Deep learning-based tunnel lining crack intelligent diagnosis and life prediction method
By combining drones and deep learning, and utilizing high-definition cameras and ultrasonic detection technology, tunnel lining cracks can be identified and predicted. This solves the problems of low efficiency and low accuracy in traditional detection methods, achieving efficient and accurate crack diagnosis and life prediction, reducing costs and optimizing maintenance strategies.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-07
- Publication Date
- 2026-03-10
AI Technical Summary
Traditional methods for detecting cracks in tunnel linings are inefficient and inaccurate, making it difficult to achieve real-time monitoring. Furthermore, manual inspection is insufficient to detect minute internal cracks, and existing technologies struggle to provide accurate crack diagnosis and lifespan prediction.
A drone equipped with a high-definition camera was used to collect images of the tunnel lining surface. Combined with ultrasonic detection equipment, internal data was obtained. Crack features were identified through deep learning algorithms, and a coupled model of concrete carbonation and corrosion was constructed for crack diagnosis and life prediction.
It enables intelligent diagnosis and accurate life prediction of tunnel lining cracks, improves detection efficiency and accuracy, reduces labor costs, optimizes maintenance plans, and lowers detection and maintenance expenses.
Smart Images
Figure CN121640324A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of tunnel safety, in particular to a tunnel lining crack intelligent diagnosis and life prediction method based on deep learning. BACKGROUND
[0002] As an important part of transportation infrastructure, the safety and durability of tunnels are crucial for ensuring smooth traffic and the safety of people's lives and property. Tunnel lining, as the main load-bearing part of the tunnel structure, is easily affected by geological conditions, traffic loads, environmental factors and other factors, and is prone to cracks and other diseases. These diseases not only reduce the carrying capacity and durability of the tunnel lining, but also may cause serious accidents such as tunnel collapse, so timely and accurate detection and diagnosis of tunnel lining cracks and prediction of their remaining life are of great significance for ensuring the safe operation of tunnels.
[0003] Traditional tunnel lining crack detection methods mainly rely on manual inspection, which is not only inefficient and subjective, but also difficult to detect internal and fine cracks. With the development of computer technology and sensor technology, image recognition and non-destructive testing-based tunnel lining crack detection methods have gradually been applied. However, these methods still have some limitations, such as poor adaptability to complex environments, low detection accuracy, and difficulty in achieving real-time monitoring.
[0004] Deep learning, as a powerful machine learning technology, has the ability to automatically extract features, process complex data and powerful nonlinear modeling, and has achieved remarkable results in image recognition, target detection and data prediction. Applying deep learning technology to tunnel lining crack detection and life prediction can effectively overcome the shortcomings of traditional methods, improve the accuracy and efficiency of detection, and achieve intelligent diagnosis and life prediction of tunnel lining cracks.
[0005] At the same time, the rapid development of unmanned aerial vehicle technology provides a new means for tunnel detection. Unmanned aerial vehicles have the advantages of strong maneuverability, flexible operation and rapid access to detection areas, and can carry high-definition cameras and other equipment to obtain image data of the tunnel lining surface, providing a rich source of data for deep learning-based crack detection. In addition, the concrete carbonation-corrosion coupling model can consider the influence of concrete carbonation and steel corrosion on the performance of tunnel lining structure, providing a more accurate theoretical basis for tunnel lining life prediction. SUMMARY
[0006] The purpose of the present application is to provide a tunnel lining crack intelligent diagnosis and life prediction method based on deep learning, which considers the influence of multiple factors on the performance of tunnel lining structure, and realizes accurate prediction of tunnel lining crack intelligent diagnosis and remaining life.
[0007] To achieve the above object, the present application provides the following scheme: A deep learning-based tunnel lining crack intelligent diagnosis and life prediction method, comprising the following steps: S1, based on the visible light camera carried by the unmanned aerial vehicle, collecting high-definition images of the tunnel lining surface along the preset flight trajectory; at the same time, using an ultrasonic detection device to obtain ultrasonic data of the internal structure of the tunnel lining; S2, inputting the high-definition images collected in S1 into a pre-trained crack identification model, the crack identification model being improved and trained based on CNN, identifying and locating cracks, and extracting surface feature parameters of the cracks; fusing the surface feature parameters with the ultrasonic data to predict the depth information of the cracks; S3, constructing a concrete carbonation-corrosion coupling model, taking the crack feature parameters obtained in S2 as inputs of the concrete carbonation-corrosion coupling model, combining tunnel environmental parameters and lining material parameters, and predicting the remaining life of the tunnel lining.
[0008] Preferably, in S1, the collected high-definition images of the tunnel lining surface are preprocessed: Firstly, a Gaussian filtering algorithm is used to remove noise in the high-definition image, the Gaussian filtering algorithm performs weighted average on each pixel point and its neighborhood pixels in the high-definition image, the core principle is to use a Gaussian function as a weight template to perform convolution operation on the image, and the formula is:
[0009] wherein, is the coordinate of the current pixel point, is the coordinate of the center pixel point of the template, is the standard deviation of the Gaussian distribution; then the number of pixels of each gray level in the high-definition image is counted through histogram equalization, the cumulative distribution function is calculated, and then the original high-definition image gray value is mapped to a new gray value according to the transformation function, to obtain a high-definition image with enhanced contrast; finally, image correction is performed, a perspective transformation-based method is adopted, feature points in the image are detected, a perspective transformation matrix between the original high-definition image and the standard high-definition image is established, and the deformed image is corrected to an orthographic image.
[0010] Preferably, in S2, the crack identification model is a model improved based on CNN, and the CNN improvement includes using a dilated convolution to expand the receptive field and introducing an attention mechanism to focus on the crack area.
[0011] Preferably, in S2, the pre-training of the crack identification model includes: First, build the dataset, collect a large number of different tunnel environments, different crack types and severity of lining images, including normal lining images and images containing cracks; manually label these images, mark the location, shape and category of the cracks, and divide the dataset into training set, validation set and test set; Set the training parameters, measure the difference between the model prediction results and the true labels based on the cross-entropy loss function; use the Adam algorithm to update the parameters of the model to minimize the loss function; set the learning rate to 0.001, the number of iterations to 100, and the batch size to 32, and adjust these parameters by the performance on the validation set.
[0012] Preferably, in S2, the surface feature parameters are fused with the ultrasonic data to predict the depth information of the cracks, specifically including: The original surface feature parameters and ultrasonic data are fused by using the data layer fusion method; the crack depth information obtained by ultrasonic detection is embedded in the corresponding image pixels in a specific encoding manner; when fusing, ensure the consistency of the image and the ultrasonic data in spatial position, and associate the ultrasonic detection points with the corresponding positions in the image by establishing a coordinate mapping relationship.
[0013] Preferably, S2 further includes: An index system for evaluating the severity of cracks is established, and a weighted summation method is used to calculate the crack severity index, the formula is:
[0014] wherein, , , L, W and D represent the length, width and depth of the crack respectively, N represents the number of cracks, D represents the crack distribution density, , , , , and w represents the weight of each factor.
[0015] Preferably, in S3, the concrete carbonation-corrosion coupling model includes: The carbonation depth prediction model has the following formula: k
[0016] wherein, is the carbonation depth, k is the carbonation coefficient, is the total time; the carbonation coefficient k is not a constant, and the calculation formula is as follows:
[0017] wherein, is a reference carbonation coefficient, is an environmental influence factor function, is a crack influence correction coefficient; a steel bar corrosion model, the formula is as follows:
[0018] wherein, is a corrosion rate, is a corrosion rate coefficient, is a corrosion start time, is a time index; based on the carbonation depth and the corrosion rate, the residual life of the tunnel lining is predicted.
[0019] Preferably, in S3, the crack characteristic parameters input into the concrete carbonation-corrosion coupling model include surface characteristic parameters of the crack and crack depth, and the surface characteristic parameters of the crack include crack length and crack width.
[0020] The application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the deep learning-based tunnel lining crack intelligent diagnosis and life prediction method according to any one of the above.
[0021] According to the specific embodiments of the application, the following technical effects are disclosed: The method of the application innovatively combines unmanned aerial vehicle image recognition and a concrete carbonation-corrosion coupling model, uses an unmanned aerial vehicle to efficiently collect tunnel lining surface images, processes and analyzes the images through a deep learning algorithm, realizes intelligent diagnosis of cracks, at the same time, evaluates the internal structure of the tunnel lining in combination with ultrasonic detection data, and uses the concrete carbonation-corrosion coupling model to consider the influence of various factors on the structural performance of the tunnel lining, realizes accurate prediction of the residual life of the tunnel lining. The scheme has broad application prospects and is expected to provide strong technical support for the safe operation and maintenance management of tunnels and improve the safety and sustainability of tunnel projects. BRIEF DESCRIPTION OF DRAWINGS
[0022] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below are only some embodiments of the application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.
[0023] Figure 1 The flowchart of the deep learning-based tunnel lining crack intelligent diagnosis and life prediction method provided by the application is shown. DETAILED DESCRIPTION
[0024] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.
[0025] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.
[0026] As shown in Figure 1 The present application provides a tunnel lining crack intelligent diagnosis and life prediction method based on deep learning, which comprises the following steps: S1, based on the visible light camera carried by the unmanned aerial vehicle, collecting high-definition images of the tunnel lining surface along the preset flight trajectory; at the same time, using ultrasonic detection equipment to obtain ultrasonic data of the internal structure of the tunnel lining; S2, inputting the high-definition images collected in S1 into a pre-trained crack identification model, the crack identification model is improved and trained based on CNN, identifying and locating cracks, and extracting surface feature parameters of the cracks; the surface feature parameters are fused with the ultrasonic data to predict the depth information of the cracks; S3, constructing a concrete carbonization-corrosion coupling model, taking the crack feature parameters obtained in S2 as the input of the concrete carbonization-corrosion coupling model, combining the tunnel environment parameters and the lining material parameters, and predicting the remaining life of the tunnel lining.
[0027] Specifically, the method of the present application comprises: A multi-rotor unmanned aerial vehicle with high mobility, stability and good endurance is selected as an image acquisition platform, for example, DJI Wenli M300 RTK unmanned aerial vehicle, which has strong flight stability and flexible control performance, and can meet the flight requirements in complex tunnel environment. According to the characteristics of narrow internal space, dim light and many obstacles in the tunnel, the flight route of the unmanned aerial vehicle is carefully planned. The flight trajectory is in the shape of "Z" or spiral, which ensures that the unmanned aerial vehicle can fully cover the tunnel lining surface while maintaining a relatively stable distance and angle with the lining surface. The flight height is generally controlled to be 1-2 meters away from the lining surface, so as to obtain clear and complete image data.
[0028] In terms of image acquisition equipment, high-resolution visible light cameras such as Sony A7R series cameras are installed, which have a resolution of over 42 million pixels, allowing clear capture of the fine cracks and disease characteristics on the surface of the tunnel lining. Camera parameters are set to ensure that the images have sufficient clarity and contrast, such as adjusting the sensitivity (ISO) between 200-800, the shutter speed between 1 / 100-1 / 1000 seconds according to the light conditions inside the tunnel, and the aperture to f / 5.6-f / 8 to obtain the best imaging effect. In order to obtain information from different angles of the tunnel lining, a multi-view image acquisition strategy is implemented. During flight, the gimbal of the UAV is controlled to rotate at certain angles in the horizontal and vertical directions, such as every 30° in the horizontal direction and every 15° in the vertical direction to collect an image, thereby constructing a comprehensive image dataset of the tunnel lining surface.
[0029] The raw images collected often contain noise, low contrast, image distortion, and other problems, so preprocessing is needed to improve image quality and provide a better data foundation for subsequent crack identification. First, a Gaussian filter algorithm is used to remove noise from the image. Gaussian filtering performs a weighted average of each pixel point and its neighborhood in the image, effectively smoothing the image while preserving the edges and details of the image. The core principle is to use a Gaussian function as a weight template to perform convolution operations on the image, with the formula:
[0030] where, is the coordinate of the current pixel point, is the coordinate of the center pixel point of the template, is the standard deviation of the Gaussian distribution, which controls the strength of the filter, the larger the value, the more obvious the filtering effect, and the more blurred the image, but the stronger the noise suppression.
[0031] Next, the histogram equalization method is used to enhance the contrast of the image. Histogram equalization re-distributes the gray values of the image by transforming the gray histogram of the image, expanding the gray range of the image to the entire gray interval, thereby enhancing the contrast of the image and highlighting the detailed features such as cracks. The basic steps are to count the number of pixels at each gray level in the image, calculate the cumulative distribution function, and then map the original image's gray values to new gray values according to the transformation function to obtain the contrast-enhanced image.
[0032] In addition, due to the attitude change of the UAV during flight, the collected images may have geometric distortions such as rotation and tilt, so image correction is needed. A perspective transformation-based method is used to detect feature points in the image, such as the edges of the tunnel lining and the corners of the wall, to establish a perspective transformation matrix between the original image and the standard image, correct the distorted image to an orthographic image, and ensure the shape and size of the objects in the image are accurate and reliable, facilitating subsequent crack analysis and measurement.
[0033] A convolutional neural network (CNN) is selected as the basic model architecture. CNN can automatically learn features in images through the combination of convolutional layers, pooling layers, and fully connected layers, and can identify and classify targets in images. In tunnel lining crack identification, the traditional CNN model is improved to better adapt to the needs of crack detection, such as using dilated convolution to expand the receptive field, allowing the model to obtain more image information and better detect long cracks; introducing an attention mechanism to make the model pay more attention to the crack area in the image and improve the accuracy of crack identification.
[0034] During model training, a high-quality dataset is first constructed. A large number of lining images with different tunnel environments, different crack types and severity are collected, including normal lining images and images containing cracks. These images are manually annotated to mark the location, shape, and category of the cracks. The dataset is divided into training, validation, and test sets, generally in the ratio of 70%, 15%, and 15%.
[0035] Training parameters are set, such as selecting an appropriate loss function (such as cross-entropy loss function) to measure the difference between the model's predicted results and the true labels; using optimization algorithms such as stochastic gradient descent (SGD), Adagrad, and Adam to update the model's parameters to minimize the loss function. Set appropriate learning rate (such as 0.001), iteration number (such as 100 times), and batch size (such as 32) and other hyperparameters, adjust these parameters based on performance on the validation set to prevent overfitting and underfitting, and improve the model's generalization ability. During training, use data augmentation techniques such as rotating, flipping, scaling, and adding noise to the image to increase data diversity and further improve the model's robustness and generalization ability.
[0036] From the recognition results of the deep learning model, the key features of the cracks are extracted. For the crack length, the continuous distribution length of the crack pixel points in the image is calculated, and the mapping relationship between the pixel and the actual size is used to convert the pixel length into the actual physical length. For the crack width, the width of the crack pixel is counted in the direction perpendicular to the crack strike, and the actual width is obtained by conversion according to the image resolution. The crack depth is estimated by combining the ultrasonic detection data, by analyzing the characteristics of the ultrasonic echo signal such as the degree of signal attenuation, reflection time, etc., and using empirical formula or machine learning model to establish the relationship between crack depth and ultrasonic signal characteristics, so as to calculate the crack depth.
[0037] In order to quantify the severity of the cracks, an index system for evaluating the severity of the cracks is established. Considering the length, width, depth of the cracks, as well as the number and distribution density of the cracks, the weighted sum method is used to calculate the crack severity index (CSI), the formula is:
[0038] wherein, , , respectively represent the length, width and depth of the crack, represents the number of cracks, represents the crack distribution density, , , , , is the weight of each factor, which is determined according to the actual engineering experience and expert opinion, and the CSI value can be used to intuitively evaluate the severity of the tunnel lining cracks, providing an important basis for subsequent life prediction and maintenance decision-making.
[0039] Ultrasonic detection uses the propagation characteristics of ultrasonic waves in concrete medium to detect crack depth and internal defects. When ultrasonic waves encounter cracks or defects, reflection, refraction and diffraction phenomena occur, causing changes in ultrasonic signal propagation time, amplitude and frequency parameters. By analyzing these changes, information such as crack depth, location and size can be inferred.
[0040] In tunnel detection, appropriate ultrasonic detection equipment is selected, such as non-metal ultrasonic detector, which has high-precision ultrasonic emission and reception functions and can accurately measure the propagation time and amplitude of ultrasonic signals. The ultrasonic probe is arranged on the surface of the tunnel lining, and methods such as opposite measurement, oblique measurement and flat measurement are usually used. Opposite measurement is to arrange the transmitting probe and the receiving probe on the two sides of the lining respectively, and to detect internal defects by measuring the direct wave of ultrasonic signals; oblique measurement is to arrange the transmitting probe and the receiving probe on the same side of the lining with a certain angle, and to detect crack depth by analyzing the refraction and reflection of ultrasonic signals; flat measurement is to arrange the transmitting probe and the receiving probe on the same side of the lining and on the same horizontal line, which is used to detect the depth of surface cracks.
[0041] In the data acquisition process, appropriate sampling frequency and sampling point number are set to ensure that the changes of ultrasonic signals can be accurately captured. The sampling frequency is generally set between 50kHz and 200kHz, and the sampling point number is adjusted according to the thickness of the tunnel lining and the detection accuracy requirements, to ensure that the collected data can fully and accurately reflect the internal condition of the tunnel lining. At the same time, in order to improve the reliability of the data, multiple repeated measurements are performed on each detection point, and the average value is taken as the final measurement result.
[0042] Fusing the image recognition results with the ultrasonic detection data can more comprehensively reflect the condition of the tunnel lining. The fusion method mainly includes data layer fusion and feature layer fusion. Data layer fusion is to directly fuse the original image data and ultrasonic data. For example, the crack depth information obtained by ultrasonic detection is embedded in the corresponding image pixels in a specific encoding manner, so that the image not only contains visual information of surface cracks, but also contains ultrasonic data information of crack depth. When performing fusion, the consistency of the image and the ultrasonic data in the spatial position needs to be ensured, and the ultrasonic detection points and the corresponding positions in the image are accurately associated through the establishment of accurate coordinate mapping relationship.
[0043] Feature layer fusion is to extract features from image and ultrasonic data respectively, and then fuse these features. Geometric features, texture features and other features of cracks are extracted from image data, and propagation time, amplitude variation and other features of ultrasonic signals are extracted from ultrasonic data. These features are spliced or weighted fused to form a feature vector containing multi-modal information. For example, principal component analysis (PCA) method is used to reduce the dimension of the fused feature vector, remove redundant information and highlight key features, to improve the efficiency and accuracy of subsequent model processing.
[0044] The fused data can provide more comprehensive information, for example, by image recognition to determine the location and surface morphology of the cracks, combined with the crack depth information obtained by ultrasonic detection, a three-dimensional model of the cracks can be established, which can more intuitively show the overall situation of the cracks, and provide a more comprehensive basis for the analysis and evaluation of the cracks.
[0045] Based on the fused data, an internal crack prediction model is constructed. Regression algorithms in machine learning, such as support vector regression (SVR), random forest regression (RFR), etc., can be used to establish the mapping relationship between ultrasonic data features, image features and crack depth. By training a large number of sample data with known crack depth, the model learns the internal law between data features and crack depth, so that it can predict the crack depth according to new fused data.
[0046] In addition, the generative adversarial network (GAN) of deep learning can also be used for internal crack prediction. The generator network generates the predicted crack image or crack depth information according to the input fused data, and the discriminator network discriminates the generated results from the real data. Through the adversarial training between the generator and the discriminator, the performance of the generator is continuously optimized, so that it can generate more accurate crack prediction results. In the training process, indicators such as mean square error (MSE) and structural similarity index (SSIM) are used to measure the difference between the prediction results and the true values, and the parameters of the model are adjusted through the back propagation algorithm to improve the prediction accuracy of the model.
[0047] There is a mutual promotion coupling mechanism between concrete carbonation and steel corrosion. On the one hand, carbonation reduces the value of concrete, destroys the passivation film on the surface of steel, and creates conditions for steel corrosion; on the other hand, the expansion stress caused by steel corrosion leads to concrete cracking, making aggressive media more easily enter the interior of concrete, accelerating the carbonation of concrete. This interaction forms a vicious cycle, continuously exacerbating the deterioration of the tunnel lining structure.
[0048] The concrete carbonation-corrosion coupling model usually uses mathematical expressions to describe this complex process. For example, in some models, the relationship between carbonation depth and time is expressed as: where is the carbonation coefficient, which is affected by the material properties of concrete (such as cement variety, water-cement ratio, admixture, etc.), environmental factors (such as concentration, humidity, temperature, etc.). For the process of steel corrosion, the corrosion rate to represent the corrosion degree of steel bars, the relationship between the corrosion rate and the carbonation depth, time and other factors can be described by establishing an electrochemical model, such as considering the influence of oxygen diffusion, ion migration and other factors on the corrosion reaction rate, and then establishing a differential equation of the corrosion rate with time, and combining the boundary conditions and initial conditions to solve, in order to realize the quantitative simulation of the corrosion process of steel bars.
[0049] Specifically, in S3, the constructed concrete carbonation-corrosion coupling model specifically includes: The carbonation depth prediction model is as follows: k
[0050] wherein, is the carbonation depth, k is the carbonation coefficient, is the total time; the carbonation coefficient k is not a constant, and the calculation formula is as follows:
[0051] wherein, is the reference carbonation coefficient, which is determined by the concrete mix proportion (water-cement ratio, cement variety, admixture, etc.), is an environmental influence factor function, which is a function based on temperature (T), relative humidity (RH) and CO2 concentration (C_CO2), is a crack influence correction coefficient, the existence of cracks can significantly increase the carbonation coefficient, and the coefficient can be associated with a crack severity index (CSI), for example wherein, is a crack influence weight coefficient.
[0052] The steel bar corrosion model is as follows:
[0053] wherein, is the corrosion rate, is the corrosion rate coefficient, which is related to the concrete resistivity, chloride ion content, oxygen supply (related to cracks), etc., is the corrosion start time, is the time index, usually taking 0.5-1.0; and the remaining life of the tunnel lining is predicted based on the carbonation depth and the corrosion rate.
[0054] In the concrete carbonation-corrosion coupling model, accurate determination of model parameters is crucial to improve the accuracy of life prediction. In terms of environmental parameters, temperature, humidity and Concentration is a key factor affecting concrete carbonation and steel corrosion. Temperature primarily affects the chemical reaction rate; generally, increased temperature accelerates both concrete carbonation and steel corrosion. Temperature sensors installed at different locations within the tunnel can be used to monitor ambient temperature in real time and obtain data on temperature changes over time. Humidity also has a significant impact on concrete carbonation and steel corrosion; suitable humidity conditions are beneficial. The dissolution and diffusion of the material, as well as the electrochemical processes of steel corrosion, were investigated. A humidity sensor was used to monitor the relative humidity inside the tunnel, and the humidity parameters in the model were determined based on the monitoring data. Concentration is a direct reactant in concrete carbonation; the higher the concentration, the faster the carbonation rate. Gas sensors can be used to monitor the carbonation process within the tunnel. Concentration is monitored to provide accurate data for the model. Concentration data.
[0055] Regarding material parameters, the concrete mix proportion is a crucial factor affecting concrete performance. Different cement types, water-cement ratios, admixtures (such as fly ash and slag powder), and additives (such as water-reducing agents and retarders) lead to variations in concrete density, pore structure, and chemical composition, thus influencing carbonation and resistance to steel corrosion. By reviewing tunnel construction data, concrete mix proportion information was obtained, and laboratory tests were conducted to determine performance indicators such as compressive strength, impermeability, and porosity to determine the appropriate values for concrete material parameters in the model. Steel reinforcement characteristics, such as type, diameter, yield strength, and corrosion potential, significantly impact the steel corrosion process. Relevant steel reinforcement parameters were obtained based on design drawings and material inspection reports.
[0056] To more accurately determine model parameters, field tests and long-term monitoring can be employed. Representative areas on the tunnel lining surface are selected, and concrete core samples are drilled. These samples are then used in a laboratory to conduct tests such as carbonation depth testing and steel reinforcement corrosion detection to obtain actual carbonation and corrosion data for verifying and calibrating the model parameters. Simultaneously, long-term monitoring points are established within the tunnel, and various sensors are installed to continuously monitor changes in environmental and structural performance parameters. Analysis of the long-term monitoring data allows for continuous optimization of model parameters, improving the model's predictive accuracy.
[0057] When incorporating crack prediction results into the coupled concrete carbonation-corrosion model, the accelerating effect of cracks on concrete carbonation and steel corrosion is considered. The presence of cracks increases the contact area between concrete and the external environment, making... Erosive media can more easily enter the interior of the concrete, thereby accelerating the carbonation rate. At the same time, the steel bars at the cracks are more prone to corrosion, and the expansion stress generated by the corrosion further aggravates the expansion of the cracks. In the model, the influence of the cracks is reflected by modifying parameters such as the carbonation coefficient and the corrosion reaction rate. For example, according to the characteristics of the crack width, length and distribution density, a relationship model between the crack and the carbonation coefficient and the corrosion rate is established, and when the crack is detected, the parameters in the coupled model are adjusted accordingly to more accurately simulate the deterioration process of the tunnel lining in the presence of cracks.
[0058] When predicting the remaining life of the tunnel lining according to the calculation results of the concrete carbonation-corrosion coupling model, a structure performance failure criterion is usually set. For example, when the steel bar corrosion rate reaches a certain degree (such as 5%-10%), or the concrete carbonation depth exceeds the thickness of the steel bar protection layer, causing the steel bar to begin to seriously corrode, and the carrying capacity of the structure decreases to a certain proportion (such as 70%-80%) of the designed carrying capacity, the tunnel lining is considered to reach the limit of service life. Through model calculation, the concrete carbonation depth, steel bar corrosion rate and structure carrying capacity at different time points are obtained, and the remaining life of the tunnel lining is determined according to the failure criterion.
[0059] However, there is a certain uncertainty in the prediction results. This is caused by various factors such as the uncertainty of model parameters, the complexity of environmental factors and the variability of the structure itself. In order to evaluate the uncertainty of the prediction results, a probability analysis method such as Monte Carlo simulation can be used. By randomly sampling the parameters in the model (such as environmental parameters, material parameters, etc.), a large number of parameter combinations are generated, and the coupled model is run multiple times using these parameter combinations to obtain a series of life prediction results. Statistical analysis of these results gives the probability distribution of the life prediction value, thereby evaluating the reliability of the prediction results and giving the confidence interval of the remaining life of the tunnel lining, providing a more scientific basis for the maintenance decision of the tunnel.
[0060] The application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the deep learning-based tunnel lining crack intelligent diagnosis and life prediction method according to any one of the above.
[0061] The method of the present application is accurate in crack identification. Through learning a large number of tunnel lining image data, the deep learning algorithm can automatically extract the characteristics of the cracks. Compared with manual visual identification and traditional image analysis methods, the deep learning-based crack identification model can more accurately detect cracks, especially fine cracks. For example, in some complex tunnel environments, manual detection may miss fine cracks with a width of less than 0.2 mm, while the deep learning-based crack identification model can achieve an accuracy of more than 95% in identifying cracks with a width of more than 0.1 mm. At the same time, the multi-source data fusion technology combines unmanned aerial vehicle image data with ultrasonic detection data to make the detection results more comprehensive and accurate, not only identifying surface cracks but also effectively predicting and evaluating internal cracks, further improving the accuracy and reliability of the detection.
[0062] In terms of direct costs, the use of unmanned aerial vehicle image recognition and intelligent diagnosis technology reduces a large amount of manual detection costs. Traditional manual detection requires a large amount of manpower, including detection personnel, safety personnel, and the like, and requires the provision of corresponding detection tools and transportation equipment. Unmanned aerial vehicle detection only requires a small number of operators for flight control and data processing, and the equipment cost is relatively low and reusable. According to estimates, for a medium-sized tunnel detection project, the use of the present solution can reduce detection costs by 30% - 50%.
[0063] In terms of indirect costs, accurate crack detection and life prediction can optimize tunnel maintenance planning. Early detection of potential problems in tunnel lining avoids large-scale repair or reconstruction due to structural damage and reduces unnecessary maintenance costs. For example, if early-stage crack defects can be detected and repaired in advance, the further expansion of cracks can be avoided, which can cause serious damage to the lining structure, thereby saving maintenance costs that can be as high as tens of millions or even hundreds of millions of yuan in the later stage. In addition, through life prediction, the maintenance time and measures of the tunnel are reasonably arranged, improving the utilization efficiency of maintenance resources and reducing the overall maintenance cost. At the same time, early detection of potential problems and avoidance of accidents also bring significant indirect economic benefits.
[0064] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be implemented by means of software and the necessary general hardware platform, or by hardware. Based on such understanding, the above technical solutions, essentially or in terms of the contribution to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, and the like, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
[0065] The principles and implementation manners of the present application are described by using specific examples in the present application, and the above examples are only used to help understand the method of the present application and its core idea; meanwhile, for the general technical personnel in the art, the specific implementation manners and application ranges will be changed according to the idea of the present application. In conclusion, the content of the present specification should not be understood as the limitation of the present application.
Claims
1. A deep learning-based intelligent diagnosis and life prediction method for tunnel lining cracks, characterized in that, The method comprises the following steps: S1, based on the visible light camera carried by the unmanned aerial vehicle, collecting high-definition images of the tunnel lining surface along the preset flight trajectory; at the same time, using an ultrasonic detection device to obtain ultrasonic data of the internal structure of the tunnel lining; S2, inputting the high-definition images collected in S1 into a pre-trained crack identification model, the crack identification model being improved and trained based on CNN, identifying and locating cracks, and extracting surface feature parameters of the cracks; fusing the surface feature parameters with the ultrasonic data to predict the depth information of the cracks; S3, constructing a concrete carbonization-corrosion coupling model, taking the crack feature parameters obtained in S2 as inputs of the concrete carbonization-corrosion coupling model, combining tunnel environmental parameters and lining material parameters, and predicting the remaining life of the tunnel lining.
2. The deep learning-based tunnel lining crack intelligent diagnosis and life prediction method according to claim 1, characterized in that, In S1, the collected high-definition images of the tunnel lining surface are preprocessed: First, the Gaussian filter algorithm is used to remove noise in the high-definition image. The Gaussian filter algorithm performs weighted average on each pixel point and its neighborhood pixels in the high-definition image. The core principle is to use a Gaussian function as a weight template to perform convolution operation on the image, and the formula is: wherein, is the coordinate of the current pixel point, is the coordinate of the center pixel point of the template, is the standard deviation of the Gaussian distribution; then the number of pixels of each gray level in the high-definition image is counted by histogram equalization, the cumulative distribution function is calculated, and then the original high-definition image gray value is mapped to a new gray value according to the transformation function, to obtain the high-definition image after contrast enhancement; finally, image correction is performed, a method based on perspective transformation is adopted, feature points in the image are detected, a perspective transformation matrix between the original high-definition image and the standard high-definition image is established, and the deformed image is corrected to an orthographic image. 3.The deep learning-based tunnel lining crack intelligent diagnosis and life prediction method according to claim 1, characterized in that, In S2, the crack identification model is a model improved based on CNN. The CNN improvement includes using a dilated convolution to expand the receptive field and introducing an attention mechanism to focus on the crack area. 4.The deep learning-based tunnel lining crack intelligent diagnosis and life prediction method according to claim 1, characterized in that, In S2, the pre-training of the crack identification model comprises: First, a data set is constructed, a large number of lining images of different tunnel environments, different crack types and severities are collected, including normal lining images and images containing cracks; the images are manually labeled to mark the position, shape and category of the cracks, and the data set is divided into a training set, a validation set and a test set; The training parameters are set, the cross-entropy loss function is used to measure the difference between the model prediction results and the true labels; the Adam algorithm is used to update the parameters of the model to minimize the loss function; the learning rate is set to 0.001, the iteration number is set to 100 times, and the batch size is set to 32. The performance on the validation set is used to adjust these parameters.
5. The deep learning-based intelligent diagnosis and life prediction method for tunnel lining cracks according to claim 1, characterized in that, In S2, the surface feature parameters and the ultrasonic data are fused to predict the depth information of the cracks, specifically including: The original surface feature parameters and the ultrasonic data are fused by using a data layer fusion method; the crack depth information obtained by ultrasonic detection is embedded into the corresponding image pixels in a specific encoding manner; when fusing, the consistency of the image and the ultrasonic data in the spatial position is ensured, and the ultrasonic detection points are associated with the corresponding positions in the image by establishing a coordinate mapping relationship.
6. The deep learning-based tunnel lining crack intelligent diagnosis and life prediction method according to claim 1, characterized in that, In S2, it also includes: An index system for evaluating the severity of cracks is established, and a weighted summation method is used to calculate the crack severity index, and the formula is: wherein, , , respectively represent the length, width and depth of the crack, represents the number of cracks, represents the crack distribution density, , , , , represents the weight of each factor.
7. The deep learning-based intelligent diagnosis and life prediction method for tunnel lining cracks according to claim 1, characterized in that, In S3, the constructed concrete carbonization-corrosion coupling model specifically includes: A carbonization depth prediction model, the formula is as follows: k wherein, is the carbonization depth, k is the carbonization coefficient, is the total time; the carbonization coefficient k is a constant, and the calculation formula is as follows: wherein, is a reference carbonation coefficient, is an environmental impact factor function, is a crack impact correction coefficient; A steel bar corrosion model, the formula is as follows: wherein, is the corrosion rate, is the corrosion rate coefficient, is the corrosion start time, is the time exponent; the remaining life of the tunnel lining is predicted based on the carbonation depth and the corrosion rate. 8.The deep learning-based tunnel lining crack intelligent diagnosis and life prediction method according to claim 1, characterized in that, In S3, the crack feature parameters input into the concrete carbonization-corrosion coupling model include the surface feature parameters of the cracks and the crack depth, and the surface feature parameters of the cracks include the crack length and the crack width. 9.A non-transitory computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by a processor to implement the deep learning-based tunnel lining crack intelligent diagnosis and life prediction method according to claims 1-7.