Early detection method suitable for dam microcrack judgment

Abstract

The invention discloses an early detection method suitable for dam micro-crack judgment, which comprises the following steps: S1, acquiring reflected light wavelength data of a sensor, and converting wavelength variation into strain data of a dam to form a first data set; s2, obtaining a speckle interference image of the dam, obtaining infinitesimal displacement information of the surface of the dam, and forming a second data set; s3, establishing a strain-displacement detection model, and obtaining a target detection data set; s4, performing imaging processing on the target detection data set, inputting the target detection data set into a pre-constructed micro-crack detection model, outputting a dam micro-crack detection result, and obtaining a micro-crack data set; and S5, performing feature extraction on the micro-crack data set to obtain crack feature data, inputting the crack feature data into a pre-constructed micro-crack time model, and judging a micro-crack development stage. According to the method, on the premise that external influence factors are discharged, the development stage of the dam micro-crack is accurately detected, and the detection accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of dam crack detection, and particularly relates to an early detection method suitable for judging micro-cracks in dams. Background Art

[0002] Micro-cracks are a common early manifestation of dam diseases. In its initial stage, the cracks are fine and develop slowly, making it difficult to be detected in time. However, if effective detection and treatment are not carried out in the early stage of micro-cracks, over time, the micro-cracks gradually expand and connect, thereby affecting key indicators such as the structural strength and anti-seepage performance of the dam. In severe cases, it may even lead to catastrophic accidents such as dam collapse.

[0003] Traditional dam micro-crack detection technologies mainly rely on manual inspections. Technicians regularly use simple tools to visually observe the dam surface with the naked eye, and rely on rich experience to identify relatively obvious cracks and measure and record basic parameters such as the length and width of the cracks. The advantage of this method is high flexibility, which can adjust the observation focus according to the actual situation at any time, and does not require complex equipment, with low costs. However, the disadvantages are also very obvious: on the one hand, the efficiency of manual inspections is low. Facing the huge surface area of large dams, a comprehensive inspection requires a large amount of time and manpower; on the other hand, the accuracy of manual detection is greatly affected by subjective factors. There are differences in the experience and judgment criteria of different technicians, and it is easy to miss small cracks. In addition, manual inspections can only detect surface cracks of the dam and are powerless for internal micro-cracks, making it difficult to meet the comprehensive and accurate requirements of dam safety monitoring.

[0004] With the development of technology, existing dam micro-crack detection technologies have overcome some of the drawbacks of traditional methods to a certain extent. Among them, ultrasonic detection technology is widely used. It uses the differences in the propagation speed and reflection characteristics of ultrasonic waves in different media to emit ultrasonic waves into the dam, and judges whether there are micro-cracks and the location and approximate size of the cracks by analyzing the reflected waves. In addition, infrared thermal imaging technology has also been gradually popularized. This technology captures the temperature distribution differences on the dam surface and identifies abnormal heat transfer areas caused by the existence of micro-cracks, thereby locating the micro-cracks. The advantages of these existing technologies are that they can achieve non-contact detection, can quickly scan large areas, improve the detection efficiency, and can detect internal micro-cracks in dams to a certain extent. However, they also have limitations: the positioning accuracy of ultrasonic detection for micro-cracks is limited. For dams with complex structures, the signals are easily interfered, resulting in misjudgments; infrared thermal imaging technology is greatly affected by factors such as environmental temperature and humidity. In an environment with drastic temperature changes or high humidity, the accuracy of the detection results will be greatly reduced. At the same time, most existing technologies can only provide static information about micro-cracks and are difficult to monitor the development trend of micro-cracks in real time.

[0005] Therefore, whether it is traditional technology or existing technology, it is difficult to fully meet the requirements of early and accurate detection of micro-cracks in dams. Summary of the Invention

[0006] Based on the above, the present application proposes an early detection method for micro-crack judgment applicable to dams, including the following steps:

[0007] S1. Lay a fiber optic sensor network, obtain the reflected light wavelength data of the sensors in real time, and convert the wavelength change amount into the strain data of the dam through a wavelength-strain conversion algorithm to form a first data set;

[0008] S2. Periodically scan the surface of the dam through laser speckle interferometry, obtain the speckle field on the surface of the dam, obtain the speckle interference image of the dam in real time, calculate the phase change of the interference fringes, obtain the micro-displacement information on the surface of the dam, and form a second data set;

[0009] S3. Perform spatio-temporal correlation analysis on the first data set and the second data set, establish a strain-displacement detection model, and fuse the first data set and the second data set according to the degree of differential correlation to form a target detection data set;

[0010] S4. Image-process the target detection data set, input it into a pre-constructed micro-crack detection model, output the dam micro-crack detection result, and obtain the micro-crack data set through the micro-crack detection result;

[0011] S5. Extract the features of the micro-crack data set, obtain the crack feature data, input it into a pre-constructed micro-crack time model, and judge the development stage of the micro-crack according to the output result of the micro-crack time model.

[0012] Preferably, after obtaining the reflected light wavelength data of the fiber optic sensor in S1, converting the wavelength change amount into the strain data of the dam through a wavelength-strain conversion algorithm is specifically as follows:

[0013] Obtain the initial reflected light center wavelength , the reflected light center wavelength collected in real time , calculate the wavelength change amount , the formula is: , through the wavelength change amount , use the wavelength-strain conversion algorithm to convert the wavelength change amount into the strain data of the dam, and the formula is: , where is the correction coefficient, is the dynamic strain influence factor, is the change rate of the wavelength change amount with time, calculate the strain values at the corresponding positions of each sensor, sort them according to the spatial distribution order of the sensors, and form a first data set.

[0014] Preferably, in S2, the laser speckle interference technology uses a high-resolution laser speckle interferometer to periodically scan the surface of the dam. Through laser irradiation, a speckle field is formed on the surface of the dam. When there is a small displacement on the surface of the dam, the speckle field will generate changes in interference fringes. Record the speckle interference images at different times, perform grayscale processing on the speckle interference images, calculate the phase change of the interference fringes through the phase-displacement conversion algorithm, obtain the small displacement information on the surface of the dam, and form the second data.

[0015] Preferably, calculating the phase change of the interference fringes through the phase-displacement conversion algorithm and obtaining the small displacement information on the surface of the dam specifically includes:

[0016] Obtain the speckle interference images at adjacent times after grayscale processing and calculate the phase difference , and the formula is: , where and are the pixel coordinates of the speckle interference images at adjacent times, is the spatio-temporal modulation coefficient. Use the phase unwrapping algorithm to process the phase difference to obtain a continuous phase distribution , and obtain the small displacement through the phase-displacement conversion algorithm , and the formula is: , where is the laser wavelength, is the order of the interference fringes. Calculate the small displacement amount of the surface of the dam corresponding to each pixel point, and integrate the displacement amounts of all pixel points to form the second data set.

[0017] Preferably, in S3, perform spatio-temporal correlation analysis on the first data set and the second data set. By establishing a strain-displacement detection model, fuse the first data set and the second data set according to the degree of differential correlation to form a target detection data set; perform spatio-temporal correlation analysis on the strain data in the first data set and the small surface displacement data in the second data set. According to the structural mechanics principle of the dam, establish a strain-displacement detection model, adopt a data fusion algorithm, and fuse the first data set and the second data set according to the degree of differential correlation to construct a target data set containing comprehensive information on strain and displacement.

[0018] Preferably, when performing spatio-temporal correlation analysis on the first data set and the second data set in S3, establishing a strain-displacement detection model to achieve data fusion specifically includes:

[0019] Obtain the strain data in the first data set and the small displacement data in the second data set at time , the formula is: , where and are the corresponding fusion coefficients respectively, is the fusion weight, and the formula is: , where and are the total numbers of strain detection points and displacement detection regions respectively, is the spatio-temporal correlation coefficient at time , where is the mean value of the strain data in the first data set, is the mean value of the micro-displacement data in the second data set, and are the corresponding standard deviations respectively. After sorting out the fusion values at all positions and times, a target detection data set is formed.

[0020] Preferably, the pre-constructed micro-crack detection model in S4 is a multi-layer convolutional neural network model, and the multi-layer convolutional neural network model includes multiple convolutional layers, pooling layers and fully connected layers; the target data set is processed into an image and converted into a data image to be input into the multi-layer convolutional neural network model. The convolutional layer of the multi-layer convolutional neural network model slides and convolves on the data image through a convolutional kernel to extract the local features of the micro-cracks in the target data set and form a micro-crack feature map; the pooling layer downsamples the convolved micro-crack feature map to reduce the data volume and retain the crack features; the fully connected layer classifies and judges the crack features after multiple convolutions and poolings, identifies the features of the micro-cracks in the target data, and outputs the judgment result of the dam micro-cracks.

[0021] Preferably, the micro-crack data set is obtained through the micro-crack detection result in S4, specifically:

[0022] S4.1. Mark the area of the micro-crack detection result. The dam is divided into multiple detection unit areas by levels. For the detection unit area, the confidence score of the micro-crack is output;

[0023] S4.2. Cluster the adjacent areas with high confidence scores into one class, representing a micro-crack cluster. Compare the output results of the multi-layer convolutional neural network model of the micro-crack cluster in different detection cycles. If there is a similar change trend of micro-crack features in multiple consecutive detection cycles, it is determined as a micro-crack area;

[0024] S4.3. Obtain the geographical coordinate information, micro-crack features and feature change trend information of the determined micro-crack area to form a micro-crack data set.

[0025] Preferably, in S5, the features of the microcrack dataset are extracted and input into a pre-constructed microcrack time model to determine the microcrack period; the feature extraction forms a composite feature set by extracting the dynamic features of the strain change rate and displacement change trend of the microcracks from the microcrack dataset, performs local feature scale transformation on the composite feature set, and inputs it into the pre-constructed microcrack time model to output the microcrack development stage.

[0026] Preferably, the pre-constructed microcrack time model is a recurrent neural network model, and the recurrent neural network model is based on the hidden state and the stage factor to output the probability distribution of the microcracks at different development stages , and the formula is: , where is the weight matrix from the hidden state to the output layer, is the weight matrix from the stage factor to the output layer, is the bias vector of the output layer; the stage factor has the formula: , where is the stage weight coefficient, is the change amount of adjacent time step features; , where is the weight matrix from the hidden state to the hidden state, is the weight matrix from the input to the hidden state, is the bias vector of the hidden layer; by obtaining the probability distribution at different development stages, the microcrack stage of the dam is determined.

[0027] Compared with the prior art, the technical solution of the present application has the following technical effects:

[0028] By combining the fiber optic sensor network and the laser speckle interferometry technology, the present invention improves the accuracy of dam microcrack detection. In the data acquisition process, the fiber optic sensor can accurately capture the subtle changes in the reflected light wavelength and use a unique wavelength-strain conversion algorithm to accurately convert the wavelength change amount into the strain data of the dam. Its calculation process fully considers factors such as the correction coefficient and the dynamic strain influence factor to ensure the accuracy of the strain data; the laser speckle interferometry technology can obtain the minute displacement information on the dam surface and accurately calculate the minute displacement from the interference fringe changes through an accurate phase-displacement conversion algorithm. After these data are subjected to spatio-temporal correlation analysis and fusion, they are input into a multi-layer convolutional neural network model for detection, effectively extracting the local features of the microcracks and performing classification and judgment, making the detection result of the microcracks more accurate, being able to accurately identify the existence of minute cracks, reducing misjudgment and missed judgment situations, and providing a reliable basis for the safety assessment of the dam.

[0029] In the feature extraction stage of the microcrack dataset of the present invention, starting from dynamic features such as the strain change rate and displacement change trend, combined with local feature scale transformation, the key information of microcrack development is obtained. The pre-constructed recurrent neural network model can fully learn this information, calculate the probability distribution of microcracks at different development stages through hidden states and stage factors, and can judge whether the microcracks are in the early development stage based on the probability distribution, and continuously monitor their development trend. It can detect microcracks in the initial stage in time, and update the model input by continuously collecting data, so as to grasp the development dynamics of microcracks in real time, provide sufficient time for dam maintenance, and take targeted measures to prevent microcracks from developing into serious diseases.

[0030] The above description is only an overview of the technical solution of the present application. In order to be able to understand the technical means of the present application more clearly, so as to be implemented in accordance with the content of the description, and in order to make the above and other purposes, features and advantages of the present application more obvious and understandable, the following takes the preferred embodiments of the present application and combines the drawings to describe in detail as follows.

[0031] According to the following detailed description of the specific embodiments of the present application in conjunction with the drawings, those skilled in the art will be more clear about the above and other purposes, advantages and features of the present application. Description of the Drawings

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to actual scale.

[0033] Figure 1 It is a flowchart of an early detection method applicable to the judgment of microcracks in a dam.

[0034] Figure 2 It is a flowchart of obtaining a microcrack dataset through microcrack detection results.

[0035] Figure 3 It is a structural diagram of a dam microcrack model.

[0036] Figure 4 It is a regional division diagram of monitoring point 3 of the dam microcrack model structure. Detailed Description of the Specific Embodiment

[0037] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this application. Apparently, the described embodiments are some, but not all, of the embodiments of this application. In the following description, specific details such as specific configurations and components are provided only to assist in a comprehensive understanding of the embodiments of this application. Therefore, those skilled in the art should understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. Additionally, descriptions of known functions and configurations are omitted for clarity and conciseness in the embodiments.

[0038] It should be understood that the phrase "one embodiment" or "the embodiment" mentioned throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, the phrase "one embodiment" or "the embodiment" that appears throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures, or characteristics can be combined in one or more embodiments in any suitable manner.

[0039] In addition, this application may repeat reference numerals and / or letters in different instances. This repetition is for the purpose of simplicity and clarity and does not in itself indicate the relationship between the various embodiments and / or arrangements discussed.

[0040] The term "and / or" in this document is merely a description of the associated relationship of the associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, B exists alone, and both A and B exist simultaneously. The term " / and" in this document describes another associated object relationship, indicating that two relationships can exist. For example, A / and B can represent: A exists alone, and both A and B exist. Additionally, the character " / " in this document generally indicates that the associated objects before and after are in an "or" relationship.

[0041] The term "at least one" in this document is merely a description of the associated relationship of the associated objects, indicating that three relationships can exist. For example, at least one of A and B can represent: A exists alone, both A and B exist simultaneously, and B exists alone.

[0042] It should also be noted that in this document, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprises", "comprising", or any other variation thereof is intended to cover a non-exclusive inclusion.

[0043] Embodiment 1

[0044] This embodiment mainly describes an early detection method suitable for judging microcracks in dams, such as Figure 1 shown, including the following steps:

[0045] S1. Lay a fiber optic sensor network, obtain the reflected light wavelength data of the sensors in real time, and convert the wavelength change amount into the strain data of the dam through the wavelength-strain conversion algorithm to form a first data set;

[0046] S2. Periodically scan the dam surface through laser speckle interferometry technology, obtain the speckle field on the dam surface, obtain the speckle interference image of the dam in real time, calculate the phase change of the interference fringes, obtain the micro displacement information on the dam surface, and form a second data set;

[0047] S3. Perform spatio-temporal correlation analysis on the first data set and the second data set, establish a strain-displacement detection model, and fuse the first data set and the second data set according to the degree of differential correlation to form a target detection data set;

[0048] S4. Image-process the target detection data set, input it into a pre-constructed microcrack detection model, output the dam microcrack detection result, and obtain the microcrack data set through the microcrack detection result;

[0049] S5. Extract the features of the microcrack data set, obtain the crack feature data, input it into a pre-constructed microcrack time model, and judge the development stage of the microcracks according to the output result of the microcrack time model.

[0050] Furthermore, after obtaining the reflected light wavelength data of the fiber optic sensor in S1, convert the wavelength change amount into the strain data of the dam through the wavelength-strain conversion algorithm, specifically:

[0051] Obtain the initial reflected light center wavelength , the reflected light center wavelength collected in real time , calculate the wavelength change amount , the formula is: , through the wavelength change amount , use the wavelength-strain conversion algorithm to convert the wavelength change amount into the strain data of the dam, the formula is: , where is the correction coefficient, is the dynamic strain influence factor, is the change rate of the wavelength change amount with time, calculate the strain values at the corresponding positions of each sensor, sort them according to the spatial distribution order of the sensors, and form a first data set.

[0052] Further, in S2, the laser speckle interference technology uses a high-resolution laser speckle interferometer to periodically scan the surface of the dam. Through laser irradiation, a speckle field is formed on the surface of the dam. When there is a small displacement on the surface of the dam, the speckle field will generate changes in interference fringes. Record the speckle interference images at different times, perform grayscale processing on the speckle interference images, calculate the phase change of the interference fringes through the phase-displacement conversion algorithm, obtain the small displacement information on the surface of the dam, and form the second data.

[0053] Further, calculate the phase change of the interference fringes through the phase-displacement conversion algorithm to obtain the small displacement information on the surface of the dam. Specifically:

[0054] Obtain the adjacent-time speckle interference images after grayscale processing and calculate the phase difference , the formula is: , where and are the pixel coordinates of the adjacent-time speckle interference images, is the spatio-temporal modulation coefficient. Use the phase unwrapping algorithm to process the phase difference to obtain a continuous phase distribution , and obtain the small displacement through the phase-displacement conversion algorithm , the formula is: , where is the laser wavelength, is the order of the interference fringes. Calculate the small displacement amount of the surface of the dam corresponding to each pixel point, and integrate the displacement amounts of all pixel points to form the second data set.

[0055] Further, in S1, perform spatio-temporal correlation analysis on the first data set and the second data set. By establishing a strain-displacement detection model, fuse the first data set and the second data set according to the degree of differential correlation to form a target detection data set; perform spatio-temporal correlation analysis on the strain data in the first data set and the small surface displacement data in the second data set. According to the structural mechanics principle of the dam, establish a strain-displacement detection model, adopt a data fusion algorithm, and fuse the first data set and the second data set according to the degree of differential correlation to construct a target data set containing comprehensive information on strain and displacement.

[0056] Further, in S3, when performing spatio-temporal correlation analysis on the first data set and the second data set, establish a strain-displacement detection model to achieve data fusion. Specifically:

[0057] Obtain the strain data in the first data set at time and the small displacement data in the second data set , calculate the fusion value of the target detection data set. The formula is: , where and are the corresponding fusion coefficients respectively, is the fusion weight, and the formula is: , where and are the total numbers of strain detection points and displacement detection regions respectively, is the spatio-temporal correlation coefficient at time , where is the mean value of strain data in the first data set, is the mean value of micro-displacement data in the second data set, and are the corresponding standard deviations respectively. After sorting out the fusion values at all positions and times, a target detection data set is formed.

[0058] Furthermore, the pre-constructed micro-crack detection model in S4 is a multi-layer convolutional neural network model, which includes multiple convolutional layers, pooling layers and fully connected layers; the target data set is processed into an image and converted into a data image and input into the multi-layer convolutional neural network model. The convolutional layer of the multi-layer convolutional neural network model slides and convolves on the data image through a convolutional kernel to extract the local features of micro-cracks in the target data set and form a micro-crack feature map; the pooling layer downsamples the convolved micro-crack feature map to reduce the data volume and retain the crack features; the fully connected layer classifies and judges the crack features after multiple convolutions and poolings to identify the features of micro-cracks in the target data and output the judgment result of the dam micro-cracks.

[0059] Furthermore, as Figure 2 shown, the micro-crack data set is obtained through the micro-crack detection result in S4, specifically:

[0060] S4.1. Mark the regional division of the micro-crack detection result, divide the dam into multiple detection unit regions by levels, and for the detection unit regions, output the confidence scores of micro-cracks;

[0061] S4.2. Cluster adjacent regions with high confidence scores into one category, representing a micro-crack cluster. Compare the output results of the multi-layer convolutional neural network model of the micro-crack cluster in different detection cycles. If similar micro-crack feature change trends appear in multiple consecutive detection cycles, it is determined as a micro-crack region;

[0062] S4.3. Obtain the geographical coordinate information, micro-crack features and feature change trend information of the determined micro-crack region to form a micro-crack data set.

[0063] In S4.1, calculate the confidence score of micro-cracks in the detection unit region by obtaining the feature difference degree Calculation of the output feature vector of the fully connected layer of the multi-layer convolutional neural network model , the formula is: , where is the mean value of the strain-displacement fusion feature vectors of all detection unit regions, is the mean value of the output layer feature vectors, is the cosine similarity, and the confidence score of the microcracks in each detection unit region is calculated.

[0064] Furthermore, in S5, the features of the microcrack data set are extracted and input into the pre-constructed microcrack time model to determine the microcrack period; the feature extraction is to extract the dynamic features of the strain change rate and displacement change trend of the microcracks from the microcrack data set to form a composite feature set, perform local feature scale transformation on the composite feature set, and input it into the pre-constructed microcrack time model to output the development stage of the microcracks.

[0065] Furthermore, the pre-constructed microcrack time model is a recurrent neural network model, and the recurrent neural network model outputs the probability distribution of the microcracks at different development stages according to the hidden state and the stage factor The formula is: , where, is the weight matrix from the hidden state to the output layer, is the weight matrix from the stage factor to the output layer, is the bias vector of the output layer; the stage factor The formula is: , where is the stage weight coefficient, is the change amount of the features of adjacent time steps; , where, is the weight matrix from the hidden state to the hidden state, is the weight matrix from the input to the hidden state, is the bias vector of the hidden layer; by obtaining the probability distribution at different development stages, the microcrack stage of the dam is determined.

[0066] This embodiment details an early detection method applicable to the judgment of dam microcracks, realizing high-precision, real-time, and dynamic monitoring of dam microcracks. By laying a fiber optic sensor network and using laser speckle interferometry technology to obtain multi-source data, and applying spatio-temporal correlation analysis and advanced model algorithms for data fusion and analysis, accurately judge the existence and development stage of microcracks, make up for the deficiencies of the existing technology, and provide strong support for the safety maintenance of dams.

[0067] Based on Embodiment 1, this embodiment details the early detection experiment of the dam microcrack judgment of this application, specifically:​

[0068] In this experiment, by comparing the effects of the technical solution of this application and the prior art in detecting microcracks in dams, the advantages of the technology of this application are verified. A medium-sized concrete dam that has been in service for many years is selected as the experimental object, and there are potential microcrack hazards to varying degrees in this dam; as Figure 3 shown, monitoring points are set at positions 1, 2, 3, 4, 5, 6, and 7 of the dam, covering key parts of the dam foundation, the middle of the dam body, and the dam crest, to simulate the complex stress conditions in the actual operating environment; during the experiment, environmental parameters such as temperature, humidity, and water level around the dam are recorded to analyze the influence of environmental factors on microcrack detection.

[0069] An optical fiber sensor network is laid at the dam monitoring points to collect the reflected light wavelength data in real time. Taking monitoring point 3 as an example, the initial center wavelength of the reflected light is 1549.987 nm, and the center wavelength data of the reflected light collected at different times are shown in Table 1;

[0070] Table 1 Center wavelength data of the reflected light at monitoring point 3 at the acquisition time

[0071] Collection time (h) Central wavelength of reflected light #timg# (nm) 1 1550.023 2 1550.031 3 1550.045 4 1550.052 5 1550.060

[0072] The surface of the dam is periodically scanned using a high-resolution laser speckle interferometer to obtain speckle interference images. The speckle interference images at adjacent times after grayscale processing are processed to calculate the phase difference and micro-displacement information; taking monitoring point 3 as an example, the relevant data at the pixel coordinates of the speckle interference images at adjacent times are shown in Table 2;

[0073] Table 2 Speckle interference image data at adjacent times of monitoring point 3

[0074] Collection time (h) #timg# #timg# #timg# #timg# #timg# 1 118.364 125.497 0.9654 0.1863 0.9542 2 122.589 128.632 0.9728 0.2257 1.1465 3 116.723 124.876 0.9685 0.2034 1.0378 4 120.456 127.589 0.9702 0.2368 1.2056 5 119.234 126.457 0.9667 0.2105 1.0789

[0075] The strain data collected by the optical fiber sensor and the displacement data obtained by the laser speckle interference technology are subjected to spatio-temporal correlation analysis to form a target detection data set; after the target detection data set is subjected to image processing, it is input into a multi-layer convolutional neural network model. The model is trained to extract features and classify and judge the input data; as Figure 4 shown, taking the areas of different monitoring points 3 as an example, divided into A1, A2, A3, A4, and A5, the microcrack confidence scores output by the model are shown in Table 3;

[0076] Table 3 Microcrack confidence of monitoring point 3

[0077] Monitoring area number Microcrack confidence score A1 0.923 A2 0.866 A3 0.529 A4 0.156 A5 0.228

[0078] According to the microcrack confidence score, the detection results are marked for regional division. Adjacent regions with a confidence score higher than 0.85 are grouped into one category, determined as the microcrack region, and a microcrack dataset is constructed. The strain change rate and displacement change trend dynamic features are extracted from the microcrack dataset and input into the pre-constructed recurrent neural network model to judge the development stage of the microcrack. Taking a certain microcrack region A1 as an example, during 5 consecutive days of monitoring, the dynamic feature data and predicted development stage of region A1 are shown in Table 4.

[0079] Table 4 Dynamic Feature Data and Predicted Development Stage of Region A1

[0080] Time (days) Strain change rate Displacement change trend Microcrack development stage 1 0.1256 0.0589 Early stage 2 0.1378 0.0692 Early stage 3 0.1834 0.1123 Transition from early stage to middle stage 4 0.2256 0.1567 Middle stage 5 0.2678 0.2012 Middle stage

[0081] Existing technology experimental process (taking ultrasonic detection technology as an example)

[0082] Use an ultrasonic detector to emit ultrasonic waves into the dam interior, and judge the presence and location of microcracks by analyzing the reflected waves. Detect in the same monitoring area and record the detection results. Due to the large influence of the complexity of the dam interior structure and interference factors on ultrasonic detection technology, misjudgments and missed judgments occur in some areas. At monitoring point 3, in monitoring area A2, there are actually microcracks, but they are not detected by ultrasonic detection; in monitoring area A4, ultrasonic detection misjudges the existence of microcracks, while actually it is only a local structural anomaly.

[0083] For the comparison of the microcrack detection accuracy of the entire dam, as shown in Table 5;

[0084] Table 5 Comparison of Microcrack Detection Accuracy between This Application and Ultrasonic Detection Technology

[0085] Comparison item Technology of this application Ultrasonic detection technology Number of microcrack areas correctly detected 35 20 Number of misjudged microcrack areas 4 8 Number of undetected microcrack areas 3 7 Accuracy rate (%) 81.40 62.50

[0086] Therefore, we can intuitively see that this application is superior to the existing technology in both misjudgment and missed judgment. At the same time, we continue the experiment and obtain the comparison of the accuracy of judging the development stage of microcracks, as shown in Table 6:

[0087] Table 6 Comparison of Microcrack Development Stage and Accuracy Rate between This Application and Ultrasonic Detection Technology

[0088] Comparison item Technology of this application Ultrasonic detection technology Number of correctly judged microcrack development stages 25 15 Number of misjudged microcrack development stages 3 5 Number of microcrack development stages that cannot be judged 2 4 Judgment accuracy rate (%) 80.65 68.18

[0089] Therefore, we can find that this application is still higher than the existing technology in judging the development stage of microcracks. At the same time, we continue to make a comparison under harsh environmental conditions, as shown in Table 7:

[0090] Table 7 Comparison of Accuracy Rates between This Application and Ultrasonic Detection Technology under Harsh Environmental Conditions

[0091] Environmental conditions Accuracy rate of technology of this application (%) Accuracy rate of ultrasonic detection technology (%) High temperature (above 38°C) 78.26 43.25 High humidity (relative humidity above 85%) 86.65 45.19 Large water level fluctuation (exceeding 6m) 86.52 55.36

[0092] From the content of the above table, it can be found that this application is also more accurate than the prior art in harsh environments. Therefore, this application is superior to the prior art in terms of the accuracy of microcrack detection, the accuracy of judging the development stage of microcracks, and the adaptability under different environmental conditions.

[0093] This embodiment details the comparison between the technical solution of this application and the existing ultrasonic detection technology in the detection of microcracks in dams, indicating that the technology of this application is superior to the prior art in terms of the accuracy of microcrack detection, the accuracy of judging the development stage of microcracks, and the adaptability under different environmental conditions. The technology of this application can more accurately detect microcracks and accurately judge their development stage, providing a reliable basis for the safety maintenance of dams, and has important practical application value and promotion significance.

[0094] The above are only the preferred embodiments of the present invention, and it does not thereby limit the protection scope of the present invention. For those skilled in the art, the present invention can have various changes and modifications; all changes, modifications, substitutions, integrations, and parameter changes made to these embodiments by means of conventional substitutions or capable of achieving the same functions without departing from the principle and spirit of the present invention fall within the protection scope of the present invention.

Claims

1. A method for early detection of microcracks in dams, characterized in that: include: S1. Lay an optical fiber sensor network to obtain the reflected light wavelength data of the sensor in real time, and convert the wavelength change into the strain data of the dam through a wavelength-strain conversion algorithm to form a first data set; S2. Periodically scan the surface of the dam using laser speckle interferometry technology to obtain the speckle field on the surface of the dam, obtain the speckle interferometry image of the dam in real time, calculate the phase change of the interference fringes, obtain the micro displacement information of the dam surface, and form a second data set; S3, performing spatiotemporal correlation analysis on the first data set and the second data set, establishing a strain-displacement detection model, and fusing the first data set and the second data set according to the degree of difference correlation to form a target detection data set; S4, image processing the target detection data set, input it into the pre-built micro-crack detection model, output the dam micro-crack detection results, and obtain the micro-crack data set through the micro-crack detection results; S5. Extract features from the microcrack data set to obtain crack feature data, input the data into a pre-built microcrack time model, and determine the microcrack development stage based on the output results of the microcrack time model.

2. The early detection method for microcracks in dams according to claim 1 is characterized in that: After the wavelength data of the reflected light from the optical fiber sensor is obtained in S1, the wavelength variation is converted into the strain data of the dam through the wavelength-strain conversion algorithm, specifically: Get the center wavelength of the initial reflected light , the center wavelength of the reflected light collected in real time , calculate the wavelength change , the formula is: , through the wavelength change , the wavelength-strain conversion algorithm is used to convert the wavelength change into the strain data of the dam. The formula is: ,in is the correction factor, is the dynamic strain influence factor, The wavelength variation rate over time is calculated, and the strain value at the corresponding position of each sensor is sorted according to the spatial distribution order of the sensors to form a first data set.

3. The early detection method for dam microcracks according to claim 2 is characterized in that: The laser speckle interferometry technology in S2 periodically scans the surface of the dam through a high-resolution laser speckle interferometer, and forms a speckle field on the surface of the dam through laser irradiation; when a small displacement occurs on the surface of the dam, the speckle field will produce interference fringe changes, and the speckle interference images at different times are recorded, and the speckle interference images are grayed. The phase change of the interference fringes is calculated through a phase-displacement conversion algorithm, and the small displacement information of the dam surface is obtained to form the second data.

4. The early detection method for microcracks in dams according to claim 3 is characterized in that: The phase change of the interference fringes is calculated by the phase-displacement conversion algorithm to obtain the micro-displacement information of the dam surface, specifically: Obtain grayscale speckle interferometry images at adjacent moments and calculate the phase difference , the formula is: ,in and is the pixel coordinate of the speckle interferometer image at adjacent moments, is the spatiotemporal modulation coefficient, and the phase difference is calculated using the phase unwrapping algorithm. Processing to obtain a continuous phase distribution , obtain small displacements through phase-displacement conversion algorithm , the formula is: ,in is the laser wavelength, The order of interference fringes is calculated, and the tiny displacement of the dam surface corresponding to each pixel is calculated. The displacements of all pixels are integrated to form the second data set.

5. The early detection method for dam microcracks according to claim 1 is characterized in that: In S3, the first data set and the second data set are subjected to spatiotemporal correlation analysis, and the first data set and the second data set are fused according to the degree of difference correlation by establishing a strain-displacement detection model to form a target detection data set; The strain data in the first data set and the surface micro-displacement data in the second data set are subjected to spatiotemporal correlation analysis. According to the structural mechanics principle of the dam, a strain-displacement detection model is established. The first data set and the second data set are fused according to the degree of difference correlation using a data fusion algorithm to construct a target data set containing comprehensive information on strain and displacement.

6. The early detection method for dam microcracks according to claim 5 is characterized in that: In S3, when performing spatiotemporal correlation analysis on the first data set and the second data set, a strain-displacement detection model is established to realize data fusion, specifically: Get The strain data in the first data set at time and the small displacement data in the second data set , calculate the fusion value of the target detection dataset , the formula is: ,in and are the corresponding fusion coefficients, is the fusion weight, the formula is: ,in and are the total number of strain detection points and displacement detection areas, respectively. for The time-space correlation coefficient is: ,in is the mean strain data of the first data set, is the mean value of the small displacement data of the second data set, and are the corresponding standard deviations respectively. After sorting the fusion values ​​of all positions and times, a target detection dataset is formed.

7. The early detection method for dam microcracks according to claim 1 is characterized in that: The pre-constructed microcrack detection model in S4 is a multi-layer convolutional neural network model, which includes multiple convolutional layers, pooling layers and fully connected layers; the target data set is imaged and converted into a data image and input into the multi-layer convolutional neural network model. The convolution layer of the multi-layer convolutional neural network model performs sliding convolution on the data image through the convolution kernel to extract the local features of the microcracks in the target data set and form a microcrack feature map; the pooling layer downsamples the microcrack feature map after convolution to reduce the data volume and retain the crack features; the fully connected layer classifies and judges the crack features after multiple convolution and pooling processes, identifies the features of the microcracks in the target data, and outputs the judgment results of the microcracks in the dam.

8. A method for early detection of microcracks in dams according to claim 1 or 7, characterized in that: In S4, a microcrack data set is obtained through microcrack detection results, specifically: S4.

1. Mark the microcrack detection results by regional division, divide the dam into multiple detection unit areas by level, and output the confidence score of the microcracks for the detection unit area; S4.

2. Cluster adjacent regions with high confidence scores into one category, representing a microcrack cluster, and compare the output results of the multi-layer convolutional neural network model of the microcrack cluster in different detection cycles. If a similar trend of microcrack feature changes is shown in multiple consecutive detection cycles, it is determined to be a microcrack region; S4.

3. Obtain the geographical coordinate information, microcrack characteristics and characteristic change trend information of the determined microcrack area to form a microcrack data set.

9. The early detection method for dam microcracks according to claim 1 is characterized in that: In the S5, the features of the microcrack data set are extracted and input into a pre-constructed microcrack time model to determine the microcrack period; the feature extraction is performed by extracting dynamic features related to the strain change rate and displacement change trend of the microcracks from the microcrack data set to form a composite feature set, and the composite feature set is transformed into a local feature scale and input into the pre-constructed microcrack time model to output the microcrack development stage.

10. The early detection method for dam microcracks according to claim 9 is characterized in that: The pre-built microcrack time model is a recurrent neural network model, which is based on the hidden state and stage factor Output probability distribution of microcracks at different development stages , the formula is: ,in, is the weight matrix from hidden state to output layer, is the weight matrix from the stage factor to the output layer, is the bias vector of the output layer; the phase factor The formula is: ,in is the stage weight coefficient, is the change of the features in adjacent time steps; ,in, is the weight matrix from hidden state to hidden state, is the weight matrix input to the hidden state, is the bias vector of the hidden layer; by obtaining the probability distribution of different development stages , determine the micro-crack stage of the dam.

Images