Tunnel deformation detection method based on convolutional neural network and machine vision sensor

Abstract

The present application relates to the field of tunnel engineering safety monitoring, and particularly relates to a tunnel deformation detection method based on convolutional neural network and machine vision sensor, comprising: collecting ring deformation data, longitudinal deformation data and cross-section displacement inclination angle data through the machine vision sensor arranged in the tunnel; applying convolutional neural network to pre-process the ring deformation data and the longitudinal deformation data; converting the cross-section displacement inclination angle data into discretized matrix data and performing visual processing based on differential geometry theory; constructing a tunnel deformation monitoring model and performing simulation calculation; determining the tunnel deformation speed by comparing the simulation calculation result with the actually collected data, the present application deeply integrates the differential geometry theory and the convolutional neural network, regards the tunnel cross-section as a two-dimensional Riemannian manifold embedded in a three-dimensional Euclidean space, accurately describes the deformation characteristics of complex surfaces, and realizes all-around and high-precision monitoring and prediction of the tunnel deformation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of tunnel engineering safety monitoring, and particularly relates to a tunnel deformation detection method based on a convolutional neural network and a machine vision sensor. BACKGROUND

[0002] As important transportation infrastructure, the structural safety of tunnels is related to people's life and property safety and national economic development. Tunnels are susceptible to deformation, cracks and other diseases during service due to various factors such as geological conditions, construction quality, service life and external environment. Traditional tunnel deformation monitoring methods mainly include manual inspection, mechanical displacement meters, strain gauges, etc. These methods generally have low monitoring accuracy, poor efficiency, limited coverage, and cannot achieve real-time monitoring.

[0003] In recent years, with the development of computer vision and deep learning technology, image-based tunnel deformation detection methods have gradually emerged. However, existing image-based detection methods still have some limitations: first, they cannot achieve all-around deformation detection and mostly only focus on deformation in a certain direction; second, they lack systematic analysis and characterization of deformation data, making it difficult to accurately reflect the overall deformation state of the tunnel; third, the deformation prediction accuracy is not high, making it difficult to discover potential risks in a timely manner.

[0004] In addition, existing methods generally use simple Euclidean geometry models to process tunnel cross-sectional deformation data, which cannot accurately describe the deformation characteristics of complex surfaces. At the same time, existing methods have limited fusion processing capabilities for multi-source heterogeneous data, making it difficult to comprehensively analyze various sensor data.

[0005] Therefore, there is an urgent need for a method that can comprehensively, accurately and efficiently monitor the deformation state of a tunnel and predict the deformation trend to improve the level of tunnel safety management. SUMMARY

[0006] The purpose of the present application is to provide a tunnel deformation detection method based on a convolutional neural network and a machine vision sensor, which realizes all-around, high-precision detection and prediction of tunnel deformation by integrating differential geometry theory and deep learning technology, and solves the technical problems of low monitoring accuracy, limited coverage and weak prediction ability in existing technologies.

[0007] The present application provides a tunnel deformation detection method based on a convolutional neural network and a machine vision sensor, which comprises:

[0008] A machine vision sensor arranged in the tunnel is used to collect tunnel ring deformation data, longitudinal deformation data and cross-sectional displacement inclination angle data;

[0009] The circumferential deformation data, the longitudinal deformation data, and the cross-sectional displacement tilt angle data are input into a computer for processing, including:

[0010] The circumferential deformation data and the longitudinal deformation data are preprocessed using a convolutional neural network in a computer.

[0011] The cross-sectional displacement tilt angle data is converted into discretized matrix data in a computer and visualized based on differential geometry theory to obtain visualized cross-sectional displacement tilt angle data.

[0012] A tunnel deformation monitoring model is constructed in a computer, and the processed circumferential deformation data, longitudinal deformation data, and visualized cross-sectional displacement tilt angle data are input into the tunnel deformation monitoring model for simulation calculation.

[0013] The tunnel deformation rate is determined by comparing the simulated calculation results with the actual collected data.

[0014] Preferably, the preprocessing of the circumferential deformation data and the longitudinal deformation data using a convolutional neural network includes:

[0015] A preprocessing network containing an encoder and a decoder is constructed and runs on the TensorFlow deep neural network platform;

[0016] Multi-layer convolutional layers are used to extract features from the image, and the extracted features are then subjected to dimensionality reduction.

[0017] The decoder outputs a matrix of the original image size, confidence scores, and bounding box data.

[0018] A detection network consisting of a residual network and a fully connected layer is constructed to extract features and output results from the preprocessed data.

[0019] Preferably, the step of converting the cross-sectional displacement tilt angle data into discretized matrix data and performing visualization processing based on differential geometry theory includes:

[0020] The tunnel cross-section is considered as a two-dimensional Riemannian manifold embedded in three-dimensional Euclidean space;

[0021] Establish a polar coordinate system (r, θ) with the center of the tunnel as the origin;

[0022] The cross-sectional displacement tilt angle data is represented as a vector field defined on the Riemannian manifold;

[0023] Calculate the geometric invariants of the Riemannian manifold, including the principal curvatures and Gaussian curvatures;

[0024] Discretize the geometric invariants into a matrix representation;

[0025] The matrix data is visualized and enhanced based on geodesic curvature flow theory to highlight areas of abnormal deformation.

[0026] Preferably, the construction of the tunnel deformation monitoring model includes:

[0027] A deformation prediction system is built based on convolutional neural networks and machine learning models;

[0028] The deformation prediction system includes a feature extraction module, a regression prediction module, and a prediction correction module;

[0029] The feature extraction module takes the circumferential deformation data, the longitudinal deformation data, and the visualized cross-sectional displacement tilt angle data as feature inputs, classifies, integrates, and extracts the features to obtain a feature sequence.

[0030] The regression prediction module predicts the tunnel deformation based on the feature sequence.

[0031] The prediction and correction module corrects the tunnel deformation based on historical deformation patterns.

[0032] Preferably, the visualization enhancement processing of the matrix data based on geodesic curvature flow theory further includes:

[0033] Acquire cross-sectional displacement and tilt angle data at multiple time points;

[0034] Based on the connection theory and covariant derivative in differential geometry, the rate of change of the cross-sectional displacement tilt angle is calculated.

[0035] Construct a change rate feature matrix and extract the main modes of deformation;

[0036] Map the geometric invariant matrix and the rate of change feature matrix to a color space;

[0037] Multi-scale analysis techniques are applied to achieve hierarchical visualization from global to local levels;

[0038] Potential risk areas are automatically identified based on curvature outliers, and risk levels are classified accordingly.

[0039] Preferably, the acquisition of tunnel circumferential deformation data, longitudinal deformation data, and cross-sectional displacement tilt angle data by machine vision sensors installed in the tunnel includes:

[0040] A machine vision sensor is installed longitudinally at predetermined intervals in the tunnel;

[0041] Each of the machine vision sensors employs an area array camera to measure the circumferential deformation and cracking of the tunnel using image comparison analysis.

[0042] Data is collected at preset time intervals, and each collection of data is treated as a set of data.

[0043] The collected data is transmitted to a network terminal server via the network.

[0044] Preferably, determining the tunnel deformation rate by comparing the simulated calculation results with the actual collected data includes:

[0045] A three-dimensional coordinate system is established in the computer. The three-dimensional coordinate system includes an X-axis, a Y-axis, and a Z-axis, where the X-axis is the longitudinal direction of the tunnel, the Y-axis is the circumferential direction, and the Z-axis is the cross-sectional direction of the tunnel.

[0046] Simulations were performed on the processed circumferential deformation data, longitudinal deformation data, and visualized cross-sectional displacement tilt angle data, respectively, with the Y-axis, X-axis, and Z-axis as the central axes, to obtain simulation data.

[0047] Simulated circumferential deformation images, simulated longitudinal deformation images, and simulated cross-sectional displacement tilt angle images at different acquisition times are plotted in the three-dimensional coordinate system.

[0048] The simulated image is compared with the actual image collected at the corresponding time point;

[0049] When the data changes of the simulated image and the acquired image at the same time are consistent, the tunnel deformation rate at this time is determined to be zero.

[0050] When there is a difference between the data of the simulated image and the acquired image at the corresponding time, the tunnel deformation rate is calculated based on the difference value.

[0051] Preferably, the regression prediction module predicts tunnel deformation based on the feature sequence, including:

[0052] Obtain the historical actual deformation corresponding to the feature sequence;

[0053] Establish a random forest model in the feature space;

[0054] The feature sequence is classified using the random forest model to obtain sub-feature sequences;

[0055] The correspondence between sub-feature sequences and actual deformation amounts is obtained by training the random forest model described above.

[0056] The deformation prediction model is obtained by weighting and reconstructing all sub-feature sequences with the actual deformation.

[0057] The feature sequence is predicted based on the deformation prediction model to obtain the corresponding predicted deformation amount.

[0058] Preferably, the prediction correction module corrects the tunnel deformation based on historical deformation development patterns, including:

[0059] Predict the deformation amount at different times by using historical deformation data of the tunnel.

[0060] The change in prediction error at the current moment is obtained based on the predicted deformation at different times.

[0061] Integrating the change in prediction error yields the range of prediction error variation.

[0062] The predicted deformation at the current moment is corrected based on the range of change of the prediction error;

[0063] When the variation range of the prediction error exceeds a preset threshold, the tunnel deformation prediction model is updated.

[0064] As a preferred option, it also includes:

[0065] Define safety thresholds and risk level standards;

[0066] The tunnel deformation status is monitored in real time and compared with the safety threshold.

[0067] When abnormal deformation is detected, an early warning mechanism is triggered;

[0068] Provides visual positioning information of the deformable area;

[0069] Generate a detailed report that includes deformation type, deformation degree, and risk assessment.

[0070] The beneficial effects of this invention include:

[0071] 1. Comprehensive monitoring: By collecting and integrating three key data points—circumferential deformation, longitudinal deformation, and cross-sectional displacement and tilt angle—comprehensive monitoring of tunnel deformation is achieved, overcoming the limitations of traditional single-dimensional detection.

[0072] 2. High-precision representation: The Riemannian manifold representation method based on differential geometry theory is adopted, which regards the tunnel cross section as a two-dimensional Riemannian manifold embedded in three-dimensional Euclidean space, accurately describing the deformation characteristics of complex surfaces and solving the problem of insufficient representation capability of traditional Euclidean geometric models.

[0073] 3. Highly efficient analysis: By designing a specific convolutional neural network structure, the efficient extraction and analysis of tunnel deformation features are achieved, which greatly improves the accuracy and efficiency of deformation recognition.

[0074] 4. Predictive capability: Combining time-series deep learning technology, it predicts future deformation trends based on historical deformation data, identifying potential risks up to 48 hours in advance, providing strong support for safety management.

[0075] 5. Cost reduction: Reduces the frequency of manual inspections, lowers maintenance costs by approximately %, and improves the reliability and coverage of safety monitoring.

[0076] 6. Engineering adaptability: The system has good adaptability to different tunnel shapes and deformation modes, and can be applied to different engineering scenarios without a lot of manual adjustment. Attached Figure Description

[0077] Figure 1 This is an overall flowchart of the tunnel deformation detection method based on convolutional neural networks and machine vision sensors according to the present invention. Detailed Implementation

[0078] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0079] Reference Figure 1 As shown, this invention provides a tunnel deformation detection method based on convolutional neural networks and machine vision sensors, comprising the following steps:

[0080] Machine vision sensors installed in the tunnel collect circumferential deformation data, longitudinal deformation data, and cross-sectional displacement tilt angle data. These data are then input into a computer for processing. A convolutional neural network is applied to preprocess the circumferential and longitudinal deformation data. The cross-sectional displacement tilt angle data is converted into discretized matrix data and visualized based on differential geometry theory to obtain visualized cross-sectional displacement tilt angle data. A tunnel deformation monitoring model is constructed in the computer, and the processed circumferential deformation data, longitudinal deformation data, and visualized cross-sectional displacement tilt angle data are input into the model for simulation calculation. By comparing the simulation calculation results with the actual collected data, the tunnel deformation rate is determined.

[0081] This invention employs a comprehensive analysis of multiple types of deformation data, providing a more complete picture of tunnel deformation compared to traditional methods. Specifically, circumferential deformation data reflects the radial changes in the tunnel cross-section, longitudinal deformation data reflects the tunnel's expansion, contraction, and displacement along the longitudinal direction, while cross-sectional displacement tilt angle data reflects the torsion and tilt of the tunnel cross-section. By comprehensively analyzing these three types of data, a complete picture of tunnel deformation can be constructed.

[0082] The specific implementation method of using a convolutional neural network to preprocess the circumferential deformation data and the longitudinal deformation data in this invention is as follows:

[0083] A preprocessing network consisting of an encoder and a decoder is constructed and runs on the TensorFlow deep neural network platform. Multi-layer convolutional layers are used to extract features from the image, and the extracted features are subjected to dimensionality reduction. The decoder outputs a matrix of the original image size, confidence scores, and bounding box data. A detection network consisting of a residual network and fully connected layers is constructed to extract features from the preprocessed data and output the results.

[0084] In a preferred embodiment of the present invention, the preprocessing network adopts the UNet architecture. The encoder consists of 5 convolutional layers, each followed by a BatchNormalization layer and a ReLU activation function. The first convolutional layer has a kernel size of 7×7, a stride of 2, and 64 output channels. The second to fifth convolutional layers all have kernel sizes of 3×3, a stride of 2, and 1, 6, 512, and 512 output channels, respectively. The decoder adopts a transposed convolutional structure, containing 5 transposed convolutional layers, corresponding to the 5 convolutional layers of the encoder, to achieve feature recovery.

[0085] The residual network in the detection network adopts the ResNet-50 architecture and contains multiple residual blocks. Each residual block contains three convolutional layers and skip connections. The fully connected layers consist of two layers. The first layer has 10 neurons, and the number of output channels in the second layer is determined according to the detection task. For classification tasks involving circumferential and longitudinal deformations, the number of output channels is set to the number of deformation types; for location regression tasks, the number of output channels is set to the number of parameters of the bounding box (usually 4, representing the coordinates of the top-left and bottom-right corners).

[0086] The convolutional neural network structure of this invention adopts a combination of encoder-decoder and residual network, which effectively solves the gradient vanishing problem in deep network training and improves the efficiency and accuracy of feature extraction.

[0087] The specific implementation method for converting the cross-sectional displacement tilt angle data into discretized matrix data and performing visualization processing based on differential geometry theory in this invention is as follows:

[0088] The tunnel cross-section is considered as a two-dimensional Riemannian manifold embedded in three-dimensional Euclidean space; a polar coordinate system (r, θ) is established with the tunnel center as the origin; the cross-section displacement tilt angle data is represented as a vector field defined on the Riemannian manifold; the geometric invariants of the Riemannian manifold, including principal curvature and Gaussian curvature, are calculated; the geometric invariants are discretized into a matrix representation; the matrix data is visualized and enhanced based on geodesic curvature flow theory to highlight abnormal deformation areas.

[0089] In a preferred embodiment of the present invention, the Riemannian manifold characterization of the tunnel cross section is performed using a parameterization method.

[0090] Assume the tunnel cross-section can be represented in polar coordinates as follows:

[0091] ,

[0092] in, Radial distance, in meters (m), representing the distance from the center of the tunnel to the measuring point; Angle, measured in radians (rad), represents the angle of the measurement point relative to a reference direction; This is a height function, with units of meters (m), representing the vertical displacement of the cross-section.

[0093] The metric tensor of this parameterized surface can be represented as:

[0094] ,

[0095] in, To measure a tensor, it is a The matrix represents the local metric on the parameterized surface; express right The partial derivative of , in dimensionless units, represents the rate of change of vertical displacement with respect to radial distance; express right The partial derivative of , in meters (m), represents the rate of change of vertical displacement with respect to angle.

[0096] Based on this metric tensor, the geometric invariants of the Riemannian manifold, including the Gaussian curvature, can be calculated. and mean curvature :

[0097] ,

[0098] ,

[0099] in, Gaussian curvature, in units of , represents the intrinsic curvature of the surface; The mean curvature is expressed in units of 1. , represents the intrinsic curvature of the surface; The second fundamental form of a surface is a The matrix represents the external geometry of the surface; Representation matrix The determinant; Represents the metric tensor The determinant; For measuring tensors The inverse matrix; Representation matrix The trace is the sum of the elements on the main diagonal.

[0100] In practical applications, the measurement points on the tunnel cross-section are discrete, thus requiring the construction of a continuous parametric surface using interpolation methods. This invention employs the radial basis function (RBF) interpolation method to interpolate the discrete measurement points into a continuous surface:

[0101] ,

[0102] in, The height function obtained by interpolation, in meters (m); The number of measurement points; The weighting coefficients, in meters (m), are determined by solving a system of linear equations. For radial functions, the present invention preferably uses a Gaussian function. ; Shape parameter, unit is m -2 The value is determined based on the point distribution density, and generally ranges from 0.1 to 10. For the first The coordinates of the measurement points; Point Time Euclidean distance, in meters (m).

[0103] The calculated geometric invariants are discretized into a matrix representation, and a matrix is ​​constructed. :

[0104] ,

[0105] in, The first geometric invariant matrix Line 1 Column elements; For point Gaussian curvature at a given point, in meters. -2 ; For point The average curvature at that point, in meters. -1 ; and Points The principal curvature at the point, in meters. -1 ; These are the coordinates of the discrete sampling points.

[0106] Visualization enhancement of matrix data is performed based on geodesic curvature flow theory, using the curvature flow equation:

[0107] ,

[0108] in, Representation of height function For time parameters The partial derivatives, in units of m / s; The Laplace-Beltrami operator is a generalization of the Laplace operator on surfaces. The weighting parameter is in m units. -2 The effect of Gaussian curvature on evolution is controlled, and its value is generally between 0.5 and 2.0. Gaussian curvature, in meters. -2 ; This is the evolution time parameter, in seconds (s).

[0109] By solving this partial differential equation, the deformation data can be enhanced and the abnormal areas highlighted. In numerical implementation, the equation is discretized using the finite difference method and solved iteratively.

[0110] This invention uses differential geometry theory to model the deformation of tunnel cross sections. Compared with the traditional Euclidean geometry method, it can more accurately describe the deformation characteristics of complex curved surfaces and provide a more reliable mathematical basis for deformation analysis.

[0111] The specific implementation method for constructing the tunnel deformation monitoring model in this invention is as follows:

[0112] A deformation prediction system is constructed based on convolutional neural networks and machine learning models. The deformation prediction system includes a feature extraction module, a regression prediction module, and a prediction correction module. The feature extraction module takes the circumferential deformation data, the longitudinal deformation data, and the visualized cross-sectional displacement tilt angle data as feature inputs, classifies, integrates, and extracts the features to obtain a feature sequence. The regression prediction module predicts the tunnel deformation based on the feature sequence. The prediction correction module corrects the tunnel deformation based on historical deformation development patterns.

[0113] In a preferred embodiment of the present invention, the feature extraction module adopts a convolutional neural network structure, including convolutional layers, pooling layers, and fully connected layers. Specifically, the convolutional layers use 3×3 convolutional kernels with a stride of 1 and padding of 1 to keep the feature map size unchanged; the pooling layers use max pooling with a 2×2 kernel size and a stride of 2 to halve the feature map size; and the fully connected layers flatten the two-dimensional feature map into a one-dimensional feature vector.

[0114] To handle multimodal data, this invention designs a feature fusion strategy. First, circumferential deformation data, longitudinal deformation data, and visualized cross-sectional displacement tilt angle data are extracted using independent convolutional networks. Then, the extracted feature vectors are concatenated into a comprehensive feature vector. Finally, a fully connected layer is used to perform dimensionality reduction and nonlinear transformation on the concatenated features to obtain the final feature sequence.

[0115] To ensure effective fusion of features from different modalities, this invention employs feature normalization technology. For the feature vector of each modality... Calculate its mean and standard deviation Then standardize:

[0116] ,

[0117] in, The standardized feature vectors are dimensionless. This is the original feature vector, and its unit varies depending on the feature type; For feature vectors The mean, in units of same; For feature vectors Standard deviation, units same.

[0118] Furthermore, to enhance the robustness of feature extraction, this invention introduces an attention mechanism. For each modality's feature map... Calculate its attention weights :

[0119] ,

[0120] in, is the attention weight matrix, which is dimensionless and takes values ​​in the range [0,1]. This is a feature map, whose dimensions are height × width × number of channels; The weight matrix contains learnable parameters. The bias vector is a learnable parameter; for A function used to normalize the output into a probability distribution.

[0121] Then, the attention weights are applied to the feature map:

[0122] ,

[0123] in, This is the weighted feature map; This represents the Hadamard product (element-wise multiplication), which is the multiplication of elements at corresponding positions.

[0124] Through the above design, the feature extraction module can effectively integrate multimodal data and extract key features of tunnel deformation, providing a reliable foundation for subsequent deformation prediction.

[0125] The visualization enhancement processing of the matrix data based on geodesic curvature flow theory in this invention also includes the following specific implementation methods:

[0126] Acquire cross-sectional displacement tilt angle data at multiple time points; calculate the rate of change of cross-sectional displacement tilt angle based on connection theory and covariant derivatives in differential geometry; construct a rate of change feature matrix to extract the main deformation modes; map the geometric invariant matrix and the rate of change feature matrix to a color space; apply multi-scale analysis technology to achieve hierarchical visualization from global to local; automatically identify potential risk areas based on curvature outliers and classify risk levels.

[0127] In a preferred embodiment of the present invention, the rate of change of the cross-sectional displacement tilt angle is calculated based on the connection theory. Assuming that the cross-sectional displacement tilt angles measured at times t and t+Δt are φ(r,θ,t) and φ(r,θ,t+Δt), respectively, their rate of change can be expressed as:

[0128] ,

[0129] in, φ represents the covariant derivative, describing the rate of change of the vector field on the manifold, in rad / s; φ(r,θ,t) is the displacement tilt angle at time t (r,θ), in radians (rad); φ(r,θ,t+Δt) is the displacement tilt angle at time t+Δt (r,θ), in radians (rad); Δt is the time interval, in seconds (s). The Christoffel notation represents a connection on a manifold, which is dimensionless; This represents the i-th component of the velocity vector, in m / s. Let be the j-th component of the displacement tilt angle vector field, in radians (rad).

[0130] Christoffel notation can be computed using a metric tensor:

[0131] ,

[0132] in, The symbol is Christoffel, dimensionless; Dimensionless, used to measure the components of a tensor; The components of the tensor inverse matrix are dimensionless. This represents the partial derivative with respect to the i-th coordinate; i, j, k, and l are indices, taking values ​​of 1 or 2, corresponding to parameters r and θ, respectively.

[0133] In practical applications, due to the finite measurement interval, the finite difference method can be used to approximate the rate of change:

[0134] ,

[0135] in, The rate of change of the displacement tilt angle is expressed in rad / s. This represents the displacement of the i-th coordinate within the time interval Δt, in meters (m).

[0136] To extract the main modes of deformation, this invention employs principal component analysis (PCA). The rate of change data is organized into a matrix. The rows represent different sampling points, and the columns represent measurements at different times.

[0137] ,

[0138] in, The rate of change matrix has dimensions of . , The number of sampling points. Number of time points; For the first A vector of the rate of change at each time point, with dimension . The unit is .

[0139] Calculate the covariance matrix:

[0140] ,

[0141] in, Let be the covariance matrix with dimension . The unit is ; The rate of change matrix; for The transpose of the matrix; This represents the number of time points.

[0142] For covariance matrix Perform eigenvalue decomposition:

[0143] ,

[0144] in, It is the covariance matrix; It is a diagonal matrix, and the diagonal elements are eigenvalues. The unit is ; is an eigenvector matrix, whose column vectors are the corresponding eigenvectors, and is dimensionless; for The transpose of .

[0145] forward Each principal component can be represented as:

[0146] ,

[0147] in, For the front A matrix composed of principal components, with dimension . ; For the first There are feature vectors with dimension . Dimensionless. Number of principal components. The choice is based on the cumulative variance explained ratio, and typically a value is chosen that maximizes the cumulative variance explained ratio. ~ of value.

[0148] The main modes of deformation can be represented as projections of the original data into the principal component space:

[0149] ,

[0150] in, The main modal matrix of the deformation has a dimension of . The unit is ; for The transpose of the matrix; This is the rate of change matrix.

[0151] For the automatic identification of risk areas, this invention employs a method based on curvature anomalies. The curvature anomaly index is defined as follows:

[0152] ,

[0153] in, For point The curvature anomaly index at the location is dimensionless. For point Gaussian curvature at , in units of ; For point The average curvature at that point, in units of ; The mean of the Gaussian curvature, in units of ; The mean of the average curvature, in units of ; The standard deviation of Gaussian curvature, in units of ; The standard deviation of the mean curvature, in units of ; This is a weighting parameter, dimensionless, typically ranging from 0.5 to 2.0, used to balance the contributions of Gaussian curvature and mean curvature.

[0154] When curvature anomaly index Exceeding the preset threshold At that time, the point is marked as a potential risk point. Threshold The choice of value is determined based on practical engineering experience, and is generally between 2.5 and 3.5. Based on the spatial distribution and degree of anomaly of risk points, risk levels can be further classified.

[0155] The dynamic feature extraction and risk identification method of the present invention can accurately capture the temporal evolution characteristics of tunnel deformation, promptly identify potential risk areas, and provide strong support for tunnel safety management.

[0156] The specific implementation method of this invention, which uses machine vision sensors installed in the tunnel to collect tunnel circumferential deformation data, longitudinal deformation data, and cross-sectional displacement tilt angle data, is as follows:

[0157] A machine vision sensor is installed at preset intervals along the longitudinal direction of the tunnel; each machine vision sensor uses an area array camera to measure the circumferential deformation and cracking of the tunnel through image comparison analysis; data is collected once at preset time intervals, and each collection of data is used as a set of data; the collected data is transmitted to a network terminal server through a network.

[0158] In a preferred embodiment of the invention, machine vision sensors are arranged every 10 meters along the longitudinal direction of the tunnel, forming a sensor network. Each sensor node includes a high-resolution area scan camera (resolution not less than 2048×1536 pixels) and an active light source compensation system. The area scan camera uses a global shutter CMOS sensor, which has high image quality and low noise level. The active light source compensation system consists of an LED array, which can provide stable illumination under different lighting conditions.

[0159] The sensor's data acquisition frequency is adaptively adjusted according to the tunnel's deformation rate. Normally, data is collected every 6 hours. When the deformation rate exceeds the warning threshold, the acquisition frequency is automatically increased to once per hour. Each acquisition includes circumferential cross-sectional images, longitudinal section images, and calibration point location information.

[0160] The acquired image data is processed using image comparison analysis. Specifically, the acquired images are first geometrically corrected and illumination equalized; then, the current image is registered with a reference image (usually the initial state or the image from the previous time point); next, the differences between the registered images are calculated, and deformation features are extracted; finally, based on the extracted features, the circumferential deformation, longitudinal deformation, and cross-sectional displacement tilt angle data are calculated.

[0161] The key to image comparison analysis lies in accurate image registration. This invention employs a feature-point-based image registration method, with specific steps including: feature point extraction (using SIFT or ORB algorithms); feature point matching (using the FLANN algorithm); estimation of the transformation matrix (using the RANSAC algorithm); image transformation and resampling. Registration accuracy is generally controlled at the sub-pixel level, corresponding to an actual size error of less than 0.5 mm.

[0162] The collected data is transmitted to the network terminal server via industrial Ethernet. To ensure the reliability of data transmission, the TCP / IP protocol is used, and data encryption and verification mechanisms are implemented. The network terminal server is responsible for the initial processing, storage, and distribution of the data, including data format conversion, timestamp addition, and outlier detection.

[0163] The sensor arrangement scheme and data acquisition and processing flow of this invention enable comprehensive and high-precision monitoring of tunnel deformation, providing a reliable data foundation for subsequent analysis.

[0164] The specific implementation method for determining the tunnel deformation rate by comparing the simulated calculation results with the actual collected data in this invention is as follows:

[0165] A three-dimensional coordinate system is established in the computer, including an X-axis, a Y-axis, and a Z-axis, where the X-axis represents the longitudinal direction of the tunnel, the Y-axis represents the circumferential direction, and the Z-axis represents the cross-sectional direction of the tunnel. Simulation calculations are performed on the processed circumferential deformation data, longitudinal deformation data, and visualized cross-sectional displacement tilt angle data, respectively, using the Y-axis, X-axis, and Z-axis as the central axes, to obtain simulated data. Simulated circumferential deformation images, simulated longitudinal deformation images, and simulated cross-sectional displacement tilt angle images at different acquisition times are plotted in the three-dimensional coordinate system. The simulated images are compared with the actual images acquired at the corresponding time points. When the data changes of the simulated image and the acquired image at the corresponding time point are consistent, the tunnel deformation rate is determined to be zero at that time. When there are differences between the simulated image and the acquired image at the corresponding time point, the tunnel deformation rate is calculated based on the difference value.

[0166] In a preferred embodiment of the present invention, the three-dimensional coordinate system is established using a right-handed Cartesian coordinate system, with the origin set at the center point of the tunnel entrance. The X-axis points longitudinally into the tunnel, the Y-axis points circumferentially (horizontally) into the tunnel, and the Z-axis points transversely (vertically) into the tunnel. The scale of the coordinate system is determined according to the tunnel dimensions; generally, the X-axis ranges from [0, L] (where L is the tunnel length), and the Y-axis and Z-axis range from [-R, R] (where R is the tunnel radius).

[0167] The processed circumferential deformation data, longitudinal deformation data, and visualized cross-sectional displacement and tilt angle data were simulated using the finite element method. Specifically, a finite element model of the tunnel was established, and the processed deformation data was used as boundary conditions to solve for the displacement field distribution. The finite element model used 8-node hexahedral elements, and the mesh size was determined according to the required computational accuracy, generally controlled between 10 and 30 cm.

[0168] To assess the deformation rate, it is necessary to calculate the difference between the simulation results and the actual collected data. Define the difference measurement function:

[0169] ,

[0170] in, The difference at time t is measured in meters (m). This represents the total number of monitoring points. This represents the measured deformation value of the i-th monitoring point at time t, in meters (m). This represents the simulated deformation value of the i-th monitoring point at time t, in meters (m). This represents the summation over all monitoring points.

[0171] The deformation rate can be defined as the derivative of the difference measure function with respect to time:

[0172] ,

[0173] in, The deformation rate at time t is expressed in m / s. Dissimilarity measurement function The derivative with respect to time t; The difference at time t; For time Time-varying measure; The time interval is expressed in seconds (s).

[0174] In practical applications, due to measurement errors and model approximations, even in a steady state, the difference measurement function... It will not be strictly zero. Therefore, a threshold is introduced. ,when At that point, the tunnel is considered to be in a stable state, with a deformation rate of zero. Threshold The selection is determined based on actual engineering experience, and is generally taken as 0.05~0.1mm / day.

[0175] When the deformation rate is not zero, further analysis of the deformation direction and distribution is required. Define the deformation direction index:

[0176] ,

[0177] in, For time The deformation direction index at time t is dimensionless and ranges from [-1, 1]. Indicates the first Each monitoring point at time The measured deformation value, in meters (m); Indicates the first Each monitoring point at time The simulated deformation value is expressed in meters (m). For the first The unit direction vector of each monitoring point is dimensionless; · represents the vector dot product. Represents the absolute value of the difference, in meters (m); This represents the summation over all monitoring points. Deformation direction index. The value is between -1 and 1, with a positive value indicating that the deformation mainly occurs along... Direction, negative values ​​indicate that deformation mainly occurs along... direction.

[0178] The above methods can accurately calculate the deformation rate and direction of the tunnel, providing a quantitative basis for safety assessment.

[0179] The specific implementation of the regression prediction module in this invention for predicting tunnel deformation based on the feature sequence is as follows:

[0180] Obtain the historical actual deformation corresponding to the feature sequence; establish a random forest model in the feature space; classify the feature sequence using the random forest model to obtain sub-feature sequences; train the random forest model to obtain the correspondence between the sub-feature sequences and the actual deformation; perform weighted reconstruction on all sub-feature sequences and the actual deformation to obtain a deformation prediction model; predict the feature sequence according to the deformation prediction model to obtain the corresponding predicted deformation.

[0181] In a preferred embodiment of the present invention, the random forest model is constructed using the following parameter settings: the number of decision trees is 100 to 500, the maximum depth of the decision trees is 10 to 20, the feature selection criterion is Gini impurity, and the sample splitting method is bootstrap.

[0182] When classifying feature sequences, the random forest model first standardizes the features:

[0183] ,

[0184] in, For the standardized first One characteristic, dimensionless; For the original number Each feature is represented by a unit, which depends on the feature type. For the first The mean of each feature, in units of same; For the first The standard deviation of each feature, in units of same.

[0185] The standardized feature sequences are classified using a random forest model to obtain sub-feature sequences. Specifically, the feature sequences are processed through each decision tree, with decisions made at the internal nodes of the tree based on the feature values, ultimately reaching a leaf node. Each leaf node corresponds to a sub-feature sequence.

[0186] To establish the correspondence between sub-feature sequences and actual deformation amounts, regression analysis was employed. For each sub-feature sequence... Establish a regression model:

[0187] ,

[0188] in, The corresponding actual deformation is expressed in meters (m). It is a regression function; Sub-feature sequence; This is the error term, and the unit is meters (m).

[0189] Regression function Various forms can be used, including linear regression, multinomial regression, and support vector regression. In this invention, support vector regression (SVR) is preferred, with the radial basis function (RBF) chosen as the kernel function.

[0190] ,

[0191] in, Let be the kernel function, representing two points in the feature space. and The similarity is dimensionless; This is the kernel parameter, and its unit is the reciprocal of the feature unit, typically taking the value as the reciprocal of the feature dimension; For feature vectors and The Euclidean distance between them is determined by the characteristic unit; It is an exponential function.

[0192] The regression results of all sub-feature sequences are weighted and reconstructed to obtain the final deformation prediction model:

[0193] ,

[0194] in, The predicted deformation is expressed in meters (m). The number of sub-feature sequences; For the first The weight coefficients of each sub-feature sequence are dimensionless and satisfy the following conditions: ; For the first Sub-feature sequences The corresponding regression function predicted value, in meters (m); This represents the summation of all sub-feature sequences.

[0195] Weighting coefficient Cross-validation was used to determine the optimal method. The training data was divided into... Fold (general) or ),exist Fold on the trained model, in the remaining Considering validation performance, calculate the validation error for each sub-model. The weighting coefficients are inversely proportional to the validation error.

[0196] ,

[0197] in, For the first The weight coefficients of each sub-feature sequence are dimensionless. For the first The validation error of each sub-model, in meters (m). The number of sub-feature sequences; This represents the summation of all sub-feature sequences.

[0198] Using the methods described above, the deformation prediction model S3 can effectively integrate the prediction results of multiple sub-models, thereby improving the accuracy and stability of the prediction.

[0199] The specific implementation method of the prediction and correction module in this invention to correct the tunnel deformation based on historical deformation development patterns is as follows:

[0200] The tunnel deformation prediction model is updated by predicting the deformation at different times based on historical deformation data. The prediction error change at the current time is obtained based on the predicted deformation at different times. The prediction error change is integrated to obtain the prediction error change range. The predicted deformation at the current time is corrected based on the prediction error change range. When the prediction error change range is greater than a preset threshold, the tunnel deformation prediction model is updated.

[0201] In a preferred embodiment of the invention, the calculation of the change in prediction error is based on a comparison between historical prediction results and actual observations. It is assumed that in time... The predicted value is The actual observed value is The prediction error is:

[0202] ,

[0203] in, For time The prediction error at that time is expressed in meters (m). For time The actual observed value at that time, in meters (m); For time The predicted value at that time is in meters (m).

[0204] The change in prediction error is defined as the time derivative of the prediction error:

[0205] ,

[0206] in, For time The change in prediction error over time, expressed in m / s; For time The prediction error at that time is expressed in meters (m). For time The prediction error at that time is expressed in meters (m). The time interval is expressed in seconds (s).

[0207] To reduce the impact of random fluctuations, a moving average is applied to the change in prediction error:

[0208] ,

[0209] in, For time The change in prediction error after smoothing, expressed in m / s; This is the window size, dimensionless, and typically ranges from 5 to 10. For time The change in prediction error over time, expressed in m / s; This represents the summation of all time points within the window; This indicates calculating the average value.

[0210] Integrating the change in prediction error yields the range of prediction error variation:

[0211] ,

[0212] in, For time The prediction error variation range over time, in meters (m); Indicates from time Time Change in prediction error after smoothing The integral; The integral time range is dimensionless and typically ranges from 10 to 30. For time The change in prediction error after smoothing, expressed in m / s; The time interval is measured in seconds (s). This represents summing over all time points within the integration range.

[0213] Based on the range of prediction error variation, the predicted deformation at the current moment is corrected:

[0214] ,

[0215] in, The corrected predicted deformation is expressed in meters (m). The original predicted deformation is expressed in meters (m). This is a correction factor, dimensionless, and typically ranges from 0.5 to 1.0. The range of prediction error variation is expressed in meters (m).

[0216] When the prediction error varies within the range When the value exceeds a preset threshold T, it indicates that the current prediction model may no longer be applicable to the latest deformation development trend, and the model needs to be updated. The threshold T is determined according to the actual engineering requirements, and is generally taken as 1.5 to 2 times the allowable error range.

[0217] The model update uses a sliding window method, retaining data from the most recent N time points for retraining. The choice of window size N needs to balance model stability and adaptability, and is generally set between 50 and 200, with the specific value determined based on the time scale of deformation development.

[0218] Through the above method, the prediction correction module can dynamically correct the prediction results according to the historical deformation development law, improving the accuracy and adaptability of the prediction.

[0219] The specific implementation manners included in the present invention are as follows:

[0220] Define a safety threshold and a risk level standard; monitor the tunnel deformation status in real time and compare it with the safety threshold; when abnormal deformation is detected, trigger an early warning mechanism; provide visual positioning information of the deformation area; generate a detailed report including the deformation type, deformation degree and risk assessment.

[0221] In a preferred embodiment of the present invention, the safety threshold and the risk level standard are determined based on engineering specifications and expert experience. Usually, the deformation level is divided into four levels: normal (green), attention (yellow), warning (orange) and danger (red).

[0222] For circumferential deformation, the safety threshold is set as follows:

[0223] Normal: deformation amount < 5 mm or deformation rate < 0.05 mm / day;

[0224] Attention: 5 mm ≤ deformation amount < 10 mm or 0.05 mm / day ≤ deformation rate < 0.1 mm / day;

[0225] Warning: 10 mm ≤ deformation amount < 20 mm or 0.1 mm / day ≤ deformation rate < 0.2 mm / day;

[0226] Danger: deformation amount ≥ 20 mm or deformation rate ≥ 0.2 mm / day;

[0227] For longitudinal deformation, the safety threshold is set as follows:

[0228] Normal: deformation amount < 10 mm or deformation rate < 0.1 mm / day;

[0229] Attention: 10 mm ≤ deformation amount < 20 mm or 0.1 mm / day ≤ deformation rate < 0.2 mm / day;

[0230] Warning: 20 mm ≤ deformation amount < mm or 0.2 mm / day ≤ deformation rate < 0.4 mm / day;

[0231] Danger: deformation amount ≥ mm or deformation rate ≥ 0.4 mm / day;

[0232] For the cross-sectional displacement tilt angle, the safety threshold is set as follows:

[0233] Normal: tilt angle < 0.5° or change rate < 0.005° / day;

[0234] Note: 0.5° ≤ tilt angle < 1.0° or 0.005° / day ≤ rate of change < 0.01° / day;

[0235] Warning: 1.0° ≤ tilt angle < 2.0° or 0.01° / day ≤ rate of change < 0.02° / day;

[0236] Hazard: Inclination angle ≥ 2.0° or rate of change ≥ 0.02° / day;

[0237] The real-time monitoring system automatically checks the latest deformation status at fixed time intervals (usually 1 hour) and compares it with safety thresholds. When the detected deformation status reaches or exceeds the attention level, the system automatically generates a warning message; when it reaches the warning level, the system triggers an alarm mechanism and automatically increases the data acquisition frequency; when it reaches the danger level, the system issues an emergency alarm to notify relevant personnel to take immediate countermeasures.

[0238] The early warning information includes: deformation location (tunnel station), deformation type (circumferential / longitudinal / cross-section), deformation amount, deformation rate, risk level, and expected development trend. Simultaneously, the system generates a visual location map of the deformation area, intuitively displaying its location and extent.

[0239] The detailed report includes: historical deformation data, deformation development curves, predicted deformation trends, risk assessment results, and recommended measures. The report is in PDF format and can be sent to relevant personnel via email or mobile application.

[0240] Through the above-mentioned early warning mechanism and report generation function, this invention can promptly detect tunnel deformation risks, provide a scientific basis for decision-making, and effectively improve the level of tunnel safety management.

[0241] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of protection of the present invention.

Claims

1. A tunnel deformation detection method based on convolutional neural networks and machine vision sensors, characterized in that, include: Machine vision sensors installed in the tunnel collect data on tunnel circumferential deformation, longitudinal deformation, and cross-sectional displacement and tilt angle. The circumferential deformation data, the longitudinal deformation data, and the cross-sectional displacement tilt angle data are input into a computer for processing, including: The circumferential deformation data and the longitudinal deformation data are preprocessed using a convolutional neural network in a computer. The cross-sectional displacement tilt angle data is converted into discretized matrix data in a computer and visualized based on differential geometry theory to obtain visualized cross-sectional displacement tilt angle data. A tunnel deformation monitoring model is constructed in a computer, and the processed circumferential deformation data, longitudinal deformation data, and visualized cross-sectional displacement tilt angle data are input into the tunnel deformation monitoring model for simulation calculation. The tunnel deformation rate is determined by comparing the simulated calculation results with the actual collected data. The process of converting the cross-sectional displacement tilt angle data into discretized matrix data and performing visualization based on differential geometry theory includes: The tunnel cross-section is considered as a two-dimensional Riemannian manifold embedded in three-dimensional Euclidean space; Establish a polar coordinate system (r, θ) with the center of the tunnel as the origin; The cross-sectional displacement tilt angle data is represented as a vector field defined on the Riemannian manifold; Calculate the geometric invariants of the Riemannian manifold, including the principal curvatures and Gaussian curvatures; Discretize the geometric invariants into a matrix representation; The matrix data is visualized and enhanced based on geodesic curvature flow theory to highlight areas of abnormal deformation. The visualization enhancement processing of the matrix data based on geodesic curvature flow theory also includes: Acquire cross-sectional displacement and tilt angle data at multiple time points; Based on the connection theory and covariant derivative in differential geometry, the rate of change of the cross-sectional displacement tilt angle is calculated. Construct a change rate feature matrix and extract the main modes of deformation; Map the geometric invariant matrix and the rate of change feature matrix to a color space; Multi-scale analysis techniques are applied to achieve hierarchical visualization from global to local levels; Potential risk areas are automatically identified based on curvature outliers, and risk levels are classified accordingly.

2. The tunnel deformation detection method based on convolutional neural networks and machine vision sensors according to claim 1, characterized in that, The preprocessing of the circumferential deformation data and the longitudinal deformation data using a convolutional neural network includes: A preprocessing network containing an encoder and a decoder is constructed and runs on the TensorFlow deep neural network platform; Multi-layer convolutional layers are used to extract features from the image, and the extracted features are then subjected to dimensionality reduction. The decoder outputs a matrix of the original image size, confidence scores, and bounding box data. A detection network consisting of a residual network and a fully connected layer is constructed to extract features and output results from the preprocessed data.

3. The tunnel deformation detection method based on convolutional neural networks and machine vision sensors according to claim 1, characterized in that, The tunnel deformation monitoring model includes: A deformation prediction system is built based on convolutional neural networks and machine learning models; The deformation prediction system includes a feature extraction module, a regression prediction module, and a prediction correction module; The feature extraction module takes the circumferential deformation data, the longitudinal deformation data, and the visualized cross-sectional displacement tilt angle data as feature inputs, classifies, integrates, and extracts the features to obtain a feature sequence. The regression prediction module predicts the tunnel deformation based on the feature sequence. The prediction and correction module corrects the tunnel deformation based on historical deformation patterns.

4. The tunnel deformation detection method based on convolutional neural networks and machine vision sensors according to claim 1, characterized in that, The process of collecting tunnel circumferential deformation data, longitudinal deformation data, and cross-sectional displacement and tilt angle data through machine vision sensors installed in the tunnel includes: A machine vision sensor is installed longitudinally at predetermined intervals in the tunnel; Each of the machine vision sensors employs an area array camera to measure the circumferential deformation and cracking of the tunnel using image comparison analysis. Data is collected at preset time intervals, and each collection of data is treated as a set of data. The collected data is transmitted to a network terminal server via the network.

5. The tunnel deformation detection method based on convolutional neural networks and machine vision sensors according to claim 1, characterized in that, The process of determining the tunnel deformation rate by comparing the simulated calculation results with the actual collected data includes: A three-dimensional coordinate system is established in the computer. The three-dimensional coordinate system includes an X-axis, a Y-axis, and a Z-axis, where the X-axis is the longitudinal direction of the tunnel, the Y-axis is the circumferential direction, and the Z-axis is the cross-sectional direction of the tunnel. Simulations were performed on the processed circumferential deformation data, longitudinal deformation data, and visualized cross-sectional displacement tilt angle data, respectively, with the Y-axis, X-axis, and Z-axis as the central axes, to obtain simulation data. Simulated circumferential deformation images, simulated longitudinal deformation images, and simulated cross-sectional displacement tilt angle images at different acquisition times are plotted in the three-dimensional coordinate system. The simulated circumferential deformation image, the simulated longitudinal deformation image, and the simulated cross-sectional displacement tilt angle image are compared with the actual images acquired at the corresponding time points; When the simulated circumferential deformation image, simulated longitudinal deformation image, and simulated cross-sectional displacement tilt angle image are consistent with the data changes at the corresponding time of the acquired image, the tunnel deformation rate at this time is determined to be zero. When there are differences between the simulated circumferential deformation image, simulated longitudinal deformation image, and simulated cross-sectional displacement tilt angle image and the data at the corresponding time of the acquired image, the tunnel deformation rate is calculated based on the difference value.

6. The tunnel deformation detection method based on convolutional neural networks and machine vision sensors according to claim 3, characterized in that, The regression prediction module predicts tunnel deformation based on the feature sequence, including: Obtain the historical actual deformation corresponding to the feature sequence; Establish a random forest model in the feature space; The feature sequence is classified using the random forest model to obtain sub-feature sequences; The correspondence between sub-feature sequences and actual deformation amounts is obtained by training the random forest model described above. The deformation prediction model is obtained by weighting and reconstructing all sub-feature sequences with the actual deformation. The feature sequence is predicted based on the deformation prediction model to obtain the corresponding predicted deformation amount.

7. The tunnel deformation detection method based on convolutional neural networks and machine vision sensors according to claim 3, characterized in that, The prediction correction module corrects the tunnel deformation based on historical deformation development patterns, including: Predict the deformation amount at different times by using historical deformation data of the tunnel. The change in prediction error at the current moment is obtained based on the predicted deformation at different times. Integrating the change in prediction error yields the range of prediction error variation. The predicted deformation at the current moment is corrected based on the range of change of the prediction error; When the variation range of the prediction error exceeds a preset threshold, the tunnel deformation prediction model is updated.

8. The tunnel deformation detection method based on convolutional neural networks and machine vision sensors according to claim 1, characterized in that, Also includes: Define safety thresholds and risk level standards; The tunnel deformation status is monitored in real time and compared with the safety threshold. When abnormal deformation is detected, an early warning mechanism is triggered; Provides visual positioning information of the deformable area; Generate a detailed report that includes deformation type, deformation degree, and risk assessment.

Images