A shield tunnel thermal disaster diagnosis method and system based on deep learning

Abstract

The present invention discloses a deep learning-based shield tunnel thermal hazard diagnosis method and system, relating to the fields of information technology and tunnel safety monitoring technology. The method comprises the following steps: collecting infrared images of the temperature field on the inner curved surface of a shield tunnel segment; performing color matching on the filtered infrared images to obtain a matching temperature cloud map; inputting the temperature cloud map into a trained thermal-stress diagnostic model to output a stress cloud map; the model training process comprises: constructing a numerical model that simulates the thermal performance of a shield tunnel under the coupled effects of heat, permeability, and force; obtaining temperature and stress cloud maps of the temperature field on the inner curved surface of the shield tunnel segment through transient simulation calculations based on the numerical model; using the temperature and stress cloud maps to train a convolutional neural network-based thermal-stress diagnostic model; and, based on the stress cloud map and in combination with preset diagnostic criteria, determining the shield tunnel thermal hazard warning level and issuing a timely warning. The present invention can provide accurate early warnings for the safety and stability of shield tunnel structures.

Description

Technical Field

[0001] The present invention relates to the fields of information technology and tunnel safety monitoring technology, and in particular to a shield tunnel thermal disaster diagnosis method and system based on deep learning. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] With the accelerating pace of urbanization, urban traffic congestion is becoming increasingly severe, prompting the rational development and utilization of underground space. Currently, advancements in shield tunneling technology are providing solid technical support for the rapid expansion of urban networks. However, compared to the surface environment, subway tunnels accumulate significant amounts of heat due to the continuous operation of electrical equipment and environmental control systems, as well as the heat dissipated by passengers. Due to the unique characteristics of underground space, this heat is difficult to dissipate quickly, resulting in long-term and persistent thermal pollution within the tunnel. As thermal pollution accumulates within the tunnel, the soil and air temperatures surrounding the tunnel rise. This persistent high temperature environment impacts the shield tunnel segments, generating thermal stresses. When coupled with structural forces, these stresses can cause deformation of the segments. Minor deformations can lead to cracks in the segments. If these conditions persist for a long time and are not promptly detected and repaired, they can even lead to collapse of the tunnel structure, posing a threat to urban safety. Therefore, the management and maintenance of the thermal environment within subway tunnels is a crucial measure to ensure urban safety and requires careful attention and the development of scientific response strategies.

[0004] At present, the severity of thermal disasters in shield tunnels is usually diagnosed by testing the effectiveness of the tunnel segment structure in a thermal environment (such as whether there are deformations, cracks, etc.). The current detection of tunnel structures still remains at the stage of complex and tedious manual inspection. This traditional method of manual inspection has many limitations: (1) Manual inspection usually requires a lot of time, manpower and material resources, resulting in high costs; (2) It is impossible to monitor the temperature field and structural stress changes in the tunnel, resulting in the inability to timely warn of potential structural problems; (3) The tunnel area covered by manual inspection is relatively small, and the detection efficiency is low; (4) The tunnel environment is hot, humid and closed, which seriously reduces the cognitive ability of personnel. Moreover, due to the limitations of the cognitive ability of the inspectors, there are risks such as missed detection and false detection, resulting in low accuracy of the final detection.

[0005] In addition, although an intelligent detection method for thermal disasters in shield tunnels has been proposed, that is, judging the severity of thermal disasters by identifying cracks on the surface of the segments, this method can only detect existing cracks on the surface of the segments and cannot identify cracks inside the segments and potential crack occurrence areas. Its consideration is incomplete, resulting in poor accuracy in the final judgment of the severity of thermal disasters and difficulty in achieving accurate diagnosis. Summary of the Invention

[0006] In order to address the deficiencies of the above-mentioned prior art, the present invention provides a shield tunnel thermal disaster diagnosis method and system based on deep learning. By simulating the thermal performance of the shield tunnel under the action of heat-permeability-force multi-field coupling, the corresponding temperature cloud map and stress cloud map are obtained, and the correspondence between the temperature cloud map and the stress cloud map is learned using a deep learning algorithm. Then, an infrared image of the actual shield tunnel segment structure is obtained, and a temperature cloud map is obtained after color map matching. Combined with the learned correspondence, the actual stress condition of the segment structure is clarified, thereby timely warning of potential structural problems, accurate diagnosis of shield tunnel thermal disasters, and accurate and timely warning of the safety and stability of the shield tunnel structure.

[0007] In a first aspect, the present invention provides a shield tunnel thermal disaster diagnosis method and system based on deep learning.

[0008] A deep learning-based shield tunnel thermal disaster diagnosis method includes:

[0009] Collect infrared images of the temperature field on the inner curved surface of shield tunnel segments and conduct preliminary screening of the collected images;

[0010] Perform chromaticity matching on the filtered infrared images to obtain a matching temperature cloud map;

[0011] The temperature cloud map is input into the trained thermal-stress prediction and diagnosis model, and the corresponding stress cloud map is output. The model training process includes: constructing a numerical model that simulates the thermal performance of the shield tunnel under the multi-field coupling of heat, permeation, and force; based on the numerical model, obtaining the temperature cloud map and stress cloud map of the temperature field on the inner curved surface of the shield tunnel segment through transient simulation calculation; using the temperature cloud map and stress cloud map to train the thermal-stress diagnosis model based on the convolutional neural network;

[0012] Based on the stress cloud map and combined with the preset diagnostic basis, the thermal disaster warning level of the shield tunnel is determined and timely warning is issued.

[0013] In the second aspect, the present invention provides a shield tunnel thermal disaster diagnosis system based on deep learning.

[0014] A deep learning-based shield tunnel thermal disaster diagnosis system, including:

[0015] Infrared image acquisition module, used to collect infrared images of the temperature field of the inner arc surface of the shield tunnel segment and perform preliminary screening of the collected images;

[0016] The temperature cloud map matching module is used to perform color matching on the filtered infrared image to obtain a matching temperature cloud map;

[0017] The stress cloud map acquisition module is used to input the temperature cloud map into the trained thermal-stress diagnostic model and output the corresponding stress cloud map. The model training process includes: constructing a numerical model that simulates the thermal performance of the shield tunnel under the multi-field coupling of heat, permeation, and force; based on the numerical model, obtaining the temperature cloud map and stress cloud map of the temperature field on the inner curved surface of the shield tunnel segment through transient simulation calculation; and using the temperature cloud map and stress cloud map to train the thermal-stress diagnostic model based on the convolutional neural network.

[0018] The thermal disaster diagnosis module is used to determine the thermal disaster warning level of shield tunnels and issue timely warnings based on stress cloud maps and preset diagnostic criteria.

[0019] In a third aspect, the present invention also provides an electronic device comprising: a memory for storing executable instructions; and a processor for implementing the above-mentioned deep learning-based shield tunnel thermal disaster diagnosis method when executing the executable instructions stored in the memory.

[0020] In a fourth aspect, the present invention also provides a computer-readable storage medium storing executable instructions for causing a processor to execute the executable instructions to implement the above-mentioned deep learning-based shield tunnel thermal disaster diagnosis method.

[0021] In a fifth aspect, the present invention also provides a computer program product, which includes executable instructions, and the executable instructions are stored in a computer-readable storage medium; wherein, when the processor of the electronic device reads the executable instructions from the computer-readable storage medium and executes the executable instructions, the above-mentioned deep learning-based shield tunnel thermal disaster diagnosis method is implemented.

[0022] One or more of the above technical solutions have the following beneficial effects:

[0023] The present invention provides a shield tunnel thermal disaster diagnosis method and system based on deep learning. By simulating the thermal performance of the shield tunnel under the action of heat-permeability-force multi-field coupling, corresponding temperature cloud maps and stress cloud maps are obtained. The correspondence between the temperature cloud map and the stress cloud map is learned using a deep learning algorithm. Then, an infrared image of the actual shield tunnel segment structure is obtained. After color map matching, a temperature cloud map is obtained. Combined with the learned correspondence, the actual stress condition of the segment structure is clarified. This can timely warn of potential structural problems, realize the diagnosis of shield tunnel thermal disasters, and realize accurate and timely warning of the safety and stability of the shield tunnel structure. The method avoids the problems of traditional manual detection and diagnosis methods such as large consumption of manpower and material resources and poor detection accuracy. Compared with the existing intelligent detection methods, the method cannot consider the internal characteristics of the segment. The method predicts the overall stress condition of the segment through the segment surface temperature field, can perform thermal disaster diagnosis and warning in a more comprehensive and in-depth manner, effectively improve the efficiency and accuracy of shield tunnel thermal disaster diagnosis, and ensure the safety and stability of the shield tunnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0025] Figure 1 This is an overall flow chart of the deep learning-based shield tunnel thermal disaster diagnosis method according to an embodiment of the present invention;

[0026] Figure 2 Schematic diagram of infrared image collection by a rail vehicle in a shield tunnel according to an embodiment of the present invention;

[0027] Figure 3 : This is a temperature cloud diagram of the temperature field of the semi-ring tunnel segment in an embodiment of the present invention;

[0028] Figure 4 The stress cloud diagram of thermal stress and structural mechanics of the semi-ring tunnel segment in the embodiment of the present invention;

[0029] Figure 5 This is a temperature distribution cloud diagram of the temperature field on the inner curved surface of a semi-annular tunnel segment in an embodiment of the present invention;

[0030] Figure 6 This is a stress distribution cloud diagram of thermal stress and structural mechanics on the inner arc surface of a semi-annular tunnel segment in an embodiment of the present invention;

[0031] Figure 7 This is a schematic diagram of the structure of the heat-stress diagnosis model based on the U-Net architecture adopted in an embodiment of the present invention.

[0032] Among them: 1. Tunnel segments; 2. Infrared cameras; 3. Rail cars. DETAILED DESCRIPTION

[0033] It should be noted that the following detailed descriptions are exemplary only and are intended to describe specific embodiments and provide further explanation of the present invention, and are not intended to limit the exemplary embodiments according to the present invention. Unless otherwise indicated, all technical and scientific terms used herein have the same meanings as those commonly understood by those of ordinary skill in the art to which the present invention belongs. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0034] Example 1

[0035] This embodiment provides a shield tunnel thermal disaster diagnosis method based on deep learning, such as Figure 1 As shown, the specific steps include:

[0036] Step S1, collecting infrared images of the temperature field of the inner arc surface of the shield tunnel segment, and performing preliminary screening on the collected images;

[0037] Step S2: performing color matching on the filtered infrared image to obtain a matching temperature cloud map;

[0038] Step S3: inputting the temperature cloud map into the trained thermal-stress diagnostic model and outputting the corresponding stress cloud map; wherein the model training process includes: constructing a numerical model simulating the thermal performance of the shield tunnel under the multi-field coupling of heat, permeation, and force; obtaining the temperature cloud map and stress cloud map of the temperature field of the inner arc surface of the shield tunnel segment through transient simulation calculation based on the numerical model; and using the temperature cloud map and stress cloud map to train the thermal-stress diagnostic model based on the convolutional neural network;

[0039] Step S4: Based on the stress cloud map and in combination with the preset diagnostic basis, the thermal disaster warning level of the shield tunnel is determined and a timely warning is issued.

[0040] The following content provides a more detailed introduction to the shield tunnel thermal disaster diagnosis method based on deep learning proposed in this embodiment.

[0041] In step S1, infrared images of the temperature field of the inner arc surface of the shield tunnel segment are collected, and the collected infrared images are preliminarily screened.

[0042] In this embodiment, the image acquisition device is an infrared imaging acquisition device, which is used to collect infrared thermal images (abbreviated as infrared images) of the temperature field of the inner arc surface of the shield tunnel segment. Figure 2As shown, the image acquisition device consists of a railcar 3 and an infrared camera 2 (also referred to as an infrared imager, infrared equipment, or infrared thermal imager). The railcar 3, equipped with the infrared camera 2, continuously operates on subway tracks within a shield tunnel to capture infrared images and other data of the tunnel segments 1. The infrared thermal imager has sufficient sampling accuracy (less than ±0.5°C) and is suitable for the tunnel temperature sampling range (0-50°C or greater). The captured infrared thermal images have sufficient resolution, such as 1280×102.

[0043] Furthermore, in the above-mentioned image acquisition process, continuous acquisition or intermittent acquisition can be adopted, that is, the infrared thermal imager can continuously acquire infrared thermal images as the rail car moves, or it can stop for a certain time (such as 10s) after each advance of a certain depth, and acquire images during this time; in addition, the rail car is equipped with a rotating device that can enable the infrared thermal imager 2 (i.e., infrared device) to rotate 360 ​​degrees, so that during the acquisition process, the infrared device can rotate around the horizontal axis to collect a complete temperature cloud map of the shield tunnel temperature field.

[0044] During the image acquisition process, the railcar body avoids the imaging range of the infrared thermal imager to prevent the body from appearing in the image. In addition, when the infrared device rotates one circle along the horizontal axis, there must be no structural parts blocking the plane where it is located to ensure the continuity and integrity of the shooting. The railcar is equipped with a controller, a motor, and a battery, etc. The battery provides sufficient power for the infrared device, controller, motor, etc. The controller is equipped with a remote communication module, or a wireless data transmission terminal (such as Bluetooth, etc.) electrically connected to the controller is set to remotely control the start and stop of the motor and infrared device through the controller, thereby realizing remote control of the railcar and image acquisition.

[0045] Based on the above design, after image acquisition is completed by the image acquisition device, the infrared thermal imager transmits the acquired infrared image to the cloud or remote terminal through the controller's remote communication module or wireless data transmission terminal. Model detection is then performed on the computer in the cloud or remote terminal, thereby realizing the diagnosis and timely warning of tunnel thermal disasters.

[0046] As a specific implementation method, Figure 2As shown, after the train stops operating, the wheels of railcar 3 are aligned with the subway tracks and the railcar 3 is placed on the tracks. Railcar 3 is equipped with a rotating device. One end of the rotating rod is fixedly connected to the rotating device, and the other end is fixedly mounted with an infrared device (i.e., a high-definition infrared camera 2). A controller is added to control the operation of the rotating device, thereby controlling the infrared device's 360° rotation. Furthermore, the infrared device, motor, controller, etc. are connected to a battery to ensure power is supplied to the equipment. After installation, the rotating device with the infrared camera 2 is checked for proper operation. After the railcar reaches the designated location, the controller is remotely controlled to control the operation of the rotating device, and the 360° rotating infrared device is used to capture continuous infrared thermal images of each segment in the tunnel.

[0047] After installation and inspection, the detection system is turned on to check whether the wireless data transmission terminal connection is normal, thus ensuring the smooth operation of the entire diagnostic process. After the instrument inspection is completed, the rail car is remotely started. Through the rail car's built-in PLC control system (based on the controller), it is set to pause for approximately 30 seconds when reaching the predetermined position, that is, the center of the circular tunnel, to allow the infrared camera to complete a panoramic capture of the temperature field of the entire circular tunnel. The infrared camera then accurately detects the arc surface temperature of the tunnel segment inside the tunnel.

[0048] After the infrared images are collected, the diagnostic method proposed later can be used to diagnose and warn of thermal disasters. The warning results can be used to guide actual engineering projects, conduct tunnel maintenance, and ensure the safety and reliability of shield tunnels.

[0049] As another implementation method, to improve acquisition efficiency, infrared cameras can be fixed, with multiple railcars operating simultaneously within a certain tunnel section to capture images. Specifically, the required fixed positions for the infrared cameras and the number of railcars can be determined based on the number of segments in the shield tunnel cross-section. Each infrared camera is fixed at a specific position, capturing only an infrared image of the temperature field on the curved surface of the segment corresponding to the current position. In this embodiment, six railcars with different infrared camera positions are used, and they are started sequentially until they operate simultaneously to maintain acquisition efficiency. This approach allows calculations to be performed on only a single segment, reducing computer performance requirements and facilitating widespread use.

[0050] Furthermore, after collecting and obtaining infrared images of the temperature field on the curved surface of the shield tunnel segment, the collected infrared images are effectively screened. For the segment at each position, infrared images with high image clarity and no obstructions are screened out through manual screening and other methods.

[0051] In step S2, the filtered infrared image is subjected to chromaticity matching to obtain a matching temperature cloud map.

[0052] Specifically, in combination with the above steps, the infrared image of the temperature field of the inner arc surface of a single tunnel segment collected by the infrared camera is used to obtain the temperature cloud map after chromaticity matching. Figure 5 As shown in Figure 2, after the diagnostic model, the stress cloud diagram of the thermal stress and structural force of the arc surface of the single tunnel segment is obtained, as shown in Figure 2. Figure 6 shown.

[0053] Among them, chromaticity matching is to match the temperature reflected at the same position of the infrared thermal image (i.e., infrared image) and the temperature cloud map. In this embodiment, before using the infrared imager to capture the infrared image of the temperature field on the curved surface of the shield tunnel segment, the color-temperature legend is adjusted to be consistent with the legend of the simulated temperature cloud map, thereby completing rapid chromaticity matching; as another implementation method, the ratio of the color-temperature legend of a certain infrared thermal image and the temperature cloud map can also be obtained first, and then the ratio is used to adjust the color-temperature legend of each infrared thermal image obtained, thereby obtaining the corresponding temperature cloud map after chromaticity matching.

[0054] Through the color map matching in step S2 above, the temperature represented by the color in the acquired infrared image can be made consistent with the temperature represented by the color in the input image during training in the subsequent diagnostic model, thereby ensuring the accuracy of subsequent detection and diagnosis.

[0055] In step S3, the temperature cloud map is input into the trained thermal-stress diagnostic model, and the corresponding stress cloud map is output; wherein, the model training process includes: constructing a numerical model that simulates the thermal performance of the shield tunnel under the multi-field coupling of heat, permeation and force; based on the numerical model, through transient simulation calculation, obtaining the temperature cloud map and stress cloud map of the temperature field of the arc surface inside the shield tunnel segment; using the temperature cloud map and stress cloud map to train the thermal-stress diagnostic model based on the convolutional neural network.

[0056] Specifically, the training process of the heat-stress diagnostic model based on convolutional neural networks consists of five parts: dataset construction, data cleaning, model selection, model training, model verification, and performance evaluation. The training process specifically includes:

[0057] Step S3.1: Build a heat-stress diagnostic model based on convolutional neural network.

[0058] Specifically, a heat-stress diagnosis model is built based on a convolutional neural network (CNN). The convolutional layer is used to extract the image features of the input temperature cloud map, and the corresponding stress cloud map is output through the fully connected layer after learning. The specific structure of the network model can be selected and adjusted according to specific circumstances, such as the size and number of input images (i.e., model selection). In this embodiment, a deep learning model based on the U-Net architecture is used to build a heat-stress diagnosis model. The U-Net architecture is as follows: Figure 7As shown in the figure, as one of the commonly used convolutional neural network architectures, it can efficiently extract image features and restore detail information through the encoder-decoder structure and jump connection, thereby achieving high-precision image processing.

[0059] The specific structure of the heat-stress diagnosis model based on the U-Net architecture includes an input layer, an encoder, a bottleneck layer, a decoder, and an output layer connected in sequence, where:

[0060] (1) Input layer: The temperature cloud image is input through the input layer. The default size of the temperature cloud image is set to (256, 256, 3), which means that the input image size is 256×256 pixels and is an RGB three-channel image. Preferably, normalization processing is added to the input layer to facilitate the effectiveness of subsequent data analysis and feature extraction.

[0061] (2) Encoder: It consists of multiple convolution blocks connected in sequence. Each convolution block includes two convolution layers as downsampling layers. Each convolution block is followed by a Dropout layer and a maximum pooling layer.

[0062] Among them, the convolution layer adopts the Conv2D layer, each Conv2D layer has a specified number of filters (such as 64, 128, 256, 512, 1024, etc.), and the convolution layer is provided with a Mish activation layer, a regularization layer, and a Dropout layer. In this embodiment, the activation function adopted by the activation layer is the Mish activation function, which adjusts the scale of the feature map after convolution, keeps the size of the feature map unchanged, and applies L2 regularization to reduce overfitting; a Dropout layer is added after each convolution block, and some neurons are randomly discarded with a probability of 0.5 to prevent overfitting of the model and improve the generalization ability of the model; the maximum pooling layer adopts the MaxPooling2D layer with a step size of 2 to reduce the size of the feature map.

[0063] Furthermore, in the above two-dimensional convolution (i.e., Conv2D layer), each element y(i, j) of the feature map Y is extracted and output, which can be expressed by the following formula:

[0064] y(i,j)=(X*K)(i,j)=∑m∑n[x(i+m,j+n)*k(m,n)];

[0065] Where X represents the input image or feature map; K represents the convolution kernel; x(i+m,j+n) represents the pixel value of the input image at position (i+m,j+n); k(m,n) represents the weight of the convolution kernel at position (m,n); and * represents the convolution operation.

[0066] In the above maximum pooling layer (MaxPooling2D), each element y(i,j) of the output feature map Y can be expressed by the following formula:

[0067] y(i,j)=maxmx(m,n),n∈R(i,j);

[0068] Among them, x(m,n) represents the pixel value of the input feature map at position (m,n); R(i,j) represents the pooling window at position (i,j).

[0069] The above Mish activation function is:

[0070] f(x)=x·tanh(softplus(x));

[0071] softplus(x)=log(1+e x );

[0072] Among them, softplus() is the activation function, x represents the input feature map, and f(x) represents the output feature map.

[0073] (3) Bottleneck layer: It consists of two stacked convolutional layers. The number of filters in each convolutional layer is 1024, and it also uses the Mish activation function and L2 regularization.

[0074] (4) Decoder: It consists of multiple convolution blocks connected in sequence. However, unlike the encoder, each convolution block includes multiple upsampling layers. The upsampling layer uses the Conv2DTranspose layer and is concatenated with the multiple downsampling layers of the encoder to gradually restore the size of the feature map.

[0075] Among them, the upsampling layer uses the Conv2DTranspose layer, also known as the transposed convolution layer, to increase the size of the feature map; the upsampled feature map is merged with the feature map extracted by the corresponding downsampling layer of the encoder through a skip connection (concatenate) operation, which helps to restore detail information; each upsampling layer is followed by two convolution layers for ReLU activation and L2 regularization; finally, a Dropout layer is added after each convolution block.

[0076] Furthermore, the above-mentioned upsampling layer (Conv2DTranspose layer) is implemented by repeating elements to output each element y(i, j) of the feature map Y; the splicing operation is implemented by the above-mentioned jump connection, which can be expressed as splicing the two feature maps X1 and X2 together along a certain dimension d to form a new feature map Y.

[0077] (5) Output layer: The Conv2D layer is used, the number of filters is 3 (corresponding to the three RGB channels), and the activation function uses the sigmoid function to output the probability that each pixel belongs to the target category.

[0078] After the above sigmoid activation function, each element y(i,j) of the output feature map Y can be expressed by the following formula:

[0079] y(i,j)=1 / [e -x(i,j) 1;

[0080] Among them, x(i,j) represents the result after the convolution operation of the output layer.

[0081] Step S3.2: Construct a numerical model to simulate the thermal performance of a shield tunnel under the multi-field coupling of heat, penetration and force.

[0082] Specifically, COMSOL Multiphysics software was used to simulate the thermal performance of shield tunnels under the coupled effects of heat, permeability, and force. First, a physical model was constructed based on existing projects. Then, the corresponding physical parameters were obtained through field testing. This physical field was then integrated into the physical model to create a numerical simulation model (referred to as the numerical model). Furthermore, the simulation results were analyzed using an expert knowledge base and objective indicators. The thermal disaster warning level was then appropriately classified based on the simulation results and the actual situation.

[0083] Step S3.3: Based on the numerical model, through transient simulation calculation, obtain the temperature cloud map and stress cloud map of the temperature field on the inner arc surface of the shield tunnel segment.

[0084] Based on the numerical simulation model described above, which has been validated in actual engineering projects, a data set was constructed through transient simulation calculations. Specifically, to obtain a sufficient number of data sets, temperature and stress contours were saved at set intervals (e.g., one hour) during the transient simulation calculations. The total simulation duration should be the entire life cycle of the shield tunnel.

[0085] Preferably, the acquired data is preprocessed, including: reading a number of images (256*256*3) and normalizing them so that the value of each pixel is kept between 0 and 1; then dividing the data set into a training set and a test set with a division ratio of 5:1.

[0086] Step S3.4: Use the temperature cloud map and the stress cloud map to train a heat-stress diagnosis model based on a convolutional neural network.

[0087] In this embodiment, based on the temperature cloud map and the stress cloud map, each pixel in the cloud map is annotated with a temperature and stress label. The temperature cloud map is used as the input image of the model, and the stress cloud map is used as the output image of the model. A heat-stress diagnostic model based on a convolutional neural network is trained to establish a mapping between the temperature and stress of each pixel in the cloud map.

[0088] In the above training process, by optimizing the model test, it is continuously iterated and trained until the preset requirements are met, such as reaching the set number of iterations or minimizing the loss function. At this time, the training is completed and the parameters and hyperparameters of the model (including convolutional layer parameters, fully connected layer parameters, etc.) are determined. In the optimization process, optimization algorithms such as SGD and Adam optimizer and loss functions such as mean_squared_error can be used to compile the model. Among them, Adam optimizer is an adaptive learning rate method based on first-order and second-order moment estimation, and mean square error loss function is commonly used for regression problems, including pixel-level prediction in image segmentation tasks. Preferably, in order to further improve the accuracy of the model, RMSProp, Namdam or LAMB optimizer can be used for optimization.

[0089] Furthermore, in this embodiment, considering that the model prediction focuses on the similarity between the visual characteristics of the cloud image (such as edges, textures, local structures, etc.) and the real cloud image, the structural similarity index (SSIM) is added, and the hybrid loss function is defined by combining MSE and SSIM. The formula is:

[0090] Loss = αMSE + β(1-SSIM);

[0091]

[0092] SSIM(x,y)=(2μ x μ y +C1)(2σ xy +C2)(μ x2 +μ y2 +C1)(σ x2 +σ y2 +C2);

[0093] In the above formula, α and β represent weight parameters; μ x and μ y Represents the mean of image x and y, representing brightness information; σ x2 and σ y2 Represents the variance of image x and y, respectively, representing the contrast information (i.e., the intensity of the change in brightness and darkness of the image); σ xy Represents the covariance of images x and y, representing the structural similarity between the two images; C1 and C2 are constants used to avoid the denominator being 0, usually C1 = (k1L) 2 , C2=(k2L) 2 , L is the range of pixel values ​​(for example, for an 8-bit depth image, L = 255), k1 and k2 are constants less than 1, usually k1 = 0.01, k2 = 0.03.

[0094] Considering that increasing the number of model parameters can improve the model's expressiveness and learning ability, but too many parameters may also lead to model overfitting and increased computational costs, while sufficient and diverse data can prevent model overfitting, but too much data will lead to increased computational costs. Therefore, through the above training, a balance is found between the number of model parameters and the amount of data to ensure a balance between model performance and computational costs.

[0095] Furthermore, the confusion matrix is ​​used as an indicator to evaluate the performance of the model, such as accuracy, and the model with better performance is selected as the diagnostic model of this embodiment.

[0096] As another implementation, the aforementioned model can be adapted from other models with image processing capabilities, such as convolutional neural networks. The matched temperature contour map is input into a trained heat-stress diagnostic model, which outputs a predicted stress contour map. The final diagnostic result is then generated based on this predicted image.

[0097] In step S4, based on the stress cloud map and combined with the preset diagnostic basis, the shield tunnel thermal disaster warning level is determined and a timely warning is issued.

[0098] like Figure 3 and Figure 4 The temperature and stress cloud maps of the tunnel segments shown in the figure can be used to determine the specific stress conditions of the segments through the thermal stress and structural mechanics of the entire tunnel segment and its coupling with the surrounding soil, so as to determine whether the segments are in danger of cracking.

[0099] Specifically, based on the predicted stress cloud map, the stress corresponding to each pixel in the map is clarified, and combined with the preset diagnostic basis, the overall warning level of the area where the pixel is located is determined. This is used to judge the warning level of thermal disasters in shield tunnels and issue timely warnings. Among them, the diagnostic basis is to divide the warning level into four levels: blue, yellow, orange and red, and their corresponding levels are level IV, level III, level II and level I, including:

[0100] (1) Blue (Level IV) warning: The segment stress is about to reach the tensile limit (the standard tensile strength of C50 concrete is 2.64 MPa), and the tunnel temperature needs to be controlled;

[0101] (2) Yellow (Level III) warning: The segment stress has reached its resistance limit and requires rapid cooling, otherwise the segment will crack.

[0102] (3) Orange (Level II) Warning: Segment cracking is minor and presents a risk. Repair and inspection should be arranged immediately.

[0103] (4) Red (Level I) warning: The cracks in the segments are severe and may cause an accident. The tunnel is shut down for maintenance.

[0104] Example 2

[0105] This embodiment provides a shield tunnel thermal disaster diagnosis system based on deep learning, including:

[0106] Infrared image acquisition module, used to collect infrared images of the temperature field of the inner arc surface of the shield tunnel segment and perform preliminary screening of the collected images;

[0107] The temperature cloud map matching module is used to perform color matching on the filtered infrared image to obtain a matching temperature cloud map;

[0108] The stress cloud map acquisition module is used to input the temperature cloud map into the trained thermal-stress diagnostic model and output the corresponding stress cloud map. The model training process includes: constructing a numerical model that simulates the thermal performance of the shield tunnel under the multi-field coupling of heat, permeation, and force; based on the numerical model, obtaining the temperature cloud map and stress cloud map of the temperature field on the inner curved surface of the shield tunnel segment through transient simulation calculation; and using the temperature cloud map and stress cloud map to train the thermal-stress diagnostic model based on the convolutional neural network.

[0109] The thermal disaster diagnosis module is used to determine the thermal disaster warning level of shield tunnels and issue timely warnings based on stress cloud maps and preset diagnostic criteria.

[0110] Example 3

[0111] This embodiment provides an electronic device, including: a memory for storing executable instructions; and a processor for implementing the above method provided in this embodiment when executing the executable instructions stored in the memory.

[0112] Example 4

[0113] This embodiment further provides a computer-readable storage medium storing executable instructions. When the executable instructions are executed by a processor, the processor will be caused to execute the above method provided in this embodiment.

[0114] Example 5

[0115] This embodiment provides a computer program product including executable instructions, which are computer instructions stored in a computer-readable storage medium. When a processor of an electronic device reads the executable instructions from the computer-readable storage medium and executes the executable instructions, the electronic device performs the method provided in this embodiment.

[0116] The steps involved in the above embodiments 2 to 5 correspond to those in embodiment 1. For detailed implementation, please refer to the relevant description of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media that includes one or more instruction sets; it should also be understood to include any medium that can store, encode, or carry an instruction set for execution by a processor and cause the processor to perform any method of the present invention.

[0117] Those skilled in the art will appreciate that the modules or steps of the present invention described above can be implemented using a general-purpose computer device. Alternatively, they can be implemented using program code executable by a computing device, which can then be stored in a storage device and executed by the computing device. Alternatively, they can be fabricated into separate integrated circuit modules, or multiple modules or steps can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0118] The above description is only a preferred embodiment of the present invention. Although the specific implementation of the present invention is described in conjunction with the accompanying drawings, it does not limit the scope of protection of the present invention. Those skilled in the art should understand that on the basis of the technical solution of the present invention, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present invention.

Claims

1. A shield tunnel thermal disaster diagnosis method based on deep learning, characterized in that: include: Collect infrared images of the temperature field on the inner curved surface of shield tunnel segments and conduct preliminary screening of the collected images; Perform chromaticity matching on the filtered infrared images to obtain a matching temperature cloud map; The temperature cloud map is input into the trained thermal-stress prediction and diagnosis model, and the corresponding stress cloud map is output. The model training process includes: constructing a numerical model that simulates the thermal performance of the shield tunnel under the multi-field coupling of heat, permeation, and force; based on the numerical model, obtaining the temperature cloud map and stress cloud map of the temperature field on the inner curved surface of the shield tunnel segment through transient simulation calculation; using the temperature cloud map and stress cloud map to train the thermal-stress diagnosis model based on the convolutional neural network; Based on the stress cloud map and the preset diagnostic basis, the early warning level of thermal disasters in shield tunnels can be determined and timely warnings can be issued; The heat-stress diagnosis model based on convolutional neural network is built using U-Net architecture, including input layer, encoder, bottleneck layer, decoder and output layer; The encoder includes a plurality of sequentially connected convolution blocks, each of which includes two convolution layers as downsampling layers, and each convolution block is followed by a Dropout layer and a maximum pooling layer; the convolution layer adopts a Conv2D layer, each Conv2D layer has a set number of filters, and the convolution layer is followed by a Mish activation layer, a regularization layer, and a Dropout layer; The decoder includes a plurality of sequentially connected convolution blocks, each of which includes a plurality of upsampling layers, and the upsampling layers are jump-connected with a plurality of downsampling layers of the encoder to gradually restore the size of the feature map.

2. A shield tunnel thermal disaster diagnosis method based on deep learning according to claim 1, characterized in that: A transient simulation calculation is performed on the thermal performance of the shield tunnel under the coupled effects of heat, penetration, and force. During the transient simulation calculation process, temperature cloud maps and stress cloud maps of the temperature field on the curved surface of the tunnel segment are obtained at set time intervals. The duration of the transient simulation calculation process is the entire life cycle of the shield tunnel.

3. The shield tunnel thermal disaster diagnosis method based on deep learning according to claim 1, characterized in that: Before using an infrared imager to capture the infrared image of the temperature field on the inner curved surface of a shield tunnel segment, the color-temperature legend is adjusted to be consistent with the legend of the simulated temperature cloud map to achieve rapid color matching.

4. The shield tunnel thermal disaster diagnosis method based on deep learning according to claim 1, characterized in that: The ratio of the color-temperature legend of the infrared image and the temperature cloud map is obtained, and the color-temperature legend of each infrared image obtained is adjusted using the ratio to obtain a temperature cloud map after chromaticity matching.

5. The shield tunnel thermal disaster diagnosis method based on deep learning according to claim 1, characterized in that: Based on the numerical model, through transient simulation calculation, the temperature cloud map and stress cloud map of the temperature field on the inner arc surface of the shield tunnel segment are obtained; Based on the temperature cloud map and stress cloud map, the temperature and stress of each pixel in the cloud map are annotated. The temperature cloud map is used as the input image of the model, and the stress cloud map is used as the output image of the model to train the heat-stress diagnosis model based on the convolutional neural network. The model training is completed through continuous iterative training until the set number of iterations is met or the loss function is minimized, and the mapping between the temperature and stress of each pixel in the cloud map is established; among them, the loss function is a hybrid loss function based on the mean square error index MSE and the structural similarity index SSIM.

6. A shield tunnel thermal disaster diagnosis system based on deep learning, characterized in that: include: Infrared image acquisition module, used to collect infrared images of the temperature field of the inner arc surface of the shield tunnel segment and perform preliminary screening of the collected images; The temperature cloud map matching module is used to perform color matching on the filtered infrared image to obtain a matching temperature cloud map; The stress cloud map acquisition module is used to input the temperature cloud map into the trained thermal-stress diagnostic model and output the corresponding stress cloud map. The model training process includes: constructing a numerical model that simulates the thermal performance of the shield tunnel under the multi-field coupling of heat, permeation, and force; based on the numerical model, obtaining the temperature cloud map and stress cloud map of the temperature field on the inner curved surface of the shield tunnel segment through transient simulation calculation; and using the temperature cloud map and stress cloud map to train the thermal-stress diagnostic model based on the convolutional neural network. Thermal disaster diagnosis module, which is used to determine the shield tunnel thermal disaster warning level and issue timely warnings based on stress cloud diagrams and preset diagnostic criteria; The heat-stress diagnosis model based on convolutional neural network is built using U-Net architecture, including input layer, encoder, bottleneck layer, decoder and output layer; The encoder includes a plurality of sequentially connected convolution blocks, each of which includes two convolution layers as downsampling layers, and each convolution block is followed by a Dropout layer and a maximum pooling layer; the convolution layer adopts a Conv2D layer, each Conv2D layer has a set number of filters, and the convolution layer is followed by a Mish activation layer, a regularization layer, and a Dropout layer; The decoder includes a plurality of sequentially connected convolution blocks, each of which includes a plurality of upsampling layers, and the upsampling layers are jump-connected with a plurality of downsampling layers of the encoder to gradually restore the size of the feature map.

7. An electronic device, characterized in that: include: a memory for storing executable instructions; The processor is configured to implement the shield tunnel thermal disaster diagnosis method based on deep learning as described in any one of claims 1 to 5 when executing the executable instructions stored in the memory.

8. A computer-readable storage medium, characterized in that Executable instructions are stored, which are used to cause the processor to execute the executable instructions to implement the deep learning-based shield tunnel thermal disaster diagnosis method described in any one of claims 1 to 5.

9. A computer program product, characterized in that The computer program product includes executable instructions stored in a computer-readable storage medium; When the processor of the electronic device reads the executable instructions from the computer-readable storage medium and executes the executable instructions, the deep learning-based shield tunnel thermal disaster diagnosis method described in any one of claims 1 to 5 is implemented.

Images