A lining fire damage grade evaluation processing method based on multi-source data fusion

Abstract

The application provides a lining fire damage level evaluation processing method based on multi-source data fusion, and relates to the fields of tunnel engineering safety and fire damage evaluation. Through obtaining fire data of a target lining, infrared thermal images and visible light images in the fire data are subjected to image fusion processing to generate a fusion image with a temperature label; a U-NET network model is used to perform damage level zone division processing on the fusion image to obtain a plurality of target damage level zones and the highest temperature corresponding to each target damage level zone; for each target damage level zone, a component damage index corresponding to the target damage level zone is obtained; based on a preset DNN network model, the damage degree corresponding to the target damage level zone and the corresponding repair priority are determined. The problems of incomplete and low-efficiency artificial detection of damage levels, focus on a single result and independent data of existing evaluation models are solved, and the technical effects of multi-angle evaluation, improved accuracy and efficiency are achieved.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering safety and fire damage assessment, and in particular to a method for assessing the fire damage level of tunnel linings based on multi-source data fusion. Background Technology

[0002] Tunnel safety has always been a crucial issue in the construction and operation of transportation infrastructure. With the continuous expansion of the scale of tunnel projects and the rapid increase in operational mileage in my country, the risk of fire accidents inside tunnels has also risen.

[0003] Currently, the main approach relies on manual on-site inspection, using visual observation, rebound hammers, and ultrasonic equipment to record and qualitatively describe defects such as lining cracks, spalling, and water leakage. Alternatively, a combination of sensing, numerical simulation, and artificial intelligence technologies can be used to generate assessment and treatment models for damage assessment and treatment of target linings under fire conditions.

[0004] However, manual on-site inspection has shortcomings such as strong subjectivity, low efficiency, and limited coverage, making it difficult to conduct comprehensive inspections in harsh environments such as high temperatures and toxic fumes after a fire. Existing assessment and treatment models mostly focus on a single form of damage (such as the blast radius), lacking a comprehensive characterization and unified quantitative framework for multiple types of damage. Data generated by different detection methods (such as manual visual inspection, rebound hammer, and ultrasound) are independent and lack an effective data fusion mechanism, failing to leverage the complementary advantages of multi-source heterogeneous data and resulting in low accuracy. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a method for assessing the fire damage level of lining based on multi-source data fusion, so as to solve the problems of insufficient comprehensiveness and low efficiency in the assessment of the target lining level after manual fire detection in the prior art, and the problems of low accuracy caused by the overly singular and scattered data in the existing damage level assessment.

[0006] In a first aspect, embodiments of the present invention provide a method for assessing the fire damage level of linings based on multi-source data fusion, comprising: Fire data of the target lining is acquired, and the infrared thermal image and visible light image in the fire data are fused to generate a fused image with temperature labels. Using the U-NET network model, the fused image is divided into damage level regions to obtain multiple target damage level regions and the highest temperature corresponding to each target damage level region. The processing for each target damage level area is as follows: Based on the correlation table between the target damage level area and the damage parameter, the target loss parameter corresponding to the target damage level area is obtained; Based on the highest temperature corresponding to the target damage level zone, determine the material damage index corresponding to the target damage level zone, and based on the material damage index and the target loss parameter, determine the component damage index of the target damage level zone. The component damage index is evaluated using a pre-trained DNN network model to obtain the damage level corresponding to the target damage level area and the repair priority corresponding to the damage level.

[0007] In conjunction with the first aspect, embodiments of the present invention provide a first possible implementation of the first aspect, wherein image fusion processing is performed on the infrared thermal image and visible light image in the fire data to generate a fused image with temperature labels, including: Based on the scale-invariant feature transformation processing method, feature extraction processing is performed on the infrared thermal image and the visible light image to obtain common stable features; Calculate the homography transformation matrix of the common stable features, and based on the homography transformation matrix, map the infrared temperature field in the infrared thermogram to the coordinates of the visible light image to obtain the fused image with temperature labels.

[0008] In conjunction with the first aspect, this invention provides a second possible implementation of the first aspect, wherein the fused image is subjected to damage level region segmentation processing using a U-NET network model to obtain multiple target damage level regions, including: Based on the convolution operation of the U-NET network model, multiple shallow feature maps related to the semantics of lining damage are obtained; Using the encoder in the U-NET network model, each shallow feature map is downsampled to obtain a deep feature map related to the semantics of lining damage. Using the decoder corresponding to the encoder, each of the depth feature maps is upsampled and fused with the shallow feature maps associated with the deep feature maps to generate a high-dimensional feature map semantically related to lining damage. Each of the high-dimensional feature maps is then processed as follows: Calculate the target feature value of the target pixel in the high-dimensional feature map, match the target feature value with the damage threshold range, and determine the target damage level corresponding to the target pixel; The multiple target pixels corresponding to the target damage level are taken as the target damage level area, and the maximum value of the temperature label under the target damage level area is taken as the highest temperature of the target damage level area.

[0009] In conjunction with the first aspect, this embodiment of the invention provides a third possible implementation of the first aspect, wherein the target damage level area includes: a bursting area, a cracking area, and a leakage area; then, obtaining the target loss parameters corresponding to the target damage level area includes: For the bursting zone, the bursting depth and bursting volume are calculated based on the point cloud data corresponding to the bursting zone and the surface of the first target lining. For the crack area, the target high-dimensional feature map corresponding to the crack area is subjected to grayscale and denoising processing, and the average width and length of the crack in the high-dimensional feature map after grayscale and denoising processing are calculated based on edge algorithm and skeleton extraction algorithm. For the leakage area, obtain the leakage area corresponding to the leakage area and the area of ​​the second target lining surface, and calculate the ratio of the leakage surface set to the area of ​​the second target lining surface to obtain the leakage area ratio of the lining surface. The first lining surface and the second lining surface are different.

[0010] In conjunction with the first aspect, this invention provides a third possible implementation of the first aspect, wherein the material damage index includes: concrete elastic modulus loss coefficient, concrete compressive strength residual coefficient, and steel reinforcement yield strength reduction coefficient; then, determining the material damage index corresponding to the target damage level zone based on the highest temperature includes: The highest temperature in the target damage level area is matched with a preset temperature range to determine the target temperature range to which the highest temperature belongs. For the material damage index values ​​corresponding to the target range, determine the concrete elastic modulus loss coefficient and the concrete compressive strength residual coefficient; Based on the temperature value of the reinforcing steel, the reduction factor for the yield strength of the reinforcing steel is obtained using the following formula:

[0011] in, Indicates the reduction factor of the yield strength of steel bars. Indicates the temperature of the reinforcing steel.

[0012] In conjunction with the first aspect, this invention provides a fourth possible implementation of the first aspect, wherein determining the component damage index of the target damage level zone based on the material damage index and the target loss parameter includes: Obtain the critical value of the preset weight and the target loss parameter; The component indexes for the target damage level zone are calculated using the following formula:

[0013]

[0014] in, to This indicates the preset weight, and H represents the burst depth. This indicates the critical value of the burst depth. Indicates the burst volume. This represents the critical value of the burst volume. Indicates the average width of the crack. This represents the critical value of the average crack width. Indicates the crack length. This represents the critical value for crack length. This indicates the proportion of the leakage area to the total cross-sectional area. Indicates the elastic loss coefficient of concrete. This represents the residual compressive strength coefficient of concrete.

[0015] In conjunction with the first aspect, embodiments of the present invention provide a fifth possible implementation of the first aspect, further comprising: Obtain point cloud data from the fire data; Based on the external parameters corresponding to the target device that acquires the visible light image, the point cloud data is mapped one by one into two-dimensional pixel coordinates; For each of the two-dimensional pixel coordinates, the nearest target pixel is determined on the fused image, and the temperature label corresponding to the target pixel is assigned to the point cloud data corresponding to the two-dimensional pixel coordinates to obtain a lining three-dimensional real scene model with temperature label. The damage level and repair priority corresponding to each target damage level area are mapped onto the three-dimensional real scene model and displayed in a visual interface.

[0016] Secondly, embodiments of the present invention provide a lining fire damage level assessment and processing device based on multi-source data fusion, comprising: The acquisition module is used to acquire fire data of the target lining, and perform image fusion processing on the infrared thermal image and visible light image in the fire data to generate a fused image with temperature labels. The segmentation module is used to perform damage level region segmentation processing on the fused image using the U-NET network model, to obtain multiple target damage level regions and the highest temperature corresponding to each target damage level region. The processing module is used to process each target damage level area as follows: based on the correlation table between the target damage level area and the damage parameter, obtain the target loss parameter corresponding to the target damage level area; determine the material damage index corresponding to the target damage level area according to the highest temperature corresponding to the target damage level area, and determine the component damage index of the target damage level area according to the material damage index and the target loss parameter; evaluate the component damage index using a pre-trained DNN network model to obtain the damage degree corresponding to the target damage level area, and determine the repair priority corresponding to the damage degree.

[0017] Thirdly, embodiments of the present invention also provide an electronic device, comprising: At least one processor; and at least one memory communicatively connected to the processor, wherein: the memory stores program instructions executable by the processor, and the processor invokes the program instructions to perform the lining fire damage level assessment processing method as described in any of the first aspects of the present invention.

[0018] This invention acquires fire data of the target lining, performs image fusion processing on the infrared thermal image and visible light image in the fire data to generate a fused image with temperature labels; uses the U-NET network model to divide the fused image into damage level zones, obtaining multiple target damage level zones and the highest temperature corresponding to each target damage level zone; the processing for each target damage level zone is as follows: based on the correlation table between target damage level zones and damage parameters, the target loss parameter corresponding to the target damage level zone is obtained; based on the highest temperature corresponding to the target damage level zone, the material damage index corresponding to the target damage level zone is determined, and based on the material damage index and the target loss parameter, the component damage index of the target damage level zone is determined; the component damage index is input into a preset DNN network model to obtain the damage degree corresponding to the target damage level zone and the repair priority corresponding to the damage degree.

[0019] The embodiments of this invention bring the following beneficial effects: avoiding problems such as incomplete detection, low efficiency, and strong subjectivity caused by relying on manual on-site detection; at the same time, the three-layer collaborative fusion framework of data level, feature level, and decision level established by this invention solves the problem that each data source operates independently and the complementary advantages are difficult to bring into play in the prior art; and based on the complementary advantages between data, the accuracy of target lining damage level assessment under fire can be improved; in addition, the processing method provided by this invention can also improve processing efficiency.

[0020] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0022] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0023] Figure 1 A schematic flowchart of the method for assessing and processing the fire damage level of lining provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a lining fire damage level assessment and treatment device provided in an embodiment of the present invention; Figure 3 A block diagram of an exemplary electronic device provided in an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] Tunnel safety has always been a crucial issue in the construction and operation of transportation infrastructure. With the continuous expansion of tunnel engineering scale and the rapid increase in operational mileage in my country, the risk of fire accidents within tunnels has also risen. Tunnel fires are often caused by vehicle collisions, spontaneous combustion of goods, or electrical faults. High temperatures can lead to various types of damage to the tunnel lining concrete, including bursting, strength deterioration, cracking, and leakage, severely weakening the structural load-bearing capacity and even causing tunnel collapse, threatening people's lives and property. Therefore, accurate, rapid, and comprehensive characterization and assessment of target lining damage after a fire is key to ensuring safe tunnel operation and developing repair strategies.

[0026] Existing technologies primarily rely on the following two methods for evaluation and processing: Method 1: Primarily relies on manual on-site inspection, using visual observation, rebound hammers, ultrasonic equipment, etc., to record and qualitatively describe defects such as lining cracks, spalling, and water leakage. However, these methods suffer from drawbacks such as high subjectivity, low efficiency, and limited coverage, making it difficult to conduct comprehensive inspections in harsh environments such as high temperatures and toxic fumes after a fire.

[0027] Method 2: With the development of sensing, numerical simulation and artificial intelligence technologies, some studies have begun to introduce multi-source data acquisition, finite element analysis and comprehensive evaluation models, but the following problems exist: the types of lining damage after a fire are diverse, and existing methods mostly focus on a single damage form (such as the range of bursting), lacking a comprehensive characterization and unified quantitative framework for multiple types of damage such as bursting, cracking, strength deterioration, and leakage; the data generated by different detection methods (such as manual visual inspection, rebound hammer, ultrasonic waves, etc.) are independent and lack an effective data fusion mechanism, which cannot give full play to the complementary advantages of multi-source heterogeneous data.

[0028] Based on this, embodiments of the present invention provide a method for assessing the fire damage level of linings based on multi-source data fusion, which can solve the problems of incomplete, inefficient, and subjective manual on-site detection of lining fire damage in existing technologies. At the same time, it can solve the problems that existing assessment models only focus on a single damage result, and the data are independent and lack an effective data fusion mechanism, which cannot give full play to the complementary advantages of multi-source heterogeneous data and has insufficient accuracy. In addition, existing assessment models mostly rely on manually set index weights and thresholds, which are highly subjective and have poor adaptability, and the overall level of intelligence is insufficient, making it difficult to meet the engineering needs of rapid and accurate post-fire assessment.

[0029] To facilitate understanding of this embodiment, a method for assessing and processing the damage level of lining fires based on multi-source data fusion, as disclosed in this embodiment of the invention, will first be described in detail, with reference to... Figure 1 , Figure 1 This is a schematic flowchart illustrating the method for assessing and processing the fire damage level of linings provided in an embodiment of the present invention. Figure 1 As shown, the method includes: Step 101: Obtain fire data of the target lining, perform image fusion processing on the infrared thermal image and visible light image in the fire data, and generate a fused image with temperature labels.

[0030] It should be noted that the executing entity in this embodiment of the invention is an assessment and processing device, or a device that carries or installs a lining fire damage level assessment and processing device, such as a server. This embodiment of the invention uses a lining fire damage level assessment and processing device as an example for explanation; for ease of explanation, it will be referred to as the assessment and processing device in the following simplification.

[0031] The method provided by this invention is mainly applied in underground structures such as tunnels, primarily for assessing the damage level of the lining after a fire. Taking a tunnel as an example, the tunnel is equipped with an infrared temperature imager (with parameters such as wavelength 8-14μm and temperature resolution ≤0.05℃), a three-dimensional laser scanner (with parameters such as scanning speed ≥500,000 points / second and ranging accuracy ±1mm), and a visible light high-definition camera (with parameters such as resolution ≥20 million pixels and a ring-shaped supplementary light).

[0032] Optionally, the infrared thermal imager, 3D laser scanner, and visible light high-definition camera are integrated and installed on the same acquisition platform that can move longitudinally along the tunnel (such as a railcar or self-propelled trolley), with the platform's travel speed controlled at 2-5 km / h.

[0033] In the same post-fire scenario, all three devices simultaneously triggered data acquisition at the same time and cross-sectional location. During acquisition, the infrared thermal imager collected temperature field data of the lining surface along three height lines: the tunnel arch, the left and right arch waists, and the left and right sidewalls, with no fewer than 9 measuring points at each cross-section. The 3D laser scanner acquired point cloud data of the entire tunnel cross-section, with a point cloud density ≥1000 points / m²; and acquired images of the lining surface, requiring each image to cover a 1m×1m area with an overlap rate ≥30%.

[0034] In a real-world scenario, after a tunnel fire is extinguished and the ambient temperature drops below 40°C and the concentration of harmful gases meets safety standards (CO concentration below 24 ppm, O2 content not less than 19.5%), the following three types of detection equipment are used to collect data along the longitudinal direction of the tunnel at 5-meter intervals. This can be understood as collecting fire data on the target lining.

[0035] These three types of sensors upload the collected fire data to the assessment and processing device, which then performs multi-source data fusion processing to enhance the complementary advantages between the data and improve the accuracy of data processing. First, the infrared thermal image and visible light image in the fire data need to be fused to generate a fused image with temperature labels.

[0036] The evaluation processing device needs to perform preprocessing operations on infrared thermal images, visible light images, and 3D point cloud data to remove redundant and outlier data, thereby improving processing efficiency and accuracy.

[0037] Optionally, spatiotemporal synchronization of data from various sensors is performed: using the laser scanner's clock as a reference, synchronous acquisition of data from the infrared thermal imager and the visible light camera is achieved via an FPGA synchronization trigger board, with a time difference ≤10ms. The raw data is preprocessed: the infrared thermal image undergoes median filtering (3×3 window size) to remove noise; the visible light image undergoes histogram equalization to enhance contrast; the point cloud data undergoes statistical filtering (50 neighboring points, standard deviation factor 1.0) to remove outliers, and then is registered to the tunnel design coordinate system using an iterative nearest-point algorithm.

[0038] When the preprocessed fire data is acquired, the evaluation processing device can perform feature extraction processing on the infrared thermogram and the visible light image using a scale-invariant feature transformation processing method to obtain common stable features; calculate the homography transformation matrix of the common stable features, and based on the homography transformation matrix, map the infrared temperature field in the infrared thermogram to the coordinates of the visible light image to obtain a fused image with temperature labels.

[0039] Specifically, feature points are extracted from the preprocessed infrared thermal image and the visible light image using scale-invariant feature transformation, and then registered using a homography matrix to generate a temperature-visible light fused image for each cross-section. Optional fusion conditions: the number of feature points in the overlapping area of ​​the two images is ≥50, and the registration error is ≤2 pixels.

[0040] The cross-section is pre-defined for the structure of each tunnel. Different tunnels have different cross-sections, which are determined by specific circumstances. The fire data collected in this embodiment of the invention are all collected based on different cross-sections.

[0041] After acquiring the fused image with temperature labels, the evaluation processing device needs to perform a division process including graded regions, which will be further explained in step 102.

[0042] Step 102: Using the U-NET network model, the fused image is divided into damage level regions to obtain multiple target damage level regions and the highest temperature corresponding to each target damage level region.

[0043] It should be noted that the U-NET network model is deployed in the evaluation processing device for image segmentation. Optionally, based on the convolution operation of the U-NET network model, multiple shallow feature maps related to the semantics of lining damage are obtained; using the encoder in the U-NET network model, each shallow feature map is downsampled to obtain a deep feature map related to the semantics of lining damage; using the decoder corresponding to the encoder, each deep feature map is upsampled, and the shallow feature maps associated with the deep feature maps are fused to generate a high-dimensional feature map related to the semantics of lining damage. For each high-dimensional feature map, the following processing is performed: the target feature value of the target pixel in the high-dimensional feature map is calculated, and the target feature value is matched with the damage threshold interval to determine the target damage level corresponding to the target pixel; multiple target pixels corresponding to the target damage level are taken as the target damage level region, and the maximum value of the temperature label under the target damage level region is taken as the highest temperature of the target damage level region.

[0044] Specifically, a pre-trained U-Net deep learning model (with a ResNet34 encoder and a 5-layer upsampling decoder) is used to perform semantic segmentation on the fused image, outputting pixel-level segmentation results to obtain multiple target damage level regions. The IoU threshold in the U-Net deep learning model is set to 0.7.

[0045] For example, the input fused image of the U-NET network model is 512×512 pixels in size and has 4 channels (R, G, B channels plus a temperature pseudo-color channel).

[0046] The encoder (downsampling path) consists of four stages, each containing two 3×3 convolutions (stride 1, padding 1, batch normalization + ReLU) and one 2×2 max pooling (stride 2) to obtain depth feature maps.

[0047] Specific output dimensions for each stage: Stage 1: Input 512×512×4 → After two convolutions (output channels 64) → Output 512×512×64 → After pooling 256×256×64; Stage 2: Input 256×256×64 → Two convolutions (128 output channels) → 256×256×128 → Pooled 128×128×128; Stage 3: Input 128×128×128 → Two convolutions (output channels 256) → 128×128×256 → Pooled 64×64×256; Stage 4 (bottleneck): Input 64×64×256 → Two convolutions (output channels 512) → 64×64×512 → Pooled 32×32×512.

[0048] The decoder (upsampling path) consists of four stages to obtain high-dimensional feature maps: Stage 1: Upsampling (transposed convolution 2×2, stride 2) transforms 32×32×512 into 64×64×256, which is then concatenated with the feature map from stage 3 of the encoder (64×64×256+64×64×256=64×64×512), followed by two more convolutions (output channel 256) into 64×64×256.

[0049] Stage 2: Upsample to 128×128×128, stitch encoder stage 2 (128×128×128) → 128×128×256, two convolution outputs 128×128×128.

[0050] Stage 3: Upsample to 256×256×64, stitch encoder stage 1 (256×256×64) → 256×256×128, two convolutions output 256×256×64.

[0051] Stage 4: Upsample to 512×512×32, and stitch together the shallow feature map after downsampling of the original input image. In fact, the U-Net decoder stitches together the shallow feature map corresponding to the deep feature map of the encoder in the first stage in the final stage.

[0052] The process of obtaining shallow feature maps involves determining whether the original input image needs to be first mapped to 32 channels via a 1×1 convolution to obtain shallow feature maps. A common approach is to transform the input image (512×512×4) into 512×512×32 via a 1×1 convolution (stride 1), and then concatenate it with the upsampled result. Alternatively, the output of the first stage of the encoder (512×512×64) can be used directly, but the number of channels needs to be reduced.

[0053] The final output layer: a 1×1 convolution (stride 1) maps 512×512×32 to 512×512×4 (corresponding to 4 damage categories), and the classification probability of each pixel is obtained by Softmax.

[0054] The specific processing steps for each convolutional layer (including kernel size, stride, padding, activation function, normalization, etc.) are as follows: all convolutional layers (except transposed convolution and output layers) use: kernel size 3×3, stride 1, padding 1 (keeping the feature map size unchanged), followed by batch normalization, and the activation function is ReLU.

[0055] Max pooling layer: pooling window 2×2, stride 2 (no overlap), no padding, size halved.

[0056] Transposed convolution (upsampling): Convolution kernel size 2×2, stride 2, output channels halved (e.g. 512→256), no padding, used to enlarge feature map size.

[0057] The splicing operation connects the feature maps of the corresponding stages of the encoder with the upsampled feature maps along the channel dimension.

[0058] Output layer: 1×1 convolution, stride 1, no padding, output channels equal to the number of classes (4), no batch normalization and ReLU are used, and logits are output directly.

[0059] The adjustment method of the step size of the U-NET network and its impact on the feature map size can be understood as follows: Downsampling stage: Each max pooling (step size 2) halves the width and height of the feature map, so from 512→256→128→64→32.

[0060] Upsampling stage: Each transposed convolution (stride 2) doubles the width and height of the feature map, from 32→64→128→256→512.

[0061] The stride of the convolutional layer is fixed at 1, and the size is not changed; padding with 1 ensures that the output size is equal to the input size.

[0062] Therefore, the output size of the entire U-NET network is the same as the input size (512×512), achieving pixel-level semantic segmentation.

[0063] In summary, various target damage level areas can be obtained. Optionally, the target damage level areas involved in the embodiments of the present invention include, but are not limited to: bursting area, crack area, leakage area and intact area.

[0064] After acquiring the target damage level areas, the assessment and processing device needs to evaluate the damage degree of each target damage level area one by one. First, it is necessary to acquire the loss parameters of each target damage level area. Step 103 provides a detailed explanation of this processing step.

[0065] Step 103: For each target damage level area, process as follows: Based on the correlation table between target damage level areas and damage parameters, obtain the target loss parameters corresponding to the target damage level area.

[0066] Optionally, taking the aforementioned example, the target damage level area includes: a bursting area, a crack area, and a leakage area. The target loss parameters corresponding to the target damage level area are obtained as follows: For the bursting area, the bursting depth and bursting volume are calculated based on the point cloud data corresponding to the bursting area and the surface of the first target lining. For the crack area, the target high-dimensional feature map corresponding to the bursting area is subjected to grayscale and denoising processing, and the average width and length of the cracks in the high-dimensional feature map after grayscale and denoising processing are calculated based on edge algorithms and skeleton extraction algorithms. For the leakage area, the leakage area corresponding to the leakage area and the area of ​​the second target lining surface are obtained, and the ratio of the leakage surface set to the surface area of ​​the second target lining is calculated to obtain the leakage area ratio of the lining surface. The first lining surface and the second lining surface are different.

[0067] Specifically, for the blast zone, the method for calculating the blast depth (H) and volume (V) by combining the 3D point cloud data from the fire data is as follows: The operation to obtain the burst depth H is as follows: From the high-dimensional feature map, extract all three-dimensional point cloud data corresponding to the burst area. This three-dimensional point cloud data is a subset of the three-dimensional point cloud data acquired by the laser scanner. The coordinates of each point in this point cloud dataset are denoted as (…). , , ), where the Z-axis is the radial direction of the tunnel (whether it points to the center of the tunnel is positive or negative needs to be pre-calibrated).

[0068] Furthermore, a ring-shaped adjacent intact area (10cm-20cm beyond the blast boundary) is selected around the blast zone, and the point cloud of all intact lining surfaces within this area is extracted. The average radial coordinate of these points is then calculated. As: the radial position of the original lining surface when undamaged.

[0069] For each point within the burst zone, calculate its radial depth difference Δ. = - , among which when Less than Δ A positive value indicates a depressed state in the burst area.

[0070] The blast depth H is taken as the maximum value of the depth difference among all points within the blast zone, i.e., H = max(Δ ), optionally, the average depth can also be used, and the method of obtaining it becomes H = average depth - average depth of adjacent intact areas, then the average depth is the average depth of all points within the blast zone. The average depth of the adjacent intact area is the average depth. The average burst depth is obtained by subtracting the two. The maximum depth is usually used to characterize the most severe damage.

[0071] The procedure for obtaining the burst volume V is as follows: The burst area is meshed on the lining surface (two-dimensional manifold), with the mesh step size being the average spacing of the point cloud (e.g., 1cm × 1cm). Delaunay triangulation or regular grid projection can be used. For each mesh cell (small rectangle or triangle), its area is obtained as dA (unit: m²), and the radial depth difference corresponding to that cell is the average of the depth differences of all point clouds within that cell, Δ. Alternatively, the depth difference at the center of the element can be obtained through interpolation. The burst volume element of this tiny element is dV = Δ ×dA; then, perform volume integration on all mesh elements in the entire burst zone: V=Σ(Δ ×d The unit is cubic meters (m³). When Δ A negative value indicates that the bursting zone is in a raised state, and it is not included in the calculation; only the concave part is calculated.

[0072] To ensure computational accuracy, the point cloud density must be ≥1000 points / m², and the boundaries of the burst zone must be precisely defined by semantic segmentation.

[0073] This can be understood as acquiring the point cloud data corresponding to the burst zone and the surface of the first target lining, and calculating the radial distance from each point cloud data point to the surface of the first target lining. The surface of the first target lining can be obtained by fitting the point cloud of the surrounding intact area, such as by fitting a local plane or quadratic surface using the least squares method. The maximum radial distance is then taken as the depth of the burst zone.

[0074] The point cloud data of the burst zone is projected onto the fitted first target lining surface, and the burst volume is obtained using the trapezoidal rule or triangular mesh integration. Specifically, the burst zone is divided into multiple tiny triangular faces, each forming a triangular prism with the original surface; the volumes are calculated and then summed.

[0075] For the crack area, before performing edge detection and skeleton extraction algorithms, the fused image corresponding to the crack area needs to be processed into a grayscale image, and median filtering or Gaussian filtering should be used to remove image noise to improve processing accuracy. Subsequently, histogram equalization or adaptive contrast enhancement methods are used to improve the grayscale difference between the crack and the background lining surface.

[0076] Optionally, embodiments of the present invention may employ the Canny edge detection algorithm to extract crack edges. Specifically, this includes: smoothing the image using Gaussian filtering, calculating the image grayscale gradient magnitude and direction, thinning the edges using non-maximum suppression, and connecting the crack edges using a double thresholding method to obtain a binary image of the crack edges. After obtaining the binary image of the crack edges, closing operations and hole filling are performed on the crack region to form a continuous crack region mask.

[0077] Subsequently, a skeleton extraction algorithm was used to refine the crack region, compressing it into a centerline skeleton of single-pixel width. The total crack length L was calculated based on the number of skeleton pixels and the actual pixel size. For horizontally and vertically adjacent skeleton pixels, the length was calculated as one pixel; for diagonally adjacent skeleton pixels, the length was calculated as... Calculation based on pixel size. The total length of the crack can be obtained using the following formula:

[0078] Where L is the total length of the crack, in meters; The number of skeleton pixels connected horizontally or vertically; represents the number of diagonally connected skeleton pixels; r represents the actual length of a single pixel, in meters per pixel. The actual pixel length r can be determined through camera calibration, a scale reference, or the registration relationship between the image and the 3D point cloud.

[0079] The average crack width W is calculated based on the relationship between the area of ​​the crack region and the length of the crack skeleton. First, the number of crack pixels in the crack region mask is counted. And based on the actual pixel area obtained. Calculate the actual area of ​​the crack using the following formula:

[0080] Based on this, the average crack width is obtained by calculating the ratio of crack area to total crack length, as shown in the following formula:

[0081] Where W is the average crack width, in meters (m); it can be converted to mm in the output. Optionally, when multiple cracks exist within the same component or cross-section in the same crack zone, the length of each crack is calculated separately. and average width The equivalent average crack width of the component or section is calculated by weighting it by length.

[0082] Correspondingly, the total length of the crack is:

[0083] in, The equivalent average crack width, This represents the total length of the crack. (Calculated) and These are used as the crack width and crack length parameters in the calculation of the component damage index, respectively.

[0084] Optionally, the crack width can be obtained by calculating the normal width of each pixel of the skeleton after masking and refining the crack, and taking the average value to obtain the average width of the crack area.

[0085] For the leakage area, the evaluation and processing device determines the high-dimensional feature map corresponding to the leakage area, and determines the leakage region to be analyzed corresponding to the second target lining surface based on the high-dimensional feature map. Then, the lining surface image is determined; a semantic segmentation model is used to perform pixel-level classification on the second target lining surface image, outputting a binary mask image of the leakage area, where pixels in the leakage area are assigned a value of 1, and pixels in the non-leakage area are assigned a value of 0. To reduce the influence of noise and isolated false positives, morphological opening and closing operations are sequentially performed on the binary mask to remove isolated spots with an area smaller than a preset threshold and fill the voids inside the leakage area.

[0086] After obtaining the binary mask of the leakage area, the high-dimensional feature map corresponding to the leakage area is registered with the three-dimensional point cloud or the coordinate system of the design section corresponding to the leakage area, and the pixels of the leakage area are converted into the actual spatial area.

[0087] For example, if a two-dimensional image is used for calculation, the actual area s_pixel corresponding to a single pixel is determined based on the camera calibration parameters and the image scale, and the leakage area is calculated using the following formula:

[0088] in, This represents the area of ​​the leakage zone, in m². This represents the number of pixels in the leakage area. This represents the actual area corresponding to a single pixel, expressed in m² / pixel.

[0089] For example, if 3D point cloud computing is used, the leakage area mask is mapped to the corresponding point cloud region, and the surface area of ​​the leakage area is calculated according to the point cloud meshing or triangulation reconstruction method. Second target: lining surface area. The actual unfolded area of ​​the detectable region of the lining surface within the current detection section can be taken, or the lining surface area of ​​the corresponding analysis unit of that section can be taken. For detection units divided according to longitudinal spacing Δl, the arc length of the inner contour of the lining can be used. Calculation with detection length Δl:

[0090] in, The cross-section is used to evaluate the inner surface profile length of the second target lining, in meters (m); Δl is the distance between adjacent test cross-sections or the longitudinal length represented by this cross-section, in meters. Ultimately, the percentage of leakage area on the surface of the second target lining is... Calculate using the following formula:

[0091] It should be noted that when multiple leakage areas exist within the same cross-section of a tunnel, the area of ​​each leakage area should be calculated separately, and then the total leakage area of ​​the cross-section should be obtained by summing them up.

[0092]

[0093] in, Let be the area of ​​the i-th leakage zone.

[0094] After obtaining the target damage parameters under each target damage level zone, the evaluation and processing device adopts the following processing steps to obtain the component damage index of the target damage level zone.

[0095] Step 104: Determine the material damage index corresponding to the target damage level area based on the highest temperature corresponding to the target damage level area, and determine the component damage index of the target damage level area based on the material damage index and the target loss parameter.

[0096] Before obtaining component damage indicators, material damage indicators must be obtained.

[0097] Optionally, material damage indicators include: concrete elastic modulus loss coefficient, concrete compressive strength residual coefficient, and steel reinforcement yield strength reduction coefficient; the process for obtaining the material damage indicators corresponding to each target damage level zone is as follows.

[0098] Match the highest temperature in the target damage level zone with the preset temperature range to determine the target temperature range to which the highest temperature belongs; determine the concrete elastic modulus loss coefficient and the concrete compressive strength residual coefficient for the material damage index values ​​corresponding to the target range; and obtain the steel reinforcement yield strength reduction coefficient based on the steel reinforcement temperature value using the following formula: ,in, Indicates the reduction factor of the yield strength of steel bars. Indicates the temperature of the reinforcing steel.

[0099] It should be noted that the pre-established database of high-temperature mechanical properties of concrete in the evaluation and treatment device is established by those skilled in the art based on actual experience values, but is not limited to other treatment methods.

[0100] For example, for each target damage level zone, assume the corresponding highest temperature Based on a pre-established database of high-temperature mechanical properties of concrete, the following material damage indices were calculated: Regarding the coefficient of elastic modulus loss of concrete When 400℃ < At ≤1000℃, ;when At ≤400℃, =0; when At >1000℃, =0.9.

[0101] Residual coefficient of concrete compressive strength :when At ≤400℃, =1; when 400℃ < At ≤800℃, When 800℃ < At ≤1000℃, ;when At >1000℃, =0.3.

[0102] Reduction factor of steel bar yield strength : Using formula ,in Let the temperature of the reinforcing steel be the temperature of the concrete at that location, assuming the temperature of the reinforcing steel is equal to the temperature of the concrete at that location (with a difference of ≤50℃). Unit: °C.

[0103] After obtaining the material damage index corresponding to each target damage level zone, the component damage index for each target damage level zone can be obtained in the following manner.

[0104] Optionally, obtain the critical values ​​of preset weights and target loss parameters; calculate the component indices of the target damage level zone based on the following formula:

[0105]

[0106] in, to This indicates the preset weight, and H represents the burst depth. This indicates the critical value of the burst depth. Indicates the burst volume. This represents the critical value of the burst volume. Indicates the average width of the crack. This represents the critical value of the average crack width. Indicates the crack length. This represents the critical value for crack length. This indicates the proportion of the leakage area to the total cross-sectional area. Indicates the elastic loss coefficient of concrete. This represents the residual compressive strength coefficient of concrete.

[0107] For example, =50mm, t=0.05m³ / m², =3mm, =5m, weights w1~w7 are determined using the analytic hierarchy process (matrix order 7, consistency ratio CR < 0.1) based on the tunnel's importance level, with a default value of 1 / 7 for each. When the calculated result exceeds 1.0, it is set to 1.0. Simultaneously, the weights are adjusted based on the fire temperature distribution at the component's location (which can be quickly interpolated using a finite element proxy model): for every 100℃ increase in the highest temperature of the fire-exposed surface, the weight is increased by 10%, but the total remains normalized.

[0108] It should be noted that the evaluation and processing device has preset weight values ​​and critical values ​​for the target loss parameters, which are set by those skilled in the art based on practical experience, and are used to obtain component damage indicators.

[0109] Understandably, each target damage level zone has a number of associated cross-sections, and each cross-section includes, but is not limited to, the following components: arch crown, left arch waist, right arch waist, left side wall, right side wall, etc. The damage indicators for each component are obtained according to the aforementioned steps for obtaining and processing component damage indicators.

[0110] After obtaining the component damage index for each target damage level area, the degree of damage for each target damage level area is determined based on the following steps.

[0111] Step 105: Use a pre-trained DNN network model to evaluate the component damage index, obtain the damage level corresponding to the target damage level area, and the repair priority corresponding to the damage level.

[0112] It should be noted that the DNN network used in this embodiment of the invention is a pre-set network model. The network structure of the pre-trained DNN network model is as follows: 50 neurons in the input layer, 3 hidden layers (with 128, 64, and 32 neurons respectively, and ReLU activation function), and 1 neuron in the output layer (linear activation). The loss function is mean squared error, the optimizer is Adam, the learning rate is 0.001, the number of training epochs is 500, the number of early stopping epochs is 50, and the batch size is 32.

[0113] For training the DNN network model, optionally, 2000 tunnel lining fire damage models are built using the finite element software ANSYS. Different damage combinations are randomly generated (burst depth 0-100mm, crack width 0-5mm, leakage area 0-0.3, elastic modulus loss 0-0.9, etc.), and the overall safety factor under each working condition is calculated. Each model contains 10 cross-sections, each cross-section has 5 components, for a total of 50 input features. These 50 input features are then used to train the DNN network model.

[0114] After obtaining the trained DNN network model, validation is required. Optionally, 10-fold cross-validation can be used, requiring R² ≥ 0.95 on the test set and mean absolute percentage error ≤ 5%.

[0115] Once the verification process meets the preset rules, the trained DNN network model is used as the pre-trained DNN network model. Before using a pre-trained DNN network model to determine the degree of damage, it is necessary to determine the number of cross sections associated with the target damage level area and the number of components under the cross sections. By multiplying, the dimension of the component damage index vector of the target damage level area can be obtained.

[0116] Subsequently, a finite element proxy model is constructed using a pre-trained DNN network. The evaluation processing device inputs the obtained component exponential vector into the pre-trained DNN network to obtain the overall remaining bearing capacity coefficient of the tunnel in the target damage level area.

[0117] Based on the preset interval values ​​of the damage level zone by the damage level processing device, the degree of damage to which the overall remaining bearing capacity coefficient of the tunnel in the target damage level zone belongs is determined.

[0118] Based on the damage level of each target damage level area, corresponding repair priorities are generated in descending order of damage level.

[0119] Specifically, the component damage indicators for each damage level zone are input into the trained DNN model to quickly output the overall remaining bearing capacity coefficient R of the tunnel. Then, the comprehensive damage level is classified according to the R value: R ≥ 0.9 is intact, 0.7 ≤ R < 0.9 is slightly damaged, 0.5 ≤ R < 0.7 is moderately damaged, and R < 0.5 is severely damaged. The repair priority ranking is also output simultaneously.

[0120] The embodiments provided by this invention offer a complete technical process from data acquisition, multi-source fusion, multi-scale characterization to comprehensive evaluation. This enables rapid and accurate assessment of tunnel lining damage after a fire, obtaining precise evaluation results and providing a reliable basis for tunnel safety emergency response and repair decisions.

[0121] This avoids relying on manual on-site verification, saving manpower and improving processing efficiency and accuracy while avoiding the influence of subjectivity. By fusing data from multiple data sources and comprehensively utilizing the complementary advantages of each data source, data can be processed from a holistic perspective, improving the accuracy of the assessment. Based on the processing flow provided in this embodiment, the target lining can be processed automatically, improving processing efficiency.

[0122] Secondly, a damage feature map covering the surface and interior of the target lining was established, and deep learning and image segmentation technology were used to automatically identify the geometric parameters (area, depth, width) and physical properties (temperature field distribution, material wave velocity degradation) of various types of damage, so as to realize the quantitative and standardized description of multiple types of damage and provide a unified damage index basis for comprehensive evaluation.

[0123] Furthermore, a three-tiered collaborative fusion framework—"data-level, feature-level, and decision-level"—is established to overcome the problem of existing technologies where data sources operate independently and their complementary advantages are difficult to leverage. This is achieved through a multi-level data fusion architecture for tunnel fire damage assessment. Data-level fusion registers infrared temperature fields with visible light images to generate fused images with temperature labels; convolutional neural networks are used to determine the target damage level area and the corresponding damage degree; this improves the accuracy and robustness of damage characterization and assessment. This invention departs from existing methods that focus solely on a single scale or empirical formula, establishing a multi-scale correlation assessment system that ranges from microscopic material degradation to macroscopic component damage and ultimately to the overall structural performance. At the material scale, the loss coefficient of concrete's elastic modulus and the reduction in steel yield strength are obtained through infrared thermal field and acoustic wave inversion. At the component scale, the burst depth, crack density, and leakage area of ​​various parts of the lining are quantified based on 3D point cloud and image recognition results. At the structural scale, using the outputs from the material and component scales as input, a finite element proxy model (trained with a deep neural network) is used to rapidly calculate the remaining bearing capacity (damage level) and safety factor of the overall lining. This multi-scale model achieves decoupling and transmission of damage parameters across different scales, enabling rapid and accurate output of the comprehensive damage level and repair priority of tunnel lining after a fire, providing a scientific basis for emergency decision-making and reinforcement design.

[0124] In one alternative embodiment, after assessing the target lining damage level and generating the corresponding repair priority, the results can be displayed on a visualization interface.

[0125] Optionally, point cloud data is acquired from fire data; based on the external parameters corresponding to the target device in the acquired visible light image, the point cloud data is mapped one by one to two-dimensional pixel coordinates; for each two-dimensional pixel coordinate, the nearest target pixel is determined on the fused image, and the temperature label corresponding to the target pixel is assigned to the point cloud data corresponding to the two-dimensional pixel coordinate to obtain a three-dimensional real-scene model of the lining with temperature labels; the damage degree and repair priority corresponding to each target damage level area are mapped to the three-dimensional real-scene model and displayed in a visualization interface.

[0126] Specifically, each pixel in the fused image will be converted into three-dimensional point coordinates (X, Y, Z) in the world coordinate system through inverse perspective projection transformation, based on the camera's intrinsic parameter matrix and image coordinates (u, v), combined with the depth information (Z value) provided by the point cloud data acquired by the 3D laser scanner and registered to the tunnel design coordinate system, and the correspondence between image pixels and three-dimensional spatial points will be established.

[0127] Then, for each point in the 3D point cloud, using the calibrated camera extrinsic parameters (rotation matrix R and translation vector T), the point is projected from world coordinates to image pixel coordinates (u', v') to find its nearest neighbor pixel position in the fused image.

[0128] Finally, the nearest neighbor interpolation algorithm is used to directly assign the temperature value of the corresponding pixel to the 3D point cloud. If multiple pixels map to the same cloud point data, the average temperature is taken; if a cloud point has no directly corresponding pixel (e.g., the point cloud density is higher than the image resolution), the temperature value of the pixel closest to the projection point in the image is used as the assignment. This algorithm is fast and suitable for temperature assignment of large-scale point clouds.

[0129] In summary, the final output is a lining 3D reality model with temperature labels, where each point has 3D coordinates (X, Y, Z) and temperature attribute (T).

[0130] Furthermore, the damage level corresponding to each target damage level area is marked on the BIM model, and based on the color coding algorithm, the damage level of each target damage level area is displayed on the three-dimensional real scene model in the form of color depth.

[0131] Optionally, green can be used for intact, yellow for minor damage, orange for moderate damage, and red for severe damage.

[0132] In a preferred embodiment, the assessment processing device can generate an assessment report, displayed on a visual interface. The report includes: target damage level zones, material damage indices associated with each target damage level zone, component damage indices, and damage severity. The report is output in PDF format and can also be exported as a .shp or .dxf file for loading by a geographic information system.

[0133] In this embodiment, the damage level of each damage level area of ​​the target lining is displayed to the user through a visual interface by the constructed three-dimensional real-scene model. This provides an intuitive display of the damage level and a convenient window for determining the subsequent repair priority. At the same time, the generated damage report can also be displayed on the visual interface to provide relevant personnel with multiple ways to obtain damage level determination.

[0134] In another optional implementation, the temperature gradient of the target damage level area is obtained based on the temperature label under the target damage level area; the target damage parameters, maximum temperature and temperature gradient of the target damage level area are input into the pre-constructed DS evidence theory framework to obtain the probability value of each target damage level area; the target damage level area corresponding to the maximum probability value is determined as the enhanced repair type.

[0135] It should be noted that the pre-constructed DS evidence theory framework can define a complete set of all possible propositions (i.e., the overall damage state of the lining).

[0136] Specifically, the assessment and processing device first establishes an identification framework, which includes four categories: intact, slightly damaged, moderately damaged, and severely damaged. For each sensor (infrared thermal imaging, laser scanner, and visible light high-definition camera), its identification accuracy for different damage levels is obtained beforehand through historical testing or on-site calibration. For example, infrared thermal imaging has an accuracy rate of 85% for high-temperature damage, visible light high-definition cameras have an accuracy rate of 80% for crack damage, and laser scanners have an accuracy rate of 90% for crater depth.

[0137] The assessment processing device determines the temperature gradient of the target damage level area based on the temperature label of the target level area, and inputs this temperature gradient into the pre-constructed DS evidence theory framework to perform the following operations: The evaluation processing device assigns a basic probability value (i.e., reliability) to each proposition in the recognition framework based on its own detection results and historical accuracy. Then, it triggers the DS framework to fuse the probability values ​​of multiple sensors using the Dempster combination rule to obtain the comprehensive reliability of each proposition after fusion. Finally, it outputs the final probability of the damage level zone from the recognition framework according to the decision rule (selecting the proposition corresponding to the maximum probability).

[0138] For example, for the current target damage level area to be evaluated, each sensor device (infrared imager, visible light high-definition camera, and laser scanner) outputs the most probable damage level judgment based on its own detection data (e.g., infrared judgment of "moderate damage"). Then the basic probability assignment function m... i Calculated according to the following rules: The reliability is determined by assigning the sensor's historical accuracy to the level it identifies: m i (The level determined by the sensor) = historical accuracy; the remaining confidence (1 - historical accuracy) is evenly distributed among the other three levels in the recognition frame: m i(For each other level) = (1 - historical accuracy) / (number of propositions in the frame - 1) = (1 - historical accuracy) / 3; If the sensor cannot give a clear judgment, then all levels are assigned uniform reliability, that is, the reliability of each level is 1 / 4. The sum of the basic probability assignments of all sensors is 1.

[0139] The reliability of each sensor's output is quantified, and the sensor's uncertainty is expressed as a reliability distribution. Then, the reliability of multiple sensors is fused using Dempster's combination rule to obtain a comprehensive probability distribution. This allows for the comprehensive utilization of multi-source information during decision-making, reduces the impact of misjudgments from a single sensor, and improves the accuracy and robustness of the overall damage level assessment.

[0140] In summary, the processing device obtains the damage probability of each damage level area and determines the target damage level area corresponding to the maximum damage probability as the enhanced repair type.

[0141] This invention, from a decision-level perspective, utilizes the DS evidence theory to perform confidence-weighted fusion of damage level judgments from different sensor devices. This framework fully leverages the complementary advantages of various data types, significantly improving the accuracy and robustness of damage characterization.

[0142] To verify the effectiveness of the embodiments of the present invention, it was applied in the post-fire assessment of a highway tunnel. The tunnel was a two-lane highway tunnel with C30 concrete lining. The fire lasted for approximately 1.5 hours, with the fire source located 800m from the tunnel entrance, and the highest temperature reaching approximately 950℃. The method of the present invention was compared with the traditional manual inspection + finite element method.

[0143] Damage identification capability: This invention automatically identified 23 burst zones (burst depth range 8-45mm, average depth 22mm), 126 cracks (average width 1.2mm, total length 89m), and a leakage area of ​​approximately 32m²; traditional manual inspection only detected 18 burst zones and 89 cracks, with missed detection rates of 21.7% and 29.4%, respectively. This invention has a burst depth measurement error ≤ ±3mm (compared to actual measurements) and a crack width error ≤ ±0.1mm.

[0144] Efficiency Assessment: From data collection to outputting a comprehensive evaluation report, this invention takes approximately 3.5 hours (including 1.5 hours for on-site scanning and 2 hours for data processing and analysis); traditional methods require 3 professionals to conduct on-site testing for 2 days, plus 2 days for modeling and analysis, totaling 4 days (approximately 96 hours). This represents an efficiency improvement of approximately 27 times.

[0145] Assessment accuracy: The overall residual bearing capacity coefficient R predicted by this invention is 0.62 (moderate damage), while the actual residual bearing capacity coefficient verified by subsequent core sampling and load tests is 0.59, with a relative error of 5.1%; the traditional method predicts a value of 0.71 (slight damage), with a relative error of 20.3%. The assessment results of this invention are more consistent with the actual situation.

[0146] Economic benefits: The total cost of the traditional method is approximately 420,000 yuan (labor costs, equipment rental, testing, etc.); the cost of the method of this invention is approximately 140,000 yuan (mainly the one-time investment in equipment and a small amount of labor), saving about 67%. At the same time, due to the rapid assessment, the tunnel was reopened to traffic 5 days earlier, reducing indirect economic losses by approximately 6 million yuan.

[0147] Social impact: The assessment results provide a clear three-dimensional visualization of the damage distribution and a clear ranking of repair priorities, providing a clear basis for emergency reinforcement decisions, avoiding blind or insufficient reinforcement, and enhancing the safety confidence of the operation and management unit.

[0148] In summary, the specific embodiments of this invention clearly demonstrate the complete technical path from multi-source data acquisition, three-layer fusion, multi-scale characterization to comprehensive evaluation. Comparative data from the examples fully demonstrate its feasibility, efficiency, accuracy, and significant economic and social benefits in post-fire tunnel lining damage assessment.

[0149] Figure 2 This is a schematic diagram of a lining fire damage level assessment and treatment device provided in an embodiment of the present invention. Figure 2 As shown, the device 20 includes: an acquisition module 201, a division module 202, and a processing module 203.

[0150] The acquisition module 201 is used to acquire fire data of tunnel lining, perform image fusion processing on infrared thermal images and visible light images in the fire data, and generate a fused image with temperature labels. The segmentation module 202 is used to perform damage level region segmentation on the fused image using the U-NET network model, to obtain multiple target damage level regions and the highest temperature corresponding to each target damage level region. The processing module 203 is used to process each target damage level area as follows: based on the correlation table between the target damage level area and the damage parameter, the target loss parameter corresponding to the target damage level area is obtained; based on the highest temperature corresponding to the target damage level area, the material damage index corresponding to the target damage level area is determined, and based on the material damage index and the target loss parameter, the component damage index of the target damage level area is determined; the component damage index is input into a preset DNN network model to obtain the damage degree corresponding to the target damage level area, and the repair priority corresponding to the damage degree is determined.

[0151] The device provided in this application embodiment has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0152] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0153] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0154] Figure 3 A block diagram of an exemplary electronic device provided in an embodiment of the present invention. Figure 3 The electronic device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0155] like Figure 3 As shown, the electronic device is represented in the form of a general-purpose computing device. The components of the electronic device may include, but are not limited to: one or more processors 410, memory 430, and communication bus 440 connecting different system components (including memory 430 and processing unit 410).

[0156] Communication bus 440 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. For example, these architectures include, but are not limited to, Industry Standard Architecture (ISA) buses, Micro Channel Architecture (MAC) buses, Enhanced ISA buses, Video Electronics Standards Association (VESA) local buses, and Peripheral Component Interconnect (PCI) buses.

[0157] Electronic devices typically include a variety of computer-readable media. These media can be any available media that can be accessed by the electronic device, including volatile and non-volatile media, and removable and non-removable media.

[0158] Memory 430 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The electronic device may further include other removable / non-removable, volatile / non-volatile computer system storage media. Memory 430 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of this application.

[0159] A program / utility having a set (at least one) of program modules can be stored in memory 430. Such program modules include—but are not limited to—an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. The program modules typically perform the functions and / or methods described in the embodiments of this application.

[0160] Processor 410 executes various functional applications and data processing by running programs stored in memory 430, such as implementing embodiments of this application. Figure 1 The method provided in the illustrated embodiment.

[0161] This application provides a non-transitory computer-readable storage medium that stores computer instructions, which cause the computer to execute embodiments of this application. Figure 1 The method provided in the illustrated embodiment.

[0162] The aforementioned computer-readable storage medium may be any combination of one or more computer-readable media. A computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), or flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium that contains or stores a program that may be used by or in connection with an instruction execution system, apparatus, or device.

[0163] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including—but not limited to—electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of transmitting, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0164] The program code contained on a computer-readable medium may be transmitted using any suitable medium, including—but not limited to—wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0165] Computer program code for performing the operations of the embodiments of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0166] The foregoing has described specific embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0167] In the description of the embodiments of this application, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the embodiments of this application. In the embodiments of this application, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in the embodiments of this application, as well as the features of different embodiments or examples.

[0168] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of embodiments of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0169] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order according to the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0170] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0171] It should be noted that the terminals involved in the embodiments of this application may include, but are not limited to, personal computers (PCs), personal digital assistants (PDAs), wireless handheld devices, tablet computers, mobile phones, MP3 players, MP4 players, etc.

[0172] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0173] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0174] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0175] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present application should be included within the scope of protection of the present application.

Claims

1. A method for assessing and processing the fire damage level of linings based on multi-source data fusion, characterized in that, include: Fire data of the target lining is acquired, and the infrared thermal image and visible light image in the fire data are fused to generate a fused image with temperature labels. Using the U-NET network model, the fused image is divided into damage level regions to obtain multiple target damage level regions and the highest temperature corresponding to each target damage level region. The processing for each target damage level area is as follows: Based on the correlation table between the target damage level area and the damage parameter, the target loss parameter corresponding to the target damage level area is obtained; Based on the highest temperature corresponding to the target damage level zone, determine the material damage index corresponding to the target damage level zone, and based on the material damage index and the target loss parameter, determine the component damage index of the target damage level zone. The component damage index is evaluated using a pre-trained DNN network model to obtain the damage level corresponding to the target damage level area and the repair priority corresponding to the damage level.

2. The method for assessing and treating the fire damage level of linings according to claim 1, characterized in that, The infrared thermal image and visible light image in the fire data are subjected to image fusion processing to generate a fused image with temperature labels, including: Based on the scale-invariant feature transformation processing method, feature extraction processing is performed on the infrared thermal image and the visible light image to obtain common stable features; Calculate the homography transformation matrix of the common stable features, and based on the homography transformation matrix, map the infrared temperature field in the infrared thermogram to the coordinates of the visible light image to obtain the fused image with temperature labels.

3. The method for assessing and treating the fire damage level of linings according to claim 1, characterized in that, Using the U-NET network model, the fused image is processed to divide into damage level regions, obtaining multiple target damage level regions, including: Based on the convolution operation of the U-NET network model, multiple shallow feature maps related to the semantics of lining damage are obtained; Using the encoder in the U-NET network model, each shallow feature map is downsampled to obtain a deep feature map related to the semantics of lining damage. Using the decoder corresponding to the encoder, each of the depth feature maps is upsampled and fused with the shallow feature maps associated with the deep feature maps to generate a high-dimensional feature map semantically related to lining damage. Each of the high-dimensional feature maps is then processed as follows: Calculate the target feature value of the target pixel in the high-dimensional feature map, match the target feature value with the damage threshold range, and determine the target damage level corresponding to the target pixel; The multiple target pixels corresponding to the target damage level are taken as the target damage level area, and the maximum value of the temperature label under the target damage level area is taken as the highest temperature of the target damage level area.

4. The method for assessing and treating the fire damage level of linings according to claim 1, characterized in that, The target damage level area includes: a bursting zone, a cracked zone, and a leakage zone; then, the target loss parameters corresponding to the target damage level area are obtained, including: For the bursting zone, the bursting depth and bursting volume are calculated based on the point cloud data corresponding to the bursting zone and the surface of the first target lining. For the crack area, the target high-dimensional feature map corresponding to the crack area is subjected to grayscale and denoising processing, and the average width and length of the crack in the high-dimensional feature map after grayscale and denoising processing are calculated based on edge algorithm and skeleton extraction algorithm. For the leakage area, obtain the leakage area corresponding to the leakage area and the second target lining surface area, and calculate the ratio of the leakage surface set to the second target lining surface area to obtain the leakage area ratio of the lining surface. The first lining surface and the second lining surface are different.

5. The method for assessing and treating the fire damage level of linings according to claim 1, characterized in that, The material damage indicators include: concrete elastic modulus loss coefficient, concrete compressive strength residual coefficient, and steel reinforcement yield strength reduction coefficient; then, determining the material damage indicators corresponding to the target damage level zone based on the highest temperature includes: The highest temperature in the target damage level area is matched with a preset temperature range to determine the target temperature range to which the highest temperature belongs. For the material damage index values ​​corresponding to the target range, determine the concrete elastic modulus loss coefficient and the concrete compressive strength residual coefficient; Based on the temperature value of the reinforcing steel, the reduction factor for the yield strength of the reinforcing steel is obtained using the following formula: in, Indicates the reduction factor of the yield strength of steel bars. Indicates the temperature of the reinforcing steel.

6. The method for assessing and treating the fire damage level of linings according to claim 1, characterized in that, The step of determining the component damage index of the target damage level zone based on the material damage index and the target loss parameter includes: Obtain the critical value of the preset weight and the target loss parameter; The component indexes for the target damage level zone are calculated using the following formula: in, Indicates the component damage index. to This indicates the preset weight, and H represents the burst depth. This indicates the critical value of the burst depth. Indicates the burst volume. This represents the critical value of the burst volume. Indicates the average width of the crack. This represents the critical value of the average crack width. Indicates the crack length. This represents the critical value for crack length. This indicates the proportion of the leakage area to the total cross-sectional area. Indicates the elastic loss coefficient of concrete. This represents the residual compressive strength coefficient of concrete.

7. The method for assessing and treating the fire damage level of linings according to any one of claims 1-6, characterized in that, Also includes: Obtain point cloud data from the fire data; Based on the external parameters corresponding to the target device that acquired the visible light image, the point cloud data is mapped one by one into two-dimensional pixel coordinates; For each two-dimensional pixel coordinate, the nearest target pixel is determined on the fused image, and the temperature label corresponding to the target pixel is assigned to the point cloud data corresponding to the two-dimensional pixel coordinate to obtain a lining three-dimensional real scene model with temperature label. The damage level and repair priority corresponding to each target damage level area are mapped onto the three-dimensional real scene model and displayed in a visual interface.

8. The method for assessing and treating the fire damage level of linings according to any one of claims 1-6, characterized in that, Also includes: Based on the temperature label under the target damage level area, obtain the temperature gradient of the target damage level area; The target damage parameters, the highest temperature, and the temperature gradient of the target damage level area are input into the pre-constructed DS evidence theory framework to obtain the probability value of each target damage level area. The target damage level area corresponding to the highest probability value is identified as the enhanced repair type.

9. A device for assessing and processing the fire damage level of linings based on multi-source data fusion, characterized in that, include: The acquisition module is used to acquire fire data of the target lining, and perform image fusion processing on the infrared thermal image and visible light image in the fire data to generate a fused image with temperature labels. The segmentation module is used to perform damage level region segmentation processing on the fused image using the U-NET network model, to obtain multiple target damage level regions and the highest temperature corresponding to each target damage level region. The processing module is configured to process each target damage level region as follows: based on the correlation table between the target damage level region and the damage parameter, obtain the target loss parameter corresponding to the target damage level region; determine the material damage index corresponding to the target damage level region based on the highest temperature corresponding to the target damage level region, and determine the component damage index of the target damage level region based on the material damage index and the target loss parameter; evaluate the component damage index using a pre-trained DNN network model to obtain the damage degree corresponding to the target damage level region and the repair priority corresponding to the damage degree.

10. An electronic device, characterized in that, include: At least one processor; as well as At least one memory communicatively connected to the processor, wherein: The memory stores program instructions that can be executed by the processor, which can invoke the program instructions to perform the lining fire damage level assessment method as described in any one of claims 1 to 8.

Images