Mechanics-driven method, device and equipment for geometric correction of road hidden diseases and storage medium

Abstract

This invention discloses a method, device, equipment, and storage medium for geometric correction of hidden road defects based on mechanical drive. The method includes: constructing a finite element simulation sample library for road defects; extracting and fusing multidimensional features from the sample library to obtain a fused feature vector; using this vector to construct a dataset for training to obtain a proxy model for defect parameter inversion; locating abnormal areas of the road medium using 3D ground-penetrating radar and activating the corresponding strain sensor array; acquiring measured strain data and fusing it into a feature vector to be identified; inputting this vector into the proxy model for inversion; and outputting the geometric parameters of hidden road defects. This effectively combines the advantages of 3D ground-penetrating radar and strain sensor arrays, compensating for the shortcomings of single detection methods, and effectively supplementing the inaccuracy of ground-penetrating radar in detecting the boundaries of hidden defects. Through multidimensional feature extraction and fusion and neural network model training, the deep information of strain data is accurately mined, improving the estimation accuracy of geometric parameters of hidden road defects.

Description

Technical Field

[0001] This invention relates to the field of road engineering, and in particular to a method, apparatus, equipment and storage medium for geometric correction of hidden road defects based on mechanical drive. Background Technology

[0002] With the deepening of the construction of "smart highways," higher demands are being placed on the perception capabilities of road infrastructure and the digitalization and precision of maintenance management. In the construction of smart highways, achieving accurate perception and quantitative assessment of hidden defects in road structures is key to supporting scientific and digital maintenance decision-making.

[0003] Currently, while 3D ground-penetrating radar (GPR) is the mainstream method for traditional non-destructive testing of hidden road defects and can detect underground anomalies, it lacks sensitivity for early-developing, small-scale defects and struggles to accurately estimate their geometric dimensions and volume. This leads to reliance on experience-based judgments for determining maintenance workload and budget, hindering the refinement of maintenance management. Meanwhile, strain sensor arrays widely deployed in smart highways can capture the real dynamic mechanical response of road structures under vehicle loads in real time with high sensitivity, theoretically providing a rich data source for defect identification. However, existing monitoring modes are mostly limited to simple single-point threshold alarms, only determining "whether there is an anomaly," and have not effectively analyzed the deeper information such as defect type, size, and severity contained in the spatial distribution of the strain field, resulting in the data value not being fully explored. Therefore, a technique for correcting the geometric boundaries of hidden defects is urgently needed to improve the accuracy of GPR in identifying the geometric information of hidden defects. Utilizing highly sensitive sensor data can compensate for the inaccuracies of single-radar boundary detection. Summary of the Invention

[0004] The main objective of this invention is to provide a mechanically driven method, device, equipment, and storage medium for geometric correction of hidden road defects. This invention aims to solve the problems in the prior art where three-dimensional ground-penetrating radar is difficult to accurately identify the geometric morphology of hidden road defects, and the massive data of strain sensor arrays has not been deeply mined, resulting in the inability to accurately quantify the geometric parameters of hidden road defects.

[0005] To achieve the above objectives, this invention provides a mechanically driven geometric correction method for hidden road defects, the method comprising the following steps:

[0006] Construct a finite element simulation sample library for road defects;

[0007] Based on the aforementioned road disease finite element simulation sample library, feature extraction and fusion of multi-dimensional feature vectors are performed to obtain fused feature vectors.

[0008] A dataset is constructed based on the fused feature vectors, and an original neural network model is trained based on the dataset to obtain a disease parameter inversion proxy model. The disease parameter inversion proxy model is a fully connected multilayer perceptron network architecture.

[0009] The abnormal medium area under the road surface is identified by three-dimensional ground penetrating radar, and the corresponding strain sensor array node below the abnormal medium area is activated.

[0010] The strain data monitored by the strain sensor array nodes is acquired, and the strain data is subjected to feature extraction and fusion to obtain the fused feature vector to be identified.

[0011] The feature vector to be identified is input into the disease parameter inversion proxy model for inversion operation, the target geometric parameters are output, and the original geometric parameters are corrected based on the target geometric parameters to obtain the corrected road hidden disease geometric parameters.

[0012] Optionally, the construction of the finite element simulation sample library for road defects includes:

[0013] The input variable space is sampled to generate multiple simulation conditions, and a road ontology model for each simulation condition is constructed. The input variable space includes the baseline geometric parameters of the defects, the depth of the defects, the irregular disturbance coefficient, and the lateral offset of the load.

[0014] In the road body model, a disease center point and a reference radius are set. An irregular disease boundary surface is generated based on the disease center point and the reference radius. Based on the irregular disease boundary surface, an irregular road hidden disease model is constructed in the road body model.

[0015] Set standard moving loads and lateral offsets for the road body model;

[0016] Calculate the geometric parameter labels of the road hidden disease model, wherein the geometric parameter labels include the equivalent volume of the disease, the shape factor, and the burial depth of the disease;

[0017] The dynamic calculations are performed and the response data of the strain sensor array nodes are extracted. The response data is then associated with the geometric parameter labels of the road defects to construct a finite element simulation sample library for road defects.

[0018] Optionally, the step of setting a disease center point and a reference radius in the road body model, generating an irregular disease boundary surface based on the disease center point and the reference radius, and constructing an irregularly shaped road hidden disease model in the road body model based on the irregular disease boundary surface includes:

[0019] The road body model is structurally divided from top to bottom to form multiple structural layers, and each structural layer is assigned material properties. The structural layers include surface layer, base layer, subbase layer and soil base layer, and the material properties include elastic modulus, Poisson's ratio and material density.

[0020] Boundary conditions are applied to the road body model, and the road body model is meshed. The key areas in the road body model are divided into fine meshes, and the far-field areas are divided into sparse meshes. The fine meshes and the sparse meshes are smoothly connected through transition meshes. The key areas are the layout areas of strain sensor array nodes and the disease generation areas.

[0021] Within the road body model, a disease center point is set, a local spherical coordinate system is established, and the surface radius function of the reference ellipsoid is defined.

[0022] A spatial random noise field is superimposed on the reference radius of the reference ellipsoid to generate an irregular defect boundary surface, as shown in the following formula:

[0023]

[0024] in, Represents the irregular boundary surface of disease. Represents the surface radius function. This represents a normalized three-dimensional random noise function. This represents the irregular disturbance coefficient. and These represent the azimuth and polar angles of the local spherical coordinate system, respectively.

[0025] The material properties of the units in the road body model that are surrounded by the irregular defect boundary surface are reduced, and an irregular road hidden defect model is formed based on the units after the material property reduction.

[0026] Optionally, the multidimensional feature vector includes spatial feature sub-vectors, frequency domain feature sub-vectors, and shape-sensitive feature sub-vectors; the step of extracting features based on the road distress finite element simulation sample library and fusing the multidimensional feature vectors to obtain the fused feature vector includes:

[0027] Based on the road disease finite element simulation sample library, effective data segments are segmented, and key measurement points in the effective data segments are identified;

[0028] Based on the key measurement points, spatial domain features, frequency domain features, and shape-sensitive features are extracted from the effective data segments to obtain spatial feature sub-vectors, frequency domain feature sub-vectors, and shape-sensitive feature sub-vectors.

[0029] The spatial feature vector, frequency domain feature vector, and shape-sensitive feature vector are fused to obtain a fused feature vector.

[0030] Optionally, the effective data segment includes a simulated effective data segment and a measured effective data segment; the step of segmenting the effective data segment based on the road distress finite element simulation sample library and identifying key measurement points in the effective data segment includes:

[0031] For the simulation data in the finite element simulation sample library of road defects, the data segments before and after a preset time are extracted as the effective simulation data segments, based on the time when the load passes through the center of the road body model.

[0032] For the response data of the strain sensor array nodes in the road defect finite element simulation sample library, a sliding window is used to monitor the continuous time-series response data. When the strain amplitude of any sensor node in the strain sensor array exceeds a preset strain threshold, it is marked as a vehicle entry event.

[0033] Based on the peak time of the vehicle entry event monitoring response, and taking the peak time as the center, a time window of a preset duration is extracted from the response data as the effective data segment of a single vehicle passage.

[0034] The strain sensor array node with the largest absolute value of the response peak within the simulated effective data segment and the measured effective data segment is defined as the main response node and its position is recorded to complete the identification of key measurement points.

[0035] Optionally, the step of extracting spatial domain features, frequency domain features, and shape-sensitive features from the effective data segment based on the key measurement points to obtain spatial feature sub-vectors, frequency domain feature sub-vectors, and shape-sensitive feature sub-vectors includes:

[0036] Based on the key measuring points, the instantaneous strain values ​​at the moment when each main response node in the effective data segment reaches the strain peak are extracted to form an instantaneous strain vector;

[0037] The instantaneous strain vector is normalized to obtain the spatial feature sub-vector;

[0038] Obtain the strain time history data of the master response node, and apply Fast Fourier Transform to transform the strain time history data from the time domain to the frequency domain to obtain the spectral coefficients, referring to the following formula:

[0039]

[0040] in, Represents the spectral coefficients. Indicates the frequency domain point number. This indicates the number of sampling points in the strain time history data. This represents the discretized strain time history data of the master response node. Indicates the sampling point number. Represents the imaginary unit;

[0041] Based on the aforementioned spectral coefficients, the normalized energy proportions of the low-frequency band, mid-frequency band, and high-frequency band are calculated to form frequency domain feature vectors, as shown in the following formula:

[0042]

[0043]

[0044] in, Represents the frequency domain eigenvectors. This represents the total energy of the frequency domain data of the strain at the master response node. Indicates frequency, This represents the energy integral value within each frequency band. Indicates low frequency band. Indicates the mid-frequency band. Indicates high frequency band;

[0045] Calculate the ratio of the longitudinal strain sensor reading to the transverse strain sensor reading at the master response node, and use the ratio as a shape-sensitive feature sub-vector.

[0046] Optionally, the step of constructing a dataset based on the fused feature vectors and training an original neural network model based on the dataset to obtain a disease parameter inversion proxy model includes:

[0047] A dataset is constructed based on the fused feature vectors and the corresponding geometric parameter labels of road defects in the finite element simulation sample library, and the dataset is divided into a training set, a validation set, and a test set.

[0048] An original neural network model is generated based on a fully connected multilayer perceptron network architecture. The original neural network model includes an input layer, multiple hidden layers, and an output layer. The dimension of the input layer matches the dimension of the fused feature vector, and the dimension of the output layer matches the dimension of the disease geometric parameter label.

[0049] The mean squared error with second-order regularization is set as the loss function. The backpropagation algorithm combined with the adaptive moment estimation optimizer is used to iteratively update the network parameters of the original neural network model using the training set, and the model training effect is monitored through the validation set.

[0050] When the loss value calculated by the loss function on the validation set no longer decreases, or when the number of model training iterations reaches the preset number of iterations, stop model training and save the current optimal network parameters to obtain the disease parameter inversion proxy model.

[0051] Furthermore, to achieve the above objectives, this invention also proposes a mechanistic-driven road hidden defect geometric correction device that applies the mechanistic-driven road hidden defect geometric correction method described above. The mechanistic-driven road hidden defect geometric correction device includes:

[0052] The sample library construction module is used to build a sample library for finite element simulation of road defects;

[0053] The feature extraction module is used to extract features based on the road disease finite element simulation sample library and fuse multi-dimensional feature vectors to obtain fused feature vectors;

[0054] The model training module is used to construct a dataset based on the fused feature vectors and train the original neural network model based on the dataset to obtain a disease parameter inversion proxy model. The disease parameter inversion proxy model is a fully connected multilayer perceptron network architecture.

[0055] The monitoring module is used to identify abnormal areas of the medium under the road surface using three-dimensional ground penetrating radar and activate the corresponding strain sensor array nodes below the abnormal areas of the medium.

[0056] The data acquisition module is used to acquire strain data monitored by the strain sensor array nodes, extract and fuse the strain data to obtain a fused feature vector to be identified.

[0057] The geometric parameter output module is used to input the fused feature vector to be identified into the disease parameter inversion proxy model for inversion operation, output the target geometric parameters, and correct the original geometric parameters based on the target geometric parameters to obtain the corrected road hidden disease geometric parameters.

[0058] Furthermore, to achieve the above objectives, this application also proposes a mechanically driven road hidden defect geometric correction device, the device comprising: a memory, a processor, and a mechanically driven road hidden defect geometric correction program stored in the memory, the processor being used to run the mechanically driven road hidden defect geometric correction program, the computer program being configured to implement the steps of the mechanically driven road hidden defect geometric correction method as described above.

[0059] In addition, to achieve the above objectives, this application also proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the mechanically driven geometric correction method for hidden road defects as described above.

[0060] This invention constructs a finite element simulation sample library for road defects, extracts features from this library, fuses multidimensional feature vectors to obtain fused feature vectors, constructs a dataset based on these fused feature vectors, and trains an original neural network model on this dataset to obtain a defect parameter inversion proxy model. The defect parameter inversion proxy model is a fully connected multilayer perceptron network architecture. It uses 3D ground-penetrating radar to identify abnormal areas of the medium beneath the road surface, activates the corresponding strain sensor array nodes below these abnormal areas, acquires strain data monitored by the strain sensor array nodes, extracts and fuses features from the strain data to obtain a fused feature vector to be identified, inputs this fused feature vector to the defect parameter inversion proxy model for inversion calculation, outputs target geometric parameters, and then... This invention modifies the original geometric parameters to obtain corrected geometric parameters for hidden road defects. It effectively taps into the deeper value of strain sensor array monitoring data, breaking the limitation of existing sensors that are only used for single-point threshold alarms. By integrating the advantages of 3D ground-penetrating radar and strain sensors, it compensates for the shortcomings of single detection methods and effectively addresses the inaccuracy of ground-penetrating radar in detecting the boundaries of hidden defects. Through strain data-driven precise correction of defect geometric parameters, it improves the accuracy of identifying the geometric morphology of hidden defects. By combining finite element simulation with neural networks, it constructs an efficient defect parameter inversion proxy model, replacing time-consuming finite element iterative calculations, significantly improving the efficiency of defect detection and parameter inversion. By achieving precise quantification of the geometric parameters of hidden road defects, it provides a scientific basis for calculating maintenance workload and formulating maintenance plans. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 This is a schematic diagram of the structure of a mechanically driven geometric correction device for hidden road defects in the hardware operating environment involved in the embodiments of the present invention.

[0063] Figure 2 This is a flowchart illustrating the first embodiment of the mechanically driven geometric correction method for hidden road defects according to the present invention.

[0064] Figure 3 This is a flowchart illustrating the second embodiment of the mechanically driven geometric correction method for hidden road defects of the present invention.

[0065] Figure 4This is a flowchart illustrating the third embodiment of the mechanically driven geometric correction method for hidden road defects of the present invention.

[0066] Figure 5 This is a structural block diagram of the first embodiment of the mechanically driven geometric correction device for hidden road defects according to the present invention.

[0067] 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

[0068] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0069] Reference Figure 1 , Figure 1 This is a schematic diagram of the structure of a mechanically driven geometric correction device for hidden road defects, which is part of the hardware operating environment involved in the embodiments of the present invention.

[0070] like Figure 1 As shown, the mechanically driven road hidden defect geometry correction device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen and an input unit such as a keyboard; the user interface 1003 may also include standard wired and wireless interfaces. The network interface 1004 may optionally include standard wired and wireless interfaces (such as Wireless-Fidelity (Wi-Fi) interfaces). The memory 1005 may be high-speed random access memory (RAM) or stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0071] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on mechanically driven road hidden defect geometry correction devices, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0072] like Figure 1As shown, the memory 1005, which is a computer-readable storage medium, may include an operating system, a network communication module, a user interface module, and a mechanics-driven geometric correction program for hidden road defects.

[0073] exist Figure 1 In the mechanically driven road hidden defect geometry correction device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and memory 1005 in the mechanically driven road hidden defect geometry correction device of the present invention can be set in the mechanically driven road hidden defect geometry correction device. The mechanically driven road hidden defect geometry correction device calls the mechanically driven road hidden defect geometry correction program stored in the memory 1005 through the processor 1001 and executes the mechanically driven road hidden defect geometry correction method provided in the embodiment of the present invention.

[0074] This invention provides a mechanically driven geometric correction method for hidden road defects, referring to... Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the mechanically driven geometric correction method for hidden road defects according to the present invention.

[0075] In this embodiment, the mechanically driven geometric correction method for hidden road defects includes the following steps:

[0076] Step S10: Construct a finite element simulation sample library for road defects.

[0077] It should be understood that the executing entity of this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or a terminal electronic device capable of realizing the above functions. The following description uses a mechanically driven road hidden defect geometric correction device (hereinafter referred to as the correction device) as an example to illustrate this embodiment and the following embodiments.

[0078] It should be noted that the road distress finite element simulation sample library can be a standardized database formed by simulating the dynamic response of hidden road distress under different working conditions using finite element simulation technology, integrating simulation data with corresponding distress geometric parameter labels, and used for subsequent feature extraction and model training. It contains input parameters, sensor response data and real geometric properties of distress for multiple sets of simulation working conditions.

[0079] In some embodiments, the correction device can utilize finite element analysis software to construct a three-dimensional solid constitutive model of the road, specifying the model's length, width, and height dimensions, and dividing it into four structural layers: surface layer, base layer, subbase layer, and subgrade. Material properties based on recommended values ​​are assigned to each structural layer, and the model's governing equations are set according to the general equations of structural dynamics. Fixed constraints are applied to the bottom of the model, and normal displacement constraints are applied to the lateral surfaces to simulate actual roadbed support conditions. A three-dimensional eight-node reduced integral hexahedral element is used to divide the mesh; a fine mesh is used for the sensor array deployment area and the defect generation area, while a sparse mesh is used for the far-field region, smoothly connected through a transition mesh. The method of "baseline morphology + random disturbance" is used to simulate irregular hidden defects. A center point of the defect is set, a base ellipsoid is established, and spatial random noise is superimposed to generate the defect boundary, reducing the material properties of the defect area unit. A standard single-axle dual-wheel set moving load is set to simulate the vehicle driving process and introduce lateral offset to cover different wheel track conditions. For each simulation condition, the equivalent volume and shape factor of the defect are calculated. A sampling method is used to cover key variables such as base geometric parameters, defect burial depth, and disturbance coefficient. N simulation conditions are automatically generated through parameterized scripts. Sensor time history response data is extracted and associated with labels to finally form a finite element simulation sample library of road defects.

[0080] Step S20: Based on the road defect finite element simulation sample library, perform feature extraction and fuse multi-dimensional feature vectors to obtain fused feature vectors.

[0081] It should be noted that a multidimensional feature vector can be a high-dimensional vector that integrates three types of features: spatial domain, frequency domain, and correlation. It contains not only the spatial distribution information of the disease, but also the information on the changes in the mechanical properties of the medium caused by the disease (frequency domain) and the shape information of the disease (correlation), and can comprehensively characterize the disease features.

[0082] In some embodiments, the correction device can extract multidimensional features from a finite element simulation sample library, including spatial features, frequency domain features, and shape features; use feature extraction algorithms such as wavelet transform and principal component analysis (PCA) to screen and reduce the dimensionality of various features and remove redundant information; and use a serial fusion method to integrate the screened multidimensional feature vectors into a unified fused feature vector to ensure that the fused features can fully reflect the essential attributes of the disease.

[0083] It is understandable that this embodiment effectively mines key features related to diseases in simulation samples, removes invalid and redundant information, reduces feature dimensionality, and improves the efficiency of subsequent model training. Through multi-dimensional feature fusion, it makes up for the deficiency of single-dimensional features in reflecting disease information incompletely, enhances the feature's ability to represent disease parameters, and provides high-quality input data for subsequent model training.

[0084] Step S30: Construct a dataset based on the fused feature vector, and train the original neural network model based on the dataset to obtain a disease parameter inversion proxy model. The disease parameter inversion proxy model is a fully connected multilayer perceptron network architecture.

[0085] It should be noted that the disease parameter inversion surrogate model refers to a model that, after being trained on a dataset, can quickly invert and output the corresponding geometric parameters of the disease based on the input fused feature vector, and is used to replace complex finite element simulation calculations.

[0086] In some embodiments, the correction device uses fused feature vectors as input data and corresponding simulated disease geometric parameters (such as equivalent volume, shape, and burial depth of the disease) as labels. It divides the data into training, validation, and test sets according to a preset ratio (e.g., 7:2:1) to construct a complete dataset. A fully connected multilayer perceptron (MLP) is selected as the original neural network model. The number of neurons in the input, hidden, and output layers is set, and the activation function, optimizer, and loss function are determined. The model is iteratively trained using the training set, the model hyperparameters are adjusted using the validation set, and the model performance is verified using the test set until the model converges and its generalization ability meets the standard, ultimately obtaining a disease parameter inversion proxy model.

[0087] Furthermore, to improve the model's generalization ability and prediction accuracy, step S30 above may include:

[0088] Step S301: Construct a dataset based on the fused feature vector and the corresponding geometric parameter labels of road defects in the finite element simulation sample library, and divide the dataset into a training set, a validation set and a test set;

[0089] Step S302: Generate an original neural network model based on a fully connected multilayer perceptron network architecture. The original neural network model includes an input layer, multiple hidden layers, and an output layer. The dimension of the input layer matches the dimension of the fused feature vector, and the dimension of the output layer matches the dimension of the disease geometric parameter label.

[0090] Step S303: Set the mean squared error with second-order regularization as the loss function, use the backpropagation algorithm combined with the adaptive moment estimation optimizer, use the training set to iteratively update the network parameters of the original neural network model, and monitor the model training effect through the validation set.

[0091] Step S304: When the loss value calculated by the loss function on the validation set no longer decreases, or when the number of model training iterations reaches the preset number of iterations, stop model training and save the current optimal network parameters to obtain the disease parameter inversion proxy model.

[0092] In the specific implementation, a training set is constructed based on the fusion feature vectors extracted from the simulation sample library. :

[0093]

[0094] Wherein, input vector For the first The fused feature vector of each sample includes normalized spatial features, frequency domain energy features, and longitudinal and transverse strain ratios.

[0095] Output Labels For the first The actual disease parameters corresponding to each sample For equivalent volume Shape factor burial depth The vector formed.

[0096] The dataset is randomly divided into training, validation, and test sets.

[0097] Construct a fully connected multilayer perceptron network. The network consists of one input layer, several hidden layers, and one output layer. The mathematical model for forward propagation is defined as follows:

[0098]

[0099] in, For the first The layer's output vector; and The first Layer weight matrix and bias vector; It is a non-linear activation function.

[0100] To ensure the model is accurate in prediction while also possessing robustness against noise (preventing overfitting), a banding method is used. Regularized mean squared error as the target loss function :

[0101]

[0102] in, This represents the number of samples in the current batch. For the model's predicted output, For truth labels; This is the regularization coefficient, used to constrain the weight magnitude and improve the model's robustness to random noise added to the simulation sample library.

[0103] The backpropagation algorithm combined with the Adam optimizer is used to optimize the network parameters. Perform iterative updates. The parameter update formula is:

[0104]

[0105] in: For learning rate, and These are the first and second moments of the gradient, respectively.

[0106] When the loss function on the validation set no longer decreases or reaches the preset number of iterations, training is stopped, the optimal model parameters are saved, and the final disease inversion proxy model is obtained.

[0107] Step S40: Identify the abnormal medium area under the road surface using three-dimensional ground penetrating radar, and activate the corresponding strain sensor array node below the abnormal medium area.

[0108] It should be noted that abnormal media areas can be areas where the physical properties of the media under the road surface are significantly different from those of the normal structural media, usually corresponding to hidden road defects (such as cavities, looseness, cracks, etc.).

[0109] The nodes of the strain sensor array can be sensor units arranged in the interior of the road structure (such as the base layer or subgrade) according to a preset pattern to monitor the strain changes of the road structure. Multiple nodes form an array, which can realize regional monitoring. When not activated, it is in a low-power standby state.

[0110] In practice, vehicle-mounted 3D ground-penetrating radar is used to periodically inspect roads and identify abnormal areas in the underlying media. The planar coordinates of these abnormal areas in the radar image are then determined. This activates the corresponding strain sensor array node below the region, putting it into a high-frequency waiting-to-trigger state.

[0111] In some embodiments, a three-dimensional ground-penetrating radar (GPR) is used to comprehensively scan the road surface. The radar emits high-frequency electromagnetic waves and receives reflected signals from different media beneath the road surface. By analyzing the amplitude and phase changes of the reflected signals, areas with physical properties (dielectric constant, density, etc.) that differ from those of normal road surface media (surface layer, base layer, subgrade) are identified, i.e., abnormal media areas. Based on the coordinates of the abnormal areas located by the three-dimensional GPR, an activation signal is sent through the control system to activate the corresponding nodes in the pre-arranged strain sensor array directly below the abnormal area, putting them into real-time monitoring mode and preparing to collect strain data. This enables rapid and non-destructive positioning of hidden defects under the road surface, avoiding the inefficiency caused by blind monitoring. By accurately activating the sensor nodes corresponding to the abnormal areas, the ineffective operation of sensors in irrelevant areas is reduced, energy consumption is lowered, and the subsequently collected strain data is highly targeted and directly related to the defect areas, improving the effectiveness of the data.

[0112] Step S50: Obtain the strain data monitored by the strain sensor array nodes, extract and fuse the strain data to obtain the fused feature vector to be identified.

[0113] It should be noted that strain data can be the strain change data of the pavement structure under its own weight, vehicle load, etc., which can be monitored by the strain sensor array nodes and can reflect the stress state of the pavement structure and the impact of defects on the structure.

[0114] The fusion feature vector to be identified refers to the fusion feature vector obtained after feature extraction and fusion of actual monitored strain data. It is used as input to the disease parameter inversion proxy model to realize the inversion of the actual disease geometric parameters.

[0115] In practice, when a vehicle passes through the aforementioned monitoring area, the sensor array automatically records complete time-history strain data. Through the analysis of Perform effective segmentation, maximum value normalization, Fourier transform, and longitudinal and transverse strain ratio calculation to generate the fused feature vector to be identified. The feature vector to be identified and fused contains the spatial distribution, frequency domain, and shape of the current disease under actual load.

[0116] In some embodiments, the correction device acquires strain data monitored by the activated strain sensor array nodes in real time through a data acquisition module; preprocesses the acquired strain data by using methods such as filtering, denoising, and normalization to eliminate invalid data caused by environmental interference and equipment errors; extracts mechanical features (such as strain peak value, strain gradient, strain rate of change, etc.) from the preprocessed strain data through feature extraction algorithms (such as wavelet transform and principal component analysis); integrates the extracted strain features into a fusion feature vector to be identified, ensuring that its format is consistent with the fusion feature vector obtained in step S20, and adapts to the subsequent proxy model input.

[0117] Step S60: Input the fused feature vector to be identified into the disease parameter inversion proxy model for inversion operation, output the target geometric parameters, and correct the original geometric parameters based on the target geometric parameters to obtain the corrected road hidden disease geometric parameters.

[0118] It should be noted that the target geometric parameters refer to the geometric parameters obtained by inverting through the defect parameter inversion proxy model, which can accurately reflect the actual size, location, shape, etc. of hidden road defects.

[0119] The original geometric parameters refer to the uncorrected geometric parameters of hidden road defects obtained through preliminary identification by three-dimensional ground-penetrating radar, which contain a certain degree of identification error.

[0120] The corrected geometric parameters of hidden road defects refer to the geometric parameters that, after being corrected by the target geometric parameters, can truly and accurately reflect the actual state of the defects and can be used for defect assessment and treatment.

[0121] Understandably, this embodiment achieves accurate correction of the original geometric parameters, solving the problem of large errors and insufficient accuracy of geometric parameters in the initial identification of 3D ground-penetrating radar, and greatly improving the identification accuracy of geometric parameters of hidden road defects; through rapid inversion using a proxy model, the correction process is ensured to be efficient, avoiding complex manual calculations and repeated detections, and improving the efficiency of obtaining defect geometric parameters.

[0122] In the specific implementation, the feature vector to be identified and fused is input into the trained disease parameter inversion surrogate model. The model performs forward inference operations and outputs the predicted disease parameter estimates in real time.

[0123]

[0124] in: For mechanical equivalent volume, For shape factor, For burial depth.

[0125] The system outputs the corrected mechanical equivalent volume of the defect. .

[0126] In some embodiments, the correction device inputs the fused feature vector to be identified into a trained disease parameter inversion proxy model. The model outputs the corresponding target geometric parameters (i.e., geometric parameters reflecting the actual state of the disease) through nonlinear mapping operations of an internal fully connected multilayer perceptron. The device obtains the original geometric parameters (such as the approximate size and location of the abnormal area) obtained from the preliminary identification by 3D ground-penetrating radar. The target geometric parameters are compared with the original geometric parameters to analyze the deviation between the two. The difference correction method is used to adjust the original geometric parameters based on the target geometric parameters to eliminate the error in the original identification. Finally, the corrected geometric parameters of the hidden road disease that can truly reflect the actual situation of the disease are obtained.

[0127] This embodiment constructs a finite element simulation sample library for road defects. Based on this library, features are extracted and multi-dimensional feature vectors are fused to obtain fused feature vectors. A dataset is constructed based on these fused feature vectors, and an original neural network model is trained on the dataset to obtain a defect parameter inversion proxy model. The defect parameter inversion proxy model is a fully connected multilayer perceptron network architecture. It uses 3D ground-penetrating radar to identify abnormal areas of the medium under the road surface and activates the corresponding strain sensor array nodes below these abnormal areas. The strain data monitored by the strain sensor array nodes is acquired, and features are extracted and fused from the strain data to obtain a fused feature vector to be identified. This fused feature vector is input into the defect parameter inversion proxy model for inversion calculation, outputting target geometric parameters. The geometric parameters are corrected from the original geometric parameters to obtain the corrected geometric parameters of hidden road defects. This embodiment effectively taps into the deep value of strain sensor array monitoring data, breaking the limitation of existing sensors that are only used for single-point threshold alarms. It integrates the advantages of three-dimensional ground-penetrating radar and strain sensors, making up for the shortcomings of single detection methods and effectively supplementing the inaccuracy of ground-penetrating radar in detecting the boundaries of hidden defects. Through strain data-driven precise correction of defect geometric parameters, the recognition accuracy of hidden defect geometric morphology is improved. By combining finite element simulation and neural networks, an efficient defect parameter inversion proxy model is constructed to replace time-consuming finite element iterative calculations, which can significantly improve the efficiency of defect detection and parameter inversion. By realizing the precise quantification of road hidden defect geometric parameters, a scientific basis is provided for the calculation of maintenance engineering volume and the formulation of maintenance plans.

[0128] refer to Figure 3 , Figure 3 This is a flowchart illustrating the second embodiment of the mechanically driven geometric correction method for hidden road defects according to the present invention.

[0129] Based on the first embodiment described above, in this embodiment, step S10 further includes:

[0130] Step S101: Sample the input variable space to generate multiple simulation conditions and construct the road ontology model for each simulation condition.

[0131] It should be noted that the input variable space refers to the set of key parameters that affect the simulation effect of hidden road defects, including the baseline geometric parameters of the defects, the depth of the defects, the irregularity disturbance coefficient, and the lateral offset of the load. The simulation case refers to a specific combination of parameters in the input variable space, corresponding to a specific defect and load condition.

[0132] In the specific implementation, a road constitutive model is constructed, including: based on the theory of elastic layered systems, a three-dimensional solid model is constructed using finite element analysis software. Its dimensions are defined as length X (m), width Y (m), and height Z (m), serving as the road body model.

[0133] In some embodiments, the correction device uses a sampling method to perform full-coverage sampling of the input variable space to establish N simulation conditions.

[0134] Step S102: Set the center point and reference radius of the defects in the road body model, generate an irregular defect boundary surface based on the center point and reference radius, and construct an irregular road hidden defect model in the road body model based on the irregular defect boundary surface.

[0135] It should be noted that the center point of the defect refers to a pre-defined coordinate point in the road model that serves as the reference point for the defect's location, used to determine the approximate location of the defect. The reference radius can be a set basic dimension parameter of the defect, used to control the approximate scale of the defect.

[0136] It should be noted that irregular defect boundary surfaces refer to irregularly shaped surfaces generated based on a reference radius and an irregular disturbance coefficient, used to define the extent and morphology of defects. Irregularly shaped road hidden defect models refer to finite element models constructed within the road body model based on irregular defect boundary surfaces, simulating the morphology of actual road hidden defects (such as irregular voids and non-through cracks).

[0137] Furthermore, in order to improve the realism of the simulation and thus realize the simulation of diseases with different degrees of irregularity and enrich the diversity of samples, the above step S102 may include:

[0138] Step S1021: Divide the road body model into multiple structural layers from top to bottom, and assign material properties to each structural layer. The structural layers include surface layer, base layer, subbase layer and soil base layer. The material properties include elastic modulus, Poisson's ratio and material density.

[0139] Step S1022: Apply boundary conditions to the road body model and set the mesh for the road body model. Divide the key area in the road body model into a fine mesh and divide the far field area into a sparse mesh. The fine mesh and the sparse mesh are smoothly connected by a transition mesh. The key area is the layout area of ​​the strain sensor array nodes and the area where defects are generated.

[0140] Step S1023: Set the center point of the defect inside the road body model, establish a local spherical coordinate system and define the surface radius function of the reference ellipsoid;

[0141] Step S1024: Superimpose a spatial random noise field on the reference radius of the reference ellipsoid to generate an irregular disease boundary surface;

[0142] Step S1025: Reduce the material properties of the units in the road body model that are surrounded by the irregular defect boundary surface, and form an irregular road hidden defect model based on the units after reducing the material properties.

[0143] In practice, the correction equipment, based on the design of typical road surface structures, divides the road body model from top to bottom into four distinct structural layers: surface layer, base layer, subbase layer, and subgrade.

[0144] Based on indoor tests or recommended values, the elastic modulus of typical materials is assigned to the surface layer, base layer, subbase layer, and subgrade, respectively. Poisson's ratio and density Parameters, etc. To simulate the dynamic response caused by vehicle movement, the model's governing equations follow the general equations of structural dynamics:

[0145]

[0146] in: , , These are the mass, damping, and stiffness matrices, respectively. , , These are the nodal displacement, velocity, and acceleration vectors, respectively. The moving vehicle load vector varies with time and space.

[0147] Fixed constraints (constraints on all degrees of freedom) are applied to the bottom of the model, and normal displacement constraints are applied to the four lateral surfaces of the model (vertical movement is allowed, horizontal movement is restricted) to reasonably simulate the infinite domain support conditions of the actual roadbed.

[0148] Three-dimensional eight-node reduced integral hexahedral elements are selected to improve computational efficiency and accuracy. The "sensor array deployment area" and "disease generation area" are defined as key regions, with a fine mesh size set, while the far-field region is set with a sparse mesh. The two are smoothly connected by a transition mesh.

[0149] The roughness and randomness of real road defects are simulated using a "baseline shape" and "random perturbation" method. A defect center point is set within the road model. To establish a local spherical coordinate system, first define a reference ellipsoid, whose surface radius function is... satisfy:

[0150]

[0151] in The diseases are respectively in The length of the semi-axis in the direction.

[0152] An irregular disease boundary surface is generated by superimposing a spatial random noise field on the reference radius, as shown in the following formula:

[0153]

[0154] in, Represents the irregular boundary surface of disease. Represents the surface radius function. This represents a normalized three-dimensional random noise function. This represents the irregular disturbance coefficient. and These represent the azimuth and polar angles of the local spherical coordinate system, respectively.

[0155] The material properties of the units in the model that are surrounded by irregular disease boundary surfaces are reduced to simulate hidden road diseases.

[0156] Step S103: Set standard moving load and lateral offset for the road body model.

[0157] It should be noted that the standard moving load refers to the load that simulates actual vehicle traffic, including parameters such as axle load, wheel pressure, and moving speed, used to simulate the effect of vehicles on the road. The lateral load offset refers to the lateral distance of the moving load relative to the road centerline, used to simulate the condition of vehicles deviating from the lane centerline during travel, making the simulation more closely resemble actual traffic scenarios.

[0158] In the implementation, to simulate the effect of real traffic flow on the road surface and cover the uncertainty of vehicle trajectories, a standard single-axle dual-wheel load is used. The tire contact patch shape is simplified to two rectangular or circular uniformly distributed load areas, and the contact pressure is set to 0.7 MPa. A moving load subroutine is written to make the aforementioned contact pressure surface move longitudinally along the road at a constant speed, simulating the complete process of a vehicle passing over the sensor. To address the issue that the vehicle may not be directly pressing against the sensor, a lateral offset is introduced into the simulation. That is, the lateral coordinate of the load centerline. Not just fixed at 0, but at The variation within the interval simulates the distribution of the vehicle's wheel tracks.

[0159] Step S104: Calculate the geometric parameter labels of the hidden road defects model, wherein the geometric parameter labels include the equivalent volume of the defects, the shape factor, and the depth of the defects.

[0160] It should be noted that the disease geometric parameter labels are a set of parameters used to characterize the geometric features of the disease. They serve as label data for subsequent model training and are associated with the response data. These parameters include the disease equivalent volume, shape factor, and disease burial depth.

[0161] The equivalent volume of a disease can be the volume of an irregularly shaped disease converted into a regular geometric shape (such as a sphere or cylinder), used to uniformly represent the scale of the disease.

[0162] The shape factor can be a parameter used to describe the degree of irregularity in the shape of a disease. It is calculated by the ratio of the actual surface area of ​​the disease to the surface area of ​​an equivalent regular geometric body. The larger the ratio, the more irregular the shape.

[0163] The depth of a road surface defect can be the vertical distance from the center of the defect to the road surface, and it is a key parameter for characterizing the location of the defect.

[0164] In the specific implementation, for each generated sample, the equivalent volume and shape factor are calculated, and its true geometric properties are used as the output label of the neural network. Volume integration is performed on the irregular diseased area to obtain the equivalent volume ( )for:

[0165]

[0166] Define shape factor ( () represents the horizontal projected area with vertical projected area The ratio is used to characterize the degree of flatness of the disease:

[0167]

[0168] in: The horizontal projected area, This represents the vertical projected area.

[0169] Step S105: Run dynamic calculations and extract response data from the strain sensor array nodes, associate the response data with the geometric parameter labels of the road defects, and construct a finite element simulation sample library for road defects.

[0170] It should be noted that dynamic calculations can be based on finite element models to simulate the dynamic mechanical response of road structures under moving loads and calculate the changes in parameters such as strain and stress over time.

[0171] It should be noted that the response data of the strain sensor array nodes can be the strain change data monitored by each node in the strain sensor array under load, which is used to reflect the impact of the defects on the mechanical structure of the road.

[0172] In the specific implementation, a parameterized script is used to automatically generate N simulation conditions in a loop. For each condition, dynamic calculations are performed to extract the time history response data of the sensor nodes throughout the entire load movement process, and the corresponding tags are associated with them. , and store in the disease characteristic simulation database.

[0173] This embodiment ensures the diversity and representativeness of simulation conditions through scientific sampling, covering different scales, forms, and load conditions of defects, avoiding model bias caused by a single sample. The constructed irregular defect model more closely resembles the real morphology of hidden defects on actual roads, improving the realism of the simulation samples. The setting of standard moving loads ensures the standardization and realism of strain response calculations, closely reflecting the stress conditions of actual roads. Combined with the lateral offset of the load, different vehicle driving conditions are simulated, enriching the strain response scenarios of the defect model and making the extracted response data more comprehensive. By calculating standardized defect geometric parameter labels, the geometric features of irregular defects are accurately quantified, providing clear label data for subsequent feature extraction and model training. Through dynamic calculations, realistic strain response data is obtained, accurately capturing the impact of defects on the mechanical performance of road structures. The response data is associated with defect geometric parameter labels to construct a complete and reliable simulation sample library.

[0174] refer to Figure 4 , Figure 4 This is a flowchart illustrating the third embodiment of the mechanically driven geometric correction method for hidden road defects according to the present invention.

[0175] Based on the first embodiment described above, in this embodiment, the multidimensional feature vector includes a spatial feature sub-vector, a frequency domain feature sub-vector, and a shape-sensitive feature sub-vector. Step S20 further includes:

[0176] Step S201: Based on the road defect finite element simulation sample library, segment the effective data segment and identify the key measurement points in the effective data segment.

[0177] It should be noted that the effective data segment refers to the data segment selected from the response data of the simulation sample library that can reflect the impact of defects on the mechanical structure of the road and has practical analytical value, while redundant data segments without defect impact are eliminated. The key measuring point refers to the strain sensor array node in the effective data segment with the most significant strain response (prominent strain peak value and strain change rate), whose monitoring data can most intuitively and accurately reflect the mechanical impact characteristics of defects.

[0178] In some embodiments, strain response data (including time series data) for each simulation condition are retrieved from the finite element simulation sample library of road defects. The original response data is segmented using the sliding window method. By setting a threshold, segments with strain values ​​exceeding the threshold are selected as valid data segments (strain response data corresponding to the defect-affected area), and invalid data segments without defect influence are removed. Based on the strain peak value and strain change rate of the valid data segments, key measuring points are identified, and 3-5 strain sensor array nodes with the largest strain peak value and the most significant strain change rate are selected as key measuring points to capture the core influence area data of the defect on the mechanical response of the road structure.

[0179] Furthermore, in order to accurately extract valid data segments and locate key measurement points, step S201 above may include:

[0180] Step S2011: For the simulation data in the road defect finite element simulation sample library, based on the time when the load passes through the center of the road body model, extract the data segments before and after a preset time as the effective simulation data segments.

[0181] Step S2012: For the response data of the strain sensor array nodes in the road defect finite element simulation sample library, a sliding window is used to monitor the continuous time-series response data. When the strain amplitude of any sensor node in the strain sensor array exceeds the preset strain threshold, it is marked as a vehicle entry event.

[0182] Step S2013: Based on the peak time of the vehicle entry event monitoring strain, take the peak time of the strain as the center, extract the time window before and after the preset duration in the response data, as the measured effective data segment of a single vehicle passage;

[0183] Step S2014: Define the strain sensor array node with the largest absolute value of the response peak within the simulated effective data segment and the measured effective data segment as the main response node and record its position to complete the identification of key measurement points.

[0184] In the specific implementation, for simulation data, the time when the load passes through the center of the model is used as the benchmark, and data before and after T seconds are extracted; for measured data, continuous time-series data collected by the strain sensor array is used. Sliding window monitoring is performed; if the strain amplitude of any sensor in the array exceeds a preset threshold... This is labeled "Vehicle Entry Event". A time window of T seconds before and after the peak time is extracted as the valid data segment for a single vehicle passage. Within this time window, the sensor node with the largest absolute value of the response peak is identified and defined as the main response node, and its position is recorded as... .

[0185] Step S202: Based on the key measurement points, extract the spatial domain features, frequency domain features and shape-sensitive features from the effective data segments to obtain spatial feature sub-vectors, frequency domain feature sub-vectors and shape-sensitive feature sub-vectors.

[0186] It should be noted that spatial domain features can be features extracted based on the spatial positional relationship and strain distribution law of key measuring points, reflecting the spatial range and intensity of the impact of the disease on the road structure.

[0187] Frequency domain features can be features extracted after converting time-domain strain signals into frequency-domain signals, reflecting the frequency distribution characteristics of strain response under the influence of disease.

[0188] Shape-sensitive features can be features extracted based on the waveform morphology of strain time series. They can sensitively capture strain waveform distortion caused by disease and reflect the morphological characteristics of disease.

[0189] Furthermore, in order to accurately mine features in each dimension, step S202 above may include:

[0190] Step S2021: Based on the key measuring points, extract the instantaneous strain values ​​of each main response node in the effective data segment at the moment when the strain peak is reached, and form an instantaneous strain vector;

[0191] Step S2022: Perform maximum value normalization on the instantaneous strain vector to obtain spatial feature sub-vectors;

[0192] Step S2023: Obtain the strain time history data of the master response node, and apply Fast Fourier Transform to convert the strain time history data from the time domain to the frequency domain to obtain the spectral coefficients;

[0193] Step S2024: Calculate the normalized energy proportions of the low-frequency band, mid-frequency band, and high-frequency band based on the spectral coefficients to form frequency domain feature sub-vectors;

[0194] Step S2025: Calculate the ratio of the longitudinal strain sensor reading to the transverse strain sensor reading of the master response node, and use the ratio as the shape-sensitive feature sub-vector.

[0195] In the specific implementation, to eliminate the direct influence of vehicle load on strain amplitude, only the spatial distribution pattern reflecting the extent of the damage is retained, and a normalized spatial vector is constructed. Assuming the sensor array contains N sensors, the peak value of all sensors at the main response node is extracted. The instantaneous strain values ​​form a vector. The vector is then normalized to its maximum value to obtain the spatial feature sub-vectors. :

[0196]

[0197] Given that pavement structure defects alter its local damping and equivalent stiffness, leading to changes in the frequency components of the dynamic response, time-history data from the principal response nodes are selected. The fast Fourier transform is applied to convert it from the time domain to the frequency domain, as shown in the following formula:

[0198]

[0199] in, Represents the spectral coefficients. Indicates the frequency domain point number. This indicates the number of sampling points in the strain time history data. This represents the discretized strain time history data of the master response node. Indicates the sampling point number. It represents the imaginary unit.

[0200] Define total energy and calculate low-frequency band. Mid-frequency band and high frequency band The normalized energy percentage constitutes the frequency domain feature vector, as shown in the following formula:

[0201]

[0202]

[0203] in, Represents the frequency domain eigenvectors. This represents the total energy of the frequency domain data of the strain at the master response node. Indicates frequency, This represents the energy integral value within each frequency band. Indicates low frequency band. Indicates the mid-frequency band. Indicates the high-frequency band.

[0204] To capture the shape factors of defects (e.g., distinguishing between flattened voids and deep cavities), the Poisson effect characteristics of asphalt pavements are utilized. Longitudinal strain sensor readings at the main response location are selected. and transverse strain sensor readings The ratio of the two is calculated as the shape-sensitive feature sub-vector. :

[0205]

[0206] Step S203: Fuse the spatial feature vector, frequency domain feature vector and shape-sensitive feature vector to obtain a fused feature vector.

[0207] In the specific implementation, the feature sub-vectors of the above three dimensions are concatenated to form the high-dimensional fused feature vector that is finally input into the neural network. :

[0208]

[0209] in, Represents the fused feature vector. This indicates a splicing operation. Indicates the feature dimension.

[0210] This embodiment achieves accurate data screening through effective data segmentation and key measurement point identification, improving the efficiency and targeting of feature extraction. Through multi-dimensional feature extraction, it comprehensively captures the spatial, frequency domain, and shape features of diseases, enriching the representational dimensions of features. Through feature fusion, it integrates features from various dimensions to form a fusion feature vector that comprehensively reflects the essential attributes of diseases, solving the problems of traditional feature extraction being singular, highly redundant, and lacking in representational ability, and providing high-quality feature input for subsequent dataset construction and model training.

[0211] Furthermore, embodiments of the present invention also propose a computer-readable storage medium storing a mechanics-driven geometric correction program for hidden road defects. When the mechanics-driven geometric correction program for hidden road defects is executed by a processor, it implements the steps of the mechanics-driven geometric correction method for hidden road defects as described above.

[0212] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0213] The aforementioned computer-readable storage medium may be included in a mechanically driven road hidden defect geometry correction device; or it may exist independently and not assembled into a mechanically driven road hidden defect geometry correction device.

[0214] Furthermore, this invention also proposes a computer program product, including a mechanics-driven geometric correction program for hidden road defects. When the mechanics-driven geometric correction program for hidden road defects is executed by a processor, it implements the steps of the mechanics-driven geometric correction method for hidden road defects as described above.

[0215] The specific implementation of the computer program product of the present invention is basically the same as the embodiments of the above-mentioned mechanically driven geometric correction method for hidden road defects, and will not be described again here.

[0216] Reference Figure 5 , Figure 5 This is a structural block diagram of the first embodiment of the mechanically driven geometric correction device for hidden road defects according to the present invention.

[0217] like Figure 5 As shown, the mechanically driven geometric correction device for hidden road defects proposed in this embodiment of the invention includes:

[0218] Sample library construction module 10 is used to construct a sample library for finite element simulation of road defects;

[0219] Feature extraction module 20 is used to extract features based on the road disease finite element simulation sample library and fuse multi-dimensional feature vectors to obtain fused feature vectors;

[0220] The model training module 30 is used to construct a dataset based on the fused feature vector and train the original neural network model based on the dataset to obtain a disease parameter inversion proxy model. The disease parameter inversion proxy model is a fully connected multilayer perceptron network architecture.

[0221] The monitoring module 40 is used to identify abnormal media areas under the road surface using three-dimensional ground penetrating radar and activate the corresponding strain sensor array nodes below the abnormal media areas.

[0222] The data acquisition module 50 is used to acquire strain data monitored by the strain sensor array nodes, extract and fuse the strain data to obtain a fused feature vector to be identified.

[0223] The geometric parameter output module 60 is used to input the fused feature vector to be identified into the disease parameter inversion proxy model for inversion operation, output the target geometric parameters, and correct the original geometric parameters based on the target geometric parameters to obtain the corrected road hidden disease geometric parameters.

[0224] This embodiment constructs a finite element simulation sample library for road defects. Based on this library, features are extracted and multi-dimensional feature vectors are fused to obtain fused feature vectors. A dataset is constructed based on these fused feature vectors, and an original neural network model is trained on the dataset to obtain a defect parameter inversion proxy model. The defect parameter inversion proxy model is a fully connected multilayer perceptron network architecture. It uses 3D ground-penetrating radar to identify abnormal areas of the medium under the road surface and activates the corresponding strain sensor array nodes below these abnormal areas. The strain data monitored by the strain sensor array nodes is acquired, and features are extracted and fused from the strain data to obtain a fused feature vector to be identified. This fused feature vector is input into the defect parameter inversion proxy model for inversion calculation, outputting target geometric parameters. The geometric parameters are corrected from the original geometric parameters to obtain the corrected geometric parameters of hidden road defects. This embodiment effectively taps into the deep value of strain sensor array monitoring data, breaking the limitation of existing sensors that are only used for single-point threshold alarms. It integrates the advantages of three-dimensional ground-penetrating radar and strain sensors, making up for the shortcomings of single detection methods and effectively supplementing the inaccuracy of ground-penetrating radar in detecting the boundaries of hidden defects. Through strain data-driven precise correction of defect geometric parameters, the recognition accuracy of hidden defect geometric morphology is improved. By combining finite element simulation and neural networks, an efficient defect parameter inversion proxy model is constructed to replace time-consuming finite element iterative calculations, which can significantly improve the efficiency of defect detection and parameter inversion. By realizing the precise quantification of road hidden defect geometric parameters, a scientific basis is provided for the calculation of maintenance engineering volume and the formulation of maintenance plans.

[0225] The mechanically driven road hidden defect geometric correction device provided in this application, employing the mechanically driven road hidden defect geometric correction method described in the above embodiments, can solve the technical problem of mechanically driven road hidden defect geometric correction. Compared with the prior art, the beneficial effects of the mechanically driven road hidden defect geometric correction device provided in this application are the same as those of the mechanically driven road hidden defect geometric correction method provided in the above embodiments, and other technical features in the mechanically driven road hidden defect geometric correction device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0226] It should be understood that the above are merely illustrative examples and do not constitute any limitation on the technical solution of the present invention. In specific applications, those skilled in the art can make settings as needed, and the present invention does not impose any restrictions on this.

[0227] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this invention. In practical applications, those skilled in the art can select some or all of the workflow to achieve the purpose of this embodiment according to actual needs, and no restrictions are imposed here.

[0228] In addition, for technical details not described in detail in this embodiment, please refer to the mechanically driven geometric correction method for hidden road defects provided in any embodiment of the present invention, which will not be repeated here.

[0229] 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, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0230] It should be noted that the user information (including but not limited to user device information, user personal information, user location information, user behavior information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0231] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0232] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory / random access memory, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0233] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A method for geometric correction of hidden road defects based on mechanical drive, characterized in that, The method includes: Construct a finite element simulation sample library for road defects; Based on the aforementioned road disease finite element simulation sample library, feature extraction and fusion of multi-dimensional feature vectors are performed to obtain fused feature vectors. A dataset is constructed based on the fused feature vectors, and an original neural network model is trained based on the dataset to obtain a disease parameter inversion proxy model. The disease parameter inversion proxy model is a fully connected multilayer perceptron network architecture. The abnormal medium area under the road surface is identified by three-dimensional ground penetrating radar, and the corresponding strain sensor array node below the abnormal medium area is activated. The strain data monitored by the strain sensor array nodes is acquired, and the strain data is subjected to feature extraction and fusion to obtain the fused feature vector to be identified. The feature vector to be identified is input into the disease parameter inversion proxy model for inversion operation, and the target geometric parameters are output. The original geometric parameters are then corrected based on the target geometric parameters to obtain the corrected road hidden disease geometric parameters. The original geometric parameters are the road hidden disease geometric parameters obtained by preliminary identification by three-dimensional ground penetrating radar and have not been corrected. The multidimensional feature vector includes spatial feature sub-vectors, frequency domain feature sub-vectors, and shape-sensitive feature sub-vectors; the step of extracting features based on the road distress finite element simulation sample library and fusing the multidimensional feature vectors to obtain the fused feature vector includes: Based on the road disease finite element simulation sample library, effective data segments are segmented, and key measurement points in the effective data segments are identified; Based on the key measurement points, spatial domain features, frequency domain features, and shape-sensitive features are extracted from the effective data segments to obtain spatial feature sub-vectors, frequency domain feature sub-vectors, and shape-sensitive feature sub-vectors. The spatial feature vector, frequency domain feature vector, and shape-sensitive feature vector are fused to obtain a fused feature vector.

2. The method for geometric correction of hidden road defects based on mechanical drive as described in claim 1, characterized in that, The construction of the finite element simulation sample library for road defects includes: The input variable space is sampled to generate multiple simulation conditions, and a road ontology model for each simulation condition is constructed. The input variable space includes the baseline geometric parameters of the defects, the depth of the defects, the irregular disturbance coefficient, and the lateral offset of the load. In the road body model, a disease center point and a reference radius are set. An irregular disease boundary surface is generated based on the disease center point and the reference radius. Based on the irregular disease boundary surface, an irregular road hidden disease model is constructed in the road body model. Set standard moving loads and lateral offsets for the road body model; Calculate the geometric parameter labels of the road hidden disease model, wherein the geometric parameter labels include the equivalent volume of the disease, the shape factor, and the burial depth of the disease; The dynamic calculations are performed and the response data of the strain sensor array nodes are extracted. The response data is then associated with the geometric parameter labels of the road defects to construct a finite element simulation sample library for road defects.

3. The method for geometric correction of hidden road defects based on mechanical drive as described in claim 2, characterized in that, The process involves setting a disease center point and a reference radius in the road body model, generating an irregular disease boundary surface based on the disease center point and the reference radius, and constructing an irregularly shaped road hidden disease model in the road body model based on the irregular disease boundary surface, including: The road body model is structurally divided from top to bottom to form multiple structural layers, and each structural layer is assigned material properties. The structural layers include surface layer, base layer, subbase layer and soil base layer, and the material properties include elastic modulus, Poisson's ratio and material density. Boundary conditions are applied to the road body model, and the road body model is meshed. The key areas in the road body model are divided into fine meshes, and the far-field areas are divided into sparse meshes. The fine meshes and the sparse meshes are smoothly connected through transition meshes. The key areas are the layout areas of strain sensor array nodes and the disease generation areas. Within the road body model, a disease center point is set, a local spherical coordinate system is established, and the surface radius function of the reference ellipsoid is defined. A spatial random noise field is superimposed on the reference radius of the reference ellipsoid to generate an irregular defect boundary surface, as shown in the following formula: in, Represents the irregular boundary surface of disease. Represents the surface radius function. This represents a normalized three-dimensional random noise function. This represents the irregular disturbance coefficient. and These represent the azimuth and polar angles of the local spherical coordinate system, respectively. The material properties of the units in the road body model that are surrounded by the irregular defect boundary surface are reduced, and an irregular road hidden defect model is formed based on the units after the material property reduction.

4. The method for geometric correction of hidden road defects based on mechanical drive as described in claim 1, characterized in that, The effective data segment includes a simulation effective data segment and a measured effective data segment; the step of segmenting the effective data segment based on the road distress finite element simulation sample library and identifying key measurement points in the effective data segment includes: For the simulation data in the finite element simulation sample library of road defects, the data segments before and after a preset time are extracted as the effective simulation data segments, based on the time when the load passes through the center of the road body model. For the response data of the strain sensor array nodes in the road defect finite element simulation sample library, a sliding window is used to monitor the continuous time-series response data. When the strain amplitude of any sensor node in the strain sensor array exceeds a preset strain threshold, it is marked as a vehicle entry event. Based on the peak time of the vehicle entry event monitoring response, and taking the peak time as the center, a time window of a preset duration is extracted from the response data as the effective data segment of a single vehicle passage. The strain sensor array node with the largest absolute value of the response peak within the simulated effective data segment and the measured effective data segment is defined as the main response node and its position is recorded to complete the identification of key measurement points.

5. The method for geometric correction of hidden road defects based on mechanical drive as described in claim 4, characterized in that, The step of extracting spatial domain features, frequency domain features, and shape-sensitive features from the effective data segment based on the key measurement points to obtain spatial feature sub-vectors, frequency domain feature sub-vectors, and shape-sensitive feature sub-vectors includes: Based on the key measuring points, the instantaneous strain values ​​at the moment when each main response node in the effective data segment reaches the strain peak are extracted to form an instantaneous strain vector; The instantaneous strain vector is normalized to obtain the spatial feature sub-vector; Obtain the strain time history data of the master response node, and apply Fast Fourier Transform to transform the strain time history data from the time domain to the frequency domain to obtain the spectral coefficients, referring to the following formula: in, Represents the spectral coefficients. Indicates the frequency domain point number. This indicates the number of sampling points in the strain time history data. This represents the discretized strain time history data of the master response node. Indicates the sampling point number. Represents the imaginary unit; Based on the aforementioned spectral coefficients, the normalized energy proportions of the low-frequency band, mid-frequency band, and high-frequency band are calculated to form frequency domain feature vectors, as shown in the following formula: in, Represents the frequency domain eigenvectors. This represents the total energy of the frequency domain data of the strain at the master response node. Indicates frequency, This represents the energy integral value within each frequency band. Indicates low frequency band. Indicates the mid-frequency band. Indicates high frequency band; Calculate the ratio of the longitudinal strain sensor reading to the transverse strain sensor reading at the master response node, and use the ratio as a shape-sensitive feature sub-vector.

6. The method for geometric correction of hidden road defects based on mechanical drive as described in any one of claims 1 to 3, characterized in that, The process of constructing a dataset based on the fused feature vectors and training an original neural network model based on the dataset to obtain a disease parameter inversion proxy model includes: A dataset is constructed based on the fused feature vectors and the corresponding geometric parameter labels of road defects in the finite element simulation sample library, and the dataset is divided into a training set, a validation set, and a test set. An original neural network model is generated based on a fully connected multilayer perceptron network architecture. The original neural network model includes an input layer, multiple hidden layers, and an output layer. The dimension of the input layer matches the dimension of the fused feature vector, and the dimension of the output layer matches the dimension of the disease geometric parameter label. The mean squared error with second-order regularization is set as the loss function. The backpropagation algorithm combined with the adaptive moment estimation optimizer is used to iteratively update the network parameters of the original neural network model using the training set, and the model training effect is monitored through the validation set. When the loss value calculated by the loss function on the validation set no longer decreases, or when the number of model training iterations reaches the preset number of iterations, stop model training and save the current optimal network parameters to obtain the disease parameter inversion proxy model.

7. A mechanically driven road hidden defect geometric correction device that applies the mechanically driven road hidden defect geometric correction method according to any one of claims 1 to 6, characterized in that, The device includes: The sample library construction module is used to build a sample library for finite element simulation of road defects; The feature extraction module is used to extract features based on the road disease finite element simulation sample library and fuse multi-dimensional feature vectors to obtain fused feature vectors; The model training module is used to construct a dataset based on the fused feature vectors and train the original neural network model based on the dataset to obtain a disease parameter inversion proxy model. The disease parameter inversion proxy model is a fully connected multilayer perceptron network architecture. The monitoring module is used to identify abnormal areas of the medium under the road surface using three-dimensional ground penetrating radar and activate the corresponding strain sensor array nodes below the abnormal areas of the medium. The data acquisition module is used to acquire strain data monitored by the strain sensor array nodes, extract and fuse the strain data to obtain a fused feature vector to be identified. The geometric parameter output module is used to input the fused feature vector to be identified into the disease parameter inversion proxy model for inversion operation, output the target geometric parameters, and correct the original geometric parameters based on the target geometric parameters to obtain the corrected road hidden disease geometric parameters.

8. A mechanically driven geometric correction device for hidden road defects, characterized in that, The mechanically driven road hidden defect geometric correction device includes: a memory, a processor, and a mechanically driven road hidden defect geometric correction program stored in the memory. The processor is used to run the mechanically driven road hidden defect geometric correction program, which is configured to implement the mechanically driven road hidden defect geometric correction method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a mechanics-driven geometric correction program for hidden road defects, which, when executed by a processor, implements the mechanics-driven geometric correction method for hidden road defects as described in any one of claims 1 to 6.

Images