present invention also provides an intelligent detection system for hidden structural defects in underground engineering. Based on the intelligent detection method for hidden structural defects in underground engineering described above, the system includes:

[0147] Data acquisition module 1 is used to acquire electromagnetic reflection wave data, elastic stress wave data and apparent resistivity data of the concealed structure of underground engineering, and record the spatial pose coordinates and timestamps when acquiring each data.

[0148] Spatiotemporal registration module 2 is used to perform spatiotemporal registration of the electromagnetic reflection wave data, elastic stress wave data and apparent resistivity data based on the spatial pose coordinates and timestamps of the various data, using a dynamic time warping algorithm, in order to construct a multimodal data tensor;

[0149] Feature inversion module 3 is used to input the multimodal data tensor into a preset physical information neural network to perform defect feature inversion and output the defect parameters of the underground engineering concealed structure. The defect parameters include defect location coordinates, geometric dimensions and medium physical properties. The loss function of the preset physical information neural network includes a partial differential equation residual loss term constructed from Maxwell's equations describing the propagation law of electromagnetic waves and the wave equation describing the propagation law of elastic waves.

[0150] The digital twin construction module 4 is used to convert the defect location coordinates, geometric dimensions, and medium physical properties into three-dimensional point cloud data, and map the three-dimensional point cloud data into a preset building information model of the underground project to construct a digital twin of the underground structure containing defect attributes; wherein, the defect attributes are composed of the defect location coordinates, geometric dimensions, and medium physical properties;

[0151] The assessment and early warning module 5 is used to extract the defect location coordinates, geometric dimensions and medium physical properties in the digital twin of the underground structure as boundary weakening and material deterioration conditions into the finite element analysis solver, calculate the bearing capacity assessment value of the hidden structure of the underground project at the defect location coordinates, and compare it with the preset safety threshold to generate a structural safety early warning signal.

[0152] Each of the above modules is used to perform the corresponding steps in the intelligent detection method for hidden structural defects in underground engineering. The specific implementation methods are as described in the above method embodiments, and will not be repeated here.

[0153] like Figure 3 As shown, the present invention also provides a computer device, which may be a server, and its internal structure may be as follows: Figure 3 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores all data required for the intelligent detection method of hidden structural defects in underground engineering. The network interface is used for communication with external terminals via a network connection. The computer program is executed by the processor to implement the intelligent detection method of hidden structural defects in underground engineering.

[0154] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer equipment on which the present application is applied.

[0155] An embodiment of this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements any of the above-described intelligent detection methods for hidden structural defects in underground engineering.

[0156] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by hardware related to computer program instructions. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the above method embodiments. Any references to memory, storage, databases, or other media provided in this application and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM), such as dynamic RAM (used as main storage) or static RAM (commonly used as cache memory). By way of illustration and not limitation, RAM has various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), and Rambus DRAM (RDRAM).

[0157] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0158] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for intelligent detection of hidden structural defects in underground engineering, characterized in that, include: S1. Collect electromagnetic reflection wave data, elastic stress wave data and apparent resistivity data of the concealed structure of underground engineering, and record the spatial pose coordinates and timestamps when collecting each data. S2. Based on the spatial pose coordinates and timestamps of the various data, the electromagnetic reflection wave data, elastic stress wave data and apparent resistivity data are spatiotemporally registered using a dynamic time warping algorithm to construct a multimodal data tensor. S3. Input the multimodal data tensor into a preset physical information neural network to perform defect feature inversion and output the defect parameters of the underground engineering concealed structure. The defect parameters include defect location coordinates, geometric dimensions and medium physical properties. The loss function of the preset physical information neural network includes a partial differential equation residual loss term constructed from Maxwell's equations describing the propagation law of electromagnetic waves and the wave equation describing the propagation law of elastic waves. S4. The defect location coordinates, geometric dimensions, and medium physical properties are converted into three-dimensional point cloud data, and the three-dimensional point cloud data is mapped to a preset building information model of the underground project to construct a digital twin of the underground structure containing defect attributes; wherein, the defect attributes are composed of the defect location coordinates, geometric dimensions, and medium physical properties; S5. Extract the defect location coordinates, geometric dimensions, and medium physical properties in the digital twin of the underground structure as boundary weakening and material deterioration conditions into the finite element analysis solver, calculate the bearing capacity assessment value of the hidden structure of the underground project at the defect location coordinates, and compare it with the preset safety threshold to generate a structural safety early warning signal.

2. The intelligent detection method for concealed structural defects in underground engineering according to claim 1, characterized in that, S1 specifically includes: S101. Deploy a multi-source sensor array and a combined positioning and timing system on the mounting platform, and use a hardware-level clock synchronization mechanism as the master control clock to synchronously send trigger commands to the multi-source sensor array; wherein, the multi-source sensor array includes a ground-penetrating radar antenna array, an acoustic sensor, and a high-density electrical resistivity tomography electrode array. S102. Under the joint control of the trigger command, the multi-source sensor array is used to perform physical field excitation and reception on the concealed structure of the underground project, and electromagnetic reflection wave data, elastic stress wave data and apparent resistivity data are acquired respectively; wherein, the ground penetrating radar antenna array acquires the electromagnetic reflection wave data, the acoustic sensor acquires the elastic stress wave data, and the high-density electrical resistivity meter electrode array acquires the apparent resistivity data. S103. The combined positioning and timing system is used to calculate the spatial pose coordinates of the platform in real time, and the spatial pose coordinates and master control timestamp acquired in the same trigger cycle are used as data tag headers and bound to the corresponding electromagnetic reflection wave data, elastic stress wave data and apparent resistivity data to output a single acquisition data frame.

3. The intelligent detection method for concealed structural defects in underground engineering according to claim 1, characterized in that, S2 specifically includes: S201. Based on the spatial pose coordinates, construct a three-dimensional spatial grid with a unified resolution. According to the pre-calibrated extrinsic parameter matrix, transform the acquisition origin of each data into the three-dimensional spatial grid. Use the Kriging spatial interpolation algorithm or the cubic spline interpolation algorithm to resample and project the discrete electromagnetic reflection wave data, elastic stress wave data and apparent resistivity data onto the same spatial grid nodes to complete the lateral spatial coordinate alignment. S202. Using the spatially aligned electromagnetic reflection wave data as a reference sequence, the cumulative cost matrix of the elastic stress wave data and apparent resistivity data with the reference sequence is calculated using a dynamic time warping algorithm, so as to perform nonlinear time stretching and compression on the elastic stress wave data and apparent resistivity data to complete longitudinal waveform feature registration. S203. The spatiotemporally registered electromagnetic reflection wave data, elastic stress wave data, and apparent resistivity data are respectively subjected to dimensional normalization processing and stacked according to the channel dimension to construct the multimodal data tensor.

4. The intelligent detection method for concealed structural defects in underground engineering according to claim 3, characterized in that, S202 specifically includes: The time series of the electromagnetic reflection wave data is extracted as the reference sequence, and the time series of the elastic stress wave data is extracted as the sequence to be normalized. Calculate the Euclidean distance between corresponding data points of the reference sequence and the sequence to be normalized to construct a distance matrix; Based on the distance matrix, the cumulative cost matrix is ​​calculated using the state transition equation, and the regularized path with the minimum total cumulative distance is found by backtracking through a dynamic programming algorithm. Based on the regularization path, the sequence to be regularized is locally extended or compressed nonlinearly so that its peak and trough features are aligned with the reference sequence on the equivalent physical interface; and the same dynamic time regularization algorithm is used to regularize the one-dimensional depth profile data of the apparent resistivity data to the reference sequence.

5. The intelligent detection method for concealed structural defects in underground engineering according to claim 1, characterized in that, S3 specifically includes: S301. Construct a physical information neural network that includes a data input layer, multiple hidden layers and a physical parameter output layer. Input the multimodal data tensor into the physical information neural network to extract physical coupling features and output the predicted proxy physical field value and the defect parameters to be inverted. S302. Construct a total loss function based on the predicted proxy physical field values, the actual measured physical field values, and the partial differential equations of physical laws; wherein, the total loss function is a weighted combination of the data fitting loss term and the residual loss term of the partial differential equations; S303. An optimization algorithm is used to perform backpropagation updates of network weights to minimize the total loss function. When the total loss function converges, the defect parameters are extracted from the physical parameter output layer, and the defect parameters are logically compared and verified with a preset physical constant threshold range. After passing the logical comparison and verification, the parameters are stored in the database.

6. The intelligent detection method for concealed structural defects in underground engineering according to claim 5, characterized in that, In S302, the total loss function The calculation formula is: in, and For adaptive dynamic weighting coefficients, For data fitting loss term, This is the residual loss term in the partial differential equation; The data fitting loss term The calculation formula is: The residual loss term of the partial differential equation Electromagnetic wave residual term and elastic wave residuals Composition; wherein, the electromagnetic wave residual term The calculation formula is: The elastic wave residual term The calculation formula is: In the above formula, This represents the total number of registration data points actually sampled by the sensor. and These represent the electric field intensity predicted by the network and the actual measured electric field intensity at the i-th data point, respectively. and These are the elastic displacement predicted by the network and the actual measured elastic displacement at the i-th data point, respectively. The number of points to be randomly selected; The Laplace operator is used; t is the time variable. This represents the second-order partial derivative with respect to time; The permeability of free space, It is the vacuum permittivity; Let be the relative permittivity of the network output at the j-th collocation point; Let be the medium density output by the network at the j-th collocation point; Let be the elastic modulus of the medium output by the network at the j-th collocation point.

7. The intelligent detection method for concealed structural defects in underground engineering according to claim 1, characterized in that, S4 specifically includes: S401. Based on the defect location coordinates and the geometric dimensions, reconstruct the three-dimensional geometric bounding box and perform meshing discretization to generate a basic spatial point cloud, and bind the medium physical properties as a physical scalar field to each point of the basic spatial point cloud to generate the three-dimensional point cloud data. S402. Obtain the building information model of the pre-built underground project, calculate the coordinate transformation matrix between the coordinate system of the sensing device and the reference coordinate system of the building information model, and use the coordinate transformation matrix to perform a spatial affine transformation on the three-dimensional point cloud data to obtain the target point cloud data registered in the reference coordinate system. S403. Create a custom attribute extension set corresponding to the target point cloud data in the building information model, write the target point cloud data and the defect attribute into the custom attribute extension set, and establish a spatial topological dependency relationship with the original hidden structural components through spatial intersection determination, thereby constructing the digital twin of the underground structure containing the defect attribute.

8. The intelligent detection method for concealed structural defects in underground engineering according to claim 1, characterized in that, S5 specifically includes: S501. Import the digital twin of the underground structure containing defect attributes into the preset finite element analysis solver, and perform three-dimensional solid mesh generation on the hidden structure based on the defect location coordinates and geometric dimensions. S502. The physical properties of the medium are converted into material degradation constitutive parameters of the finite element element, and the initial stiffness matrix of the finite element element in the defect region is degraded and reconstructed based on the material degradation constitutive parameters to obtain the damage stiffness matrix. S503. Apply the actual boundary load under the operating state of the underground project to the outer boundary of the finite element model, call the finite element analysis solver to solve the displacement field and stress field of the overall structure, and calculate the bearing capacity assessment value of the hidden structure at the coordinates of the defect location. S504. The load-bearing capacity assessment value is compared with the preset safety threshold. If the load-bearing capacity assessment value does not meet the preset safety threshold, a structural safety early warning signal containing the early warning level and defect location is generated.

9. The intelligent detection method for concealed structural defects in underground engineering according to claim 8, characterized in that, Specifically, S502 includes: Define a damage reduction factor generated by the dynamic mapping of the physical properties of the medium. Based on the damage reduction factor, reconstruct the initial stiffness matrix of the finite element elements within the defect region using the following formula: in, Here is the damage stiffness matrix. This is the damage reduction factor. This is the initial stiffness matrix.

10. An intelligent detection system for concealed structural defects in underground engineering, based on the intelligent detection method for concealed structural defects in underground engineering as described in any one of claims 1 to 9, characterized in that, The system includes: The data acquisition module is used to collect electromagnetic reflection wave data, elastic stress wave data and apparent resistivity data of the concealed structure of underground engineering, and to record the spatial pose coordinates and timestamps when collecting each data item. The spatiotemporal registration module is used to perform spatiotemporal registration of the electromagnetic reflection wave data, elastic stress wave data and apparent resistivity data based on the spatial pose coordinates and timestamps of the various data, using a dynamic time warping algorithm, in order to construct a multimodal data tensor. The feature inversion module is used to input the multimodal data tensor into a preset physical information neural network to perform defect feature inversion and output the defect parameters of the underground engineering concealed structure. The defect parameters include defect location coordinates, geometric dimensions and medium physical properties. The loss function of the preset physical information neural network includes a partial differential equation residual loss term constructed from Maxwell's equations describing the propagation of electromagnetic waves and the wave equation describing the propagation of elastic waves. A digital twin construction module is used to convert the defect location coordinates, geometric dimensions, and medium physical properties into three-dimensional point cloud data, and map the three-dimensional point cloud data into a preset building information model of the underground engineering to construct a digital twin of the underground structure containing defect attributes; wherein, the defect attributes are composed of the defect location coordinates, geometric dimensions, and medium physical properties; The assessment and early warning module is used to extract the location coordinates, geometric dimensions, and physical properties of the medium of the defect in the digital twin of the underground structure as boundary weakening and material deterioration conditions and input them into the finite element analysis solver. It calculates the bearing capacity assessment value of the hidden structure of the underground project at the defect location coordinates and compares it with a preset safety threshold to generate a structural safety early warning signal.

Images